[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

[K12OSN] Capacity analysis - cron, SNMP, or what?

new subject alert.

On Wed, 13 Nov 2002, Jan Wilson wrote:

> workstation, it MAY be adequate.  Best advice:  stick as much RAM into
> the server as possible, and use the 256 MB + 50 MB per workstation as
> a very crude minimum.
> access or CPU speed.  For a task that takes 5 seconds with two
> workstations online, if it take 10 seconds it's either CPU or HD
> access.  If it takes 3 minutes, it's RAM  ;-)

hmmm, got me thinking ... We have run variations of this
capacity analysis thread a couple of times, and from several

Assuming an otherwise properly configured host (DNS is a real
killer when it is wrong), the major operative variables to
watch (off the top of my head) are:

   process load level -- how many processes are ready  
and waiting (i.e., not blocked), thread and handle depletion

   swap thrashing -- how often is the system having to push 
stuff back and forth from swap

   i/o blocking -- can the system get content from the drives 
ina timely fashion

   network latencies -- collisions, aggregate bandwidth 
available actually in use

iostat, vmstat, netstat, ifconfig, procinfo, free, w, and the 
systat suite each watch parts of these items.

I will take a stab at setting up a cronnable tool for data
collection, without major loading, in a form parsable by rrd
or mrtg.  SNMP collection is another route.  The thought would
be to produce a package to install, which gathers, and then
yields a report which identifies the 'top offender' so that it
can be addressed to a lay admin.

The stats also would permit a 'real' answer to the capacity
analysis questions we see.  

Also, there is a lot of prior art, and I see no need to
reinvent the wheel.  RRD seems like an early leader, but it is
somewhat 'touchy feely' pictures.  OpenNMS is too young, and
somewhat dificult to get running.

Does anyone else have any favorite metrics to include on that
list? A missing low level data collection, or end user
presentation tool?

-- Russ Herrold

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]