Ravikesh Chandra
2009-Apr-12 22:08 UTC
[Xen-users] Dom0 vs native performance: does it look right?
Hi folks,
Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and I''ve just
been testing the performance with the Unixbench test. What I found is
that running the test on the Dom0 (under Xen kernel) resulted in only
50% performance of native! I appreciate there is always going to be
overheads with virtualisation but I just wanted to check with the list
if these numbers were expected? If that doesn''t seem normal any
suggestions on what I could look at?
Native:
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 34739474.5 922.0
Double-Precision Whetstone 83.1 1368.3 164.7
Execl Throughput 188.3 28715.8 1525.0
File Copy 1024 bufsize 2000 maxblocks 2672.0 508398.0 1902.7
File Copy 256 bufsize 500 maxblocks 1077.0 158973.0 1476.1
File Read 4096 bufsize 8000 maxblocks 15382.0 3337799.0 2169.9
Pipe-based Context Switching 15448.6 1703323.3 1102.6
Pipe Throughput 111814.6 5979578.6 534.8
Process Creation 569.3 80856.4 1420.3
Shell Scripts (8 concurrent) 44.8 6240.0 1392.9
System Call Overhead 114433.5 5937086.6 518.8
========
FINAL SCORE 985.7
Dom0:
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 33159482.3 880.1
Double-Precision Whetstone 83.1 1399.8 168.4
Execl Throughput 188.3 8307.4 441.2
File Copy 1024 bufsize 2000 maxblocks 2672.0 269880.0 1010.0
File Copy 256 bufsize 500 maxblocks 1077.0 72967.0 677.5
File Read 4096 bufsize 8000 maxblocks 15382.0 2160271.0 1404.4
Pipe-based Context Switching 15448.6 615502.1 398.4
Pipe Throughput 111814.6 2306450.5 206.3
Process Creation 569.3 16938.2 297.5
Shell Scripts (8 concurrent) 44.8 2537.3 566.4
System Call Overhead 114433.5 1956760.1 171.0
========
FINAL SCORE 448.8
Regards
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
Stefan de Konink
2009-Apr-12 22:41 UTC
Re: [Xen-users] Dom0 vs native performance: does it look right?
Ravikesh Chandra wrote:> Hi folks, > > Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and I''ve just > been testing the performance with the Unixbench test. What I found is > that running the test on the Dom0 (under Xen kernel) resulted in only > 50% performance of native! I appreciate there is always going to be > overheads with virtualisation but I just wanted to check with the list > if these numbers were expected? If that doesn''t seem normal any > suggestions on what I could look at?no tls cflag is a start. Stefan _______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ravikesh Chandra
2009-Apr-12 22:51 UTC
Re: [Xen-users] Dom0 vs native performance: does it look right?
Hi Stefan, Thanks for your reply. However could you please elaborate further on your statement? I have the file "/etc/ld.so.conf.d/kernelcap-2.6.18-128.1.6.el5xen.conf" which contains "hwcap 0 nosegneg".. is that what you mean? I am running a 64-bit installation btw. Thanks in advance. 2009/4/13 Stefan de Konink <stefan@konink.de>:> Ravikesh Chandra wrote: >> >> Hi folks, >> >> Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and I''ve just >> been testing the performance with the Unixbench test. What I found is >> that running the test on the Dom0 (under Xen kernel) resulted in only >> 50% performance of native! I appreciate there is always going to be >> overheads with virtualisation but I just wanted to check with the list >> if these numbers were expected? If that doesn''t seem normal any >> suggestions on what I could look at? > > no tls cflag is a start. > > > Stefan > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users >_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Peter Booth
2009-Apr-13 00:00 UTC
Re: [Xen-users] Dom0 vs native performance: does it look right?
If you look at the wikipedia page for xen it suggests that the performance cost of enabling vanilla tls is 50%. It''s worth trying the tls disabled configuration and seeing if this changes things On Apr 12, 2009, at 6:51 PM, Ravikesh Chandra <rcha108@aucklanduni.ac.nz> wrote:> Hi Stefan, > > Thanks for your reply. However could you please elaborate further on > your statement? > > I have the file > "/etc/ld.so.conf.d/kernelcap-2.6.18-128.1.6.el5xen.conf" which > contains "hwcap 0 nosegneg".. is that what you mean? I am running a > 64-bit installation btw. > > Thanks in advance. > > > 2009/4/13 Stefan de Konink <stefan@konink.de>: >> Ravikesh Chandra wrote: >>> >>> Hi folks, >>> >>> Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and I''ve just >>> been testing the performance with the Unixbench test. What I found >>> is >>> that running the test on the Dom0 (under Xen kernel) resulted in >>> only >>> 50% performance of native! I appreciate there is always going to be >>> overheads with virtualisation but I just wanted to check with the >>> list >>> if these numbers were expected? If that doesn''t seem normal any >>> suggestions on what I could look at? >> >> no tls cflag is a start. >> >> >> Stefan >> >> _______________________________________________ >> Xen-users mailing list >> Xen-users@lists.xensource.com >> http://lists.xensource.com/xen-users >> > > _______________________________________________ > Xen-users mailing list > Xen-users@lists.xensource.com > http://lists.xensource.com/xen-users_______________________________________________ Xen-users mailing list Xen-users@lists.xensource.com http://lists.xensource.com/xen-users
Ravikesh Chandra
2009-Apr-13 00:40 UTC
Re: [Xen-users] Dom0 vs native performance: does it look right?
Thanks for your input Peter, Tom, Stefan.
