[Asterisk-Users] Opinions on renice or turning off swap or ramdis k as swap?

[Peter Svensson]

On Thu, 25 Nov 2004, Colin Anderson wrote:

> I have 4 gig in my * box. I'm tuning for performance and I'd like to ask
> opinions:
> 
> 1. asterisk -p == renice -20 ?? 

The -p option sets asterisk to realtime priority if possible. This is 
different from the traditional unix nice levels. A program with realtime 
priorities will _never_ be preempted by a normal program. A program with 
the traditional unix nicelevel of -20 will give up some time slots to 
lower priority programs.

This has advantages (scheduling of packets are much better e.g.) and 
disadvantages (a broken asterisk server will leave the system impossible 
to log in to). The disadvantage can be soften by keeping a shell open with 
realtime priority (higher) around.

Running with -p does help a lot for VoIP stuff.

Note that this is user-space (pseudo) realtime stuff. Lowlatency patches 
for the kernel can give better response time for the zaptel drivers as 
well. This is orthogonal.

> 2. I've turned off swap with no apparent ill effects. Can anyone commment on
> long term effects with moderate load (say, 30 SIP phones / 2-3K calls /day)

Turning off swap means that the only page-freeing operation left is demand 
paging of binaries. I.e. program text pages (the actual program that is) 
are discarded and reloaded. This can be a rather bad idea.

Asterisk should stay resident in memory under normal circumstances. There 
can be a problem if you are running an ide disk and have unmasq irq turned 
off. In that case disk accesses will definitly hurt you. For any hardware 
from this side of the millenium unmasq irq should be safe.

> 3. Can anyone comment on using ramdisk as swap and whether this is a good
> idea or bad idea?

That is a bad idea. Use the ram as ram instead.

> I'm using 2.6 kernel. I've modified the PCI latency in rc.local:
> 
> setpci -v -s <my T100P address> latency_timer=ff

Are you sure that is a good idea? That only makes the T100 hold the bus 
longer. If my memory serves me the T100P is not bus mastering. I don't 
think the pci latency value for it is relevant, but I am not sure. I do 
believe the latency values of the other devices is at least as important. 
The T100P does not need to transfer many bytes at a time, but it does need 
to transfer them as soon as possible when they are ready.

Peter