Statistics From Our GP Boxes

[stephen]

Scott;

Here are some stats from our GP boxes; after Looking at this any one have any ideas servers constantly go down several times a day eventually they recover but they deny any port 25 connections. We are doing this via DNS Round Robin, was trying to determine how hard it would be to load balance :/

Since May 25, 2005

Gamera
Deleted-Email= 1,206,916
Virus-Infected= 138,162
Msg-delivered-to-sd= 465,835
Gamera-Total-Email-Handled= 1,810,913

Since May 22

Mothra
Deleted-Emails= 1,617,070
Virus-Infected= 100,171
Msg-delivered-to-sd= 340,463
Mothra-Total-Email-Handled= 2,057,704

Rodan
Deleted-Emails= 1,262,898
Virus-Infected= 133,968
Msg-delivered-to-sd= 490,838
Rodan-Total-Email-Handled= 1,887,704


Grand-Total Email processed by Monsters= 5,756,321

Addendum;
Rough Estimation
Over 32 Days is; 179,885 messages processed every day
Over 24 hours is; 7,495 messages processed every hour
Over 60 Minutes; 125 messages processed Every Minute

Now too me the most staggering statistic is this is for 2 domains, there are only around 3000 email accounts.

MiND BLOWING.

<root@gamera:~> sysinfo-cmd.pl
os[Linux 2.4.20-31.9 - Red Hat Linux release 9 (Shrike)] up[ 50 days, 18 hours, 05 minutes] cpu[AMD Sempron(tm) 2800+, 1996.398 MHz (3984.58 bogomips)] mem[ 1293.86/2016.68 MB (64.2%)] video[S3 Inc. ViRGE/DX or /GX (rev 1). at ( bits)]
<root@gamera:~>

<root@mothra:~> sysinfo-cmd.pl
os[Linux 2.6.10-1.770_FC2smp - Fedora Core release 2 (Tettnang)] up[ 14 days, 20 hours, 59 minutes] cpu[Intel(R) Pentium(R) 4 CPU 3.00GHz SMP (2 processors), 3001.167 MHz (11927.54 bogomips)] mem[ 1934.12/2017.77 MB (95.9%)] video[ at ( bits)]
<root@mothra:~>

[root@rodan ~]# sysinfo-cmd.pl
os[Linux 2.6.10-1.770_FC3 - Fedora Core release 3 (Heidelberg)] up[ 14 days, 19 hours, 01 minutes] cpu[AMD Sempron(tm) Processor 3100+, 1799.823 MHz (3579.9 bogomips)] mem[ 1909.26/2027.57 MB (94.2%)] video[ at ( bits)]
[root@rodan ~]#

[scott]

wow those are some great numbers there, my first guess is that qmail-scanner is overwhelming the box, for the record the highest I've ever tested PG was at a rate of 40,000 messages a day, so as far as I know, you're running the most high-end PG servers I know of.

So on to tuning, perl is very sensitive to the amount of L2 cache, the more you have the better its going to perform. Software options, you can limit the number of connections qmail will spawn in total in tcpserver from /service/smtpd/run, the line:
readdefault concurrency concurrencysmtpd 20

The default is to allow 20 concurrent connections, perhaps dialing that down to 10 would help.

Next up, make sure you've got a local instance of named running on the box, and that /etc/resolv.conf is configured to use 127.0.0.1 as the first name server.

Last suggestion, this is something Ive only just started expirementing with, set up a dedicated spamd box. I'm still tinkering with the design, but it seems to have made a dramatic increase in performance now that I dont have multiple perl scripts fighting for the CPU.

PS - great names, you can call your Bale (thats what you call a cluster of turtles btw) "Monster Island"