RE: RE: Updated XenMon Patch

RE: RE: Updated XenMon Patch

[Jeffrey Buell]

> > My experiences so far:
> >
> > I tried this release on Changeset 7480 running on a 2P Xeon HT box.  No
> > issues patching or building Xen and the tools.  XenMon doesn't give
detailed
> > performance or profiling data, but it does give some very useful
high-level
> > CPU information with little effort by the user.
> 
> Thanks for your comments Jeff.
> 
> What kinds of performance/profiling data would you like to see
> instead? Atleast as far as the scheduler is concerned, I think its
> hard to do any better -- XenMon already reports things such as per
> domain blocking time, waiting time, number of memory page exchanges,
> number of times a VM is scheduled. It'll be very useful if you can
> give some concrete examples of the kinds of metrics you would want in
> such a tool.
> 
> Thanks,
> Diwaker

Diwaker,

I didn't mean to imply that XenMon is deficient in any way.  Only that it is
a different kind of performance tool than, say, a function-level tool like
Xenoprof.  XenMon works well for me now, the only suggestion I have is to
improve the formatting of the notlive output:  some numbers are printed with
12 significant figures and the columns are not well-aligned.

Jeff

Re: RE: Updated XenMon Patch

[Rob Gardner]

Jeffrey Buell wrote:

> Diwaker,
>
>I didn't mean to imply that XenMon is deficient in any way.  Only that it is
>a different kind of performance tool than, say, a function-level tool like
>Xenoprof.  XenMon works well for me now, the only suggestion I have is to
>improve the formatting of the notlive output:  some numbers are printed with
>12 significant figures and the columns are not well-aligned.
>  
>

Jeff,

The idea of the not-live output was that you'd read that stuff with a 
script and parse it, so little attention was given to making it look 
pretty. But I'll see what we can do to neaten it up a bit. Thanks again 
for using the tool and providing feedback!

Rob

Re: RE: Updated XenMon Patch

[Diwaker Gupta]

> I didn't mean to imply that XenMon is deficient in any way.  Only that it is
> a different kind of performance tool than, say, a function-level tool like
> Xenoprof.  XenMon works well for me now, the only suggestion I have is to
> improve the formatting of the notlive output:  some numbers are printed with
> 12 significant figures and the columns are not well-aligned.

And I didn't think you did! :-) Its really good to see someone
actually using the tool, and so I was just interested in getting some
more feedback from you. Like Rob said, we didn't really pay much
attention to notlive output. Those 12 digits are a Python thing --
numbers are rarely that precise. We'll try to improve it. Thanks
again!

I think XenMon and XenoProf are complementary, and useful for
different kinds of analysis. Neither is a panacea for profiling by
itself.

Diwaker
--
Web/Blog/Gallery: http://floatingsun.net

RE: Updated XenMon Patch

[Jeffrey Buell]

> (This is a repeat of the XenMon patch from last week, but includes an 
> important bug fix. Please take a look...)
> 
> Attached is the second release of XenMon, which is a unique performance 
> monitoring tool for Xen. Instead of using hypervisor calls to get domain 
> information, we use the xentrace facility to provide fine-grained 
> monitoring of various metrics (see README below).

My experiences so far:

I tried this release on Changeset 7480 running on a 2P Xeon HT box.  No
issues patching or building Xen and the tools.  XenMon doesn't give detailed
performance or profiling data, but it does give some very useful high-level
CPU information with little effort by the user.

Jeff

Re: RE: Updated XenMon Patch

[Diwaker Gupta]

> My experiences so far:
>
> I tried this release on Changeset 7480 running on a 2P Xeon HT box.  No
> issues patching or building Xen and the tools.  XenMon doesn't give detailed
> performance or profiling data, but it does give some very useful high-level
> CPU information with little effort by the user.

Thanks for your comments Jeff.

What kinds of performance/profiling data would you like to see
instead? Atleast as far as the scheduler is concerned, I think its
hard to do any better -- XenMon already reports things such as per
domain blocking time, waiting time, number of memory page exchanges,
number of times a VM is scheduled. It'll be very useful if you can
give some concrete examples of the kinds of metrics you would want in
such a tool.

Thanks,
Diwaker
--
Web/Blog/Gallery: http://floatingsun.net