[imp] IMP 4.0 slower than 3.0

[Didi Rieder]

Quoting Steve Lidie <sol0 at Lehigh.EDU>:

> If/when someone else makes I4/H3 production, please feel free to
> contact me off-list before/during/after to discuss performance issues.
> As it is, I dare not try again until I have more data, yet cannot
> collect data w/o trying a production mode run again, which I
> won't/can't because the semester is starting.

Hi Steve,

I put a short hardware overview of our system into Horde's wiki:
http://wiki.horde.org/display.php?page=Deployments
Look at Graz University of Technology.

we are running Horde 3 / Imp 4.1-cvs

as you can see, we are using the following Hardware:

webservers:
   - Sun Fire V280R, 2 x 900 Mhz UltraSparc III, 4 Gb Ram
   - Sun Fire V480R, 2 x 900 Mhz UltraSparc III, 4 Gb Ram
   Both running under Sparc Solaris 9 with:
      - apache 1.3.33
      - php-4.3.10
      - Zend Performanc Suite for PHP accelleration

   Both are behind a cisco loadbalancer (CSS11503), which has a crypto-card for
   SSL-accelleration. This means the SSL connections are handled by this card
   and terminated there, the loadbalancer then speaks normal HTTP to 
the backend
   servers.

mySQL:
    - Sun Fire V480R, 4 x 1200 Mhz UltraSparc III, 8 Gb Ram
    running under Sparc Solaris 9 with:
      - mySLQ 4.0.21 (query cache enabled)

    This server is used for the horde preferences, vfs, addressbooks... system

imapproxy:
     we have installed the up-imapproxy on the webservers, for imap-connection
     caching

imap-servers:
     - Sun Fire V280R, 2 x 900 Mhz UltraSparc III, 4 Gb Ram
     - Sun Fire V480R, 2 x 1200 Mhz UltraSparc III, 8 Gb Ram (runs the mySLQ-db
       for horde, see above)
    running under Sparc Solaris 9 with:
        - cyrus-imapd 2.2.10
        - sendmail-8.13.2
        - mimedfang
        - spamassassin
        - clamd

     The V280 serves for about 2500 employees
     The V480 serves for about 9000 students

     Both servers store the mail on a SAN to which they are connected via FC.

There are also about 30 small servers from institutes which can be 
accessed via
the same Horde/IMP instance, but the are not producing much load because the
are not heavily used.

The system handles about 1800 concurrent users, with a acceptable speed 
(message
viewing takes about 1 second)

I should also mention that in the past we were running the system without the
cisco loadbalancer, we did just a DNS round-robin for load sharing between the
2 webservers. SSL was then also done be the webservers themselves. The speed
was still acceptable.

The biggest performance boot can be gained by a php accellerator, like ZPS,
turck-cache, phpa and so on. An imapproxy is also a good idea.

Hope that helps...

Didi

-- 
-------------------------
Didi Rieder
adrieder at sbox.tugraz.at
PGPKey ID: 3431D0B0
-------------------------