posted 09-17-2002 11:51 AM
quote:
Originally posted by DA:
I would like to know what are best counters for capacity planing
CPU
Memory
Disk
NetworkThis will help me my capaciyt panning project.
I just went through a very similar exercise and yes the counters mentioned in this thread will indeed help you understand where capacity use of your application is now.
However, capacity planning goes beyond the now and looks into the future.
The easiest (and the most problematic) way to plan for increase in usage is to do straight extrapolations. For instance, you have information on system usage running a simulation of 100 users. Your systems are server a, b and c.
Now you want to plan for 200 users. You can calculate how many servers of the same configuration of a,b and c you would need to support that load. But you do not know how many servers you would need if you changed configured like doubled processors, got faster processors etc.
You simply cannot make assumption that a two-processor machine will do double the work unless you have done testing on a two-processor machine with the same load.
Another important issue in capacity planning is to identify the bottleneck in your system and when the bottleneck will cause you to upgrade your capacity. In my case we quickly found (by extrapolation) that our network bandwidth to the Internet would never be enough for the future plans of the company, so we upgraded from 1.5 Mbs to 10 Mbs pipe.
Secondly we found that our web servers were just doing fine but our business servers were used the heaviest. Initial extrapolations showed that we would have to bring in 6-10 (see how fuzzy the predictions become!) times the amount of servers to cope with the anticipated load.
This triggered a re-architecting of our Internet environment that included load balancing and clustering, App center, network appliances, SSL accellerators etc.
An initial environment will be created according to this new architecture and we will be load testing/monitoring and capacity planning again to refine the needs.
Another challenge in our environment is that we run many applications and that we have to make a choice which set of applications we use for simulating a load with (sadly we lack the resources to do all). Some applications are big resource users and others are pretty light. Related to this is the simulation and the type of scenarios you want to run. Your scenarios should resemble actual usage as closely as possible. Check out Scott Barber's articles on this (important) topic.
I have included my set of counters and guidelines that we use when monitoring. This is from our internal performance testing manual.
I hope this rambling post helps you.
------------------
Roland
[This message has been edited by rstens (edited 09-17-2002).]
posted 09-16-2002 01:55 PM
