Crystal 8.5 ad probably 8.0 as well, have memory usage limitations that are application specific.
Memory usage for the APS.exe is a combination of logged in users, number of report instances, number of recurring instances and number of existing user accounts.
Above 1.6GB of memory usage, Windows NT / 2000 the OS gets flakey. CE 8.5 will not use more than 2GB of RAM and without a bootloader entry, Windows 2000 will not use more than 3GB of total memory.
We put the switch in, but when taking in just over 100,000 instances, the APS.exe service eventually shuts down on its own. It gets to just above 1.8GB of memory usage. We backed it down to 60,000 instances and it settled in at around 1.2GB to 1.4GB of memory, and runs fine.
Our reasoning for this is that we have over 100,000 report instances, 15,000+ users and over 10,000 recurring reports. Pulling the metrics on the system at the default value of 10,000 instance limit usually fails due to the 9 minute default timeout in CE. So when trying to manage objects, users or pull metrics, it would take from 6 to 9 minutes to get the data, if it returned at all.
Our thinking was to pull all of the instances into memory and therefore speed up the use of these tools, and make them functional again. At the 60,000 level. It definitely speed up the tools to 1 minute or less.
Our hardware is as follows:
2 x Web servers (Dual 2.8GHZ Xeon, 4GB RAM)
2 x WCS servers - (Quad Xeon 2.0GHZ, 4GB RAM)
2 x APS servers (Clustered) - (Quad Xeon 2.0GHZ, 4GB RAM)
2 x JSPS servers - (Quad Xeon 2.0GHZ, 4GB RAM)