[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: Much Slower than 7.2?

On Tue, Dec 24, 2002 at 11:07:39AM -0600, Ronald W. Heiby wrote:

> Hash: SHA1
> Monday, December 23, 2002, 2:38:57 PM, Ronald wrote:
> > As an experiment, I've copied my raw data and compilation app to
> > another system with 1 GB of RAM, a 4 GB swap partition, and SCSI
> > drives that show about 50 MB/sec buffered read throughput. It will be
> > interesting to see how fast that finishes. The original system is
> > an HP desktop box, running with a 1.6 GHz P4. This other system is a
> > Dell server configuration running a pair of 1 GHz P3 (Coppermine)
> > processors. (The data compile process is not multi-threaded.) Anyway,
> > I would guess that if cpu bound, the experiment will take longer, but
> > if memory/disk/swap bound, the experiment will take less time.
> The experiment concluded. The dual P3 system with SCSI drives is
> running RH 8.0 kernel 2.4.18-14smp, versus the P4 system running RH
> 8.0 kernel 2.4.18-18.8.0. The elapsed time was 11 hours 7 minutes.
> For the experiment, ran the data compilation under "time -v", and
> learned that the job got 5% of the cpu time during its run. This says
> to me that cpu time is not the main factor in performance. There were
> several million page faults. Since the same executable was used both
> places, it's not likely to be a compiler problem. (The executable was
> *not* statically linked.)
> In summary, using the same hardware (P4) on ext2 filesystems, my job
> went from less than 7 hours to more than 48 hours. At the same time,
> the job runs in about 11 hours on a dual P3 with SCSI drives and a bit
> more RAM. The differences I see that might be accounting for the big
> performance hit:
> 1) Kernel Version (memory management / paging, code and algorithms)
> 2) 8.0 increased memory footprint vs 7.2 (???) leading to more paging
> Any ideas?

Is your process doing lots of disk I/O? If so, tell us about the
filesystem. Is it ext2 or ext3? Or something else? I've seen filesystem
choice give drastic performance changes on high disk I/O processes.

Javier Gostling                        Av. Kennedy 5757, of. 1502
Ingeniero de Sistemas                  Las Condes, Santiago, Chile
Virtualia S.A.                         Fono: +56 (2) 202-6264 x 130
jgostling virtualia cl                 Fax:  +56 (2) 342-8763

Attachment: pgp00066.pgp
Description: PGP signature

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]