vm.dirty_bytes should be set to something sane

Bug #908279 reported by Daniel Levine on 2011-12-23
8
This bug affects 1 person
Affects Status Importance Assigned to Milestone
linux-lts-backport-oneiric (Ubuntu)
Undecided
Unassigned

Bug Description

Linux has developed a reputation for sluggish performance under heavy disk I/O, e.g. during system updates. This seems to be mostly due to its inability to distinguish urgent tasks from I/O hogs, resulting in preferential treatment of hogs (and thus sluggishness under load). This applies even on very high-end computers.

On a server, this (probably) doesn't matter a great deal. On a desktop, it matters a lot, because users will generally not tolerate sluggishness, especially on computers that are supposed to be fast (as a famous company from Redmond discovered when they released a shiny new OS in 2007). Since Ubuntu is primarily a desktop distribution, and currently the most popular, I think it would be a good idea to try and solve this issue.

The only solution that I know to work reliably is to set vm.dirty_bytes (and vm.dirty_background_bytes) to something relatively low. This forces programs to make smaller and more frequent writes to the disk, preventing processes from hogging huge amounts of I/O bandwidth. In my experience, 4 MB (4194304 bytes) is a sane value; it increases responsiveness under load (and even during desktop startup) greatly, while only reducing throughput by a few MB/s on modern computers. Since I care more about latency than throughput, I consider this a decent tradeoff.

The only major issue I can see with this change is that it might increase wear on SSDs; but I'm not sure how much, or if it's worth making an exception for SSDs.

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers