CNFS buffers bigger than 2 GB under Linux ?

Nicholas Geovanis nickgeo at merle.acns.nwu.edu
Mon Mar 18 21:35:03 UTC 2002


On 18 Mar 2002 dsr+inn at mail.lns.cornell.edu wrote:

> 
> Russ Allbery <rra at stanford.edu> writes:
> > Antoine Delvaux <antoine.delvaux at belnet.be> writes:
> > >> Well, you can just use large disk files.
> > 
> > > Yes, but this will slow down innd I think, no ?  That's why I wanted to
> > > use raw devices.
> > 
> > It may slow it down a bit, but we've not run bechmarks.  I doubt it will
> > be particularly noticeable.
> 
> I haven't benchmarked, but my guess would be that there should be very
> little slowdown with an extent based file system (e.g.  SGI's xfs or
> Veritas VxFS).  

FWIW, last I looked, raw devices are not supported by INN on HPUX 10.20,
so we use "real" files for cycbuffs on the transport server. Total innd
throughput is about 1x10**6 (one million) incoming articles per day and
about 3x10**6 outgoing articles per day. This is an HP9000/813 with 128MB
and two NICs; a four-year-old, 120 MHz, single-CPU, PA-RISC machine.
Generally, load average is below 0.60 or 0.70.

Also FWIW, Veritas VxFS is a performance disaster on HPUX with the
traditional or cycbuff news storage. There is an HP white-paper on
filesystem performance which states that small-block I/O is faster
on HFS than VxFS, and (here at least) about 80% of news articles are
smaller than 4K. So for both reader and transport machines, all news
filesystems are HFS here.

* Nick Geovanis              "The nuclear bomb. Does that bother you?
| IT Computing Svcs              I just want you to think big"
| Northwestern Univ                 - Pres. Richard M. Nixon
| n-geovanis at nwu.edu                    April 25, 1972
+------------------->



More information about the inn-workers mailing list