-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

David Sparks writes:
> > oh, btw, I've just released IPC::DirQueue 0.05, which includes support for
> > the enqueue_sub() API as used in http://taint.org/wk/QpsmtpdIpcDirqueue .
> 
> I'm getting ~5% (whoopee!) better throughput by calling the
> enqueue_string method with a full message.  Tested with 11k and 50k
> messages.  This is with the queue living on a NFS mounted partition and
> mounted locally.

Hmm!  I'd guess that it's faster, because when you enqueue as a string it
performs 1 write() call, compared to 2 with the sub variant -- combining
the write() system calls into one has a big effect that way when NFS is
involved.

> 11k message throughput (50k had a similar ratio):
> enqueue_sub local:    5500 msg/min
> enqueue_sub nfs:      2800 msg/min
> 
> enqueue_string local: 6000 msg/min
> enqueue_string nfs:   2900 msg/min
> 
> Unless there is a misteak :) with the way I'm doing it.

Nah, looks correct to me.  I'd expect to see spamtrap going faster, given
what Matt describes.

The key thing about dq is that it won't go much faster on a single box;
however, if you add more boxes writing to the one backend, it'll still
work, since they won't scribble on each other's stuff ;)

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFCaalZMJF5cimLx9ARAu/3AJ9PWw7/TF2WjkbCswWCbi+rXziDdgCghwO+
A3+DBhj7zE9u9C1UZK+Z02o=
=8g0A
-----END PGP SIGNATURE-----

Reply via email to