Workaround has been applied:-
Minus 50 sec
-bash-3.2$ cat fork1.output
# bin/../bin-i86pc/fork -E -C 200 -L -S -W -N fork_100 -B 100 -C 100
prc thr usecs/call samples errors cnt/samp
fork_100 1 1 23881.74167 86 0 100
#
# STATISTICS usecs/call (raw) usecs/call (outliers removed)
# min
22510.65833 22510.65833
# max 49583.97639 26053.60615
# mean 24982.76488 24009.66137
# median 24098.16940 23881.74167
# stddev 3179.35577
686.60767
# standard error 314.80313 74.03881
# 99% confidence level 732.23208 172.21427
# skew 5.00473 0.60015
# kurtosis 33.19530 0.17513
# time correlation
-15.29000 -0.54277
#
# elasped time 257.34507
# number of samples 86
# number of outliers 16
# getnsecs overhead 238
#
# DISTRIBUTION
# counts usecs/call
means
# 1 22500.00000 |**** 22510.65833
# 0 22590.00000 | -
# 0 22680.00000
| -
# 0 22770.00000 | -
# 3 22860.00000 |************
22890.55975
# 0 22950.00000 | -
# 3 23040.00000 |************ 23074.55564
# 1 23130.00000
|**** 23178.31415
# 2 23220.00000 |******** 23264.49284
# 4 23310.00000 |**************** 23376.40680
# 4 23400.00000
|**************** 23453.45574
# 4 23490.00000 |**************** 23523.64573
# 7 23580.00000 |**************************** 23635.30316
# 8 23670.00000 |******************************** 23710.76531
# 5 23760.00000 |********************
23788.50820
# 4 23850.00000 |**************** 23893.68383
# 3 23940.00000 |************ 23973.25051
# 4 24030.00000 |**************** 24078.65052
# 3 24120.00000
|************ 24141.85327
# 3 24210.00000 |************ 24246.67291
# 4 24300.00000 |**************** 24364.04465
# 5 24390.00000 |********************
24444.50472
# 4 24480.00000 |**************** 24539.64735
# 1 24570.00000 |**** 24621.17191
# 2 24660.00000 |********
24739.20456
# 1 24750.00000 |**** 24760.38002
# 0 24840.00000 | -
# 0 24930.00000
| -
# 3 25020.00000 |************ 25076.92854
# 1 25110.00000 |**** 25182.04928
# 1
25200.00000 |**** 25276.34825
#
# 5 > 95% |******************** 25573.33034
#
# mean of 95% 23913.13860
# 95th %ile 25295.36178
--- On Mon, 5/5/08, Juergen Keil <[EMAIL PROTECTED]> wrote:
From: Juergen Keil <[EMAIL PROTECTED]>
Subject: Re: [xen-discuss] How to install Solaris 0508 HVM DomU at SNV87
To:
[EMAIL PROTECTED]
Cc: [email protected]
Date: Monday, May 5, 2008, 12:43 PM
> We are done on SNV87 HVM DomU.
> Service ppd-cache-update now online.
> I gonna wait for completition on S10U5 HVM DomU
Can you please try to run the fork_100 test from
the libMicro-0.4.0 benchmark on that SNV87 HVM DomU?
The source of the benchmark is available for download here:
http://opensolaris.org/os/project/libmicro/
http://opensolaris.org/os/project/libmicro/files/libmicro-0.4.0.tar.gz
% wget http://opensolaris.org/os/project/libmicro/files/libmicro-0.4.0.tar.gz
% gunzip< libmicro-0.4.0.tar.gz |tar xf -
% cd libMicro-0.4.0
% make
% bin/fork -E -C 200 -L -S -W -N fork_100 -B 100 -C 100 > /tmp/fork.output
Running: fork_100 for 4.77470 seconds
% cat /tmp/fork.output
# bin/../bin-i86pc/fork -E -C 200 -L -S
-W -N fork_100 -B 100 -C 100
prc thr usecs/call samples errors cnt/samp
fork_100 1 1 280.76964 100 0 100
#
# STATISTICS usecs/call (raw) usecs/call (outliers removed)
# min 237.29929 237.29929
# max 949.35702 424.32036
# mean 306.45514 298.44325
# median 281.65932 280.76964
# stddev 79.72480 44.62684
# standard error 7.89393 4.46268
# 99% confidence level 18.36128 10.38020
# skew 5.29799 1.00536
# kurtosis 38.85951 0.12177
# time correlation -1.47891 -1.07615
#
# elasped time 4.76910
# number
of samples 100
# number of outliers 2
# getnsecs overhead 399
#
# DISTRIBUTION
# counts usecs/call means
# 1 234.00000 |** 237.29929
# 7 240.00000 |************** 242.43814
# 2 246.00000 |**** 249.26708
# 1 252.00000 |** 256.03853
# 3 258.00000 |****** 262.67013
# 16 264.00000 |******************************** 267.68925
# 10 270.00000 |******************** 273.27753
# 12 276.00000 |************************ 279.59795
# 7 282.00000 |************** 284.16094
#
5 288.00000 |********** 290.65786
# 2 294.00000 |**** 297.36328
# 3 300.00000 |****** 302.75731
# 1 306.00000 |** 309.20215
# 2 312.00000 |**** 315.67402
# 1 318.00000 |** 318.00967
# 1 324.00000 |** 326.74916
# 3 330.00000 |****** 331.74546
# 3 336.00000 |****** 340.82672
# 3 342.00000 |****** 342.74536
# 5 348.00000 |********** 350.05268
# 2 354.00000 |****
357.77666
# 0 360.00000 | -
# 1 366.00000 |** 369.75543
# 0 372.00000 | -
# 1 378.00000 |** 383.21981
# 3 384.00000 |****** 386.35274
#
# 5 > 95% |**********
408.37425
#
# mean of 95% 292.65741
# 95th %ile 397.26077
On an "AMD Athlon(tm) 64 X2 Dual Core Processor 6400+" /
metal / snv_89, the fork benchmark runs for 4.77 seconds
(that's the output included above). Under xVM / Xen
and on current Intel / AMD processors it should complete
in 20-30 seconds.
I've seen cases where the fork_100 benchmark used > 700 seconds,
when I
ran it in a 32-bit PV domU.
--------------------------------------------------------------------
I do have a workaround for this xen / opensolaris performance
problem[*]; it's a modified /usr/lib/libc/libc_hwcap3.so.1
shared C library (see the attachment). It was compiled from
current opensolaris source (post snv_89), but seems to work OK
under snv_81. Under xVM you should find a lofs mount like this
in "df -h" output:
# df -h
Filesystem size used avail capacity Mounted on
...
/usr/lib/libc/libc_hwcap3.so.1
6.1G 4.1G 1.9G 69% /lib/libc.so.1
...
Now try this in your domU:
# gunzip < libc_hwcap3.tar.gz | ( cd /tmp; tar xfv - )
x libc_hwcap3.so.1, 1646444 bytes, 3216 tape blocks
# mount -O -F lofs /tmp/libc_hwcap3.so.1 /lib/libc.so.1
Now run the fork_100 benchmark. It might run much
faster.
And after "umount /lib/libc.so.1", the fork_100 should become slow.
===
[*] http://www.opensolaris.org/jive/thread.jspa?threadID=58717&tstart=0 |
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now.
_______________________________________________
xen-discuss mailing list
[email protected]
