Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[CaLViN]

perhaps the one in vienna, next summer, thats a short trip for us...


ok with msg size of 5kb you could have about 2.5 billion mails per month if you saturate your uplink with 100% all the time, thats pretty good, how many destination domains do you have?


I think our main performance problem at the moment is the sheer amount of different destination domains...

Nearly 20k domains.
No cables or specific domains.
If you are sending to cables or such small sized receivers,then you will probably need more power as it becomes a matter of resources for pmta.

[CaLViN]

@caLVin, @tobias, you guys rock. Thanks Brothers....Stay well and safe.....

I guess you were wondering about the maximum ip limits for pmta and here you have found your answer.
Wanna buy us a coke ?

[Tobias Herkula]

okey 20k sound like very specific set of destination domains, my destination domain cache is currently at 4.5millions, so I think thats clearly the reason for the difference in our performance numbers...

I'm fine If Fred could show me all the cool places in SanFran next february

[CaLViN]

okey 20k sound like very specific set of destination domains, my destination domain cache is currently at 4.5millions, so I think thats clearly the reason for the difference in our performance numbers...

4.5 million domains ? Oh thats ... !
I guess you are not running an ESP ?

[Tobias Herkula]

I'm working for an ESP, and we have a big amount of B2B customers worldwide and I think thats the reason for the big amount of different destination domains. B2C traffic goes normally to hotmail, yahoo, gmail, gmx, web.de, mail.ru and aol.


sorry for totally hijacking your thread...

[CaLViN]

I'm working for an ESP, and we have a big amount of B2B customers worldwide and I think thats the reason for the big amount of different destination domains. B2C traffic goes normally to hotmail, yahoo, gmail, gmx, web.de, mail.ru and aol.


sorry for totally hijacking your thread...

Well,
B2B is another story.
How do you manage inboxing ?
What about blacklists ?

[Tobias Herkula]

I have written a realtime log analyzer that get's it's infos from a account file directive that pipes to netcat. as soon there is the word blacklist or related stuff in in there I get a nagios alert, thats the reason I'm still awake, I currently have some issues with a mail admin, thats still using rfc-ignorant rbl and as I'm bases in germany my PTR-RR all have .de as domain and .de is completly blocked by rfc-ignorant because the denic doesn't allow rfc compliant whois querys (because of german data protection laws).

So I'm actually rerouting traffic and working on some config changes...

--

I don't have much problems with inbox placement, cause of almost 99% percent of our traffic is DOI...

[rg]

You are using an i7? What HDD setup do you have, I often ran into IO Problems as soon as the PMTA has to queue some mails. And I already use a RAID10 with 4x SAS 15k HDDs, thinking about SSDs or FusionIO cards...

Any experience or recommendation with RAID 50 for PMTA server?

[rg]

96 /24 Networks? Holy ****, on one PMTA? This must be a motherf***ing monster machine.

I have 7 /24 on 4 PMTAs and thats enough to send Billions of mails per month. What volume are you sending on so many IPs?

@caLVin, @tobias - Are you using or recommend some tools / gui to manage the IPs... getting active, inactive and moving in/out of vmtas? Ips number not as high... but the process of managing ips is manual and cumbersome.
Thanks

[Tobias Herkula]

Any experience or recommendation with RAID 50 for PMTA server?

RAID 5 needs additional computations for the parity data and so it's slower if you are not using a raid controller with a dedicated cpu, a RAID 10 brings enough time to react on disk failures, but it depends. I could understand it, if your queue is often very high and you have a lot of mails in the queue for a long time. But in the end, the RAID 10 brings more performance on write and read ops...

@caLVin, @tobias - Are you using or recommend some tools / gui to manage the IPs... getting active, inactive and moving in/out of vmtas? Ips number not as high... but the process of managing ips is manual and cumbersome.

I can only speak for me, I'm not using any GUI tools to manage the IP stack, we use puppet to manage everything in the production system and If i really need to change settings it's only a commit and a puppet run away. I see no need to change IP settings often, if a customer needs a dedicated pool it's one time change and I don't have much problems with RBL and such stuff. So my config changes are very limited...

[rg]

RAID 5 needs additional computations for the parity data and so it's slower if you are not using a raid controller with a dedicated cpu, a RAID 10 brings enough time to react on disk failures, but it depends. I could understand it, if your queue is often very high and you have a lot of mails in the queue for a long time. But in the end, the RAID 10 brings more performance on write and read ops...

you got our queue make up...
last year we lost 2 disks in succession in RAID5...
configuring new server... thought I get reaction. thanks @tobias

[CaLViN]

I am using SSD ones for queue.
I dont have a necessity for better performance now but i would use raid10 if i needed to.

[rg]

I can only speak for me, I'm not using any GUI tools to manage the IP stack, we use puppet to manage everything in the production system and If i really need to change settings it's only a commit and a puppet run away. I see no need to change IP settings often, if a customer needs a dedicated pool it's one time change and I don't have much problems with RBL and such stuff. So my config changes are very limited...

Thanks @tobias We do some custom reporting to compare performance... and move things around a bit more...potentially time to look into puppet :-)

[rg]

I am using SSD ones for queue.
I dont have a necessity for better performance now but i would use raid10 if i needed to.

thanks @calvin - seems the convergence is raid 10...

[Teneo]

I know logically this doesn't make any sense, but we have seen some pretty dramatic increase in performance when reconfiguring our SAN Array in RAID5 compared to 10. I know, again, the theory is just the opposite but practically it runs better for the load that PMTA has.

We are currently running on dedicated RAID5 SSD Array with Intel X7 series for queueing and on SAS SAN with RAID 5 for logging and accounting files.