|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
Subject: Re: high capacity configuration nightmares
From: Wietse Venema (wietse
porcupine.org)Date: Thu Oct 26 2000 - 13:30:51 CDT
- Next message: Len Conrad: "other than 25"
- Previous message: Drew Bloechl: "Re: high capacity configuration nightmares"
- In reply to: Pere Camps: "Re: high capacity configuration nightmares"
- Next in thread: Pere Camps: "Re: high capacity configuration nightmares"
- Next in thread: Drew Bloechl: "Re: high capacity configuration nightmares"
- Reply: Wietse Venema: "Re: high capacity configuration nightmares"
- Reply: Pere Camps: "Re: high capacity configuration nightmares"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Pere Camps:
> time ./smtp-source -s 50 -l 10000 -m 1000 -c -f me
me.com -t
> deleteme
localhost localhost
>
> syslog: async
> filesystem: sync
>
> machine: dual pentium III, 733 Mhz, .5 GB RAM, compaq smart
> array RAID 5 controller.
>
> maximum smtp connections in postfix: 300
>
> sessions time
> -------- ----
> 50 2m10s
> 100 1m39s
> 150 1m20s
With 20 messages/second, that is twice as slow as my measurements
with single disk SCSI system.
For first-hand experiences, type "raid 5 performance" into google
and see for yourself why RAID 4/5 write performance sucks so badly.
RAID 5 write performance sucks, especially for random writes. And
that is exactly what Postfix spends all its time waiting for.
RAID 5 random write performance is slower than with a single-disk
architecture, because to write a data block, RAID 5 also has to read
then update a parity block. read+update is slower than just write.
RAID 5 is OK for read-mostly stuff like web servers. Not for mail.
Wietse
- Next message: Len Conrad: "other than 25"
- Previous message: Drew Bloechl: "Re: high capacity configuration nightmares"
- In reply to: Pere Camps: "Re: high capacity configuration nightmares"
- Next in thread: Pere Camps: "Re: high capacity configuration nightmares"
- Next in thread: Drew Bloechl: "Re: high capacity configuration nightmares"
- Reply: Wietse Venema: "Re: high capacity configuration nightmares"
- Reply: Pere Camps: "Re: high capacity configuration nightmares"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]