Ok, thanks for the answer. Just another question, is also true for packets with two VLANs because i'm seeing that they only arrive on the first core.
Thanks, Sorin On Fri, Apr 6, 2012 at 8:05 PM, Tantilov, Emil S <[email protected]> wrote: > Hi Sorin, > > Unfortunately there is no way to parse the headers past the first. > > One thing you may want to check out is using ethtool's ntuple option that > allows you to set filters and steer traffic to different queues. However > since your traffic is tunneled - I am not sure if this would help much. > > Thanks, > Emil > >>-----Original Message----- >>From: Sorin Dumitru [mailto:[email protected]] >>Sent: Wednesday, March 28, 2012 2:03 AM >>To: Tantilov, Emil S >>Cc: [email protected] >>Subject: Re: [E1000-devel] ipip tunnels on 82599EB with ixgbe driver >> >>We are using version 2.0.84.9 of the out of tree driver. >> >>On Wed, Mar 28, 2012 at 2:55 AM, Tantilov, Emil S >><[email protected]> wrote: >>> Sorry for the late reply. >>> >>> Could you provide the exact version of the driver you are using? >>> >>> Is it the built-in driver in the kernel or an out-of-tree from SF? >>> >>> Thanks, >>> Emil >>> >>>>-----Original Message----- >>>>From: Sorin Dumitru [mailto:[email protected]] >>>>Sent: Thursday, March 22, 2012 12:44 AM >>>>To: [email protected] >>>>Subject: [E1000-devel] ipip tunnels on 82599EB with ixgbe driver >>>> >>>>Hi, >>>> >>>>We are trying to run a large number of ip in ip(ipv6 over ipv4 or ipv4 >>>>ove ipv6) tunnels over intel cards >>>>with the ixgbe driver. This is what lspci is showing about these cards: >>>>04:00.0 Ethernet controller: Intel Corporation 82599EB 10-Gigabit >>>>Network Connection (rev 01) >>>>04:00.1 Ethernet controller: Intel Corporation 82599EB 10-Gigabit >>>>Network Connection (rev 01) >>>> >>>>I think that the hardware isn't able to parse the packets after the >>>>first IP header so the only hashing it >>>>uses to distribute packets to receive queues is done based on the >>>>first IP header. Because that header >>>>is the same for all packets, all packets will arrive on the same cpu, >>>>which doesn't scale very well for us. >>>>I've attached what /proc/interrupts is showing, as you can see all >>>>packets arrive on CPU6. >>>> >>>>Is there some way we can convince the hardware to distribute these >>>>packets on all cpus? If we could at least >>>>include the second ip header, we would get a much better distribution >>>>because there are many users on these >>>>tunnels. I know we can do this from software on linux, but we are >>>>using linux 2.6.32, which doesn't have this >>>>option and upgrade isn't really an option right now. >>>> >>>>We are also using a rather old driver(i think it's a version from >>>>early 2010), have any improvements >>>>been made in newer versions on this issue? >>>> >>>>Thanks, >>>>Sorin ------------------------------------------------------------------------------ For Developers, A Lot Can Happen In A Second. Boundary is the first to Know...and Tell You. Monitor Your Applications in Ultra-Fine Resolution. Try it FREE! http://p.sf.net/sfu/Boundary-d2dvs2 _______________________________________________ E1000-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/e1000-devel To learn more about Intel® Ethernet, visit http://communities.intel.com/community/wired
