Here are the results of a simple "performance experiment".
I performed a series of kernel compiles "make -j3 bzImage" on my labtop for different configurations against 2.6.8.1. Times are reported "total (sys)" in seconds. "*" indicates after reboot.
vanilla CKRM-E16 + cpu-v8 CKRM-E16+cpu-v8+numtasks 336.75(32.01)* 337.92(31.96)* 332.29(31.87) 332.64(31.56) 332.71(31.48) 332.68(31.60) 332.99(31.53) 333.18(31.86) 332.44(31.83 332.53(31.56) 332.72(31.79) 332.81(31.65)
In the CKRM case, no only the CPU controller was compiled into the kernel (as required). numtasks was dynamically loaded. No other controllers were enabled.
Based on these result, I don't see in this workload any overhead for CKRM, the cpu scheduler or the numtasks controller.
The memory controller is next to test in this series. But I simply wanted to get a few potential culprits off the most-wanted list.
-- Hubertus
Marc E. Fiuczynski wrote:
I am not using any rules at all nor have I created any classes in /rcfs/taskclass.
-----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] Behalf Of Hubertus Franke Sent: Tuesday, September 28, 2004 3:12 PM To: Marc E. Fiuczynski Cc: ckrm-Tech Subject: Re: [ckrm-tech] performance and stability issues with CKRM
Can you provide the rule set you are using as a shell script. If you don't use any rbce rules please indicate so as well. I'll check on the ckrm + CPU issue.
Marc E. Fiuczynski wrote:
Hello,
I have the latest CKRM patches for all controllers applied to my tree. Specifically, I am using cpu v7, io v4, and mem v1 controllers
with e16rc1
which we released over the past few days --- i.e., the most
recent code that
addresses most of the outstanding issues. My kernel is
configured with all
things CKRM related statically compiled into my kernel, but I
did not select
the socket controller nor the LRU ordering options for the memory controller. Note that my kernel is based off of Fedora Core 2
1.521 (linux
2.6.8.1), rather than just plain vanilla 2.6.8.1 from kernel.org.
As I actually enjoy eating my own dogfood, I run this kernel on my Dell Inspiron 5150 laptop and then use it for kernel compilations,
email (running
MS Outlook via cxoffice), etc.. This way I get a good feel for its base performance -- particularly interactivity -- and forces me to
immediately
address stability issues. By base performance I am referring to using a system that does not create any classes in /rcfs/taskclass;
everything is
running in the base class.
In terms of performance, my general sense is that the systems
with the CKRM
controllers feels sluggish compared to running a FC2 1.521.
Compiles of the
kernel appear to take a longer --- I've reported on this before. Unfortunately, due to lack of time, I don't have hard statistics to back this up. Moreover, I don't have a good sense which CKRM controller is causing the sluggishness.
In terms of stability, it appears the i/o controller causes the
system to
hang while doing a kernel compile (make -j 3). Of course, after
several hard
reboots the filesystem was corrupted. Trying to fix the filesystem (ext) with the same kernel, fsk would fail with a SIG_USR1 about 50%
through the
check. My conjecture is that something is failing in the
cfq-iosched.c code
that reflects this signal back to fsck. My reasonig for this is that running the same fsck on a system booted with a vanilla FC2 1.521 kernel succeeded to clean the filesystem. Until the I/O scheduler
attains better
stability, I plan to simply switch back to the vanilla FC2 1.521 cfq-iosched.c.
I'd like to hear about other experiences with CKRM so far.
Best, Marc
------------------------------------------------------- This SF.net email is sponsored by: IT Product Guide on ITManagersJournal Use IT products in your business? Tell us what you think of
them. Give us
Your Opinions, Get Free ThinkGeek Gift Certificates! Click to
find out more
http://productguide.itmanagersjournal.com/guidepromo.tmpl _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
------------------------------------------------------- This SF.net email is sponsored by: IT Product Guide on ITManagersJournal Use IT products in your business? Tell us what you think of them. Give us Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more http://productguide.itmanagersjournal.com/guidepromo.tmpl _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
------------------------------------------------------- This SF.net email is sponsored by: IT Product Guide on ITManagersJournal Use IT products in your business? Tell us what you think of them. Give us Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more http://productguide.itmanagersjournal.com/guidepromo.tmpl _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
------------------------------------------------------- This SF.net email is sponsored by: IT Product Guide on ITManagersJournal Use IT products in your business? Tell us what you think of them. Give us Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out more http://productguide.itmanagersjournal.com/guidepromo.tmpl _______________________________________________ ckrm-tech mailing list https://lists.sourceforge.net/lists/listinfo/ckrm-tech
