Hi Marco:
There are a number of places this could be, and not all of them are Prime Infrastructure's fault.
First, does an SNMPWalk of the cpmCPU* objects from the CISCO-PROCESS-MIB agree with the output of "show memory statistics history". If they don't, then the problem is in the SNMP agent on the device.
If they do match, when you change the Memory Utilization portlet from a graph to a chart, are all the numbers the same or are they different? These numbers are polled by Prime Infrastructure from the device and stored in a database table. The chart view is the portlet looking directly into the database tables, while the graph view involves drawing and that sort of thing. If the numbers in the chart don't match what those from your SNMPWalk earlier show, then something's up with the polling and how Prime Infrastructure is writing those values to the database tables every time the polling task runs.
There are also situations where folks who install their own certificates into Prime Infrastructure to avoid security messages have gotten the internal certificates out of sync, so the portlet is not able to get logged into the database to get real data, so it kind of sticks on what it last had.
In any of these situations, it's probably something you would want to open a TAC Service Request to have looked into.
If you don't have the new Update-1 patch for Prime Infrastructure 1.3.0.20 installed (filename PI_1.3.0.20_Update_1-12.tar.gz) installed, I'd suggest you get it. While it's not going to specifically address this issue, there are a lot of really good fixes in it.
