Whoops, the TeamQuest webinar link should be
http://www.teamquest.com/resources/webinars/display/30/index.htm
Without the www, it goes nowhere (;-))
--dave
--
This message posted from opensolaris.org
___
perf-discuss mailing list
perf-discuss@opensola
Sar is a good starting point: it collects information about
everything except the network.
sar -o sar.raw 10 30 will collect 5 minutes of data,
sar -A -f sar.raw >sar.txt will convert it into
analyzable form: do this on the machine where the
data was collected, after the individual load test
If the two programs don't happen to have been started at
the same time, use the same sampling interval and have
the same sampling overheads, the numbers will always
be slightly different. You'd need to do a sensitivity
test to see how many decimal places this will affect...
--dave
This messag
Hmmn... if you're snooping on the same machine, the
act of snooping, in promiscuous mode, may have had
an effect on the interface.
Mind you, it isn't **supposed** to, but the observer
effect applies to hardware as well as people (;-))
Snoop from a different machine on the same wire
(you may need
Ok, that passes the back-of-the-envelope test (;-))
Now on to how to measure and mitigate.
The first part is to measure: what metric will
- tell you if we've fixed it
- help the developers quantify the problem?
If we can't measure it, we can't manage it!
--dave
This message posted from open
Ok, how do we quantify this? If it causes a 1%
slowdown on rare occasions, it's probably not
a big deal, but a 300% slowdown at some
critical moment is a different thing entirely!
--dave
This message posted from opensolaris.org
___
perf-discuss mai
Memory usage can be controlled globally or
by project on Solaris 9 with Solaris Resource
Manager (SRM).
The reference I usually use is "Solaris Containers-
Resource Management and Solaris Zones", although
it's predominantly about Solaris 10.
This message posted from opensolaris.org
_
