Package: linux-2.6 Version: 2.6.26-15 Severity: important Tags: patch Phoronix benchmarked kernels from 2.6.24-29: "When it came to the SQLite performance, a serious performance regression began with the Linux 2.6.26 kernel and ended with the Linux 2.6.29 release. Normally it required 27~28 seconds to perform 12,500 database insertions using SQLite, but with the Linux 2.6.26 through 2.6.28 kernel releases it took 109 seconds! Fortunately, this regression is now fixed."
>From 78f707bfc723552e8309b7c38a8d0cc51012e813 Mon Sep 17 00:00:00 2001 From: Jens Axboe <jens.ax...@oracle.com> Date: Tue, 17 Feb 2009 13:59:08 +0100 Subject: [PATCH] block: revert part of 18ce3751ccd488c78d3827e9f6bf54e6322676fb The above commit added WRITE_SYNC and switched various places to using that for committing writes that will be waited upon immediately after submission. However, this causes a performance regression with AS and CFQ for ext3 at least, since sync_dirty_buffer() will submit some writes with WRITE_SYNC while ext3 has sumitted others dependent writes without the sync flag set. This causes excessive anticipation/idling in the IO scheduler because sync and async writes get interleaved, causing a big performance regression for the below test case (which is meant to simulate sqlite like behaviour). ---- test case ---- int main(int argc, char **argv) { int fdes, i; FILE *fp; struct timeval start; struct timeval end; struct timeval res; gettimeofday(&start, NULL); for (i=0; i<ROWS; i++) { fp = fopen("test_file", "a"); fprintf(fp, "Some Text Data\n"); fdes = fileno(fp); fsync(fdes); fclose(fp); } gettimeofday(&end, NULL); timersub(&end, &start, &res); fprintf(stdout, "time to write %d lines is %ld(msec)\n", ROWS, (res.tv_sec*1000000 + res.tv_usec)/1000); return 0; } ------------------- Thanks to sean.wh...@apcc.com for tracking down this performance regression and providing a test case. Signed-off-by: Jens Axboe <jens.ax...@oracle.com> --- fs/buffer.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/fs/buffer.c b/fs/buffer.c index 665d446..62b57e3 100644 --- a/fs/buffer.c +++ b/fs/buffer.c @@ -3108,7 +3108,7 @@ int sync_dirty_buffer(struct buffer_head *bh) if (test_clear_buffer_dirty(bh)) { get_bh(bh); bh->b_end_io = end_buffer_write_sync; - ret = submit_bh(WRITE_SYNC, bh); + ret = submit_bh(WRITE, bh); wait_on_buffer(bh); if (buffer_eopnotsupp(bh)) { clear_buffer_eopnotsupp(bh); -- 1.6.2.1 -- System Information: Debian Release: squeeze/sid APT prefers testing APT policy: (900, 'testing'), (500, 'unstable'), (1, 'experimental') Architecture: amd64 (x86_64) Kernel: Linux 2.6.29-1-amd64 (SMP w/2 CPU cores) Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8) Shell: /bin/sh linked to /bin/bash -- To UNSUBSCRIBE, email to debian-kernel-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org