Hi there. That is an interesting problem you describe. No obvious explanation for it comes to my mind. Over Ethernet I'd typically expect to see higher goodput from UDP than TCP, especially where packet loss is involved. I would recommend collecting some traffic stats from either endpoint & try and isolate where exactly the packets are being dropped. netstat -i is a quick way to check for tx/rx drops at either end. capturing a trace with wireshark may provide some helpful clues. I'd also check the system log for any error messages that might be relevant...possibly from the NIC...normally...not as certain about behavior of the virtual hardware in this regard. did you use iperf for you TCP and UDP tests? MikeR. On Sat, Mar 6, 2010 at 3:02 PM, mpsingh <mpaul7 [ at ] gmail [ dot ] com> wrote: > Hi all, > > I am experimenting with virtual machines existing in the same physical > machine. > > My configuration is:- Native: opensuse,/xen and tool used is "iperf" > > I am getting TCP throughput of around 8Gbps between two VM. but the UDP > throughput is around 1.53 Gbps at data rate of 3000Mbps. but with this the > data loss is around 20-30% > > I tinkered with linux kernel settings to tune the TCP sockets > r|w _mem = 4096 12288 262144 > r|w _ default = 12288 > r\w_max = 2262144 > RTT = 0.096 ms > txqueuelen = 10000 > backlog queue= 2000 > > the above setting are done in both VMs and in the Dom0. > > Earlier I got UDP performance of 1.8 Gbps with negligible data loss. I dont > know those settings in the kernel. > I am not getting the optimum settings to get around this problem. > > Thanking you in anticipation > -- > M P Singh > Carleton University, Ottawa > +1 613 422 8074 > _______________________________________________ > Linux mailing list > Linux [ at ] lists [ dot ] oclug [ dot ] on [ dot ] ca > http://oclug.on.ca/mailman/listinfo/linux >