Core Migration

[tommy-u]

Running memcached while shortcutting reads and writes without taskset occasionally barfs. Is this related to core migration? Could this be related to failure to task switch the GS register on core migration?

Should we work around this? Solve it in Symlib? Solve it on a kernel path?

[handong32]

Why would you not want to run taskset for your experiments? Did you pin the network queues to specific cores/disable hyperthreads? Technically, it should be pinned all the way from network receive to processing on the specific core, unless the kernel is somehow scheduling network processing received on one queue to a separate core. But that'd violate your initial network queue-> core pinning

[tommy-u]

Good q. In our development, memcached runs fine when taskset to a single core. When it can migrate, we run into trouble (probably even if we taskset it to 2 cores). This is a correctness issue, not a performance one. Once we've nailed the correctness problem, we'll need your net expertise for tuning the apps. @ryanbsull probably has more details