Hello
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
http://www.crisp.demon.co.uk/tools.html
Thanks! Antonello
On Tue, 21 Dec 2010, Antonello Piemonte wrote:
To: centos@centos.org From: Antonello Piemonte apiemont@googlemail.com Subject: [CentOS] I/O size distribution?
Hello
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
This might do it. It's in the updates repo.
Name : systemtap Arch : i386 Version : 1.1 Release : 3.el5_5.3 Size : 6.3 M Repo : installed Summary : Instrumentation System URL : http://sourceware.org/systemtap/ License : GPLv2+ Description: SystemTap is an instrumentation system for systems running Linux 2.6. : Developers can write instrumentation to collect data on the operation : of the system.
Kind Regards,
Keith Roberts
On Tue, 2010-12-21 at 23:01 +0100, Antonello Piemonte wrote:
Hello
IOTOP from here [1]. From a very quick glance of the Specfile it should build with no problems under EL5.
John
On 12/21/2010 05:01 PM, Antonello Piemonte wrote:
Hello
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
If you want to look at the specific run of an application and not the system as a whole or on a continual basis, then you can use strace.
However, this can produce a great deal of data. So I wrote a simple tool to scan through the strace output file and produce some statistics. You can download it at:
http://clusterbuffer.wetpaint.com/page/Strace+Analyzer
The version that is on the website is currently a bit dated with a few bugs in it. I should be posting the latest version in the next couple of weeks.
Thanks!
Jeff
Greetings,
On Tue, Dec 21, 2010 at 10:01 PM, Antonello Piemonte apiemont@googlemail.com wrote:
Hello
I have read that under Solaris one can use DTrace to get I/O request
I have heard that Dag or rf has it?
No guarantees!
Regards
Rajagopal
......
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
You can try SystemTap: http://sourceware.org/systemtap/
It seems to be roughly analagous to dtrace and is aimed squarly at the linux kernel.
Thanks for all the suggestions.
I had a look at systemtap but I have the feeling that in it's current state it is aimed more at kernel developers rather than average admins like me :-)
Still, I'll keep an eye on it also give the fact that there seem to be now new documentation being written about it
http://docs.redhat.com/docs/en-US/Red_Hat_Enterprise_Linux/5/pdf/SystemTap_B...
On the other end, I have discovered blktrace and related tools, and this seems more palatable to me (subjective, I know ...). This example in particular seems to work for me
# blktrace /dev/sda -a issue -a complete -w 60 -o - | blkiomon -I 4 -h -
thought to share in case anyone finds it useful.
cheers A.
On Wed, 22 Dec 2010, Antonello Piemonte wrote:
To: CentOS mailing list centos@centos.org From: Antonello Piemonte apiemont@googlemail.com Subject: Re: [CentOS] I/O size distribution?
Thanks for all the suggestions.
I had a look at systemtap but I have the feeling that in it's current state it is aimed more at kernel developers rather than average admins like me :-)
Still, I'll keep an eye on it also give the fact that there seem to be now new documentation being written about it
http://docs.redhat.com/docs/en-US/Red_Hat_Enterprise_Linux/5/pdf/SystemTap_B...
On the other end, I have discovered blktrace and related tools, and this seems more palatable to me (subjective, I know ...). This example in particular seems to work for me
# blktrace /dev/sda -a issue -a complete -w 60 -o - | blkiomon -I 4 -h -
thought to share in case anyone finds it useful.
Can I use this to monitor which files are being written to disk?
Keith
@ Charles Polisher
yes, seems that avgrq-sz does provide the info I needed, thanks for pointing that out!
@ Keith
I am not sure if blktrace can monitor which files being written to disk or if it is the best tool for that. But you could check out the inotify-tools (iwatch and related) which go on that directions. An example usage is here
http://openquery.com/blog/tool-day-inotify
or iwatch
http://iwatch.sourceforge.net/index.html
cheers Antonello
Antonello Piemonte wrote:
Hello
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
Maybe the average request size from iostat -x could supply part of the /global/ piece?
On Tue, Dec 21, 2010 at 10:01 PM, Antonello Piemonte apiemont@googlemail.com wrote:
Hello
I have read that under Solaris one can use DTrace to get I/O request size distribution on a global scale (also on a per process/pid basis). See for example
http://prefetch.net/articles/observeiodtk.html
Can anyone recommend an alternative to get similar information under CentOS? I looked into dtrace for linux but it seems still work in progress, even putting aside CDDL issues ...
http://www.crisp.demon.co.uk/tools.html
Thanks! Antonello _______________________________________________ CentOS mailing list CentOS@centos.org http://lists.centos.org/mailman/listinfo/centos
Nowhere near as advanced, but give dstat a try.