# memory handling

Memory handling is the allocation, reallocation, and freeing of memory for programs at the operating-system and programming levels.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## Uber Cuts Translation Cache Memory by up to 90% While Keeping P99 Latency Below 1 MS

DevFeed: [Uber Cuts Translation Cache Memory by up to 90% While Keeping P99 Latency Below 1 MS](<https://devfeed.tech/articles/uber-cuts-translation-cache-memory-by-up-to-90-while-keeping-p99-latency-below-1-ms-146238.md>)

Original publisher: [Read original article](<https://www.infoq.com/news/2026/11/uber-go-translation-cache/>)

Author: Leela Kumili

Published: 2026-11-20T14:38:00Z

Content type: news

Language: en

Sources: [InfoQ](<https://devfeed.tech/sources/infoq.md>)

Topics: [Cache](<https://devfeed.tech/topics/cache.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [cost-optimization](<https://devfeed.tech/topics/cost-optimization.md>)

Tags: [architecture](<https://devfeed.tech/tags/architecture.md>), [architecture-design](<https://devfeed.tech/tags/architecture-design.md>), [cache](<https://devfeed.tech/tags/cache.md>), [caching](<https://devfeed.tech/tags/caching.md>), [concurrency](<https://devfeed.tech/tags/concurrency.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [development](<https://devfeed.tech/tags/development.md>), [distributed-systems](<https://devfeed.tech/tags/distributed-systems.md>), [garbage-collection](<https://devfeed.tech/tags/garbage-collection.md>), [garbagecollection](<https://devfeed.tech/tags/garbagecollection.md>), [go](<https://devfeed.tech/tags/go.md>), [go-language](<https://devfeed.tech/tags/go-language.md>), [hit](<https://devfeed.tech/tags/hit.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [hybrid-architecture](<https://devfeed.tech/tags/hybrid-architecture.md>), [hybrid-work](<https://devfeed.tech/tags/hybrid-work.md>), [keys](<https://devfeed.tech/tags/keys.md>), [latency](<https://devfeed.tech/tags/latency.md>), [low-latency](<https://devfeed.tech/tags/low-latency.md>), [memory](<https://devfeed.tech/tags/memory.md>), [news](<https://devfeed.tech/tags/news.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [overhead](<https://devfeed.tech/tags/overhead.md>), [p99](<https://devfeed.tech/tags/p99.md>), [performance-tuning](<https://devfeed.tech/tags/performance-tuning.md>), [reduce](<https://devfeed.tech/tags/reduce.md>), [storage](<https://devfeed.tech/tags/storage.md>), [translation](<https://devfeed.tech/tags/translation.md>), [uber](<https://devfeed.tech/tags/uber.md>), [uber-go-translation-cache](<https://devfeed.tech/tags/uber-go-translation-cache.md>), [values](<https://devfeed.tech/tags/values.md>)

### AI overview

Uber reduced memory use in its Go translation cache by keeping keys and frequently accessed values in memory while storing other values in disk-backed files. Replacing sync.Map with a map protected by an RWMutex cut object allocations and improved CPU utilization by more than 30% in an affected service, but did not address dataset size or refreshes that temporarily held two full caches. The hybrid design shards storage across 256 locks and files, uses compact hash-based keys with collision fallback, and preloads recently used entries during refresh. Reported results include memory falling from about 2.7 GB to 750 MB, cache hit rates above 99.9%, and p99 disk-backed miss latency below 100 microseconds. In one workload, around 30,000 of more than four million entries were in memory.

### Source excerpt

Uber redesigned its Go-based translation cache to reduce its initial memory footprint by more than 90% while maintaining a cache hit rate above 99.9% and p99 disk-backed cache miss latency below 100 microseconds. The hybrid architecture combines in-memory keys and frequently accessed values with disk-backed storage, reducing garbage-collection overhead and memory consumption. By Leela Kumili

## Red Hat Working To Improve Memory Tiering For QEMU VMs To Save RAM

DevFeed: [Red Hat Working To Improve Memory Tiering For QEMU VMs To Save RAM](<https://devfeed.tech/articles/red-hat-working-to-improve-memory-tiering-for-qemu-vms-to-save-ram-133972.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/QEMU-VM-Memory-Tiering>)

Author: Michael Larabel

Published: 2026-11-14T13:12:22Z

Content type: news

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [qemu](<https://devfeed.tech/topics/qemu.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [NVMe](<https://devfeed.tech/topics/nvme.md>), [DRAM price increase](<https://devfeed.tech/topics/dram-price-increase.md>), [redhat](<https://devfeed.tech/topics/redhat.md>), [memory price increases](<https://devfeed.tech/topics/memory-price-increases.md>)

Tags: [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [kvm](<https://devfeed.tech/tags/kvm.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-tiering](<https://devfeed.tech/tags/memory-tiering.md>), [nvme](<https://devfeed.tech/tags/nvme.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [qemu](<https://devfeed.tech/tags/qemu.md>), [ram](<https://devfeed.tech/tags/ram.md>), [red-hat](<https://devfeed.tech/tags/red-hat.md>), [ssds](<https://devfeed.tech/tags/ssds.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

Red Hat engineer Peter Xu presented work on improving memory tiering for QEMU VMs so NVMe SSDs can replace DRAM in more cases while maintaining good performance. Options under evaluation include a new maccess() system call, QEMU-managed user-space tiering, and existing kernel APIs to migrate hotness. The existing-API approach appears to be the leading contender because it requires no kernel changes and only small QEMU changes; a proposed kernel patch to address limitations in MADV_PAGEOUT could improve it.

### Source excerpt

Red Hat engineers are working through a few possible approaches to enhance memory tiering for QEMU-based virtual machines to help in conserving memory. Ultimately they are after leveraging NVMe SSDs in more scenarios for QEMU VMs to save on RAM usage given today's sky high memory pricing...

## Help prevent out-of-memory restarts on Aurora MySQL 8.4

DevFeed: [Help prevent out-of-memory restarts on Aurora MySQL 8.4](<https://devfeed.tech/articles/help-prevent-out-of-memory-restarts-on-aurora-mysql-8-4-129584.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/database/help-prevent-out-of-memory-restarts-on-aurora-mysql-8-4/>)

Author: Aayushi Mittal

Published: 2026-11-11T16:51:16Z

Content type: tutorial

Language: en

Sources: [AWS Database Blog](<https://devfeed.tech/sources/aws-database-blog.md>)

Topics: [Amazon Aurora](<https://devfeed.tech/topics/amazon-aurora.md>), [MySQL](<https://devfeed.tech/topics/mysql.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [1-3](<https://devfeed.tech/tags/1-3.md>), [2](<https://devfeed.tech/tags/2.md>), [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [amazon-aurora](<https://devfeed.tech/tags/amazon-aurora.md>), [automatic](<https://devfeed.tech/tags/automatic.md>), [best-practices](<https://devfeed.tech/tags/best-practices.md>), [cloudwatch](<https://devfeed.tech/tags/cloudwatch.md>), [cluster](<https://devfeed.tech/tags/cluster.md>), [control](<https://devfeed.tech/tags/control.md>), [event](<https://devfeed.tech/tags/event.md>), [help](<https://devfeed.tech/tags/help.md>), [learn](<https://devfeed.tech/tags/learn.md>), [management](<https://devfeed.tech/tags/management.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [memory-pressure](<https://devfeed.tech/tags/memory-pressure.md>), [mysql](<https://devfeed.tech/tags/mysql.md>), [oom](<https://devfeed.tech/tags/oom.md>), [out-of-memory](<https://devfeed.tech/tags/out-of-memory.md>), [prevent](<https://devfeed.tech/tags/prevent.md>), [protection](<https://devfeed.tech/tags/protection.md>), [take](<https://devfeed.tech/tags/take.md>), [with](<https://devfeed.tech/tags/with.md>)

### AI overview

Aurora MySQL 8.4 enables automatic memory management by default to respond to memory pressure before the operating system kills `mysqld` and restarts the cluster. The article shows how to monitor `AuroraMemoryHealthState`, recovery status and CloudWatch counters, then use error logs and Performance Schema to identify whether connection buffers, temporary tables or another consumer is driving pressure. In its test, Aurora shed work with query declines and connection terminations while keeping the database up; disabling protection led to an OOM restart. The article recommends keeping automatic management on for most 8.4 workloads and alerting on sustained memory-health values above zero. Manual control requires explicitly setting `aurora_oom_response` on 8.4, where it is empty by default when automatic management is off; some actions have pairing or deployment limits. Serverless v2 scales capacity before terminating connections and does not support `tune` or `tune_buffer_pool`. Applications should handle error 1041.

