Speed up amavisd-new by using a tmpfs filesystem for $TEMPDIR

Prerequisite

Linux

You need a kernel that is capable of creating and using a tmpfs filesystem.

*BSD

You need a kernel that is capable of creating and using an mfs filesystem.

Why do this at all?

amavisd-new's speed may be limited by the speed and latency of your disk I/O.
By putting the temporary files into a ramdisk we eliminate precious read/write operations on a slow mechanical disk.

But is it safe?

Yes.
It is safe because amavisd-new will return a 250 OK only after it reinjected the mail back into Postfix.

Sizing

If you run n amavisd-new instances and each one accepts mails of $message_size_limit, then you need a tmpfs of size:

n * (1 + max(expansionfactor)) * $message_size_limit

Let's assume 5 instances, $message_size_limit of 10 MB = 50 MB for the original mails alone.
Now add space for the unpacked mail: 5*10 MB raw mail size become about 5*8 MB binary size , unpack at ratio 1:2, become 80 MB.
Sum: 130 MB.

How to do it?

Find out where amavisd-new has it's $TEMPBASE. Default is /var/amavis.
Create a tmpfs entry in /etc/fstab:
- On Linux:
```
/dev/shm   /var/amavis   tmpfs defaults,noexec,nodev,nosuid,size=150m,mode=750,uid=7509,gid=54322 0 0
```
  Use the uid of the user your amavisd-new runs as. As gid use the group clamd runs as. "mode=750" makes sure that other users like clamd which need to be able to read this directory are able to access it. Also choose the size sufficiently large!
- On FreeBSD versions prior to 5.x (added by Vivek Khera):
```
/dev/aacd0s2b   /var/amavis   mfs   rw,-s=65536   0   0
```
  Where the first column is the device on which your swap lives, the second the mounted directory, and the -s=NNN is the size in 512k blocks. The example creates a 32Mb MFS file system on /var/amavis. Prior to 5.x you need in your kernel config file this entry which is normally the default:
  pseudo-device vn #Vnode driver (turns a file into a device)
- On FreeBSD versions 5.x (added by Reinoud Koornstra):
```
md      /var/amavis     mfs     rw,-s64m        0       0
```
  in 5.x you need:
  device md # Memory "disks" in your kernel configfile (which is the default).
- On OpenBSD (added by Bryan Irvine):
```
/dev/wd0b /var/amavisd/tmp mfs rw,async,-s=524288 0 0
```
  or from the command line
```
mount_mfs -s 524288 /dev/wd0b /var/amavisd/tmp/
```
- On Solaris (added by WC Jones): in /etc/vfstab:
```
swap   -   /var/amavis   tmpfs   -   yes   size=200M
```
  or from the command line
```
mkdir /var/amavis
mount -F tmpfs -o size=200m swap /var/amavis
```
  But see SUNWdrvs8 here
Stop amavisd-new
mount /var/amavis
Start amavisd-new

That's it.

Does it work?

You know the drill: Check your logs!

Case study: Before and after

Before: Left side of the graphs
After: Right side of the graphs

Throughput

Received: we see an increase from 160 mails/minute to 210 m/m.
Sent: we see an increase from 60 mails/minute to 125 m/m.

Queue

The queue used to be filled with mail still left to be scanned. After the optimiziation there's no queue buildup anymore.

© by Ralf Hildebrandt
This document contains links to external information sources that I do neither monitor nor control. I explicitly disclaim any liabilities in respect to external references.
You are getting this document without any guarantees. Any methods shown above are meant as demonstration and may be wrong in some place. You may damage your system if you try to follow my hints and instructions. You do this at your own risk!

This file was last modified 17. Jan 2007 by root