On Aug 26, 2009, at 10:58 AM, Vishal Ahuja wrote:
> Hi Donour,
>
> Thanks for the reply. So what I understand is that I
> could measure the time spent in each of the syscalls - starting from
> ip_input, all the way down to e1000g0_intr. Is that correct?
Sort of, but it''s not so simple. If you use the wall clock to measure
a socket read(), then some of the time will be spend waiting for data
to arrive over the network. My impression is that you really only want
time spent on the cpu, not time blocked.
FYI: ip_input() is a kernel function, not a system call.
> Also, would you know how to take into account the copy from the
> kernel space to the user space for the packets...
Yes. I don''t have a nice example in front of me, but if you trace a
recv(), you should see a so_copy function. That''s the ticket....I
think. :-p
donour