On Thu, 1 Feb 2018, Jarle Thorsen wrote: > Ilpo Järvinen: > > I'd next try to tweak the txqueuelen (at the netvm side): > > sudo ifconfig vifxx.0 txqueuelen xxxx > > > > Appvm side (eth0) seems to have 1000 but the other side (vifxx.0) has > > only 64 by default that seems a bit small for high-performance transfers. > > Thanks a lot for your help so far! > > Unfortunately setting txqueuelen to 1000 did not make any difference...
I found this: https://wiki.xenproject.org/wiki/Xen-netback_and_xen-netfront_multi-queue_performance_testing It might be that roughly 4Gbps might be what you can get for cross-vm with one flow (but those results are quite old). I guess that there are by default two queues per vif (based on the kernel thread naming) which would explain why going beyond 2 VCPUs didn't help any. ...Now we only just need to figure out how to configure the number of queues to see if that has some impact on performance (ethtool seems to be unable to do that). ...And of course you'll need to use more than one flow to utilize all those queues anyway but I guess you tested also the inter-vm test with more than one flow? -- i. -- You received this message because you are subscribed to the Google Groups "qubes-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to qubes-users+unsubscr...@googlegroups.com. To post to this group, send email to qubes-users@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/qubes-users/alpine.DEB.2.20.1802011102270.18034%40whs-18.cs.helsinki.fi. For more options, visit https://groups.google.com/d/optout.