Red Hat Bugzilla – Bug 1282826
hackbench: modify to run within NUMA node to avoid cross-node traffic
Last modified: 2016-11-03 21:34:16 EDT
the hackbench load excercises memory allocation/free and memory copies. Originally rteval started it's loads across the entire set of cpus, but this is not consistent with good RT programming practice as it generates large amounts of cross-node memory traffic, which leads to latency spikes.
Modify hackbench to run one instance for each numa node in the system, rather than one instance for the entire system.
Created attachment 1095714 [details]
patch to prevent hackbench generating cross-numa-node traffic
Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.
For information on the advisory, and where to find the updated
files, follow the link below.
If the solution does not work for you, open a new bug report.