### Source excerpt

Out-of-memory restarts drop every connection and force an Aurora MySQL cluster restart. Learn how Aurora MySQL 8.4's automatic memory management prevents them: confirm protection is active, read the signals during a memory-pressure event, monitor OOM avoidance with CloudWatch, and decide when to take manual control.

## Googlebooks get their first big update, freeing up 3GB to 6GB of RAM

DevFeed: [Googlebooks get their first big update, freeing up 3GB to 6GB of RAM](<https://devfeed.tech/articles/googlebooks-get-their-first-big-update-freeing-up-3gb-to-6gb-of-ram-129082.md>)

Original publisher: [Read original article](<https://blog.adafruit.com/2026/11/11/googlebooks-get-their-first-big-update-freeing-up-3gb-to-6gb-of-ram/>)

Author: Anne Barela

Published: 2026-11-11T14:45:01Z

Content type: news

Language: en

Sources: [Adafruit Industries](<https://devfeed.tech/sources/adafruit-industries-makers-hackers-artists-designers-and-engineers.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [chromeos](<https://devfeed.tech/topics/chromeos.md>), [Release notes](<https://devfeed.tech/topics/release-notes.md>), [Beta channel](<https://devfeed.tech/topics/beta-channel.md>)

Tags: [acer-googlebook-14](<https://devfeed.tech/tags/acer-googlebook-14.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-features](<https://devfeed.tech/tags/ai-features.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [asus-googlebook-14](<https://devfeed.tech/tags/asus-googlebook-14.md>), [browser](<https://devfeed.tech/tags/browser.md>), [computers](<https://devfeed.tech/tags/computers.md>), [dell-xps-googlebook](<https://devfeed.tech/tags/dell-xps-googlebook.md>), [google](<https://devfeed.tech/tags/google.md>), [googlebook](<https://devfeed.tech/tags/googlebook.md>), [hp](<https://devfeed.tech/tags/hp.md>), [hp-googlebook-14](<https://devfeed.tech/tags/hp-googlebook-14.md>), [intel](<https://devfeed.tech/tags/intel.md>), [intel-core](<https://devfeed.tech/tags/intel-core.md>), [laptops](<https://devfeed.tech/tags/laptops.md>), [lenovo-googlebook-15](<https://devfeed.tech/tags/lenovo-googlebook-15.md>), [linux](<https://devfeed.tech/tags/linux.md>), [memory](<https://devfeed.tech/tags/memory.md>), [npu](<https://devfeed.tech/tags/npu.md>), [on-device-ai](<https://devfeed.tech/tags/on-device-ai.md>), [panther-lake](<https://devfeed.tech/tags/panther-lake.md>), [ram](<https://devfeed.tech/tags/ram.md>), [release-notes](<https://devfeed.tech/tags/release-notes.md>), [software-update](<https://devfeed.tech/tags/software-update.md>), [update](<https://devfeed.tech/tags/update.md>)

### AI overview

Google's first major Googlebook update frees 3GB to 6GB of RAM on the Acer Googlebook 14, Asus Googlebook 14 and Lenovo Googlebook 15, whose Intel Core Ultra 5 Panther Lake chips support the change. Google's Reddit post attributes the gain to an updated NPU driver and reworked memory management; the release notes describe it as improved on-device AI memory use. The Snapdragon-based Dell XPS Googlebook and HP Googlebook 14 are excluded from that RAM change, but all five launch models receive a browser memory fix for heavy tab loads. The update also fixes about 30 bugs, including DRM-video black screens on external monitors and Zoom crashes with background blur, and improves Linux terminal session handling and diagnostic logging.

### Source excerpt

Julio Franco at TechSpot writes that Google has pushed the first big software update for its Googlebook laptops, about a month after they went on sale. The main change frees up 3GB to 6GB of system RAM that the on-device AI features were holding onto. Google's release notes just say memory usage for on-device AI [...]

## What Memory Management Can Teach Us About Debt, Fiat Money, and Sustainability

DevFeed: [What Memory Management Can Teach Us About Debt, Fiat Money, and Sustainability](<https://devfeed.tech/articles/what-memory-management-can-teach-us-about-debt-fiat-money-and-sustainability-128460.md>)

Original publisher: [Read original article](<https://hackernoon.com/what-memory-management-can-teach-us-about-debt-fiat-money-and-sustainability?source=rss>)

Author: DoubleX

Published: 2026-11-11T07:29:53Z

Content type: article

Language: en

Sources: [HackerNoon](<https://devfeed.tech/sources/hackernoon.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [Finance](<https://devfeed.tech/topics/finance.md>), [Memory Leaks](<https://devfeed.tech/topics/memory-leaks.md>), [arc memory management](<https://devfeed.tech/topics/arc-memory-management.md>)

Tags: [1-4](<https://devfeed.tech/tags/1-4.md>), [2](<https://devfeed.tech/tags/2.md>), [2-2](<https://devfeed.tech/tags/2-2.md>), [2020](<https://devfeed.tech/tags/2020.md>), [2023](<https://devfeed.tech/tags/2023.md>), [2026](<https://devfeed.tech/tags/2026.md>), [22](<https://devfeed.tech/tags/22.md>), [3](<https://devfeed.tech/tags/3.md>), [about](<https://devfeed.tech/tags/about.md>), [can](<https://devfeed.tech/tags/can.md>), [desktop-ssd](<https://devfeed.tech/tags/desktop-ssd.md>), [economics](<https://devfeed.tech/tags/economics.md>), [fiat](<https://devfeed.tech/tags/fiat.md>), [garbage-collection](<https://devfeed.tech/tags/garbage-collection.md>), [management](<https://devfeed.tech/tags/management.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-leak](<https://devfeed.tech/tags/memory-leak.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [money](<https://devfeed.tech/tags/money.md>), [oom-killer](<https://devfeed.tech/tags/oom-killer.md>), [out-of-memory](<https://devfeed.tech/tags/out-of-memory.md>), [programming](<https://devfeed.tech/tags/programming.md>), [promises](<https://devfeed.tech/tags/promises.md>), [resource](<https://devfeed.tech/tags/resource.md>), [resource-management](<https://devfeed.tech/tags/resource-management.md>), [sustainability](<https://devfeed.tech/tags/sustainability.md>), [swapspace](<https://devfeed.tech/tags/swapspace.md>)

### AI overview

The article uses memory management as an analogy for debt, fiat money, and ecological limits. It explains programming risks such as circular-reference leaks, allocation failures, and heavy use of swap, then likens borrowing against future memory to finance that depends on future production and resource extraction. It argues that compound debt growth pressures demand and resource use, while supply-chain dependence can spread failures. Its numerical estimates and forecasts are presented as rough calculations based on figures attributed to AI, not as established measurements. The author proposes anchoring ordinary money to a disclosed basket of electricity, essential goods, services, and precious metals, with fiat money reserved for emergencies and subject to legislative and judicial review. The proposal remains oversimplified; the article acknowledges risks including corruption, security concerns, international coordination, and political abuse.

### Source excerpt

Memory management in programming is surprisingly similar to natural resource management in economics :)

## BunshinRL: Splitting Live Agent Workspaces for Memory-Efficient Agentic RL

DevFeed: [BunshinRL: Splitting Live Agent Workspaces for Memory-Efficient Agentic RL](<https://devfeed.tech/articles/bunshinrl-splitting-live-agent-workspaces-for-memory-efficient-agentic-rl-125948.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.10453>)

Author: Yutong Huang, Vikranth Srivatsa, Alex Asch, Yiying Zhang

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [Agentic AI](<https://devfeed.tech/topics/what-is-agentic-ai.md>), [virtualization](<https://devfeed.tech/topics/virtualization.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [caching](<https://devfeed.tech/tags/caching.md>), [containers](<https://devfeed.tech/tags/containers.md>), [copy-on-write](<https://devfeed.tech/tags/copy-on-write.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [environments](<https://devfeed.tech/tags/environments.md>), [latency](<https://devfeed.tech/tags/latency.md>), [learning](<https://devfeed.tech/tags/learning.md>), [linux](<https://devfeed.tech/tags/linux.md>), [linux-kernel-module](<https://devfeed.tech/tags/linux-kernel-module.md>), [memory](<https://devfeed.tech/tags/memory.md>), [prefix-caching](<https://devfeed.tech/tags/prefix-caching.md>), [process](<https://devfeed.tech/tags/process.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [rl](<https://devfeed.tech/tags/rl.md>), [rl-rollouts](<https://devfeed.tech/tags/rl-rollouts.md>), [rollback](<https://devfeed.tech/tags/rollback.md>), [sharing](<https://devfeed.tech/tags/sharing.md>), [state](<https://devfeed.tech/tags/state.md>), [tool](<https://devfeed.tech/tags/tool.md>), [workspaces](<https://devfeed.tech/tags/workspaces.md>)

### AI overview

BunshinRL addresses memory use and delays in concurrent agentic RL rollouts by sharing environment memory, caching recurring tool-call prefixes, and speculatively executing tools. Its BunshinFork mechanism makes live copies of whole container workspaces: it reconstructs process trees, shares memory copy-on-write, preserves file, communication, and GUI state, and supports reuse of cached state and rollback after incorrect speculation. The system uses Linux kernel module and CRIU modifications. The abstract reports reductions of up to 43% in peak memory and 51% in total memory, and 23% in end-to-end rollout latency.

