Re: Rewritten TCP reassembly

From: Andrew Gallatin (gallatin_at_cs.duke.edu)
Date: 12/13/04

  • Next message: Andre Oppermann: "Re: per-interface packet filters" 
    Date: Mon, 13 Dec 2004 16:04:44 -0500 (EST)
To: Andre Oppermann <andre@freebsd.org>

    Andre Oppermann writes:
    >
    > I've got some excellent review feedback from Mike Spengler and he found
    > a off-by-one queue limit tracking error.
    >
    > http://www.nrg4u.com/freebsd/tcp_reass-20041213.patch
    >

    Here are the same tests running your new patch in comparison to a
    stock 6.x kernel from Friday. It looks like everything is faster.
    I've never seen 3.7Gb/sec on FreeBSD before. Nice work!

    The "before" and "tcp_reass-20041213" kernels differ only in the
    contents of the above patch. I ran netperf with 20 times for each of
    4 socket buffer sizes (64KB, 128KB, 256KB, and 1MB) before and after
    patching. All tests were run with net.isr.enable=1, and
    machdep.cpu_idle_hlt=0. CPU was pretty much maxed thoughout. (SMP
    kernel, 1 HTT p4). All the numbers in Mb/sec. The sender was running
    linux-2.6.6 (also SMP on an identical HTT P4).

    The out-of-order accounting bug seems to be gone.

    Drew

    x before.65536
    + tcp_reass-20041213.65536
    +--------------------------------------------------------------------------+
    | + |
    | + |
    | xx + + + + +*x + ++ ++ + ++ + +++|
    ||__________________A________|______M__|_____MA________________| |
    +--------------------------------------------------------------------------+
        N Min Max Median Avg Stddev
    x 4 3121.03 3135.88 3135.27 3128.3325 8.3667691
    + 20 3126.75 3151.97 3139.41 3139.608 7.5192088
    Difference at 95.0% confidence
            11.2755 +/- 8.67923
            0.360432% +/- 0.277439%
            (Student's t, pooled s = 7.64032)
    x before.131072
    + tcp_reass-20041213.131072
    +--------------------------------------------------------------------------+
    | x x ++ xx + + + |
    |x x xxx xx*x x x+* + xxx++ + + + ++ + + + +|
    | |_________AM_______||_____________A_M_____________| |
    +--------------------------------------------------------------------------+
        N Min Max Median Avg Stddev
    x 20 3336.63 3516.95 3449.87 3447.838 52.708815
    + 20 3443.67 3717.15 3579.92 3571.8355 80.445555
    Difference at 95.0% confidence
            123.997 +/- 43.5271
            3.59638% +/- 1.26245%
            (Student's t, pooled s = 68.0063)
    x before.262144
    + tcp_reass-20041213.262144
    +--------------------------------------------------------------------------+
    | x + |
    | x x ++ x + ++ |
    |+ x x x x xx+ *x x * ++ +xx xx+ x++ + * + +|
    | |_______|____A_________A__|M__________| |
    +--------------------------------------------------------------------------+
        N Min Max Median Avg Stddev
    x 20 3197.77 3421.67 3316.64 3314.8555 59.787371
    + 20 3135.57 3479.29 3380.31 3361.5785 71.303501
    Difference at 95.0% confidence
            46.723 +/- 42.1136
            1.4095% +/- 1.27045%
            (Student's t, pooled s = 65.7979)
    x before.1048576
    + tcp_reass-20041213.1048576
    +--------------------------------------------------------------------------+
    | + + |
    | x + +x x+ x+ |
    |xx x x *x * x x * +x +* x+x+x++ x+ ++ +|
    | |_______________A|__M________AM|__________| |
    +--------------------------------------------------------------------------+
        N Min Max Median Avg Stddev
    x 20 2658.05 2732.45 2706.91 2700.1955 23.318833
    + 20 2686.35 2765.88 2722.09 2720.7805 18.752084
    Difference at 95.0% confidence
            20.585 +/- 13.5427
            0.762352% +/- 0.501546%
            (Student's t, pooled s = 21.159)
    _______________________________________________
    freebsd-net@freebsd.org mailing list
    http://lists.freebsd.org/mailman/listinfo/freebsd-net
    To unsubscribe, send any mail to "freebsd-net-unsubscribe@freebsd.org"

  • Next message: Andre Oppermann: "Re: per-interface packet filters"

    Relevant Pages

    • [PATCH] SMP alternatives
      ... Can you include the patch in -mm to give it some testing? ... This patch implements SMP alternatives, ... +/* Replace instructions with better alternatives for this CPU type. ... +void alternatives_smp_module_del ...
      (Linux-Kernel)
    • Re: umount() and NFS races in 2.4.26
      ... > after deploying an SMP machine at work, ... As James pointed out my latest autofs4 patch resolved the issue for him. ... Note that the patch for the kernel module will not apply cleanly to ... > the dentry needed for the async unlink (which cannot be removed ...
      (Linux-Kernel)
    • Re: [PATCH] i386 kexec: Allow the kexec on panic support to compile on voyager.
      ... apics. ... I simply mean a crash kernel that is able to boot. ... My patch for that got reverted because it wouldn't boot on Linus's ... SMP laptop. ...
      (Linux-Kernel)
    • Re: [VOYAGER] fix build broken by shift to smp_ops
      ... Subject: i386: move common parts of smp into their own file ... This patch conflicts a lot with James's one (which I named ... Voyager and is preferred then cool, but a) the changelog should tell us ... Part of the fix was the prerequisite patch "i386: ...
      (Linux-Kernel)
    • Re: [VOYAGER] fix build broken by shift to smp_ops
      ... Subject: i386: move common parts of smp into their own file ... This patch conflicts a lot with James's one (which I named ... Part of the fix was the prerequisite patch "i386: ...
      (Linux-Kernel)