On Thu, Aug 6, 2009 at 10:23 PM, Alexander J. Maidak<ajmai...@mchsi.com> wrote:
> I have a number of systems running Solaris zones.  I'm looking for a
> tool that will do the following:
> 1) Capture/store the cpu utilization of the global zone and all
> non-global zones
> 2) Capture/store the memory utilization of the global zone and all
> non-global zones
> 3) Post the graphics to a website that can display both historical and
> realtime data.

Last time I looked there weren't any tools that did this in an
accurate way.  Admittedly, it has been a while.  The approach I took

On every global zone:

1) Enable task accounting.  In 2005 Adrian Cockroft had a few blog
posts about this.  See http://perfcap.blogspot.com/
2) Process task accounting records every 2 minutes, summing by zone
3) Send task accounting and various kstats related to memory, network,
etc., and other data (vmstat, iostat, etc.) to a central server as
soon as the data is gathered.  This is over a TCP session.

The central server (really a zone with adequate storage)

1) Read data from clients as it comes in
2) Store a copy of the raw data just in case you need to do other
types of analysis (excel, R, etc.)
3) Update rrd files.

There is then a simple CGI program that is able to present the RRD files.

The beauty of this is that I never configure anything.  The part that
goes on every global zone is baked into the server installation
procedure.  New zones are picked up as soon as they first boot.  I am
able to store 2 minute resolution data for ~ 40 days, 10 minute
resolution for ~6 months, and 30 minute resolution for 3 years in
about 200 MB per global zone plus about 10 MB per zone.  The CGI
interface can tell me somewhat detailed data about activity that has
happened in the past few minutes and any other period up to 3 years
ago with a response time of a couple seconds or less.

I would love to hear that there is a pre-baked tool out there that can
have this level of simplicity, but have a prettier interface than the
CGI interface that I came up with.  The core of this tool was written
(pre-S10 days) over the course of a weekend due to my dissatisfaction
with the tool that I had to use that would page me all night long but
took over 10 minutes to give me historical data (last 30 minutes, did
it do the same thing last week?).  Someone I was talking to recently
appropriately tagged this as "software written out of spite."  I like
that. :)

You might get other insights at sysadmin-discuss as well.

Mike Gerdts
zones-discuss mailing list

Reply via email to