### Source excerpt

arXiv:2611.10453v1 Announce Type: new Abstract: Agentic reinforcement learning (RL) rollouts interleave model reasoning with tool calls in persistent workspaces. Scaling these rollouts incurs substantial memory costs for concurrent environments, while slow tool calls delay entire rollout batches. We present BunshinRL, an agentic RL rollout system that exploits shared environment state and recurring tool calls through three techniques: memory sharing across environments, tool-prefix caching, and speculative tool execution. These techniques build on BunshinFork, a novel live whole-workspace forking mechanism for containers. BunshinFork reconstructs process trees, shares memory copy-on-write, and preserves file, communication, and GUI state, enabling cached-state reuse and rollback after incorrect speculation. We implement BunshinRL and BunshinFork through Linux kernel module and CRIU modifications. BunshinRL reduces peak memory by up to 43%, total memory consumption by up to 51%, and end-to-end rollout latency by 23%.

## OOM: Offload-Optimized Full-Parameter MoE Training via Memory-Execution Co-Design

DevFeed: [OOM: Offload-Optimized Full-Parameter MoE Training via Memory-Execution Co-Design](<https://devfeed.tech/articles/oom-offload-optimized-full-parameter-moe-training-via-memory-execution-co-design-125366.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11364>)

Author: Ruijia Yang, Shiyuan Lin, Chengxi Liao, Zeyi Wen

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [GPU optimization](<https://devfeed.tech/topics/gpu-optimization.md>), [Multi-GPU](<https://devfeed.tech/topics/multi-gpu.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [long-context](<https://devfeed.tech/topics/long-context.md>), [High-Performance Computing](<https://devfeed.tech/topics/high-performance-computing.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [Memory Heirarchy](<https://devfeed.tech/topics/memory-heirarchy.md>), [gradient\_accumulation](<https://devfeed.tech/topics/gradient-accumulation.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [co-design](<https://devfeed.tech/tags/co-design.md>), [cost](<https://devfeed.tech/tags/cost.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dependencies](<https://devfeed.tech/tags/dependencies.md>), [execution](<https://devfeed.tech/tags/execution.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-memory](<https://devfeed.tech/tags/gpu-memory.md>), [host-memory](<https://devfeed.tech/tags/host-memory.md>), [hybrid](<https://devfeed.tech/tags/hybrid.md>), [layer](<https://devfeed.tech/tags/layer.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [model](<https://devfeed.tech/tags/model.md>), [moe](<https://devfeed.tech/tags/moe.md>), [multi-gpu](<https://devfeed.tech/tags/multi-gpu.md>), [offloading](<https://devfeed.tech/tags/offloading.md>), [oom](<https://devfeed.tech/tags/oom.md>), [optimized](<https://devfeed.tech/tags/optimized.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

OOM co-designs expert execution and memory allocation to train full-parameter MoE models under GPU and host memory limits. It plans expert residency and activation retention for expert-parallel execution, partitions GPU worksets for local execution, and, in hybrid models, retains attention results on the host and selectively regenerates layer boundaries. Calibrated cost models guide configurations, which OOM selects using short runs. On four H100s, it supports 1M-token contexts and a 122B-parameter MoE model where FSDP2, ZeRO-Offload, and MegaTrain run out of GPU or host memory; on four 24 GB RTX 4090s, it supports 512K-token contexts and 1,024 1K-token sequences per step without gradient accumulation. In matched-global-batch comparisons, it reports 1.58x and 4.08x geometric-mean throughput over CPU-offloaded FSDP2 on the two platforms, with up to 2.14x and 7.77x, while using the least host memory among evaluated systems.

### Source excerpt

arXiv:2611.11364v1 Announce Type: cross Abstract: Offloading model state to host memory and streaming layer weights let a single multi-GPU node train large mixture-of-experts (MoE) models. GPU memory then becomes a performance budget, shared by resident expert state, retained activations, and execution worksets. The best allocation depends on the expert execution path, batch size, and sequence length. We present OOM (Offload-Optimized MoE), a full-parameter MoE training system that co-designs expert execution and memory allocation. For expert-parallel execution, OOM updates resident experts on their owning GPUs and jointly plans state residency and MoE activation retention according to pipeline dependencies. For local execution, OOM bounds GPU worksets through dependency-aware partitioning; in hybrid models, it also retains attention results on the host to avoid costly recomputation and selectively regenerates layer boundaries through cheaper layers to make room. Calibrated cost models guide joint configuration within GPU and host budgets to minimize step time; OOM selects among the configured paths through short executions. On four H100s, OOM trains with 1M-token contexts, whereas FSDP2, ZeRO-Offload, and MegaTrain exhaust GPU memory at 512K. It also trains a 122B-parameter MoE model, on which all three exhaust host memory. On four 24 GB RTX 4090s, it trains 512K-token contexts and 1,024 1K-token sequences per step without gradient accumulation. Across matched-global-batch comparisons, OOM delivers 1.58x and 4.08x geometric-mean (up to 2.14x and 7.77x) throughput over CPU-offloaded FSDP2 on the two platforms while using the least host memory among the evaluated systems -- as little as 1/3.0 of FSDP2's.

## Googlebooks get their first big update, freeing up 3GB to 6GB of RAM

DevFeed: [Googlebooks get their first big update, freeing up 3GB to 6GB of RAM](<https://devfeed.tech/articles/googlebooks-get-their-first-big-update-freeing-up-3gb-to-6gb-of-ram-124138.md>)

Original publisher: [Read original article](<https://www.techspot.com/news/114498-googlebooks-get-their-first-big-update-freeing-up.html>)

Author: Julio Franco

Published: 2026-11-09T22:23:00Z

Content type: news

Language: en

Sources: [TechSpot](<https://devfeed.tech/sources/techspot.md>)

Topics: [bugfixes](<https://devfeed.tech/topics/bugfixes.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [virtual machines](<https://devfeed.tech/topics/virtual-machines.md>), [enhancing AI model accuracy](<https://devfeed.tech/topics/enhancing-ai-model-accuracy.md>)

Tags: [acer-googlebook-14](<https://devfeed.tech/tags/acer-googlebook-14.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-features](<https://devfeed.tech/tags/ai-features.md>), [ai-stack](<https://devfeed.tech/tags/ai-stack.md>), [android](<https://devfeed.tech/tags/android.md>), [asus-googlebook-14](<https://devfeed.tech/tags/asus-googlebook-14.md>), [audio](<https://devfeed.tech/tags/audio.md>), [bluetooth](<https://devfeed.tech/tags/bluetooth.md>), [browser](<https://devfeed.tech/tags/browser.md>), [bugs](<https://devfeed.tech/tags/bugs.md>), [build](<https://devfeed.tech/tags/build.md>), [changelog](<https://devfeed.tech/tags/changelog.md>), [chrome](<https://devfeed.tech/tags/chrome.md>), [first](<https://devfeed.tech/tags/first.md>), [googlebook](<https://devfeed.tech/tags/googlebook.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [intel](<https://devfeed.tech/tags/intel.md>), [laptop](<https://devfeed.tech/tags/laptop.md>), [lenovo-googlebook-15](<https://devfeed.tech/tags/lenovo-googlebook-15.md>), [memory-optimization](<https://devfeed.tech/tags/memory-optimization.md>), [operating-system](<https://devfeed.tech/tags/operating-system.md>), [panther-lake](<https://devfeed.tech/tags/panther-lake.md>), [ram](<https://devfeed.tech/tags/ram.md>), [software](<https://devfeed.tech/tags/software.md>), [update](<https://devfeed.tech/tags/update.md>), [updates](<https://devfeed.tech/tags/updates.md>)

### AI overview

Google's November Googlebook update frees 3-6GB of RAM on the Acer Googlebook 14, Asus Googlebook 14, and Lenovo Googlebook 15 by updating the NPU driver and revising memory management for on-device AI. Google's release notes describe improved memory use and multitasking stability but give no figure; the RAM savings are reported in a Google Reddit post. The Dell XPS Googlebook and HP Googlebook 14 do not receive this optimization, though all five launch models get a browser memory stability fix. The update also addresses DRM video display failures, Zoom crashes, audio and Bluetooth issues, and Linux terminal sleep, SSH forwarding, storage, and diagnostic logging behavior. It includes the November 1 Android security patch, which addresses around 60 vulnerabilities. Rollout is gradual; owners can check for build CL3B.260622.286.R2 under Settings > System > Software updates > System update.

### Source excerpt

The Googlebook November update also fixes roughly 30 bugs across all five launch machines. The memory optimization applies to the Acer Googlebook 14, Asus Googlebook 14 and Lenovo Googlebook 15, the three launch models based on Intel's Core Ultra 5 Panther Lake chips. Read Entire Article

## Memory Management in Python

DevFeed: [Memory Management in Python](<https://devfeed.tech/articles/memory-management-in-python-122569.md>)

Original publisher: [Read original article](<https://realpython.com/python-memory-management/>)

Author: Alexander VanTol

Published: 2026-11-08T14:00:00Z

Content type: article

Language: en

Sources: [Real Python](<https://devfeed.tech/sources/real-python.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [memory pool](<https://devfeed.tech/topics/memory-pool.md>), [Python](<https://devfeed.tech/topics/python.md>)

Tags: [algorithms](<https://devfeed.tech/tags/algorithms.md>), [cpython](<https://devfeed.tech/tags/cpython.md>), [data-structures-and-algorithms](<https://devfeed.tech/tags/data-structures-and-algorithms.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [python](<https://devfeed.tech/tags/python.md>)

### AI overview

The excerpt explains how CPython manages object memory in its default build, using reference counts to trigger object deallocation when a count reaches zero. Its small-block allocator organizes memory into arenas, pools, and blocks: pools group blocks by size class, track available space, and can be used, full, or empty. Freed blocks are reused within the Python process; arenas are the units that can be returned to the operating system. The allocator favors fuller arenas so that other arenas can become empty and potentially be released. The description is version-sensitive (Python 3.14) and focuses on the default GIL-enabled CPython build; it notes that the optional free-threaded build uses a different allocator.

### Source excerpt

Take a deep dive into Python memory management and learn how CPython allocates and frees memory for your objects with arenas, pools, and blocks.

## CVE-2026-123616 - imgproxy 3.3.0 before 4.0.16 Memory Exhaustion DoS via ICO Loader

DevFeed: [CVE-2026-123616 - imgproxy 3.3.0 before 4.0.16 Memory Exhaustion DoS via ICO Loader](<https://devfeed.tech/articles/cve-2026-123616-imgproxy-3-3-0-before-4-0-16-memory-exhaustion-dos-via-ico-loader-122370.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-123616>)

Author: CVEFEED.io; vulncheck.com

Published: 2026-11-08T02:16:57Z

Content type: article

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [crashes](<https://devfeed.tech/tags/crashes.md>), [cve-2026-123616](<https://devfeed.tech/tags/cve-2026-123616.md>), [dos](<https://devfeed.tech/tags/dos.md>), [evilmartians](<https://devfeed.tech/tags/evilmartians.md>), [ico](<https://devfeed.tech/tags/ico.md>), [imgproxy](<https://devfeed.tech/tags/imgproxy.md>), [loader](<https://devfeed.tech/tags/loader.md>), [memory](<https://devfeed.tech/tags/memory.md>)

### AI overview

imgproxy versions 3.3.0 through before 4.0.16 have an ICO loader flaw: it does not validate the file's data_size field before allocating memory. A crafted 23-byte ICO file declaring about 4 GB can cause memory exhaustion and worker crashes without authentication. The source rates the vulnerability 8.7 (HIGH); it does not provide remediation details.

### Source excerpt

CVE ID :CVE-2026-123616 Published : Nov. 8, 2026, 2:16 a.m. | 47 minutes ago Description :imgproxy from 3.3.0 before 4.0.16 contains an unbounded memory allocation vulnerability in the ICO loader that fails to validate the data_size field. Attackers can supply a crafted 23-byte ICO file declaring a ~4GB data_size to trigger memory exhaustion and worker crashes without authentication. Severity: 8.7 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## KDE Plasma 6.9 Will Stop Leaking Memory When Right Clicking Files/Folders On Desktop

DevFeed: [KDE Plasma 6.9 Will Stop Leaking Memory When Right Clicking Files/Folders On Desktop](<https://devfeed.tech/articles/kde-plasma-6-9-will-stop-leaking-memory-when-right-clicking-files-folders-on-desktop-122015.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/KDE-Plasma-6.9-Leak-Desktop>)

Author: Michael Larabel

Published: 2026-11-07T11:34:48Z

Content type: news

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [desktop](<https://devfeed.tech/tags/desktop.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [files](<https://devfeed.tech/tags/files.md>), [kde](<https://devfeed.tech/tags/kde.md>), [kde-plasma](<https://devfeed.tech/tags/kde-plasma.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [memory](<https://devfeed.tech/tags/memory.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [plasma](<https://devfeed.tech/tags/plasma.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>)

### AI overview

KDE developers are preparing Plasma 6.9 with several usability changes: users will be able to reposition on-screen display pop-ups, and desktop widgets will be sized to avoid cutting off their content. A fix removes a small memory leak that occurred each time users right-clicked desktop files or folders. Other changes include a KDE-style password prompt for `systemd-ask-password --no-tty` on systemd 257 and newer, and Plasma Browser Integration support for Glide. The excerpt also notes a KWin crash fix for the Plasma 6.8.3 point release when a DisplayLink screen is connected.

### Source excerpt

With Plasma 6.8 stable approaching one month old, KDE developers continue ramping up work toward Plasma 6.9. Here is some of the interesting work they've been up to so far in November...

## CVE-2026-71397 - Unbounded request body accumulation on streaming RPCs causes memory exhaustion in grpc\_server

DevFeed: [CVE-2026-71397 - Unbounded request body accumulation on streaming RPCs causes memory exhaustion in grpc\_server](<https://devfeed.tech/articles/cve-2026-71397-unbounded-request-body-accumulation-on-streaming-rpcs-causes-memory-exhaustion-in-grpc-server-121676.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-71397>)

Author: cvefeed.io

Published: 2026-11-06T21:17:10Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [gRPC](<https://devfeed.tech/topics/grpc.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [Streaming](<https://devfeed.tech/topics/streaming.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [crash](<https://devfeed.tech/tags/crash.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-71397](<https://devfeed.tech/tags/cve-2026-71397.md>), [elixir](<https://devfeed.tech/tags/elixir.md>), [elixir-grpc](<https://devfeed.tech/tags/elixir-grpc.md>), [grpc](<https://devfeed.tech/tags/grpc.md>), [grpc-server](<https://devfeed.tech/tags/grpc-server.md>), [memory](<https://devfeed.tech/tags/memory.md>), [request](<https://devfeed.tech/tags/request.md>), [server](<https://devfeed.tech/tags/server.md>), [streaming](<https://devfeed.tech/tags/streaming.md>)

### AI overview

CVE-2026-71397 (CVSS 8.2 High) lets unauthenticated clients exhaust memory and crash grpc_server through oversized client-streaming or bidirectional-streaming requests. The streaming read path does not enforce the configured max_body_size, and reading_stream/1 keeps appending received chunks until a complete length-prefixed message arrives. A client can send a declared large message across many small HTTP/2 DATA frames, causing unbounded buffering. The unary read path does enforce the cap. Affected versions are grpc_server 1.0.0 through versions before 1.0.6.

