cnfs cycbuff problem

Nicholas Geovanis nickgeo at merle.acns.nwu.edu
Tue Apr 11 22:49:40 UTC 2000


On 11 Apr 2000, Russ Allbery wrote:

> Brian Kantor <brian at UCSD.Edu> writes:
> 
> > As a data point, 'cnfsstat -a' on my news transit system ran for about
> > 88 minutes elapsed time before I killed it.  It hadn't come up with an
> > answer yet at that point.
> 
> > Ok, so I have a 400MB history file and two cnfs buffers of 16GB each it
> > had to wander around in, but still...  (BSD/OS 4.1,
> > inn-STABLE_2_2-2000-03-22_03-07, about 2.6 GB/hour incoming)
> 
> 700MB history file, 14 1GB cycbuffs, and cnfsstat -a finished nearly
> immediately for me and found the oldest article in every case.

I've only just glanced at the code, but it seems to me that results will
be very dependent on OS and architecture as well as specific inn
configuration. I'd guess that the only ways around it would be file-level
locking (unacceptable performance hit) or "real-time" info gathering and
reporting by innd (not a quick fix, I would guess, and maybe another
perfomance hit).

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

+-------------------->
| Nick Geovanis            Indeed I tremble for my country when
| IT Computing Svcs          I reflect that God is just: that his
| Northwestern Univ          justice cannot sleep forever...
V n-geovanis at nwu.edu         - Thomas Jefferson, 1781




More information about the inn-workers mailing list