Messages in this thread | | | Date | Thu, 10 Jan 2013 04:05:46 +0000 | From | Eric Wong <> | Subject | Re: [PATCH v2] fadvise: perform WILLNEED readahead asynchronously |
| |
Simon Jeons <simon.jeons@gmail.com> wrote: > On Tue, 2012-12-25 at 02:22 +0000, Eric Wong wrote: > > Please add changelog.
Changes since v1:
* separate unbound workqueue for high-priority tasks
* account for inflight readahead to avoid denial-of-service
* limit concurrency for non-high-priority tasks (1 per CPU, same as aio)
* take IO priority of requesting process into account when in workqueue.
* process queued readahead in 2M chunks to help ensure fairness between multiple requests with few CPUs/workqueues. Idle tasks get smaller 256K chunks.
* stops readahead for idle tasks on read congestion
Will try to benchmark with Postgres when I get the chance.
Any other (Free Software) applications that might benefit from lower FADV_WILLNEED latency?
| |