### Source excerpt

CVE ID :CVE-2026-71397 Published : Nov. 6, 2026, 9:17 p.m. | 53 minutes ago Description :Allocation of Resources Without Limits or Throttling vulnerability in elixir-grpc grpc_server allows unauthenticated attackers to exhaust memory and crash the node via an oversized client-streaming or bidirectional-streaming request body. The streaming read clause of 'Elixir.GRPC.Server.Adapters.Cowboy.Handler':info/3 (grpc_server/lib/grpc/server/adapters/cowboy/handler.ex) never checks the configured max_body_size. Only the unary read path enforces that cap. 'Elixir.GRPC.Server.Adapters.Cowboy':reading_stream/1 (grpc_server/lib/grpc/server/adapters/cowboy.ex) appends every received chunk to its buffer until a complete length-prefixed message arrives. A client can declare a very large message and send it across many small HTTP/2 DATA frames, and the server buffers all of it. This issue affects grpc_server: from 1.0.0 before 1.0.6. Severity: 8.2 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## CVE-2026-73095 - Apache Juneau: Denial of service via unbounded memory allocation in parsers and REST request handling

DevFeed: [CVE-2026-73095 - Apache Juneau: Denial of service via unbounded memory allocation in parsers and REST request handling](<https://devfeed.tech/articles/cve-2026-73095-apache-juneau-denial-of-service-via-unbounded-memory-allocation-in-parsers-and-rest-request-handling-119027.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-73095>)

Author: cvefeed.io

Published: 2026-11-05T12:17:50Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [apache](<https://devfeed.tech/tags/apache.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-73095](<https://devfeed.tech/tags/cve-2026-73095.md>), [denial-of-service](<https://devfeed.tech/tags/denial-of-service.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [gigabyte](<https://devfeed.tech/tags/gigabyte.md>), [handling](<https://devfeed.tech/tags/handling.md>), [input](<https://devfeed.tech/tags/input.md>), [issue](<https://devfeed.tech/tags/issue.md>), [limits](<https://devfeed.tech/tags/limits.md>), [memory](<https://devfeed.tech/tags/memory.md>), [request](<https://devfeed.tech/tags/request.md>), [rest](<https://devfeed.tech/tags/rest.md>)

### AI overview

Apache Juneau through 9.2.0 is vulnerable to denial of service from unbounded memory allocation. Its default-enabled MessagePack parser sizes buffers using input length values, so an unauthenticated client can send only a few bytes to trigger a multi-gigabyte allocation and an OutOfMemoryError. Related unbounded allocations in multipart uploads and request reader buffers are also fixed. Upgrade to Apache Juneau 9.2.1.

### Source excerpt

CVE ID :CVE-2026-73095 Published : Nov. 5, 2026, 12:17 p.m. | 2 hours, 7 minutes ago Description :Allocation of Resources Without Limits or Throttling vulnerability in Apache Juneau. The MessagePack parser, which the REST server enables by default, sizes buffers from length values in the input. An unauthenticated client can send a few bytes that force a multi-gigabyte allocation and an OutOfMemoryError, causing denial of service. Related unbounded allocations in REST request handling (multipart uploads and request reader buffers) are fixed at the same time. This issue affects Apache Juneau: through 9.2.0. Users are recommended to upgrade to version 9.2.1, which fixes the issue. Severity: 8.7 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## Where Did the Saved Bytes Go? Compact Object Headers in IntelliJ IDEA

DevFeed: [Where Did the Saved Bytes Go? Compact Object Headers in IntelliJ IDEA](<https://devfeed.tech/articles/where-did-the-saved-bytes-go-compact-object-headers-in-intellij-idea-114251.md>)

Original publisher: [Read original article](<https://blog.jetbrains.com/idea/2026/11/where-did-the-saved-bytes-go-compact-object-headers-in-intellij-idea/>)

Author: Igor Kulakov

Published: 2026-11-03T12:20:40Z

Content type: article

Language: en

Sources: [JetBrains](<https://devfeed.tech/sources/the-jetbrains-blog.md>)

Topics: [IntelliJ IDEA](<https://devfeed.tech/topics/intellij-idea.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [boundary](<https://devfeed.tech/tags/boundary.md>), [garbage-collection](<https://devfeed.tech/tags/garbage-collection.md>), [headers](<https://devfeed.tech/tags/headers.md>), [idea](<https://devfeed.tech/tags/idea.md>), [intellij-idea](<https://devfeed.tech/tags/intellij-idea.md>), [intellij-platform](<https://devfeed.tech/tags/intellij-platform.md>), [java](<https://devfeed.tech/tags/java.md>), [jvm](<https://devfeed.tech/tags/jvm.md>), [memory](<https://devfeed.tech/tags/memory.md>), [object](<https://devfeed.tech/tags/object.md>), [performance](<https://devfeed.tech/tags/performance.md>)

### AI overview

In tests inside IntelliJ IDEA, compact object headers reduced modeled heap footprints by about 6.3% for the Welcome screen and Petclinic objects and 8.2%-8.4% for monorepo objects, but produced no consistent RSS reduction at identical -Xmx. They also cut allocated volume by 8.8%-13.0%; monorepo indexing runs recorded fewer GC events and lower mean collection time and CPU, while scripted operation timings suggested only a modest speedup. The minimum tested heap that completed every repeat fell 6% for Petclinic and 17% for the monorepo; these are workload thresholds, not production recommendations. Savings depend on object shapes, fields and alignment, so smaller headers do not shrink every object or guarantee lower RSS or faster execution. Measure heap, allocation, GC, performance and heap limits separately.

### Source excerpt

AI agents may make code faster to produce, but they don't make it cheaper to run. In applications managing millions of objects, memory allocation and garbage collection still matter. The JVM has been addressing these costs on several fronts: ZGC targets garbage collection pauses, Project Leyden tackles startup and warmup, and Project Lilliput reduces the [...]

## Sculpt OS 26.10 Released - Now Powered By Genode's Custom Microkernel

DevFeed: [Sculpt OS 26.10 Released - Now Powered By Genode's Custom Microkernel](<https://devfeed.tech/articles/sculpt-os-26-10-released-now-powered-by-genode-s-custom-microkernel-109838.md>)

Original publisher: [Read original article](<https://www.phoronix.com/news/Sculpt-OS-26.10>)

Author: Michael Larabel

Published: 2026-10-31T09:37:22Z

Content type: release

Language: en

Sources: [Phoronix](<https://devfeed.tech/sources/phoronix.md>)

Topics: [Kernel](<https://devfeed.tech/topics/kernel.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [virtual machines](<https://devfeed.tech/topics/virtual-machines.md>)

Tags: [camera](<https://devfeed.tech/tags/camera.md>), [components](<https://devfeed.tech/tags/components.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [custom](<https://devfeed.tech/tags/custom.md>), [desktop-linux](<https://devfeed.tech/tags/desktop-linux.md>), [handling](<https://devfeed.tech/tags/handling.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [keyboards](<https://devfeed.tech/tags/keyboards.md>), [linux-benchmarking](<https://devfeed.tech/tags/linux-benchmarking.md>), [linux-hardware-benchmarks](<https://devfeed.tech/tags/linux-hardware-benchmarks.md>), [linux-hardware-reviews](<https://devfeed.tech/tags/linux-hardware-reviews.md>), [linux-how-to](<https://devfeed.tech/tags/linux-how-to.md>), [linux-performance](<https://devfeed.tech/tags/linux-performance.md>), [linux-server-benchmarks](<https://devfeed.tech/tags/linux-server-benchmarks.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [open-source-graphics](<https://devfeed.tech/tags/open-source-graphics.md>), [os](<https://devfeed.tech/tags/os.md>), [phoronix](<https://devfeed.tech/tags/phoronix.md>), [phoronix-test-suite](<https://devfeed.tech/tags/phoronix-test-suite.md>), [running](<https://devfeed.tech/tags/running.md>), [scheduling](<https://devfeed.tech/tags/scheduling.md>), [third-party](<https://devfeed.tech/tags/third-party.md>), [ubuntu-benchmarks](<https://devfeed.tech/tags/ubuntu-benchmarks.md>), [ubuntu-hardware](<https://devfeed.tech/tags/ubuntu-hardware.md>), [uk](<https://devfeed.tech/tags/uk.md>), [update](<https://devfeed.tech/tags/update.md>), [virtualbox](<https://devfeed.tech/tags/virtualbox.md>)

### AI overview

Sculpt OS 26.10 is the first release based on Genode's custom microkernel rather than a third-party kernel. The excerpt says the kernel was designed for Genode and Sculpt OS and addresses CPU scheduling and kernel-memory management deficiencies. The release also updates its VirtualBox components from version 6 to 7 for running in virtual machines, improves webcam handling, and adds UK keyboard support.

### Source excerpt

Sculpt OS as the general purpose operating system project born out of the Genode initiative is now using the project's own microkernel rather than a third-party microkernel...

