Hi,
We occasionally get like a gazzillion bounced spam messages, so I our
proxy qmail-ldap machine is set to have 5000 concurrents which it
promptly bounces since those accounts do not exist on our system(as a
policy we do not allow domain catchalls - specifically for this
reason). The big spike for us is nearly always that kind of bounced
spam problem.
Jimmy
Matt wrote:
Yeah.. ok that should have been "I have 6 systems".
Early morning.. sorry :)
On 3/23/07, Matt <mhoppes@gmail.com
> wrote:
Hi
List,
I have 9 mail servers, using ldap as the backend authentication
system. They all deliver to a NAS unit. Out of the 6 system, 3 are
old machines (500-800mhz), 2 are dual core xeons, and 1 is a P3
2.8Ghz. All have 1.5GB of memory, with the P3 and xeons having 2GB of
memory.
Here's my question... what would cause a high level of unprocessed
messages? On the 3 old machines I never see this problem (odd!), yet
the P3 will often end up grinding and slowing down, maxing out its
connections, and then finally start getting unprocessed messages. At
that point, I firewall off port 25 and let us deliver.
I have checked and I am running the same version of SPAMD, CLAMAV on
all the systems. My guess is that a spam run is triggering spamd to
do some extensive scanning on the mail.... but then why would the older
machines not have the same issue? Even when/if they get totally bogged
down, they have never ended up with unprocessed messages.
TO BE RESOLVED:
1 - What might be causing these unprocessed messages? Is it a bottle
neck?
2 - How do others adjust their concurrencyincomming? It seems that
even 40msg concurrent can swamp the P3 2.8GHz
, depending on the incoming mail... yet it will be fine for weeks...
likewise, I've seen the Xeons go fine with a max of 254, but that can
easily kill them if a huge spam run comes in.
Thoughts?
|
|