A Node.js implementation of the Acme Air Sample Application. With datastore support of MongoDB, Cloudant, Cassandra. With runtime support of Bluemix/CloudFoundry, Docker... With Micro-Services.
Apache License 2.0
109
stars
74
forks
source link
Seeing very low CPU utilization on Haswell Server running Ubuntu #14
I am running AcmeAir on a Haswell server with 36 cores (Xeon E5-2699 @ 2.30GHz) with 64 bit Ubuntu 16.04. I'm using the JMeter driver with default options except I set the number of threads to 36, running for 10 minutes. On the server, if I run vmstat or sar I see approximately 8-9% CPU utilization by user, 1% by system, and 90-91% CPU idle.
Setting the number of threads down to 1 drastically lowers the throughput, but raising it to higher numbers seems to have no effect somewhere in the 5-10 thread range. Both vmstat and sar report the same numbers across different runs. Is is expected of the workload? If not, would this be caused by not setting certain options to raise the CPU utilization?
I am running AcmeAir on a Haswell server with 36 cores (Xeon E5-2699 @ 2.30GHz) with 64 bit Ubuntu 16.04. I'm using the JMeter driver with default options except I set the number of threads to 36, running for 10 minutes. On the server, if I run vmstat or sar I see approximately 8-9% CPU utilization by user, 1% by system, and 90-91% CPU idle.
Setting the number of threads down to 1 drastically lowers the throughput, but raising it to higher numbers seems to have no effect somewhere in the 5-10 thread range. Both vmstat and sar report the same numbers across different runs. Is is expected of the workload? If not, would this be caused by not setting certain options to raise the CPU utilization?