Re: [v2] fadvise: perform WILLNEED readahead asynchronously

2013-01-10 Thread Eric Wong
Riccardo Magliocchetti wrote: > Hello, > > Il 25/12/2012 03:22, Eric Wong ha scritto: > > Any other (Free Software) applications that might benefit from > > lower FADV_WILLNEED latency? > > Not with fadvise but with madvise. Libreoffice / Openoffice.org have > this comment: > > // On Linux, mad

Re: [v2] fadvise: perform WILLNEED readahead asynchronously

2013-01-10 Thread Riccardo Magliocchetti
Hello, Il 25/12/2012 03:22, Eric Wong ha scritto: > Any other (Free Software) applications that might benefit from > lower FADV_WILLNEED latency? Not with fadvise but with madvise. Libreoffice / Openoffice.org have this comment: // On Linux, madvise(..., MADV_WILLNEED) appears to have the und

Re: [PATCH v2] fadvise: perform WILLNEED readahead asynchronously

2013-01-09 Thread Eric Wong
Simon Jeons wrote: > On Tue, 2012-12-25 at 02:22 +, 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, sam

Re: [PATCH v2] fadvise: perform WILLNEED readahead asynchronously

2013-01-05 Thread Simon Jeons
On Tue, 2012-12-25 at 02:22 +, Eric Wong wrote: Please add changelog. > Using fadvise with POSIX_FADV_WILLNEED can be very slow and cause > user-visible latency. This hurts interactivity and encourages > userspace to resort to background threads for readahead (or avoid > POSIX_FADV_WILLNEED

Re: [PATCH v2] fadvise: perform WILLNEED readahead asynchronously

2012-12-26 Thread Zheng Liu
On Tue, Dec 25, 2012 at 02:22:51AM +, Eric Wong wrote: > Using fadvise with POSIX_FADV_WILLNEED can be very slow and cause > user-visible latency. This hurts interactivity and encourages > userspace to resort to background threads for readahead (or avoid > POSIX_FADV_WILLNEED entirely). > > "

[PATCH v2] fadvise: perform WILLNEED readahead asynchronously

2012-12-24 Thread Eric Wong
Using fadvise with POSIX_FADV_WILLNEED can be very slow and cause user-visible latency. This hurts interactivity and encourages userspace to resort to background threads for readahead (or avoid POSIX_FADV_WILLNEED entirely). "strace -T" timing on an uncached, one gigabyte file: Before: fadvise6