# This is not the performance you were looking for: the tricks systems play on us

DevFeed: [This is not the performance you were looking for: the tricks systems play on us](<https://devfeed.tech/articles/this-is-not-the-performance-you-were-looking-for-the-tricks-systems-play-on-us-20215.md>)

Original publisher: [Read original article](<https://blog.janestreet.com/this-is-not-the-performance-you-were-looking-for-the-tricks-systems-play-on-us/>)

Author: Andy Myers

Published: 2017-04-20T00:00:00Z

Content type: tutorial

Language: en

Sources: [Jane Street](<https://devfeed.tech/sources/jane-street.md>)

Topics: [performance-engineering](<https://devfeed.tech/topics/performance-engineering.md>), [NUMA](<https://devfeed.tech/topics/numa.md>), [Processor Affinity](<https://devfeed.tech/topics/processor-affinity.md>), [Process](<https://devfeed.tech/topics/process.md>), [scheduling](<https://devfeed.tech/topics/scheduling.md>)

Tags: [affinity](<https://devfeed.tech/tags/affinity.md>), [app](<https://devfeed.tech/tags/app.md>), [background](<https://devfeed.tech/tags/background.md>), [cache](<https://devfeed.tech/tags/cache.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [latency](<https://devfeed.tech/tags/latency.md>), [linux](<https://devfeed.tech/tags/linux.md>), [linux-kernel](<https://devfeed.tech/tags/linux-kernel.md>), [network](<https://devfeed.tech/tags/network.md>), [node](<https://devfeed.tech/tags/node.md>), [numa](<https://devfeed.tech/tags/numa.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [performance](<https://devfeed.tech/tags/performance.md>), [policy](<https://devfeed.tech/tags/policy.md>), [process](<https://devfeed.tech/tags/process.md>), [processes](<https://devfeed.tech/tags/processes.md>), [ram](<https://devfeed.tech/tags/ram.md>), [scheduler](<https://devfeed.tech/tags/scheduler.md>), [server](<https://devfeed.tech/tags/server.md>), [software](<https://devfeed.tech/tags/software.md>)

## AI overview

Unexpectedly poor performance can come from deployment conditions that undo application-level optimization. The excerpt recommends verifying that the running binary and its startup-linked dynamic libraries are the expected ones, then checking CPU affinity: pinning a single-threaded app to one core may help cache locality, while a busy multithreaded app may need several cores. Affinity can also place an app near the NUMA node serving its network adapter; small per-operation latency costs can add up. Check hardware topology and account for other processes that may share pinned cores by considering their affinity too. The specific tools mentioned are for Linux, though the author says the checks should apply to modern-server operating systems generally. The excerpt ends mid-discussion, so it does not establish the remaining checks.

## Source excerpt

It's often surprising just how much software performance depends on how the software is deployed. All the time and effort you've invested in optimization can be erased by a few bad decisions in scheduler policy, affinity, or background workload on a server.