In message <38E4D8E7.9A73CF84@alumni.caltech.edu>, Dan Kegel writes:
>Takashi Richard Horikawa (horikawa@ccm.cl.nec.co.jp) wrote:
>> Since Linux 2.3.49 uses TCP/IP port number up to 65535,
>> the time length from using a socket to (re)using a socket
>> that has the same port number to that is longer than that
>> of Linux 2.2.14, which uses TCP/IP port number up to 32767.
>
>This is a benchmarking issue that happens only when you try
>to simulate too many clients with a single client machine, right?
>Did you try to tell 2.2.14 to use a larger port range, e.g.
> # echo 1024 65535 > /proc/sys/net/ipv4/ip_local_port_range
> # cat /proc/sys/net/ipv4/ip_local_port_range
> 1024 65535
>on your unpatched runs?
I have measured the SpecWeb96 performance in the way that Dan Kegel suggested.
The results was:
Linux of Linux of (Benchmark results)
Server Client Throughput Response
Machine Machine [HTTPops/s] [ms]
----------------------------------------------------
2.2.14 - 2.2.14 312 25.5
2.2.14 - 2.3.49 567 14.1
2.2.14 Patched - 2.2.14 625 12.8
2.2.14 L.P.R - 2.2.14 L.P.R 639 12.5 (New)
2.2.14 L.P.R - 2.3.49 L.P.R 623 12.8 (New)
(L.P.R means 'Larger Port Range')
As he prospected, using a larger port range brought the increase in
the SpecWeb96 performance to the same level as that with using
a patched 2.2.14.
Conclusion derived from this is that we must increase port range of
Linux by 'echo 1024 65535 > /proc/sys/net/ipv4/ip_local_port_range'
when we conduct such performance measurement as SpecWeb96 benchmark in
which a single client machine simulates too many clients.
It is true for 2.2.14 and 2.3.49.
--------
name: Takashi Richard Horikawa
E-mail: horikawa@ccm.cl.nec.co.jp
Tel.: +81-44-856-2079
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Apr 07 2000 - 21:00:09 EST