When using RH 6.2, we were able to achieve 459k messages per second for a 10 minute duration. Roughly 3.17% of the CPU
was idle. This is on a HP G8 24 cpu machine with 32 gig of memory. We are using 11 NICs. When we upgraded to RH 6.5, the best we can do now is 275k message per second with 4.38% of the CPU idle.
Traffic over the NICs uses TCP. We have done an apples-to-apples comparison by have two separate disks - one with RH 6.2 and the other with RH 6.5 everything else on the disks are identical to the best of our knowledge. We put in the 6.2 disk and get 459k. We put in the 6.4 disk and get 275k.
The majority of systems with IGB drivers will not experience this issue. Please note this issue only happens under very specific circumstances. Particularly when there are multiple IGB interfaces, in this case 8 and an many clients that are driving a multi-threaded application which uses almost 100% of the systems CPU. No performance degradation was noticed with iperf and other standard performance tools or test applications.
- Red Hat Enterprise Linux 6.5
- Intel IGB driver
- HP G8 24 cpu
Subscriber exclusive content
A Red Hat subscription provides unlimited access to our knowledgebase of over 48,000 articles and solutions.