17:11:26 <jbk> is it possible to 'bind' (for lack of a better term) a TX ring to a cpu?
18:28:05 <gitomat> [illumos-gate] 15513 zfs: Add zfs create dryrun -- Mike Gerdts <mike.gerdts⊙jc>
19:18:36 <gitomat> [illumos-gate] 15992 SMB: FileAllInformation missing FileAccessInformation -- Gordon Ross <gwr⊙rc>
19:27:02 <sommerfeld> jbk: didn't know the answer to that for illumos but went looking (haven't found it yet).  a strict cpu-> tx queue mapping creates the potential for reordering in the driver if a flow switches tx queues because the sender switched cpus.
19:28:44 <sommerfeld> (for TCP, reordering is not a correctness problem as the receiver puts things back in the right order, but it can induce performance blips and spurious retransmissions).
20:15:09 <jimklimov> jbk: "One ring to bring them all, and in the darkness bind them" - *that* lack of better term? ;)