Regarding TLS well, is there any way to easily verify it is the
culprit? Reading the Wiki + mailing list searches seem to imply that I
have already disabled this? (with /etc/ld.so.conf.d/xen.conf file
containing "hwcap 0 nosegneg" as reported
http://www.xen-support.com/?p=180)
Here is my output from ldconfig -v (ref
http://www.mail-archive.com/fedora-xen@redhat.com/msg00041.html c.f.
XenFAQ)
# ldconfig -v -p 2>&1 | grep libc.so
libc.so.6 (libc6,x86-64, OS ABI: Linux 2.6.9) => /lib64/libc.so.6
libc.so.6 (libc6, hwcap: 0x0018000000000000, OS ABI: Linux
2.6.9) => /lib/i686/nosegneg/libc.so.6
libc.so.6 (libc6, OS ABI: Linux 2.6.9) => /lib/libc.so.6
That looks OK to me?
Cheers
2009/4/13 Peter Booth <peter_booth@mac.com>:> If you look at the wikipedia page for xen it suggests that the performance
> cost of enabling vanilla tls is 50%.
>
> It''s worth trying the tls disabled configuration and seeing if
this changes
> things
>
>
>
> On Apr 12, 2009, at 6:51 PM, Ravikesh Chandra
<rcha108@aucklanduni.ac.nz>
> wrote:
>
>> Hi Stefan,
>>
>> Thanks for your reply. However could you please elaborate further on
>> your statement?
>>
>> I have the file
>> "/etc/ld.so.conf.d/kernelcap-2.6.18-128.1.6.el5xen.conf"
which
>> contains "hwcap 0 nosegneg".. is that what you mean? I am
running a
>> 64-bit installation btw.
>>
>> Thanks in advance.
>>
>>
>> 2009/4/13 Stefan de Konink <stefan@konink.de>:
>>>
>>> Ravikesh Chandra wrote:
>>>>
>>>> Hi folks,
>>>>
>>>> Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and
I''ve just
>>>> been testing the performance with the Unixbench test. What I
found is
>>>> that running the test on the Dom0 (under Xen kernel) resulted
in only
>>>> 50% performance of native! I appreciate there is always going
to be
>>>> overheads with virtualisation but I just wanted to check with
the list
>>>> if these numbers were expected? If that doesn''t seem
normal any
>>>> suggestions on what I could look at?
>>>
>>> no tls cflag is a start.
>>>
>>>
>>> Stefan
>>>
>>> _______________________________________________
>>> Xen-users mailing list
>>> Xen-users@lists.xensource.com
>>> http://lists.xensource.com/xen-users
>>>
>>
>> _______________________________________________
>> Xen-users mailing list
>> Xen-users@lists.xensource.com
>> http://lists.xensource.com/xen-users
>
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
Boris Derzhavets
2009-Apr-13 08:50 UTC
Re: [Xen-users] Dom0 vs native performance: does it look right?
Just wondering
1.Would Xen 3.3.1 direct port to CentOS 5.3 ( 2.6.18.6-xen kernel)
behave same way ?
2. Would install native 3.1.2 Hypervisor improve the numbers ?
Boris.
--- On Sun, 4/12/09, Ravikesh Chandra <rcha108@aucklanduni.ac.nz> wrote:
From: Ravikesh Chandra <rcha108@aucklanduni.ac.nz>
Subject: [Xen-users] Dom0 vs native performance: does it look right?
To: "xen-users@lists.xensource.com"
<xen-users@lists.xensource.com>
Date: Sunday, April 12, 2009, 6:08 PM
Hi folks,
Firstly I am running CentOS 5.3 with Xen 3.3.1 (gitco) and I''ve just
been testing the performance with the Unixbench test. What I found is
that running the test on the Dom0 (under Xen kernel) resulted in only
50% performance of native! I appreciate there is always going to be
overheads with virtualisation but I just wanted to check with the list
if these numbers were expected? If that doesn''t seem normal any
suggestions on what I could look at?
Native:
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 34739474.5 922.0
Double-Precision Whetstone 83.1 1368.3 164.7
Execl Throughput 188.3 28715.8 1525.0
File Copy 1024 bufsize 2000 maxblocks 2672.0 508398.0 1902.7
File Copy 256 bufsize 500 maxblocks 1077.0 158973.0 1476.1
File Read 4096 bufsize 8000 maxblocks 15382.0 3337799.0 2169.9
Pipe-based Context Switching 15448.6 1703323.3 1102.6
Pipe Throughput 111814.6 5979578.6 534.8
Process Creation 569.3 80856.4 1420.3
Shell Scripts (8 concurrent) 44.8 6240.0 1392.9
System Call Overhead 114433.5 5937086.6 518.8
========
FINAL SCORE 985.7
Dom0:
TEST BASELINE RESULT INDEX
Dhrystone 2 using register variables 376783.7 33159482.3 880.1
Double-Precision Whetstone 83.1 1399.8 168.4
Execl Throughput 188.3 8307.4 441.2
File Copy 1024 bufsize 2000 maxblocks 2672.0 269880.0 1010.0
File Copy 256 bufsize 500 maxblocks 1077.0 72967.0 677.5
File Read 4096 bufsize 8000 maxblocks 15382.0 2160271.0 1404.4
Pipe-based Context Switching 15448.6 615502.1 398.4
Pipe Throughput 111814.6 2306450.5 206.3
Process Creation 569.3 16938.2 297.5
Shell Scripts (8 concurrent) 44.8 2537.3 566.4
System Call Overhead 114433.5 1956760.1 171.0
========
FINAL SCORE 448.8
Regards
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users
_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users