Fwd: [Shr-User] SHR-unstable got a facelift. And you a christmas present....

Tue Nov 24 00:29:06 CET 2009

В Пнд, 23/11/2009 в 23:22 +0100, Martin Jansa пишет:
> If whole Thomas patch you mean drm-tracking branch from Thomas White
> then its built almost daily (sometimes even few times a day :)) here:
> http://build.shr-project.org/tests/mrmoku/kms/
> This kernel is not in shr-unstable by default just because all shr devs
> get WSOD during resume with this one, but we like it a lot.

The whole patch is simple one-liner against andy-tracking, published by
Thomas to change some FIFO depth (as far as I understood). Everything
still working after that patch, keeping nice boost. Default image was
unusable because of speed issues.
> If you want to try 2.6.31 with all Thomas's patches again
> http://build.shr-project.org/tests/mrmoku/2.6.31/
> but expect some problems with ie sound/gsm/(W|B)SOD
>
Thanks for links, I'll hope i can try them in spare time, but
unfortunately for me, I need gsm/sound and feel no need in any sort of
SOD. With QtMoko i've found a solution for death - just always keep it
plugged, now trying it with shr. ;)
I've just followed Carsten suggestion and tested several kernels with
lmbench, I wanted to publish it for interested people with separate
letter, but we started with kernel, here is it, anyway I don't know
where to proceed:
Last time I've tried to measure memory bandwidth on om, n810 and old
Celeron 600. Now I've got interesting results with om kernel, but
unfortunately didn't get n810 to my grasp to run lmbench where.
Most interesting thing is following:
*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
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ----
-----
neo Linux 2.6.29- 19.8 18.1 18.9 36.7 108.0 59.1 59.2 108.
187.6
neo_patch Linux 2.6.29- 16.5 16.6 13.1 25.1 74.4 40.7 40.8 74.4
130.7
router2 Linux 2.6.26- 46.4 49.7 33.5 119.8 295.0 78.9 57.2 294.
68.4
yes, faster kernel is in middle :). the difference between two one is
kernel, both systems were tested with qtmoko bought down and top showing
0 load except top. interesting that on both kernels to showed different
load. As i've got similar results with my copy test (40mb/s), i think
that Thomas one-line patch (second kernel) is unrelated. So question is
open: in qtmoko, with qpe.sh brought down and no active processes except
top in top, which thing slows down whole device from 1/3 to 1/2? time I
was unable to run oprofile because of some problems between daemon and
kernel, next thing i plan is to investigate this. I attached whole
results for whoose who are interested.
Btw, I successfully resisted idea to buy N900 in favor of continue using
OM after reading this: http://talk.maemo.org/showthread.php?t=31346 :)
After that, OM is evolving computer for me, n900 is not.
Gennady.
-------------- next part --------------
 L M B E N C H 2 . 0 S U M M A R Y
 ------------------------------------
Basic system parameters
----------------------------------------------------
Host OS Description Mhz
 
--------- ------------- ----------------------- ----
neo Linux 2.6.29- armv4tl-linux-gnu 389
neo_patch Linux 2.6.29- armv4tl-linux-gnu 389
router2 Linux 2.6.26- i686-pc-linux-gnu 679
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
--------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ----
neo Linux 2.6.29- 389 0.53 1.52 9.68 15.5 63.2 2.77 6.84 3335 10.K 25.K
neo_patch Linux 2.6.29- 389 0.70 3.31 18.6 35.6 89.7 5.77 16.8 4839 15.K 37.K
router2 Linux 2.6.26- 679 0.36 0.84 4.39 9.62 18.4 1.44 6.45 1136 3366 9409
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
--------- ------------- ----- ------ ------ ------ ------ ------- -------
neo Linux 2.6.29- 168.5 367.4 714.1 368.5 731.5 376.5 732.3
neo_patch Linux 2.6.29- 305.4 615.1 881.1 463.2 966.6 497.8 987.3
router2 Linux 2.6.26- 10.8 41.3 186.0 95.2 265.1 107.8 268.1
*Local* Communication latencies in microseconds - smaller is better
-------------------------------------------------------------------
Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP
 ctxsw UNIX UDP TCP conn
--------- ------------- ----- ----- ---- ----- ----- ----- ----- ----
neo Linux 2.6.29- 168.5 364.8 476. 831.9 1123. 3178
neo_patch Linux 2.6.29- 305.4 506.0 1444 1845. 2483. 4579
router2 Linux 2.6.26- 10.8 26.1 42.5 82.6 156.3 147.5 225.6 656.
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 
--------- ------------- ------ ------ ------ ------ ------- ----- ----- 
neo Linux 2.6.29- 1.469 70.0
neo_patch Linux 2.6.29- 116.0
router2 Linux 2.6.26- 344.0 1.223 3.00000
*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
--------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- -----
neo Linux 2.6.29- 19.8 18.1 18.9 36.7 108.0 59.1 59.2 108. 187.6
neo_patch Linux 2.6.29- 16.5 16.6 13.1 25.1 74.4 40.7 40.8 74.4 130.7
router2 Linux 2.6.26- 46.4 49.7 33.5 119.8 295.0 78.9 57.2 294. 68.4
Memory latencies in nanoseconds - smaller is better
 (WARNING - may not be correct, check graphs)
---------------------------------------------------
Host OS Mhz L1 $ L2 $ Main mem Guesses
--------- ------------- ---- ----- ------ -------- -------
neo Linux 2.6.29- 389 5.065 236.1 244.7 No L2 cache?
neo_patch Linux 2.6.29- 389 5.064 337.9 350.2 No L2 cache?
router2 Linux 2.6.26- 679 5.017 184.3 196.8 No L2 cache?