## The Real Python Podcast - Episode #317: Demystifying Python Memory Management & Choosing a Project Layout

DevFeed: [The Real Python Podcast - Episode #317: Demystifying Python Memory Management & Choosing a Project Layout](<https://devfeed.tech/articles/the-real-python-podcast-episode-317-demystifying-python-memory-management-choosing-a-project-layout-108362.md>)

Original publisher: [Read original article](<https://realpython.com/podcasts/rpp/317/>)

Author: Real Python

Published: 2026-10-30T12:00:00Z

Content type: article

Language: en

Sources: [Real Python](<https://devfeed.tech/sources/real-python.md>)

Topics: [project layout](<https://devfeed.tech/topics/project-layout.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [Python](<https://devfeed.tech/topics/python.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [agentic-coding](<https://devfeed.tech/tags/agentic-coding.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agent](<https://devfeed.tech/tags/ai-agent.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [browser](<https://devfeed.tech/tags/browser.md>), [cli](<https://devfeed.tech/tags/cli.md>), [code](<https://devfeed.tech/tags/code.md>), [codex](<https://devfeed.tech/tags/codex.md>), [coding](<https://devfeed.tech/tags/coding.md>), [course](<https://devfeed.tech/tags/course.md>), [layout](<https://devfeed.tech/tags/layout.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [podcast](<https://devfeed.tech/tags/podcast.md>), [project](<https://devfeed.tech/tags/project.md>), [project-layout](<https://devfeed.tech/tags/project-layout.md>), [projects](<https://devfeed.tech/tags/projects.md>), [python](<https://devfeed.tech/tags/python.md>)

### AI overview

The episode covers Python memory management and project layout, alongside community news and tools. Its memory segment introduces how Python allocates, tracks, and frees memory, in light of recent garbage collector changes and ongoing work to remove the GIL. The project layout segment addresses choosing between flat and `src/` layouts, organizing code, and applying patterns to CLI, web, and GUI apps. Other featured items include Parquet's columnar layout, encodings, zstd compression, and row group statistics; a Django profiler and a tool for identifying dependencies that block upgrades; and a discussion of Pendulum's patched `+` operator hack and the design implications of subclassing `datetime`. The excerpt lists topics but does not provide the episode's detailed arguments or recommendations.

### Source excerpt

How does Python manage memory behind the scenes? What are the common ways to lay out a Python project? Christopher Trudeau is back on the show this week with another batch of PyCoder's Weekly articles and projects.

## Rimuru: Unified Memory Management for Modern LLM Serving via Elastic Expert Offloading

DevFeed: [Rimuru: Unified Memory Management for Modern LLM Serving via Elastic Expert Offloading](<https://devfeed.tech/articles/rimuru-unified-memory-management-for-modern-llm-serving-via-elastic-expert-offloading-108894.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.40305>)

Author: Zhiyuan Fang, Dingyang Li, Jianfeng Zhong, Bin Wang, Weixi Zhu, Cunshu Ni, Zicong Hong, Wuhui Chen, Zibin Zheng

Published: 2026-10-30T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [memory pool](<https://devfeed.tech/topics/memory-pool.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [architectures](<https://devfeed.tech/tags/architectures.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attention](<https://devfeed.tech/tags/attention.md>), [building](<https://devfeed.tech/tags/building.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [combines](<https://devfeed.tech/tags/combines.md>), [concurrency](<https://devfeed.tech/tags/concurrency.md>), [copy](<https://devfeed.tech/tags/copy.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-os](<https://devfeed.tech/tags/cs-os.md>), [elastic](<https://devfeed.tech/tags/elastic.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-serving](<https://devfeed.tech/tags/llm-serving.md>), [management](<https://devfeed.tech/tags/management.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [modern](<https://devfeed.tech/tags/modern.md>), [offloading](<https://devfeed.tech/tags/offloading.md>), [serving](<https://devfeed.tech/tags/serving.md>), [unified-memory](<https://devfeed.tech/tags/unified-memory.md>)

### AI overview

Rimuru addresses fixed memory partitions in LLM serving by sharing a page pool across model weights and attention states, then dynamically reallocating pages as workload demands change. Per-tensor memory views support different page sizes and page tables; huge-page alignment and a multi-level allocator split and merge pages to reuse capacity across tensor types. A page-table-aware fused MoE kernel accesses non-contiguous expert weights without copying. Built on vLLM, Rimuru is reported to achieve up to 19.03x baseline throughput and an average 98.1% reduction in TTFT. These are the abstract's reported results; the supplied text does not provide evaluation conditions or further detail.

### Source excerpt

arXiv:2610.40305v1 Announce Type: cross Abstract: Modern LLMs strain device memory through large model weights and attention states, particularly in resource-constrained scenarios. Existing systems separately provision memory for model weights and attention states, imposing fixed fixed boundaries that cannot adapt to the changing memory demands of workloads. Under heavy workloads, insufficient state capacity limits request concurrency and reduces throughput; under light workloads, idle memory in one region cannot be reallocated to another, wasting device memory. Meanwhile, modern LLM architectures introduce heterogeneous tensors with diverse layouts and lifetimes, further complicating memory management. These problems motivate unified memory management for heterogeneous tensors. However, a shared pool alone does not make memory directly reusable across tensors with different allocation granularities and execution layouts. This paper presents Rimuru, a modern LLM serving system that unifies memory management through elastic expert offloading. Rimuru establishes a shared page pool to enables elastic memory sharing across heterogeneous tensors. To accommodate different allocation granularities, Rimuru introduces memory views that allow each tensor type to interpret the shared memory space using its own page size and page table. Furthermore, Rimuru combines huge-page alignment with a multi-level page allocator that splits and merges pages, enabling cross-type capacity reuse. A page-table-aware fused MoE kernel directly accesses non-contiguous expert weights, avoiding copy overhead. Building on these mechanisms, Rimuru employs a workload-aware adjuster that dynamically reallocates pages across tensor types according to runtime demands. Implemented on top of vLLM, Rimuru remains compatible with modern LLMs. Evaluations show that Rimuru achieves up to 19.03x baseline throughput and reduces TTFT by 98.1% on average.

## CVE-2026-109584 - HTTP/2 cookie crumb join expands a small HPACK header block into an unbounded allocation in Bandit

DevFeed: [CVE-2026-109584 - HTTP/2 cookie crumb join expands a small HPACK header block into an unbounded allocation in Bandit](<https://devfeed.tech/articles/cve-2026-109584-http-2-cookie-crumb-join-expands-a-small-hpack-header-block-into-an-unbounded-allocation-in-bandit-100466.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-109584>)

Author: cvefeed.io

Published: 2026-10-26T13:32:22Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [HTTP](<https://devfeed.tech/topics/http.md>), [Cookies](<https://devfeed.tech/topics/cookies.md>)

Tags: [bandit](<https://devfeed.tech/tags/bandit.md>), [cve-2026-109584](<https://devfeed.tech/tags/cve-2026-109584.md>), [http-2](<https://devfeed.tech/tags/http-2.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mtrudel](<https://devfeed.tech/tags/mtrudel.md>), [vulnerability](<https://devfeed.tech/tags/vulnerability.md>)

### AI overview

In Bandit versions 0.3.3 through releases before 1.14.0, an unauthenticated attacker can use a small HPACK-compressed HTTP/2 HEADERS block to trigger excessive memory allocation. The block can store a large cookie value once and reference it tens of thousands of times; Bandit then joins the cookie crumbs into one binary. A reported 49 KB block caused a default-configured server to allocate a 183 MB cookie binary before application code ran, and each concurrent stream can repeat the cost. Bandit limits compressed block size, but the decoded header-list check defaults to :infinity, so it does not constrain this amplification.

