We are running CentOS 6.3 currently. 16GB mem each. AE v9 sp11 hf2, with eyes on getting to 11.2 predominately for zero-downtime future upgrades.
We have 8 cores per app node, but most go lightly/un used. There is another thread about this
HERE. Maybe it's expected that only 1 CPU will take brunt of load, but makes our infrastructure resource folks anxious seeing 7/8 cores barely being used. The graph Chip pasted on that other thread was before we doubled our WPs. Having checked today, average usage for cpu0 has increased significantly to 40-50% (up from 20-25%), while others still hover below 5%. Not ideal, but cpu0 can be explained by doubling WPs I suppose. Other than this, can't think of any issue we have had regarding performance at the VM/OS level. Memory does seem to climb a bit over time, but as they say, unused memory is wasted memory and we're fine so far.
We run about 100,000 jobs per day.
We increased our DWPs because we were having frequent complaints of user actions taking a very long time to complete. Indeed, we noticed it too, and yes you can see this in System Overview > AE--we would see "100" for all DWPs. Part of it was a known bug in the software version we were in, which we've patched now. So, if you have a few people who ask for stats at the same time and it takes 10-20 seconds or more, those actions bottleneck your DWPs and no other user requests will be processed in the mean time. 10 DWPs + the recent patch has shown a dramatic improvement. We have 50-75 users typically.
Our WP activity spikes a bit on the hour, since many teams "Force adjustment on timeframe," but otherwise only a couple WPs seems like enough.
For us, the database seems like where we need to focus to increase performance. Even after the patch, we think user requests could be more responsive. So we're
working on tuning the DB now to get getstatistics() calls and getactivities() calls lower.We've added some indexes to improve DB Maintenance. We've shrinked some tables, and still working on reorganizing/rebuilding. Apparently it's tricky to push these kinds of things through on an active DB.
What do you mean by
administrative overhead?
Hope this helps!