Hi
I like glance, but MUCH prefer MeasureWare, not because it is better, but because you get the history of a problem/system. Similar to some of the comments above about getting used to what a system "usually" does, MeasureWare does this very quickly. I usually use the global class & extract CPU%, sys CPU% and peak disk util%. Also I plot out al the queues. Very quicly you can see what the system does over say 24 hours.
The next step I like to define application groups, for this you really have to have a relatively good knowledge of what runs as what & by whom. It may take a few itterations but when the application groups are sorted, again you can see what does what.
Lastly, this may be the hardest for some systems, you need an application "load indicator", and or an application "backlog indicator". This may be impossible to agree on some systems, but I have found that usually there is one or two overriding loading things. What I measure is
o Insert rate into a key table (simply the max unique index number)
o The number of files backlogged (ls
| wc -l)
o The number of requests/s (our application provides this figure, OR there may be a key table which you can look at isam reads etc)
o The number of items processed. (aplication provides this)
o Or anything else that may be useful, say number of users...
After this the KEY is to plot say cpu Vs load. The main idea is to get a load response graph. I have been lucky as all the systems I have done this on had one overriding loading factor, thus I only needed ONE item per system.
After the above three steps (about 3-18 month work depending on how easy it is to implement things on your systems). I've used the above to
o Data to allow us to capacitise future systems
o Problem solve and analyse
o benchmark systems before & after changes to ensure performance is OK/better
o As nice colourfull graphs for management, as proof that I do do a job and dont just guess....
Regards
Tim
-
