* [BENCHMARK] lmbench results for 5.53 mm1
@ 2002-12-26 6:31 Aniruddha M Marathe
2002-12-26 7:48 ` Andrew Morton
0 siblings, 1 reply; 3+ messages in thread
From: Aniruddha M Marathe @ 2002-12-26 6:31 UTC (permalink / raw)
To: linux-kernel; +Cc: Larry McVoy
Hi,
Benchmark didn't produce any result for local communication latency for UDP. Larry, any clue?
Here is a comparison of results of 2.5.53-mm1 and 2.5.53. The figures in the table below indicate median of 5 repetitions of tests.
2.5.53-mm1 2.5.53
==============================================================================
Processor, Processes - times in microseconds - smaller is better
1. increase in time for fork proc 415 382
2. increase in time for sh proc 8190 8088
------------------------------------------------------------------------------
Context switching - times in microseconds - smaller is better
1. 2p/64K ctxsw 6.49 7.44
2. 16p/16K ctxsw 38 44
------------------------------------------------------------------------------
*Local* Communication latencies in microseconds - smaller is better
1. RPC / UDP 61 58
------------------------------------------------------------------------------
File & VM system latencies in microseconds - smaller is better
1. mmap latency 630 610
==============================================================================
*****************************************************************************
Lmbench result
kernel 2.5.53 with mm1
*****************************************************************************
L M B E N C H 2 . 0 S U M M A R Y
------------------------------------
(Alpha software, do not distribute)
Basic system parameters
----------------------------------------------------
Host OS Description Mhz
--------- ------------- ----------------------- ----
benchtest Linux 2.5.53 i686-pc-linux-gnu 790
benchtest Linux 2.5.53 i686-pc-linux-gnu 790
benchtest Linux 2.5.53 i686-pc-linux-gnu 790
benchtest Linux 2.5.53 i686-pc-linux-gnu 790
benchtest Linux 2.5.53 i686-pc-linux-gnu 790
Processor, Processes - times in microseconds - smaller is better
----------------------------------------------------------------
Host OS Mhz null null open selct sig sig fork exec sh
call I/O stat clos TCP inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ----
benchtest Linux 2.5.53 790 0.44 0.80 28 30 32 1.28 5.47 446 1647 8100
benchtest Linux 2.5.53 790 0.44 0.82 28 30 35 1.29 5.24 373 1616 8190
benchtest Linux 2.5.53 790 0.44 0.80 28 30 32 1.29 5.47 387 1604 8190
benchtest Linux 2.5.53 790 0.44 0.82 28 30 37 1.31 5.47 430 1629 8130
benchtest Linux 2.5.53 790 0.44 0.82 28 30 35 1.29 5.46 415 1619 8231
Context switching - times in microseconds - smaller is better
-------------------------------------------------------------
Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
--------- ------------- ----- ------ ------ ------ ------ ------- -------
benchtest Linux 2.5.53 1.190 4.9100 14 6.7200 174 39 177
benchtest Linux 2.5.53 1.470 5.1100 14 7.6800 178 42 180
benchtest Linux 2.5.53 1.380 5.0600 15 6.4900 178 37 179
benchtest Linux 2.5.53 1.420 5.1000 15 6.2400 179 38 180
benchtest Linux 2.5.53 1.490 5.1000 14 6.1200 186 38 181
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
benchtest Linux 2.5.53 1.190 8.170 21 61 127 156 178
benchtest Linux 2.5.53 1.470 8.122 21 61 125 135 185
benchtest Linux 2.5.53 1.380 8.284 19 62 127 155 183
benchtest Linux 2.5.53 1.420 8.273 21 61 125 155 182
benchtest Linux 2.5.53 1.490 8.319 21 61 125 158 175
File & VM system latencies in microseconds - smaller is better
--------------------------------------------------------------
Host OS 0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
--------- ------------- ------ ------ ------ ------ ------- ----- -----
benchtest Linux 2.5.53 119 59 383 128 653 1.033 4.00000
benchtest Linux 2.5.53 122 59 384 125 631 1.014 4.00000
benchtest Linux 2.5.53 122 60 388 131 634 0.868 4.00000
benchtest Linux 2.5.53 121 59 387 128 630 0.946 4.00000
benchtest Linux 2.5.53 122 59 385 126 632 0.938 4.00000
*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
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----
benchtest Linux 2.5.53 296 134 23 298 355 124 114 356 171
benchtest Linux 2.5.53 393 134 22 290 353 123 113 354 170
benchtest Linux 2.5.53 412 136 23 294 352 123 112 354 169
benchtest Linux 2.5.53 293 134 22 285 352 123 112 353 169
benchtest Linux 2.5.53 406 137 23 293 352 123 112 353 169
Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- ---- ----- ------ -------- -------
benchtest Linux 2.5.53 790 3.807 8.8740 174
benchtest Linux 2.5.53 790 3.798 8.8720 175
benchtest Linux 2.5.53 790 3.800 8.8840 176
benchtest Linux 2.5.53 790 3.808 8.8830 176
benchtest Linux 2.5.53 790 3.798 8.8830 176
Rest of the results are not much different.
Regards,
Aniruddha Marathe
WIPRO Technologies, India
aniruddha.marathe@wipro.com
+91-80-5502001 to 2008 extn 5092
^ permalink raw reply [flat|nested] 3+ messages in thread* RE: [BENCHMARK] lmbench results for 5.53 mm1
@ 2002-12-26 8:17 Aniruddha M Marathe
0 siblings, 0 replies; 3+ messages in thread
From: Aniruddha M Marathe @ 2002-12-26 8:17 UTC (permalink / raw)
To: Andrew Morton; +Cc: linux-kernel, Larry McVoy
It was not enabled.
> From: Andrew Morton [mailto:akpm@digeo.com]
> Subject: Re: [BENCHMARK] lmbench results for 5.53 mm1
>
>
> Aniruddha M Marathe wrote:
> >
> > 1. increase in time for fork proc 415 382
> > 2. increase in time for sh proc 8190 8088
>
> pagetable sharing adds a little overhead to fork and exec in
> microbenchmarks. You should state whether it was enabled.
>
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2002-12-26 8:09 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-12-26 6:31 [BENCHMARK] lmbench results for 5.53 mm1 Aniruddha M Marathe
2002-12-26 7:48 ` Andrew Morton
-- strict thread matches above, loose matches on Subject: below --
2002-12-26 8:17 Aniruddha M Marathe
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox