Hi, I''m doing researches on remus and have a question about its network buffering. I use the code in the xen unstable tree. I''ve run a guest and remus with network buffering(without --no-net), using the default interval 200ms. Then I ping the guest on the 3rd machine(other than primary and backup). I found that the average delay is about 30~50 ms, but I think it should be larger than 100ms.(due to network buffering, the output packet could be held for 0~200ms) Did I make anything wrong with the idea of network buffering? Or did I make any mistake during the test? Thanks, -- Frank Pan Computer Science and Technology Tsinghua University _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Monday, 23 August 2010 at 13:40, Frank Pan wrote:> Hi, > > I''m doing researches on remus and have a question about its network buffering. > > I use the code in the xen unstable tree. > I''ve run a guest and remus with network buffering(without --no-net), > using the default interval 200ms. Then I ping the guest on the 3rd > machine(other than primary and backup). > I found that the average delay is about 30~50 ms, but I think it > should be larger than 100ms.(due to network buffering, the output > packet could be held for 0~200ms) > > Did I make anything wrong with the idea of network buffering? Or did I > make any mistake during the test?You''ll probably want to test with something a bit more systematic than ping. If ping sends a packet to the protected machine just before the end of an epoch, the response won''t be buffered very long at all. On the other hand, if ping sends its packet at the start of an epoch, the response will be delayed for the entire epoch length. By default, ping sends packets infrequently, and the delay between pings isn''t particularly rigid, so you''ll get jumpy results with it. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Tue, Aug 24, 2010 at 12:39 AM, Brendan Cully <brendan@cs.ubc.ca> wrote:> You''ll probably want to test with something a bit more systematic than > ping. If ping sends a packet to the protected machine just before the > end of an epoch, the response won''t be buffered very long at all. On > the other hand, if ping sends its packet at the start of an epoch, the > response will be delayed for the entire epoch length. By default, ping > sends packets infrequently, and the delay between pings isn''t > particularly rigid, so you''ll get jumpy results with it. >Thanks for reply. After some debugging I found the buffering is not working. I use an hvm guest with qemu-simulated NIC, the active interface is tapX.0, not vifX.0. I''ve modified remus code, replace "vif%d.0" with "tap%d.0", but it does not work. Do you know how to make it working? I also tried pv NIC on hvm, but it seems remus does not support "PV-on-HVM" guest. Is it hard to do, if I want to add this feature? Can you give me some advices? Thanks. -- Frank Pan Computer Science and Technology Tsinghua University _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel