specs & platform for new server

Sylvain Phaneuf sylvain.phaneuf at IMSU.OXFORD.AC.UK
Fri Nov 19 12:39:52 GMT 2004


I see your point and I don't mind the server being very busy. The only
problem is that we have left the default setting "reject MTA connections
when load is >12". So when our ISP realises that we don't accept their
messages, our messages are put on their "wait 15 min" queue. Should I
change the setting to much bigger nmbers?

Sylvain



>>> mailscanner at ECS.SOTON.AC.UK 19/11/2004 09:46:23 >>>
You have raised one interesting point: MailScanner will happily drive
your
system's load average up through the roof. Most of these will be
processes
awaiting network response. A load average of 10 or 12 doesn't really
mean
anything. What counts is the size of message batches that are being
processed. If the batches are small then you have nothing to worry
about. If
the batches start approaching 30 regularly, then you should look into
more
powerful hardware and better network setup.

Don't think your system is being hammered because it reports a load
average
over 10 most of the time. That is MailScanner working your box very
hard!

On 19/11/04 9:29 am, "Sylvain Phaneuf"
<sylvain.phaneuf at IMSU.OXFORD.AC.UK>
wrote:

> I think the number of messages a system can process is slightly
besides
> the point. The size of the messages is surely an important factor,
> especially when the system runs anti-virus checks.
>
> Far from being a Unix expert, I have asked advice from expert
> colleagues for tuning our current server (P3, 1.4 GHZ cpu, 1 GB RAM,
> RAID 1), and we have followed the suggestions given on the FAQ and
MAQ.
> Processing 35k messages a day gives us several periods of load > 12.
For
> about 5 or 6 hours during the day we receive nearly 100 messages per
> minute, the average size being 30 to 40 KB. I have difficulties
seing
> that our system should be able to handle 1 million messages per day
as
> suggested by some.
>
> Sylvain
>
>
>>>> mailscanner at CPYOU.COM 19/11/2004 01:33:59 >>>
> Since reading this and other related threads of late, I have to
revisit
> my
> server design guidelines.  I administrate 5 production MailScanner
> based
> email servers processing a total of 80-100k messages per day.  I was
> recently asked to design and possibly build a spam/virus filtering
> solution
> to easily support up to 2 million messages per day.
>
> I remember seeing guidelines (from Julian himself I believe) on this
> list
> and in the FAQ mentioning to design MailScanner running on decent
> hardware
> for approximately 50,000 messages per day.  Now that I read the
> suggestions
> and the FAQ again, I see that people are saying a well tuned
solution
> should be capable of 1 million...
>
> Most of my email servers (2-3 year old P3 and P4's typically) have
> periods
> of increased load even when processing a measley 20-30k messages per
> day, I
> doubt that the effective throughput difference makes the system
scale
> to a
> million (thats why I am asking you folks).  I have seen less load
with
> recent versions of MS and SA but not to that degree.
>
> For the first time ever I have a nearly unlimited design budget for
> this
> project (but I might not get to build it for other reasons) and am
> wondering a few things.  I was initially thinking of using a quad
> Opteron
> system with 4 gigs of ram and 8 disks in raid 10 based on the
> benchmarks I
> am seeing.  Since the solution has to be completely redundant I
would
> need
> at least 2 of these boxes...  I assume from the new comments that 2
of
> these should have no problems handling more than 2 million messages
per
> day.
>
> Alternately I would use 4 or 6 smaller dual Xeon Nocona or 2xx
series
> Opteron servers with 2 GB ram and 4 disks in raid 10 each.  I assume
> this
> would give greater CPU and I/O per email request at a lower cost.
> Would 6
> of these handle the expected peak loads?
>
> Also, should I use the Linux High Availability project to direct
SMTP
> requests evenly across all or should I simply load balance using
(even
> or
> odd) weighted MX records in DNS?  I guess the LHA solution would
> require an
> extra 2 small servers acting as external facing TCP port 25
directors
> and 2
> high I/O systems acting as internal central mailbox repository
servers
> which drives up the cost/complexity.
>
> The solution I have proposed is the standard Sendmail, Procmail,
> MailScanner, Spamassassin, a list of RBLs, DCC, Vipuls Razor, Pyzor,
> ClamAV
> with at least 2 other commercial virus scanners running on RHEL.

--
Julian Field
www.MailScanner.info
Buy the MailScanner book at www.MailScanner.info/store

PGP footprint: EE81 D763 3DB0 0BFD E1DC 7222 11F6 5947 1415 B654

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at jiscmail.ac.uk with the words:
'leave mailscanner' in the body of the email.
Before posting, read the MAQ (http://www.mailscanner.biz/maq/) and
the archives (http://www.jiscmail.ac.uk/lists/mailscanner.html).

Support MailScanner development - buy the book off the website!

------------------------ MailScanner list ------------------------
To unsubscribe, email jiscmail at jiscmail.ac.uk with the words:
'leave mailscanner' in the body of the email.
Before posting, read the MAQ (http://www.mailscanner.biz/maq/) and
the archives (http://www.jiscmail.ac.uk/lists/mailscanner.html).

Support MailScanner development - buy the book off the website!




More information about the MailScanner mailing list