| poor performance in openvz kernels... [message #11964] |
Fri, 13 April 2007 06:20  |
devonblzx
Messages: 127
Registered: December 2006
|
Senior Member |
|
|
I have a new Dual Quad-Core Clovertown (E5310) with a RAID5 and when I run unixbench (the wht variant) I get these results...
Default CentOS4 Kernel (2.6.9smp):
Start Benchmark Run: Thu Apr 12 22:01:07 EDT 2007
22:01:07 up 1 min, 1 user, load average: 0.20, 0.08, 0.03
End Benchmark Run: Thu Apr 12 22:11:14 EDT 2007
22:11:14 up 11 min, 1 user, load average: 27.25, 10.54, 4.39
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 73062884.2 1939.1
Double-Precision Whetstone 83.1 1452.6 174.8
Execl Throughput 188.3 11972.2 635.8
File Copy 1024 bufsize 2000 maxblocks 2672.0 90568.0 339.0
File Copy 256 bufsize 500 maxblocks 1077.0 22404.0 208.0
File Read 4096 bufsize 8000 maxblocks 15382.0 1112686.0 723.4
Pipe-based Context Switching 15448.6 1536903.0 994.8
Pipe Throughput 111814.6 7310880.6 653.8
Process Creation 569.3 48057.5 844.2
Shell Scripts (8 concurrent) 44.8 2787.1 622.1
System Call Overhead 114433.5 8538136.8 746.1
=========
FINAL SCORE 586.1
Kernel 2.6.18-stab027-smp:
Start Benchmark Run: Thu Apr 12 21:29:33 EDT 2007
21:29:33 up 1 min, 1 user, load average: 0.22, 0.11, 0.04
End Benchmark Run: Thu Apr 12 21:39:43 EDT 2007
21:39:43 up 11 min, 1 user, load average: 25.45, 10.04, 4.23
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 71079686.4 1886.5
Double-Precision Whetstone 83.1 1456.2 175.2
Execl Throughput 188.3 4733.3 251.4
File Copy 1024 bufsize 2000 maxblocks 2672.0 120019.0 449.2
File Copy 256 bufsize 500 maxblocks 1077.0 32769.0 304.3
File Read 4096 bufsize 8000 maxblocks 15382.0 1125105.0 731.4
Pipe-based Context Switching 15448.6 415588.7 269.0
Pipe Throughput 111814.6 6285422.1 562.1
Process Creation 569.3 11449.0 201.1
Shell Scripts (8 concurrent) 44.8 1505.2 336.0
System Call Overhead 114433.5 8894040.1 777.2
=========
FINAL SCORE 416.8
Any idea why it would be such a big performance loss? It seems like the system has trouble with Pipe-based Context Switching, Process Creation, Shell Scripts and Execl Throughput.
I know that 2.6.18 has a problem with the CPU scheduler but I tested it on the rhel4 stable openvz 2.6.9 and turned up even worse results (around 350 final score).
ByteOnSite President
[Updated on: Fri, 13 April 2007 06:23] Report message to a moderator |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| Re: poor performance in openvz kernels... [message #12536 is a reply to message #12216] |
Wed, 02 May 2007 10:51  |
dev
Messages: 1693
Registered: September 2005
Location: Moscow
|
Senior Member |
|
|
BTW, 2.6.18+ kernels have still CONFIG_UBC_DEBUG_KMEM=y which affects this microbenchmark pretty badly.
So if you have a chance, please recheck without it.
on 2.6.18-028stab031.1 kernel and 16 CPUs Intel machine we get the following:
kernel 030.1 RHEL5 030.1 (*)
FINAL SCORE 450.4 471.7 481.7
(*) - it is 030.1 kernel with CONFIG_UBC_DEBUG_KMEM=n
[Updated on: Wed, 02 May 2007 10:51] Report message to a moderator |
|
|
|
|
|
|
|
|
|
 Re: poor performance in openvz kernels... [message #37910 is a reply to message #13352] |
Sun, 01 November 2009 14:34 |
hzqbbc
Messages: 2
Registered: November 2009
|
Junior Member |
|
|
 This post is very old ( May 2007 ) but we still encounter the performance problem.
Unixbench-4.1.0-wht-2 result different from 2.6.18-128.2.1.el5.028stab064.7 and 2.6.18-128.el5 (standard RHEL5 kernel), the difference is huge.
After searching openvz forum, i think i'm closing to the answer, but not sure, the huge perf loss may due to openvz UBC accounting.
So my question is *WHEN* openvz dev team will fix this problem? dev said he would release new kernel to archive better perf but i try the latest 2.6.18-128.2.1.el5 (stab064.7) and got no improvement.
My bench result:
==============================================================
BYTE UNIX Benchmarks (Version 4.1-wht.2, 8 threads)
System -- Linux vz 2.6.18-128.el5 #1 SMP Wed Jan 21 10:41:14 EST 2009 x86_64 x86_64 x86_64 GNU/Linux
65672880 3609080 58673952 6% /
End Benchmark Run: Mon Nov 2 06:18:21 CST 2009
06:18:21 up 13 min, 2 users, load average: 31.71, 12.44, 5.31
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 36719630.6 974.6
Double-Precision Whetstone 83.1 1298.6 156.3
Execl Throughput 188.3 20191.8 1072.3
File Copy 1024 bufsize 2000 maxblocks 2672.0 177441.0 664.1
File Copy 256 bufsize 500 maxblocks 1077.0 43517.0 404.1
File Read 4096 bufsize 8000 maxblocks 15382.0 1771809.0 1151.9
Pipe-based Context Switching 15448.6 2739102.5 1773.0
Pipe Throughput 111814.6 11210018.3 1002.6
Process Creation 569.3 64281.4 1129.1
System Call Overhead 114433.5 19818827.5 1731.9
=========
FINAL SCORE 840.3
==============================================================
BYTE UNIX Benchmarks (Version 4.1-wht.2, 8 threads)
System -- Linux vz 2.6.18-128.2.1.el5.028stab064.7 #1 SMP Wed Aug 26 15:47:17 MSD 2009 x86_64 x86_64 x86_64 GNU/Linux
65672880 3648024 58635008 6% /
Start Benchmark Run: Mon Nov 2 06:22:56 CST 2009
06:22:56 up 1 min, 1 user, load average: 2.20, 0.88, 0.32
End Benchmark Run: Mon Nov 2 06:33:02 CST 2009
06:33:02 up 11 min, 2 users, load average: 27.90, 11.13, 4.73
INDEX VALUES
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 36500265.7 968.7
Double-Precision Whetstone 83.1 1295.2 155.9
Execl Throughput 188.3 3618.3 192.2
File Copy 1024 bufsize 2000 maxblocks 2672.0 167840.0 628.1
File Copy 256 bufsize 500 maxblocks 1077.0 41757.0 387.7
File Read 4096 bufsize 8000 maxblocks 15382.0 1752507.0 1139.3
Pipe-based Context Switching 15448.6 2243795.1 1452.4
Pipe Throughput 111814.6 14018287.3 1253.7
Process Creation 569.3 9315.9 163.6
Shell Scripts (8 concurrent) 44.8 1147.4 256.1
System Call Overhead 114433.5 20733248.3 1811.8
=========
FINAL SCORE 538.9
|
|
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
