>Running kernel 2.2.10-rmk2-stk1, I have noticed that context switch
>times are pretty terrible. Running lmbench to measure this reports the
For interest's sake, here's the latest in my set of irregularly-collected
lmbench results. I ran the benchmark twice on my NetWinder under both 2.2.10
and 2.3.9 and I've left in some of the historical ones as a comparison.
There are some funnies -- I've no idea what's going on with the page-fault
latency measurement under 2.3.9, and it seems strange that select has suddenly
got 200us slower. Anyway...
p.
L M B E N C H 1 . 9 S U M M A R Y
------------------------------------
(Alpha software, do not distribute)
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 inst hndl proc proc proc
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ----
alpha-lin Linux 2.1.132 226 4.1 8.5 45 70 0.39K 18.9 45 4.5K 27K 97K
arm32-net NetBSD 1.3F 224 2.2 11. 62 101 0.27K 5.3 257 15.8K 42K 74K
armv4l-li Linux 2.2.9 233 0.6 1.3 10 12 0.11K 4.5 4 3.3K 13K 58K
armv4l-li Linux 2.2.7 275 0.5 1.1 8 10 0.08K 3.8 3 3.0K 14K 26K
armv4l-li Linux 2.0.31 274 0.6 2.7 16 15 0.08K 5.5 50 3.8K 15K 66K
armv4l-li Linux 2.2.10 275 0.5 1.1 8 10 0.08K 3.7 3 2.8K 12K 51K
armv4l-li Linux 2.2.10 275 0.5 1.1 8 10 0.08K 3.7 3 2.8K 12K 51K
armv4l-li Linux 2.3.9 275 0.5 1.1 8 9 0.10K 3.7 3 2.7K 10K 41K
armv4l-li Linux 2.3.9 275 0.5 1.1 8 10 0.10K 3.7 3 2.7K 10K 41K
i586-linu Linux 2.2.6 100 4.5 2.6 22 33 0.20K 4.5 9 1.6K 8K 45K
i586-linu Linux 2.2.0-p 133 1.1 2.1 16 26 0.13K 3.7 8 1.1K 7K 39K
sparc-lin Linux 2.1.130 33 10.7 17. 157 178 1.01K 56.5 105 10.5K 63K 226K
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
--------- ------------- ----- ------ ------ ------ ------ ------- -------
alpha-lin Linux 2.1.132 34 140 699 267 1515 335 1678
arm32-net NetBSD 1.3F 381 617 821 879
armv4l-li Linux 2.2.9
armv4l-li Linux 2.2.7 162 342 768 360 777 352 779
armv4l-li Linux 2.0.31 177 348 750 354 770 378 780
armv4l-li Linux 2.2.10 157 330 731 345 746 355 755
armv4l-li Linux 2.2.10 158 329 732 331 749 345 769
armv4l-li Linux 2.3.9 160 334 732 350 735 344 768
armv4l-li Linux 2.3.9 159 335 733 335 767 355 775
i586-linu Linux 2.2.6 10 79 346 116 528 126 570
i586-linu Linux 2.2.0-p 8 75 292 82 404 104 462
sparc-lin Linux 2.1.130 517 532 1211
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
alpha-lin Linux 2.1.132 34 128 288 568 749 2265
arm32-net NetBSD 1.3F 381 875 871 1256 1383 3278
armv4l-li Linux 2.2.9 417 691
armv4l-li Linux 2.2.7 162 325 451 612 823 1792
armv4l-li Linux 2.0.31 177 359 490 709 985 1127
armv4l-li Linux 2.2.10 157 315 821 590 793 1862
armv4l-li Linux 2.2.10 158 316 894 588 770 1754
armv4l-li Linux 2.3.9 160 321 565 625 770 1747
armv4l-li Linux 2.3.9 159 325 632 598 770 1750
i586-linu Linux 2.2.6 10 40 73 168 468 869
i586-linu Linux 2.2.0-p 8 29 62 163 239 742
sparc-lin Linux 2.1.130 268 576 1094 1564 4398
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
--------- ------------- ------ ------ ------ ------ ------- ----- -----
alpha-lin Linux 2.1.132 55 22 170 40 56904 13.8K
arm32-net NetBSD 1.3F 35107 1934 5.0K
armv4l-li Linux 2.2.9 0 2 1.9K
armv4l-li Linux 2.2.7 28045 1 1.6K
armv4l-li Linux 2.0.31 65 4 156 10 27317 0.2K
armv4l-li Linux 2.2.10 25178 1 1.3K
armv4l-li Linux 2.2.10 25168 1 1.3K
armv4l-li Linux 2.3.9 3842 1 0.0K
armv4l-li Linux 2.3.9 3858 1 9.3K
i586-linu Linux 2.2.6 21500 5 2.1K
i586-linu Linux 2.2.0-p 40 6 143 12 9439 2.1K
sparc-lin Linux 2.1.130 555 100 763 234 37715 1.5K
*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
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----
alpha-lin Linux 2.1.132 15 15 8 18 34 13 14 34 36
arm32-net NetBSD 1.3F 5 5 9 14 75 52 52 72 98
armv4l-li Linux 2.2.9 -1 -1 -1 -1 -1 -1 -1 -1
armv4l-li Linux 2.2.7 10 23 18 35 88 62 62 88 116
armv4l-li Linux 2.0.31 9 15 12 36 92 65 65 92 121
armv4l-li Linux 2.2.10 10 23 19 37 92 64 64 92 121
armv4l-li Linux 2.2.10 10 24 16 37 92 64 64 92 121
armv4l-li Linux 2.3.9 10 23 18 37 92 64 64 92 121
armv4l-li Linux 2.3.9 10 23 19 37 92 64 64 92 121
i586-linu Linux 2.2.6 35 4 12 30 75 38 47 75 88
i586-linu Linux 2.2.0-p 37 24 18 35 90 39 46 90 89
sparc-lin Linux 2.1.130 7 2 3 7 18 9 10 18 16
Memory latencies in nanoseconds - smaller is better
(WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- --- ---- ---- -------- -------
alpha-lin Linux 2.1.132 226 13 137 890
arm32-net NetBSD 1.3F 224 8 334 347 No L2 cache?
armv4l-li Linux 2.2.9 233 - - - Bad mhz?
armv4l-li Linux 2.2.7 275 7 278 289 No L2 cache?
armv4l-li Linux 2.0.31 274 0 270 270 No L2 cache?
armv4l-li Linux 2.2.10 275 7 269 279 No L2 cache?
armv4l-li Linux 2.2.10 275 7 269 279 No L2 cache?
armv4l-li Linux 2.3.9 275 7 269 280 No L2 cache?
armv4l-li Linux 2.3.9 275 7 269 279 No L2 cache?
i586-linu Linux 2.2.6 100 20 123 193
i586-linu Linux 2.2.0-p 133 15 80 147
sparc-lin Linux 2.1.130 33 90 888 889 No L2 cache?
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
