I have several PowerEdge 1950 Generation III units with identical set-ups (clean Ubuntu installations, Java 8). One particular unit which happens to have 52GB of memory takes about twice as long to perform an identical task as the other units which have 16GB. Each unit has a swap space that is the same as the memory size. All hard disks are the same size and probably even the same model.
The task and data are self-contained. I send (on a local subnet) the task to each unit and wait a few minutes for the result which is sent back to the "master" application running on a different server. The task does not rely upon a database, nor internet, and only a trivial amount of hard disk access. There is no need to use the local subnet in the period when the task is "doing the work". vmstat says there is no swapping happening on any of the units.
BIOS reports the correct memory speed, correct bus speed, correct number of processors, correct cache size, and correct clock speed. Ubuntu reports the correct clock speed and the correct number of "processors" which is really the number of cores. The Intel Processor Diagnostic Tool says this unit "passes". I ran memtest86+ for a few minutes (until the end of Test # 4) and it was fine. The Dell System E-Support Tool (DSET) does not seem to do anything but collect information for several minutes.
These units are not on Dell support.
What else should be checked?
The solution was found in the users' manual. It says the 8 DIMM slots must be populated in identical pairs and the total should be either 2, 4, or 8 DIMMs for optimal performance. The population of 52GB happened to be 6 DIMMs (6 x 8GB, 2 x 2GB) so it does not provide optimal performance.