bad performance

[robvoi]

Hi.

I have the next problem. The runtime of mails is very long.
I see these errors in the logs:

B8F1F802FB: to=<mailaddress@testdomain.test>, relay=none, delay=3,
status=deferred (delivery temporarily suspended: connect to
127.0.0.1[127.0.0.1]: Connection refused)

If I drop 300 mails at once to the machine the delivery takes hours.



It's a virtual machine on ESX with SAN. 2x 3GHz CPU and 2GB mem.

What could be wrong?

Thanks in advance.
Robert

 

[Manfred Muschiol]

Throtteling

Hello,

dropping 300 mails at one time will lead into a DOS, regardsless of your hardware.

Its nearly 100% that, depending on the content of the emails, the scanners will need much memory and your maschien begins to page and swap. If swapping occurs, the performance will dramatically break down. Even if you need bigger mails with HTML and binary attachments, this will happend immediatelly. Do you have tried 300 simple mails, only with a few characters?

To avoid this paging, you can simply try to throttle the speed how mails from a single source will be acceptet. This means, that only the spammer trying to kill your maschien will be tamed, all others are still received.

Test under:

Configuration/Mailproxy/Options the settings:

Client Connection Count Limit 8
Client Connection Rate Limit 16
Client Message Rate Limit 12

This settings are enough for receiving much more than 10.000 mails per day and they ensure, that a spammer s not able to overload your maschien sending thousands of spammails to faked accounts.

The maschine running this settings has 376 MB ram and 2,8 GHz handling this volumen without any headache. delay is around 3 seconds, bigger PDF or ZIP-Files (30MB or more) need around 20 to 60 seconds to be virusscanned.
Simple text mails takes below 1 second.

The harddisk performance of this maschine is poor and she is running virtually in a 512 MB PC as VMWare-server.

Hoping this is the turnknob you need to drill down your problem.

Kind regards

Manfred

 

[robvoi]

Thanks a lot for the tips.
Yes, my test mail was a simple plain text mail with jsut a few chars.

I will test the settings rights away.

Thanks and Greets
Robert

 

[dietmar]

Hi.
B8F1F802FB: to=<mailaddress@testdomain.test>, relay=none, delay=3,
status=deferred (delivery temporarily suspended: connect to
127.0.0.1[127.0.0.1]: Connection refused)

This usually indicates that a service is not running - either the 'Policy' or 'Filter' service (check at administration/Server/Services), then restart that servive (or simple reboot an test if all services are up).

 

[robvoi]

Thanks a lot for the input. For now it works fine. A reboot and the connection limits did work for the test environment.

Off topic:

We are currently testing some spam filter product. So far I like proxmox a lot.
I would go for a live test but I have some difficulties to set up the test plan.

The reason is the limited domain/time test license.

We host about 450 domains for about 50 companies. (all part of one holding)
Some of the domains nearly don't receive mail, others a lot.

Because of the size of the environment and the amount of domains I have difficulties doing a good live test and keeping the system running live until the licenses are purchased.

To my question:
Is it possible to get a 3 month test license for the pro version? If so, whom do I need to contact?

Greets
Robert

 

[dietmar]

If I drop 300 mails at once to the machine the delivery takes hours.

It's a virtual machine on ESX with SAN. 2x 3GHz CPU and 2GB mem.

How much memory did you assign to the VM? We get quite good results assigning 1GB to the VM. But be sure that this memory is really available.
Do not assign to much/all memory to the VM, because Proxmox is optimized for performance and tries to use all available memory. So if you assign all memory to Proxmox VM the ESX server will start to swap and performance decreases.

- Dietmar

 

[robvoi]

I did dynamically assign 2GB mem to the machine. The host has enough to really provide it.

Is there a suggested setting/best practice for ESX server (on ESX side) for proxmox?

Robert

 

[dietmar]

I did dynamically assign 2GB mem to the machine. The host has enough to really provide it.

I thought you ESX Server has only 2GB RAM (see you initial post)?

How many virtual CPUs has the Proxmox VM (1 or 2)?)

Is there a suggested setting/best practice for ESX server (on ESX side) for proxmox?

No, because it depends on to many things. We usually use 1GB/1CPU, and that works quite well.

- Dietmar

 

[robvoi]

I was not really clear in my first post.

The 2x CPU and the 2GB is what has been assigned to the VM.
The ESX node has much more resources.


I will decrease the assigned resources. At least it's a test.

 

[dietmar]

Hi.
B8F1F802FB: to=<mailaddress@testdomain.test>, relay=none, delay=3,
status=deferred (delivery temporarily suspended: connect to
127.0.0.1[127.0.0.1]: Connection refused)

Please can you send me that log file completely (/var/log/syslog*). Maybe i can see the problem. (send to dietmar@proxmox.com)

- Dietmar

 

[robvoi]

Hi Dietmar.

Thanks for the offer. I did send the logs.

The problem seems to be solved now.
It looks like the dynamic memory assignment wasn't ver compatible with the proxmox VM. I did now fix reserved the memory for the VM.

My last test were miles better.

Thanks so far.
Robert

 

[tom]

I did dynamically assign 2GB mem to the machine. The host has enough to really provide it.

Is there a suggested setting/best practice for ESX server (on ESX side) for proxmox?

Robert

hi,
please take a look in the deployment guide, chapter 6.2 describing the settings for vmware server, but the RAM recommendations are the same for ESX.

you can download the documentation package here.

 

[robvoi]

Thanks a lot. I indeed did use the manual. Just didn't read this part ... shame on me

Robert

 

[dietmar]

I did send the logs.

The log shows very long processing times (> 15 minutes), which is almost always a memory problem (swapping).

I also saw a reboot without proper shutdown (you simply stopped the VM without proper shutdown) - this is not good concerning file system/database consistency (although its not the problem in your case)

Anyways, the problem should be solved with the new memory settings.

- Dietmar

 

[robvoi]

The hard reboot was the last reaction on the system failing to respond at all. (web interface)
In a live environment that's of cause not a good idea. I agree.

The memory settings seems to do fine.

Robert

 

[dietmar]

BTW: I assume that you already installed Hotfix 1702, which fixes some memory related problem.

You can also try using 2 CPUs if anything else works (should duplicate performance).

- Dietmar

 

[robvoi]

I assume the hotfix is in the download version?
I tried to apply the hotfix but got a message like "skipped, already there", or something like this. (don't exactly remember the wording).

I will monitor the performance and see if a 2nd CPU is needed.

 

[tom]

I assume the hotfix is in the download version?
I tried to apply the hotfix but got a message like "skipped, already there", or something like this. (don't exactly remember the wording).

hi,
if you use the ISO image, hotfix 1702 is not included. if you downloaded the latest preconfigure vmware virtual appliance, it is.

you can see the build number on the start page (see screenshot).

 

[robvoi]

The hotfix is/was in. (used the VM)

I did run some more stress tests. The performance is fine now. So the memory settings really make a big difference.

Thanks a lot.
Robert

 

[CADL]

The hotfix is/was in. (used the VM)

I did run some more stress tests. The performance is fine now. So the memory settings really make a big difference.

Thanks a lot.
Robert

my server of mail have two cards one internal and external direction, the server promox has one internal direction, in the default relay proxmox I have done tests with the internal direction of exim and with the external direction, but there do not happen yet for proxmox the post office, which configuration addition is necessary to do?