[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] Is CPU overhead really independent of packet size in Xen?

I'm reading some Xen performance slides which is published on XenSummit (www.xensource.com/files/summit_3/perf-isolation-gupta.pdf). The presenter claims the CPU overhead is independent of network packet size in Xen VM domains. I haven't done much research on network I/O on Xen, but I have done many experiments on disk I/O.

Based on my disk I/O experimental results, I found out, generally the CPU utilization decrease as the I/O packet size increase which is consistent with the IPv4 forwarding experiments I did before (not on Xen, on plain IA32 Linux). The only exception I noticed is DomainUF. For both small size and large size sequential I/O, the CPU utilization in DomainUF remains almost constant (100%) while CPU utilization drops for 4k and 8K random I/O.

If anyone in this mailing list has done similar experiments, please adivse the difference when considering CPU utilization for network I/O and disk I/O? or should both network I/O and disk I/O have the same curve of CPU utilization vs. Packet Size?



Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.