2.4.20rc2aa1

From: Andrea Arcangeli (andrea@suse.de)
Date: Thu Nov 21 2002 - 06:03:48 EST


Please read the comments on the 9981_elevator-lowlatency-1 patch, if you
don't want to risk bandwidth regressions in some seeking workload, make
sure to backout the 9981_elevator-lowlatency-1 patch before compiling.

URL:

        http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20rc2aa1.gz
        http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20rc2aa1/

Changelog between 2.4.20rc1aa1 and 2.4.20rc2aa1:

Only in 2.4.20rc2aa1: 00_backout-irda-trivial-1

        My gprs phone stops working with this applied (and the palm as well
        works only once in a while with it applied).

Only in 2.4.20rc1aa1: 00_extraversion-12
Only in 2.4.20rc2aa1: 00_extraversion-13
Only in 2.4.20rc1aa1: 00_read_full_page-get_block-err-1
Only in 2.4.20rc2aa1: 00_read_full_page-get_block-err-2

        Rediffed.

Only in 2.4.20rc2aa1: 00_highio-nohighmem-fix-1

        Potential performance with highio disabled or on non highio capable
        devices. (from Jens, but found some week ago by me and also more
        recently by Steve Lord on l-k)

Only in 2.4.20rc2aa1: 00_i810-unlock_page-fastcall-1

        Make sure it sees the fastcall calling conventions.

Only in 2.4.20rc2aa1: 00_intermezzo-tcgets-1
Only in 2.4.20rc2aa1: 00_presto-dentry-slab-1

        Fix compile problems.

Only in 2.4.20rc2aa1: 00_msgrcv-smp-race-1
Only in 2.4.20rc2aa1: 00_poll-smp-races-1

        Fix a few smp races.

Only in 2.4.20rc1aa1: 00_readahead-got-broken-somewhere-1
Only in 2.4.20rc2aa1: 00_readahead-got-broken-somewhere-2

        Increase the readahead a bit more, to make read against read
        faster decreasing the seeks.

Only in 2.4.20rc2aa1: 00_umount-against-unused-dirty-inodes-race-1

        The lowlatency bit in unused-dirty-inodes was racy,
        noticed by Terence Rokop at Polyserve. This fix
        allows to retain the lowlatency optimization. Other lowlatency
        patches around may need the same fix.

Only in 2.4.20rc2aa1: 55_uml-highmem-1

        Fix compile problem.

Only in 2.4.20rc1aa1: 95_fsync-corruption-fix-2

        This introduced a cache coherency problem with O_DIRECT so backed out.

Only in 2.4.20rc2aa1: 9980_fix-pausing-2

        Fix various longstanding unplugging races that could lead to long hangs
        with the disk idle and all tasks in D state. This fixes the
        get_request_wait against get_request_wait case too.

Only in 2.4.20rc2aa1: 9981_elevator-lowlatency-1

        *Very* experimental patch, that limits the queue size to a small value,
        this should provide more interactive behaviour better for a desktop
        but it will make things like dbench behave very very badly, the numbers
        will be quite horrible (i.e. low performance). OTOH I suspect the 2.5 deadline
        I/O scheduler would do very bad on dbench too if applied on top of a 2.4.
        2.5 can reorder stuff at the pagecache layer, during the writeback stage,
        2.4 only relies on the elevator instead. Anyways the patch is safe so I want
        to give it a spin knowing in advance that dbench will decrease horribly.
        If you test it let me know what you think about it. If you go read the code
        you will notice read-latency is now pointless with this patch applied
        (with a `cp /dev/zero .` in background the max number of requests in
        the queue could be even lower than the seventh-request where read-latency
        would attempt to put the read request). This should make lowmem machine beahave
        much better too, since it's less likely to block for vm. Running the contest
        benchmark on this one would be interesting too.

Andrea
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sat Nov 23 2002 - 22:00:36 EST