perfcounter interpretation

[David Becker <becker@cs.duke.edu>]

I turned on the perf counters for the first time and ran some iperf tests
to try them out.

This is with yesterday's rev of 2.0-testing with ignorebiostables=1.
iperf is run on xen0.  No xenUs exist, xend is not running, and the
etherbridge is not installed.

The 10 second tx test sent 260MB in 200K packets, and got 95K ACKs back.
xenperf says calls_to_mmu_update and num_page_updates both went up about 180K.
So I guessed these counters should go up once per tx packet.
map_domain_mem went up 95K.  That goes up once per rx packet?

Apparently not though.  The 10 sec rx test also moved 260MB, receiving
180K packets but only sending 30K ACKs (from the ifconfig stats; TCP
must be coalescing lots of ACKs whenever it can squeeze and ACK out).

This time xenperf says calls_to_mmu_update and num_page_updates again
went up 170K, which would seem to correspond more with the rx packets
than tx as I first thought.

map_domain_mem went up 85K, which doesn't match anything.

So what should the perfcounters do while xen0 streams packets?

For each test, about 100K interrupts occured (these are the real PCI intrs?)
and there were 900 apic timer interrupts, 2500 context switches, and 7000 runs
thru sceduler.


-------------------------------------------------------
The SF.Net email is sponsored by: Beat the post-holiday blues
Get a FREE limited edition SourceForge.net t-shirt from ThinkGeek.
It's fun and FREE -- well, almost....http://www.thinkgeek.com/sfshirt