### Source excerpt

CVE ID :CVE-2026-109584 Published : Oct. 26, 2026, 1:32 p.m. | 32 minutes ago Description :Improper Handling of Highly Compressed Data (Data Amplification) vulnerability in the HTTP/2 implementation of mtrudel bandit allows an unauthenticated remote attacker to exhaust server memory via a small HPACK-compressed HEADERS block. The block stores one large cookie value in the HPACK dynamic table and then references it tens of thousands of times with 1-byte indexed fields. In the reporter's run, a 49 KB block made a server with default settings allocate a 183 MB cookie binary before any application code ran, and every concurrent stream can repeat the cost. Bandit.HTTP2.Connection caps only the compressed block (max_header_block_size), and its check of the decoded list against max_header_list_size never fires because that setting defaults to :infinity. combine_cookie_crumbs/1 in lib/bandit/http2/stream.ex then joins every crumb into one binary. This issue affects bandit: from 0.3.3 before 1.14.0. Severity: 8.7 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## CVE-2026-111101 - HTTP/1.1 chunked request bodies made of 1-byte chunks amplify memory use in Bandit

DevFeed: [CVE-2026-111101 - HTTP/1.1 chunked request bodies made of 1-byte chunks amplify memory use in Bandit](<https://devfeed.tech/articles/cve-2026-111101-http-1-1-chunked-request-bodies-made-of-1-byte-chunks-amplify-memory-use-in-bandit-100470.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-111101>)

Author: cvefeed.io

Published: 2026-10-26T13:31:48Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [HTTP](<https://devfeed.tech/topics/http.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [2026](<https://devfeed.tech/tags/2026.md>), [amplify](<https://devfeed.tech/tags/amplify.md>), [bandit](<https://devfeed.tech/tags/bandit.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-111101](<https://devfeed.tech/tags/cve-2026-111101.md>), [handler](<https://devfeed.tech/tags/handler.md>), [http](<https://devfeed.tech/tags/http.md>), [http1](<https://devfeed.tech/tags/http1.md>), [limits](<https://devfeed.tech/tags/limits.md>), [list](<https://devfeed.tech/tags/list.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mtrudel](<https://devfeed.tech/tags/mtrudel.md>), [payload](<https://devfeed.tech/tags/payload.md>), [process](<https://devfeed.tech/tags/process.md>), [remote](<https://devfeed.tech/tags/remote.md>), [request](<https://devfeed.tech/tags/request.md>), [resources](<https://devfeed.tech/tags/resources.md>), [route](<https://devfeed.tech/tags/route.md>), [server](<https://devfeed.tech/tags/server.md>), [vulnerability](<https://devfeed.tech/tags/vulnerability.md>)

### AI overview

Bandit versions 0.1.0 through releases before 1.14.0 are affected by an unauthenticated resource-exhaustion vulnerability in the HTTP/1.1 chunked-body reader. The reader retains each chunk as a separate list cell and binary until joining them; read_body/2's :length limit counts payload bytes, not chunks. As a result, many 1-byte chunks can consume far more memory and CPU than their payload suggests: the excerpt reports that 48 MB of such data grew one handler process to about 500 MB and used about 6 seconds of CPU. A route need not explicitly read the body: on keep-alive connections, ensure_completed/1 drains unread bodies through the same reader.

### Source excerpt

CVE ID :CVE-2026-111101 Published : Oct. 26, 2026, 1:31 p.m. | 32 minutes ago Description :Allocation of Resources Without Limits or Throttling vulnerability in the HTTP/1.1 chunked body reader of mtrudel bandit allows an unauthenticated remote attacker to exhaust server memory and CPU via a request body made of many 1-byte chunks. The chunked reader in lib/bandit/http1/socket.ex keeps every chunk as its own list cell and binary term until IO.iodata_to_binary/1 joins them, and nothing limits the number of chunks. The :length option of read_body/2 counts only data bytes, so tiny chunks cost far more memory than their payload: 48 MB of 1-byte chunks grew one handler process to about 500 MB and took about 6 s of CPU. The route does not need to read the body, because on keep-alive connections ensure_completed/1 drains unread bodies through the same reader. This issue affects bandit: from 0.1.0 before 1.14.0. Severity: 8.7 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## CVE-2026-70388 - Memory Corruption via Integer Underflow in Message Length Validation

DevFeed: [CVE-2026-70388 - Memory Corruption via Integer Underflow in Message Length Validation](<https://devfeed.tech/articles/cve-2026-70388-memory-corruption-via-integer-underflow-in-message-length-validation-100198.md>)

Original publisher: [Read original article](<https://cvefeed.io/vuln/detail/CVE-2026-70388>)

Author: CVEFEED.io

Published: 2026-10-26T09:16:35Z

Content type: news

Language: en

Sources: [Latest High/Critical Vulnerabilitiy Feed](<https://devfeed.tech/sources/latest-high-critical-vulnerabilitiy-feed.md>)

Topics: [Memory Safety](<https://devfeed.tech/topics/memory-safety.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>)

Tags: [2-0-0](<https://devfeed.tech/tags/2-0-0.md>), [2026](<https://devfeed.tech/tags/2026.md>), [arithmetic](<https://devfeed.tech/tags/arithmetic.md>), [bypass](<https://devfeed.tech/tags/bypass.md>), [code](<https://devfeed.tech/tags/code.md>), [cve](<https://devfeed.tech/tags/cve.md>), [cve-2026-70388](<https://devfeed.tech/tags/cve-2026-70388.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-corruption](<https://devfeed.tech/tags/memory-corruption.md>), [message](<https://devfeed.tech/tags/message.md>), [published](<https://devfeed.tech/tags/published.md>), [validation](<https://devfeed.tech/tags/validation.md>)

### AI overview

The source reports that SOEM versions 2.0.0 and earlier may mishandle message lengths from a malicious network device. Arithmetic without proper validation can let negative or extremely large values bypass checks, causing writes outside allocated memory and potentially arbitrary code execution on the master. The listing rates the issue 8.3 (High), but marks remote exploitation "No"; it does not record affected products in its table or provide mitigation details.

### Source excerpt

CVE ID :CVE-2026-70388 Published : Oct. 26, 2026, 9:16 a.m. | 37 minutes ago Description :SOEM version 2.0.0 and earlier contains vulnerabilities where a malicious device on the network can send specially crafted messages that cause the master to write data outside allocated memory boundaries. The software performs arithmetic on message length values without proper validation, allowing negative or extremely large values to bypass safety checks. This can corrupt memory and potentially allow an attacker to execute arbitrary code on the master system. Severity: 8.3 | HIGH Visit the link for more details, such as CVSS details, affected products, timeline, and more...

## Allocating Off-Heap Memory: Typically 6-18x Faster in JDK 28

DevFeed: [Allocating Off-Heap Memory: Typically 6-18x Faster in JDK 28](<https://devfeed.tech/articles/allocating-off-heap-memory-typically-6-18x-faster-in-jdk-28-97194.md>)

Original publisher: [Read original article](<https://inside.java/2026/10/23/confined-pools/>)

Author: Per-Ake Minborg

Published: 2026-10-23T00:00:00Z

Content type: article

Language: en

Sources: [Inside Java](<https://devfeed.tech/sources/inside-java.md>)

Topics: [memory pool](<https://devfeed.tech/topics/memory-pool.md>), [Cache](<https://devfeed.tech/topics/cache.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [openjdk](<https://devfeed.tech/topics/openjdk.md>)

Tags: [api](<https://devfeed.tech/tags/api.md>), [arena](<https://devfeed.tech/tags/arena.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [benefit](<https://devfeed.tech/tags/benefit.md>), [cache](<https://devfeed.tech/tags/cache.md>), [faster](<https://devfeed.tech/tags/faster.md>), [improvement](<https://devfeed.tech/tags/improvement.md>), [jdk](<https://devfeed.tech/tags/jdk.md>), [jdk-28](<https://devfeed.tech/tags/jdk-28.md>), [memory](<https://devfeed.tech/tags/memory.md>), [native](<https://devfeed.tech/tags/native.md>), [performance](<https://devfeed.tech/tags/performance.md>), [project-panama](<https://devfeed.tech/tags/project-panama.md>)

### AI overview

JDK 28 pools small allocations in confined FFM arenas, cutting latency for eligible short-lived native allocations without source changes. Each platform thread lazily caches up to four 64-byte pools; arenas acquire one only on a suitable allocation, own it exclusively, and return it on close after clearing used memory. Large or over-aligned allocations use the regular path, and other arena types do not benefit. Virtual threads borrow pools through their carrier thread, adding some handoff overhead while allowing migration. Lifetime and zero-initialization guarantees remain, though closing an arena may return memory to the JDK cache instead of calling the native allocator, so malloc/free calls need not match arena allocations. M4 microbenchmarks show substantial gains for 5- and 20-byte allocations, with near parity above the pool size; whole-application gains depend on how much time is spent allocating and cleaning up around native calls.

### Source excerpt

Pooled confined arenas provide a significant performance improvement for short-lived native allocations.

