|
Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com |
Subject: RE: info requested
From: Yacketta,Ronald J (YackettaRJ
worldkitchen.com)Date: Thu Aug 24 2000 - 13:50:06 CDT
- Next message: Simon J Mudd: "Re: Broken SMTP Notifications in Postfix RPMs"
- Previous message: Wietse Venema: "STOP THAT THREAD"
- Maybe in reply to: Yacketta,Ronald J: "info requested"
- Next in thread: John Pettitt: "Headroom (was RE: info requested)"
- Maybe reply: Yacketta,Ronald J: "RE: info requested"
- Reply: John Pettitt: "Headroom (was RE: info requested)"
- Reply: Brad Knowles: "RE: info requested"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Sorry about that
I ment ~2k a minute what a MAJOR typo that was :(
that's what I get for writing email about one topic, think about another and
speaking to the other sysadmin about the project :(
man I multitask about as good as micro$oft does.
sorry about that!
-----Original Message-----
From: wietse
porcupine.org [mailto:wietse
porcupine.org]
Sent: Thursday, August 24, 2000 12:19 PM
To: postfix-users
postfix.org
Subject: Re: info requested
Yacketta,Ronald J:
> Gateway server class machine
> with 2 Intel 700+ MHZ processors
> 1gb of Ram
> a totally of 150 - 200gb of Disk space (scsi, with aha2940 UW card)
>
>
> Would this setup be able to handle the daily torment (close to ~2k
messages
> a second)
> of a mail relay?
2k messages/second, that's enough to fill most of a 100BaseT network.
What network infrastructure did you have in mind for this project?
I don't see how one can move this amount of mail over today's Internet.
2k messages/second, that's 0.5 ms for creating/writing/renaming/removing
a message file. Even with a file system that cheats you'd have to
do at least one disk access per message. Attached is some data for
a reasonably fast single-disk machine.
With 10000 RPM disks your average disk access latency is going to
be at least 3 ms (assuming disk heads move in zero time), so you
would need to spread the load over at least 10 disks or so. More
realistic is 50 disks. It would make sense to use multiple servers
instead of putting all this equipment into one box.
Obviously, one queue file per message isn't going to cut this
performance requirement. Postfix would have to switch to a format
of multiple messages per queue file, so that the file system is
less burdened with create/rename/remove operations.
> with amavis doing virus detection?
You have 0.5 ms. That is probably not enough time.
> What recommendations do you have for a fs type for the mail queue etc?
>
> We are also thinking
> of taking a E450 Solaris server with 4 gb ram, 4 processors and turning
that
> into the relay. Though our initial thoughts are "over-kill" with this one.
I don't think so.
You need to do some measurements. Postfix has smtp-source and smtp-sink
in the smtpstone source directory.
Wietse
- Next message: Simon J Mudd: "Re: Broken SMTP Notifications in Postfix RPMs"
- Previous message: Wietse Venema: "STOP THAT THREAD"
- Maybe in reply to: Yacketta,Ronald J: "info requested"
- Next in thread: John Pettitt: "Headroom (was RE: info requested)"
- Maybe reply: Yacketta,Ronald J: "RE: info requested"
- Reply: John Pettitt: "Headroom (was RE: info requested)"
- Reply: Brad Knowles: "RE: info requested"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]