New Server Worse Performance

A customer bought a new Linux server and migrated their application to the new machine.

The new machine was beefy, bought by their purchasing agent and cost about $20,000.  They bought 5 of these machines, production, test, training, disaster recovery, etc.  When they migrated to the new machines, they were surprised that the system was running slower than their old server which was 10 years old. 

They tried everything to no avail.

OmegaServe was called in to help.  The DB Health Check showed no significant deficiencies.  The memory allocation was good, the disk speed was sufficient, there was plenty of CPU available to the application.  Quite the mystery, until the “load” on the system was looked at.

The system had 32 CPUs on an Intel Linux server.  The system “load” indicated that it would require 80 CPUs to accomplish the load it was under.  Most of the CPU time was being consumed under “system” not “user”. 

Through experience, the best systems run with 60% user, 30% system %10 idle.

This machine was clearly NUMA machine, which is a topic for a different time. 

One the test machines was “borrowed” and benchmarks were run on it, where CPUs were turned off.  Without changing anything, no configuration to OpenEdge or code changes, OmegaServe was able to get performance exceeding 5 times what their old system was capable of doing by configuring the CPUs correctly.