Messages/day limit for EFA

[rg2]

Hi,

I work at a ISP and I was looking for a anti-spam gateway server. After testing several (commercial) solutions, I am happy I have found EFA. This thing is great. Congrats to the people that have put time and effort into it, specially for making it free.

Well, I have read some topics already, but I haven't found a answer to my question.

How many messages/day can a default EFA installation handle? I currently have 3 mail servers. One gets 12.000 messages/day, another gets 30.000 messages/day and the last one gets 100.000 messages/day.

At first I would make EFA filter messages to two servers, a total of 42.000 messages/day. Can a default EFA Hyper-V installation handle it? If not, can I change the number of processors and RAM to improve it or should I bouild a second EFA?

Thank you!
Rafael

[shawniverson]

Yes, you can adjust the number of procs and memory for EFA. In fact, for your higher load, I would recommend at least increasing the amount of memory.

[rg2]

Hi shawn,

Is there any indicator to show me when I should increase memory and procs? I am not sure how to interpret the load average...

Thank you for your answer and keep up the good work!

[rg2]

Hi,

This is how my server is currently running:



I have not changed default EFA configs, but I have doubled default VM processors and RAM (4 procs and 4Gb RAM).

EFA have processed 6,600 messages and its 10 am. In a 24 hour period it is processing around 16.000 messages.

Do you think the Load Average is normal? I don't know how to interpret that. Thanks!

[darky83]

I have 2 systems running that do a bunch of traffic (about 50 to 75k each day)

one as primary and the other as secondary but did some config changes to the VM to make it run, but beside that its just running the default EFA appliance.
VM is upgraded to 16 virtual cpu's and 64GB mem.

The advantage of Virtualisation is that you can overcommit, so running the system with high specs allows us to process peak usage without any delay.s

[rg2]

I have 2 systems running that do a bunch of traffic (about 50 to 75k each day)

one as primary and the other as secondary but did some config changes to the VM to make it run, but beside that its just running the default EFA appliance.
VM is upgraded to 16 virtual cpu's and 64GB mem.

The advantage of Virtualisation is that you can overcommit, so running the system with high specs allows us to process peak usage without any delay.s

Hi,

My current EFA is running with 4 CPUs and 16Gb RAM. See image below:


But I have a LOT of messages on queue. How to improve system behavior? How do I "force" that queue to send messages? A lot of messages on queue are from yestarday...


Thanks!

[darky83]

If you look in vmware how is the memory utilization there?

Also I would play around with the mailscanner childeren, it is probably faster to limit your children..( I normaly don't use more than the ammount of cores I have).
In your case it seems like the processing queue is building up, so the system is not 'fast' enough to process, however your CPU load seems to be minimal?

if you run 'top' for a while on the machine, does it show you any information on what it is doing?

[rg2]

If you look in vmware how is the memory utilization there?

Also I would play around with the mailscanner childeren, it is probably faster to limit your children..( I normaly don't use more than the ammount of cores I have).
In your case it seems like the processing queue is building up, so the system is not 'fast' enough to process, however your CPU load seems to be minimal?

if you run 'top' for a while on the machine, does it show you any information on what it is doing?

I run it on Hyper-V. Hyper-V doesn't tell me how much RAM it is using. The graph that I posted is from EFA's SNMP, so I think it is correct. EDIT: It's currently using 4Gb RAM.

I never touched children configuration, it is default.

Running top, I don't know if this behavior is normal, but MailScanner uses a lot of CPU for a few seconds, (sometimes reaching 100%) and then CPU utilization drops to zero for a few seconds. This keeps going on and on. Eventually clamad shows high CPU usage but it is rare.

I'll change core number to 8 and see if it helps. Do you recommend changing children configuration? (it is currently default). EDIT: EFA wont start up with 16 or 12 CPUs, only when I tried 8 CPUs it did start. Is any configuration required to use 16 CPUs?

Thanks a lot!

[darky83]

Default efa Children is set to 2 so as you are running with 8 I guess you changed something
If you upgrade to 8 cpu's then it should be good to keep the current values.

In top press '1' does it use all available cpu's ?

Never tried using more than 8 cpu's on a hyper-v setup will try later on.

What is the physical hardware where it runs on? is it recent?

[rg2]

Default efa Children is set to 2 so as you are running with 8 I guess you changed something
If you upgrade to 8 cpu's then it should be good to keep the current values.

In top press '1' does it use all available cpu's ?

Never tried using more than 8 cpu's on a hyper-v setup will try later on.

What is the physical hardware where it runs on? is it recent?

Maybe I changed children config but I don't remenber doing it

I'm running with 8 cpus now. It seems all CPUs are being used. If you can, please let me know if it is possible to use more than 8 cpus (on hyper-v config).

The hardware is a Dell R720 with 64Gb RAM and 2 Intel Xeon E5-2660 (32 logical processors). The server has 8 hard disks. EFA's hard disk is not shared with any other VM, but it is a simple 7200 RPM 2Tb hard disk. Do you think disk IO could be the problem?

[darky83]

If you run top there are multiple items that can go up:

from my local workstation which is doing nothing:

Code: Select all

Cpu0  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,  0.0%st

if the %wa is high it might be a disk IO bottle neck.

You might also try to run:

Code: Select all

vmstat  1 100

and

Code: Select all

iostat 1 100

this might give you some idea of what is using your systems resources.

(i'm running my high volume systems from pure SSD san storage units so never have IO issues )

[darky83]

Don't have an hyper-v system available where I can add more than 8 cpu's :S

but you might try to add numa=off to the GRUB boot.cfg

[rg2]

If you run top there are multiple items that can go up:

from my local workstation which is doing nothing:

Code: Select all

Cpu0  :  0.0%us,  0.0%sy,  0.0%ni,100.0%id,  0.0%wa,  0.0%hi,  0.0%si,  0.0%st

if the %wa is high it might be a disk IO bottle neck.

You might also try to run:

Code: Select all

vmstat  1 100

and

Code: Select all

iostat 1 100

this might give you some idea of what is using your systems resources.

(i'm running my high volume systems from pure SSD san storage units so never have IO issues )

SSD storage is a beast! $$$$$$

My system stats:

%wa is around 10%...15%.
%iowait is also around 10% and 15%.
Are those number ok or too high?

Disabling numa worked, my system is now running with 16 cpus, all of them working. Let's see if it makes a difference.

Thank you!

EDIT: Setting CPU number to 16 automatically changed children number to 16, even thought I had it set to 8 children before changing the number of CPUs. Maybe that's why I did not remember changing it the first time.