> On Jun 19, 2016, at 6:46 PM, Nigel W <[email protected]> wrote: > > On Sun, Jun 19, 2016 at 12:33 PM, Dale Ghent <[email protected]> wrote: > One thing I'd like to see people who have ixgbe hardware (X550 and anything > else) is RSS. A good smoke test is to open up a high-traffic TCP stream > between the ixgbe box and another, with the ixgbe box being the receiver. A > tool for this is iperf. > > In another window, watch interrupt count for ixgbe using 'intrstat 1' - you > should see a ton of interrupts from ixgbe landing on one particular CPU. > > Then, open up a second (or third, or fourth) such stream and you should see > the interrupts for those land on a different CPU, as show with intrstat. > > Since receive-side scaling was one of the major areas I had to re-work to > accommodate X550, this basic test would be a quick (albeit not exactly > scientific) way of making sure it's working as intended. > > It does seem to be spreading, though not evenly or to all CPUs (4 of 8) but > that may be because I am only able to put 250 Mbit/s at it or because I am in > a zone
It's not meant to spread traffic evenly; it's meant to pin rx interrupts for *a specific* connection (hashed by source IP and source port) to a particular, single CPU. Zones have no bearing on this aspect. Why only 250Mb/s? Is this a local limitation of some sort? /dale
signature.asc
Description: Message signed with OpenPGP using GPGMail
------------------------------------------- smartos-discuss Archives: https://www.listbox.com/member/archive/184463/=now RSS Feed: https://www.listbox.com/member/archive/rss/184463/25769125-55cfbc00 Modify Your Subscription: https://www.listbox.com/member/?member_id=25769125&id_secret=25769125-7688e9fb Powered by Listbox: http://www.listbox.com
