[Asterisk-Users] Asterisk Hardware Recommendation

[Matt Roth]

Thanks Daniel,

We may end up replicating your tests in order to confirm some of your 
results.  I don't know if it will be anytime soon, because we don't have 
the hardware yet. Regardless, I will share my results with the list.

Anyone out there have any ideas on why the NFS mount affected call 
quality?  It seems backwards, since it should have relieved some of the 
load from the Asterisk machine.

Matthew Roth
http://voip-info.org/tiki-index.php?page=Running%20Asterisk%20on%20Debian

Daniel Salama wrote:

> Sure.
>
> I setup a small lab on a machine with 4 T1s and 36 agents logged in.  
> The system was configured to Monitor all outbound calls as well as  
> monitor all calls distributed by Queue app (monitor-format setting in  
> queues.conf).
>
> When recording to local disk, everything was working fine. Agents 
> were  busy 99.5% and there were at least 30 calls waiting in Queue to 
> be  distributed. Average call conversation length was about 7.5 minutes.
>
> Then I mounted /var/spool/asterisk/monitor via NFS using 10/100 Fast-E.
>
> The moment we pushed the load on the Asterisk machine, everything  
> worked for about 40 seconds. Then call quality started suffering  
> significantly. Chopped audio. Bad audio. No audio. Good audio. You  
> could imagine. So we stopped the test.
>
> Then we unmounted the NFS drive and repeated the test again. 
> Everything  worked fine again.
>
> The machine we tested asterisk on is a dual Xeon 3 GHz with 2G RAM.  
> During all tests, CPU utilization was about 55% on the average (for  
> each CPU). Memory usage was under 1G.
>
> I would say I need to try more troubleshooting. Maybe there was  
> congestion on the Fast-E, although preliminary analysis indicates 
> there  were no CRC errors, collisions, or packet loss.
>
> The NFS machine was completely idle.
>
> Last, we repeated the test over a 1 hour period. This time, Monitor 
> was  recording on local drives and we were copying files every 15 
> minutes  with a background process (perl script) to NFS mount point. 
> Everything  worked fine as well.
>
> I don't know if these tests are conclusive yet. However, from the  
> results so far, I would recommend staying away from recording to NFS  
> mounted point. I will continue running simulations to see if anything  
> else can be identified.
>
> Thanks,
> Daniel