## Neutralizing PHP2Shell: Stopping Critical PHP RCEs with VMware vDefend and Avi WAF

DevFeed: [Neutralizing PHP2Shell: Stopping Critical PHP RCEs with VMware vDefend and Avi WAF](<https://devfeed.tech/articles/neutralizing-php2shell-stopping-critical-php-rces-with-vmware-vdefend-and-avi-waf-95991.md>)

Original publisher: [Read original article](<https://blogs.vmware.com/load-balancing/2026/10/22/stop-php-rces-with-vdefend-avi-waf/>)

Author: abhinav modi

Published: 2026-10-22T14:51:40Z

Content type: news

Language: en

Sources: [VMware](<https://devfeed.tech/sources/vmware-blogs.md>)

Topics: [PHP](<https://devfeed.tech/topics/php.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [cve remediation](<https://devfeed.tech/topics/cve-remediation.md>)

Tags: [avi](<https://devfeed.tech/tags/avi.md>), [avi-load-balancer](<https://devfeed.tech/tags/avi-load-balancer.md>), [cloud-infrastructure](<https://devfeed.tech/tags/cloud-infrastructure.md>), [critical](<https://devfeed.tech/tags/critical.md>), [load-balancing](<https://devfeed.tech/tags/load-balancing.md>), [php](<https://devfeed.tech/tags/php.md>), [php2shell](<https://devfeed.tech/tags/php2shell.md>), [vmware](<https://devfeed.tech/tags/vmware.md>), [vmware-vdefend](<https://devfeed.tech/tags/vmware-vdefend.md>), [waf](<https://devfeed.tech/tags/waf.md>)

### AI overview

The article describes two unauthenticated PHP RCE vulnerabilities, CVE-2026-109618 (out-of-memory corruption) and CVE-2026-109619 (stale global state), and proposes network controls to protect workloads while patches are prepared. vDefend Gateway Firewall decrypts HTTPS at Tier-0/Tier-1 boundaries and uses IDPS signatures to block exploit attempts and INFO-mode detectors for visibility; existing NSX overlay customers can enable inspection without added appliances or network redesign. Avi WAF identifies PHP versions from traffic headers, normalizes and inspects requests, applies CRS controls, and recommends CVE-specific virtual patches in Enforcement mode and IP Reputation. vDefend DFW micro-segments workloads, while ATP analyzes suspicious payloads and blocks command-and-control or reverse-shell traffic. These measures are presented as virtual patching and containment, not software fixes.

### Source excerpt

The global threat landscape shifted dramatically today following the disclosure of PHP2Shell, a pair of critical Remote Code Execution (RCE) vulnerabilities tracked as CVE-2026-109618 (OOM memory corruption) and CVE-2026-109619 (stale global state). Both carry a CVSS score of 9.2 and affect core memory handling across decades of PHP releases. Because PHP powers a vast portion ... Continued The post Neutralizing PHP2Shell: Stopping Critical PHP RCEs with VMware vDefend and Avi WAF appeared first on VMware Blogs.

## UlaCache: Unified Logical Addressing for Sparse LLM Serving

DevFeed: [UlaCache: Unified Logical Addressing for Sparse LLM Serving](<https://devfeed.tech/articles/ulacache-unified-logical-addressing-for-sparse-llm-serving-95169.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2610.29820>)

Author: Isaac Zhu, Alan Cao, Hugh Yin, Rocky Wang, Orion Wang, Hscos Zhang, Jack Li, Jason Zhao

Published: 2026-10-22T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [memory pool](<https://devfeed.tech/topics/memory-pool.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [cache](<https://devfeed.tech/tags/cache.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-serving](<https://devfeed.tech/tags/llm-serving.md>), [speculative-decoding](<https://devfeed.tech/tags/speculative-decoding.md>)

### AI overview

UlaCache manages sparse-attention KV caches through stable logical addresses that decouple allocation and selection from physical placement across GPU pools and optional CPU memory. Its shared interfaces support data-parallel and tensor-parallel attention; local GPU buffers retain selected KV across decoding steps, limiting transfers to missing entries. With CPU storage enabled, Parallel Bucketed LRU manages a GPU pool shared across requests, choosing eviction victims and updating recency metadata in parallel before KV transfers and address remapping. Under speculative decoding and matched CPU/GPU memory budgets, the abstract reports peak throughput speedups over the state-of-the-art baseline of 1.16x (DP) and 1.70x (TP) on H200, and 1.73x (DP) and 2.51x (TP) on B300.

### Source excerpt

arXiv:2610.29820v1 Announce Type: new Abstract: Long contexts increase the memory cost of the key-value (KV) cache in large language model (LLM) inference. Sparse attention reduces attention computation and memory traffic by selecting historical tokens, but the retained KV cache must remain available for later decoding steps. We present UlaCache, a KV cache management system built around a unified logical address space spanning GPU KV pools and optional CPU memory. Stable logical references let the inference engine allocate and select KV independently of physical placement. Common allocation and sparse-access interfaces support data-parallel (DP) and tensor-parallel (TP) attention. For distributed storage, local GPU buffers retain selected KV across decoding steps, so only missing entries require transfers. When CPU storage is enabled, Parallel Bucketed LRU (PB-LRU) manages a GPU pool shared across requests. It updates recency metadata and selects eviction victims in parallel. UlaCache transfers KV and then updates the address mappings. With speculative decoding enabled under matched CPU and GPU memory budgets, UlaCache achieves peak decoding throughput speedups of 1.16 $\times$ (DP) and 1.70 $\times$ (TP) on H200, and 1.73 $\times$ (DP) and 2.51 $\times$ (TP) on B300, compared with the state-of-the-art baseline.

## PHP patches released for two critical RCEs

DevFeed: [PHP patches released for two critical RCEs](<https://devfeed.tech/articles/php-patches-released-for-two-critical-rces-92810.md>)

Original publisher: [Read original article](<https://lwn.net/Articles/1102014/>)

Author: jzb

Published: 2026-10-21T13:53:18Z

Content type: release

Language: en

Sources: [LWN.net](<https://devfeed.tech/sources/lwn-net.md>)

Topics: [PHP](<https://devfeed.tech/topics/php.md>), [Vulnerabilities](<https://devfeed.tech/topics/vulnerabilities.md>), [Memory Safety](<https://devfeed.tech/topics/memory-safety.md>), [memory handling](<https://devfeed.tech/topics/memory-handling.md>), [patches](<https://devfeed.tech/topics/patches.md>)

Tags: [critical](<https://devfeed.tech/tags/critical.md>), [cve-2026-109618](<https://devfeed.tech/tags/cve-2026-109618.md>), [cve-2026-109619](<https://devfeed.tech/tags/cve-2026-109619.md>), [patches](<https://devfeed.tech/tags/patches.md>), [php](<https://devfeed.tech/tags/php.md>), [rce](<https://devfeed.tech/tags/rce.md>), [remote-code-execution](<https://devfeed.tech/tags/remote-code-execution.md>), [vulnerabilities](<https://devfeed.tech/tags/vulnerabilities.md>)

### AI overview

PHP released versions 8.5.12, 8.4.27, 8.3.36, and 8.2.35 to fix two critical vulnerabilities that could enable unauthenticated remote code execution. One involves cleanup treating partially initialized request data as valid after out-of-memory handling; the other involves a later request in the same worker using request-owned state after its memory has been freed. The source identifies all PHP installations serving web requests as at risk, including PHP-FPM, mod_php, php-cgi, FrankenPHP, and LiteSpeed, and says risk is highest for mod_php, where an exploit can run before application code. PHP versions before 8.1 no longer receive project updates; patches for PHP 7.4, 8.0, and 8.1 were provided to aid backports. Linux distributions and vendors received patches in advance, with updated packages expected soon.

### Source excerpt

The PHP project has published PHP versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35 to address two critical vulnerabilities that could allow unauthenticated remote code execution. Who is at risk: all PHP installations serving web requests, including PHP-FPM, Apache with mod_php, php-cgi, FrankenPHP and LiteSpeed. We consider the risk highest for mod_php, with a demonstrated exploit where an unauthenticated request can trigger the attack before any application code runs. The vulnerabilities: RCE via OOM corruption, CVE-2026-109618 (GHSA-w8hw-w9fx-j79w): out-of-memory handling can interrupt PHP while internal request data is only partially initialized. Request cleanup may then treat the invalid data as valid. RCE via stale global state, CVE-2026-109619 (GHSA-6wmf-4rv5-p3rp): request-owned internal state can survive after its backing memory has been freed. A later request handled by the same worker may use that stale state before normal initialization resets it. PHP versions older than 8.1 will not receive updates from the project, though the project has provided some patches for PHP 7.4, PHP 8.0, and PHP 8.1 to assist in creating backports. According to the announcement, Linux distributions and vendors received patches in advance, so updated packages should be coming soon.

[Next page](<https://devfeed.tech/topics/memory-handling.md?cursor=WyIyMDI2LTEwLTIxVDEzOjUzOjE4KzAwOjAwIiwgImRhMDQ2ODVhLWE3YmQtNGNkNy1iNDk3LTk5YTZhZGQ2NjQ1MSJd>)