Profiling nnrpd

Russ Allbery rra at stanford.edu
Mon Dec 23 02:54:53 UTC 2002


Jeffrey M Vinocur <jeff at litech.org> writes:
> On Sat, 14 Dec 2002, Jeffrey M. Vinocur wrote:

>>   4.20      0.64     0.20    tdx_search (tdx-data.c:441)
>>   3.99      0.83     0.19    tdx_search (tdx-data.c:442)

> The results for that continue to puzzle me.  I mean, assignments that
> look just like every other inexplicably take a lot of time.  Here's the
> no-nnrpdcheckart stats, and it's even odder -- look at the last few
> lines of tdx_search()!

>  15.93      0.18     0.18    tdx_search (tdx-data.c:441)
>  13.27      0.33     0.15    tdx_search (tdx-data.c:442)

This is what I'd expect.  Those lines are:

    artdata->token = entry->token;
    artdata->arrived = entry->arrived;

where entry is a pointer into memory-mapped space.  In other words, those
assignments represent disk reads.  (Although I'm surprised that all the
time isn't born by the first assignment.  I would have thought that the
first assignment would have pulled that page into memory, and then any
subsequent reads of the same page would be very quick.)

-- 
Russ Allbery (rra at stanford.edu)             <http://www.eyrie.org/~eagle/>

    Please send questions to the list rather than mailing me directly.
     <http://www.eyrie.org/~eagle/faqs/questions.html> explains why.


More information about the inn-workers mailing list