# The benefits of having data

DevFeed: [The benefits of having data](<https://devfeed.tech/articles/the-benefits-of-having-data-12438.md>)

Original publisher: [Read original article](<http://brooker.co.za/blog/2012/01/10/drive-failure.html>)

Author: Marc Brooker

Published: 2012-01-10T00:00:00Z

Content type: opinion

Language: en

Sources: [Marc Brooker's Blog](<https://devfeed.tech/sources/marc-brooker-s-blog.md>), [Marc Brooker's Blog](<https://devfeed.tech/sources/marc-brooker-s-blog-2.md>)

Topics: [data](<https://devfeed.tech/topics/data.md>), [DRIVE](<https://devfeed.tech/topics/drive.md>), [Google](<https://devfeed.tech/topics/google.md>), [systems](<https://devfeed.tech/topics/systems.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [articles](<https://devfeed.tech/tags/articles.md>), [data](<https://devfeed.tech/tags/data.md>), [drive](<https://devfeed.tech/tags/drive.md>), [engineering](<https://devfeed.tech/tags/engineering.md>), [google](<https://devfeed.tech/tags/google.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [models](<https://devfeed.tech/tags/models.md>), [testing](<https://devfeed.tech/tags/testing.md>)

## AI overview

The article compares Google and Seagate studies of hard-drive failure rates and temperature. It argues that Google's analysis of failure data from more than 100,000 drives provides a better basis for conclusions than Seagate's accelerated-aging tests and statistical models, whose assumptions are not sufficiently justified.

## Source excerpt

The benefits of having data Two ways to look at drive failures and temperature. Almost all recent articles and papers I have read on hard drive failure rates refer to either Failure Trends in a Large Disk Drive Population from Google, or Estimating Drive Reliability in Desktop Computers and Consumer Electronics Systems from Seagate. Despite both sounding and looking authoritative, these papers come to some wildly different conclusions, and couldn't be more different in their approach. How does temperature affect drive failure rate? The Seagate paper says an increase from 25C to 30C increases it by 27%. The Google paper suggests a decrease of around 10%. How can the two most widely used studies differ by so much? It's really because these papers use completely different approaches: the Google study uses simple analysis, while the Seagate paper uses powerful and sophisticated models, accelerated aging, and complex statistical tools. Despite sounding less authoritative, the Google paper is much better. The Seagate paper doesn't actually present the results of testing drives at different temperatures. Instead, all the drives were tested using a standard accelerated aging approach, in an oven heated to 42C. Another standard accelerated aging technique, the Arrhenius Model, was used to estimate the effect of temperature on failure rates. The Seagate paper goes on to use Weibull modeling, and a fairly sophisticated Bayesian approach to estimating the Weibull parameters. The underlying, and unmentioned, assumption is that the failure rate of drives is proportional to the reaction rate constant, or the speed that an unlimited chemical reaction would proceed at a given temperature. No attempt is made to justify this choice, other than appealing to standard textbooks describing the approach. The Google paper, on the other hand, doesn't use any statistical concepts that would be unfamiliar to an undergraduate engineering student. Instead, they use the failure data from over a h