[parisc-linux] LMBench results on for the C200 for HP-UX 10.20,
Linux 2.4.24-pa0, and Linux 2.6.0-pa8
Riccardo
rollei at tiscalinet.it
Wed Jan 7 02:43:59 MST 2004
Ryan Bradetich wrote:
>
> Hello all,
>
> Thought people might be intersted in the LMBench results between HP-UX
> 10.20, Linux 2.4 and Linux 2.6. All tests were performed on the same
> hardware and not optomized in any way for the tests.
Yes, very interesting. I'm using gentoo with a 2.4 kernel and felt the
box sooo slow sometimes while other times it seems reasonable.
For example doing setiathome crunching linux is 40% slower. Or
compilation is awfully slow (big projects like gcc 3.2.2 can take 2-3
times the time on hp-ux 10.20).
It seems that in some "traditional" unix areas HP-UX is faster while in
other it is slower.
> ----------------------------------------------------------------
> Host OS Mhz null null open selct sig sig fork
> exec sh
> call I/O stat clos TCP inst hndl proc
> proc proc
> --------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ----
> ---- ----
> vega HP-UX B.10.20 200 0.83 26.1 36.0 200.6 7.97 30.2 1905
> 5220 19.K
> vega HP-UX B.10.20 200 0.83 24.5 36.0 147.4 7.92 30.0 1910
> 5133 19.K
> vega HP-UX B.10.20 200 0.83 24.3 36.3 145.0 7.90 29.9 1924
> 5258 19.K
> vega Linux 2.6.0-p 200 1.46 2.45 14.4 19.3 150.3 5.65 28.9 5353
> 21.K 66.K
> vega Linux 2.6.0-p 200 1.44 2.43 14.4 18.1 5.63 28.9 5344
> 21.K 65.K
> vega Linux 2.6.0-p 200 1.46 2.45 14.4 18.1 149.8 5.65 28.8 5492
> 23.K 65.K
> vega Linux 2.4.24- 200 1.37 2.80 14.7 18.9 148.2 4.96 29.2 10.K
> 42.K 111K
> vega Linux 2.4.24- 200 1.37 2.80 14.7 18.9 148.0 4.97 29.2 11.K
> 42.K 111K
> vega Linux 2.4.24- 200 1.37 2.80 14.7 18.9 148.2 4.97 29.0 11.K
> 42.K 111K
HP-UX fork is fast! linux 2.6 seems to bring some relief here. I always
marvel how fast HP-UX is when responding to a telnet login.
Instantaneus. Probably fork() speed compensates the lower latency linux
has in other areas.
>
> *Local* Communication latencies in microseconds - smaller is better
> -------------------------------------------------------------------
> Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
> ctxsw UNIX UDP TCP conn
> --------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
> vega HP-UX B.10.20 9.110 48.7 78.7 117.6 500.2 127.5 300.2 313.
> vega HP-UX B.10.20 9.510 47.0 65.6 115.0 464.2 166.3 734.1 284.
> vega HP-UX B.10.20 9.230 47.9 77.5 112.4 506.4 127.1 325.6 300.
> vega Linux 2.6.0-p 4.830 36.9 53.3 80.7 235.8 109.6 251.7 490.
> vega Linux 2.6.0-p 4.530 31.7 53.5 140.2 230.2 108.0 286.3 469.
> vega Linux 2.6.0-p 5.670 36.6 54.0 143.1 232.5 107.9 283.5 483.
> vega Linux 2.4.24- 2.120 57.1 46.4 73.5 226.9 93.5 300.2 9958
> vega Linux 2.4.24- 3.490 60.7 46.3 73.2 238.7 97.0 299.6 9957
> vega Linux 2.4.24- 3.410 52.8 46.4 73.4 234.9 92.4 304.5 9958
seems that for AF_UNIX and TCP pipes 2.42. UDP is almost the half than
2.6
> *Local* Communication bandwidths in MB/s - bigger is better
> -----------------------------------------------------------
> Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem
> Mem
> UNIX reread reread (libc) (hand) read
> write
> --------- ------------- ---- ---- ---- ------ ------ ------ ------ ----
> -----
> vega HP-UX B.10.20 168. 102. 88.1 103.0 137.6 79.1 63.7 122.
> 102.3
> vega HP-UX B.10.20 179. 99.5 97.7 101.4 136.3 85.0 64.1 121.
> 104.7
> vega HP-UX B.10.20 180. 102. 108. 103.2 136.2 82.2 63.7 122.
> 102.8
> vega Linux 2.6.0-p 19.4 19.9 5.65 31.4 246.4 126.0 129.4 249.
> 215.2
> vega Linux 2.6.0-p 19.5 20.2 5.67 31.4 245.8 125.9 129.4 249.
> 214.8
> vega Linux 2.6.0-p 19.4 20.2 5.86 31.4 245.6 125.8 129.4 249.
> 214.6
> vega Linux 2.4.24- 19.6 19.8 9.02 32.0 246.7 125.8 129.5 249.
> 215.3
> vega Linux 2.4.24- 19.6 19.7 9.50 31.7 246.4 126.0 129.4 249.
> 215.1
> vega Linux 2.4.24- 19.6 19.8 9.40 35.5 246.2 125.9 129.4 249.
> 214.9
traditional pipe, af_unix and TCP are impressive for hp-ux. Really up to
ten times the bandwidth?. But memory is clear an advantage for linux.
Thanks for the testing.
-R
More information about the parisc-linux
mailing list