[Asterisk-Users] cheap gig switch? smc, netgear, or 3com?

[Scott Laird]

On Oct 20, 2004, at 4:17 PM, Jay Wilton wrote:
>
> What is the most important feature for VoIP quality:
> latency, qos, vlan?  I'm leaning towards least latency with
> qos and/or vlans at the linux router.  Might be my best
> shot for an inexpensive gig switch ($100).
>
> I have only seen the qos (802.1p) in the Netgear gs608, but
> reviews of Netgear are terrifying (too light, breaks, hot,
> no support, warranty is a problem after 90 days).

Frankly, the easiest way to get good voice quality in a small network 
is to overprovision everything.  Never run any link that will have to 
carry voice traffic over 20-30% utilization, and you'll probably never 
see a problem.  My gut instinct says that the biggest problem is going 
to be inter-switch links; if everything runs into a single switch, 
particularly a GigE switch, I just can't see problems developing, 
unless you're pushing so much traffic that your cheap switch runs 
completely out of juice.

If overprovisioning is too expensive (or if a choppy call every few 
days is unacceptable), *then* QoS starts to come into play.  As long as 
VoIP packets get priority, it shouldn't really matter how full the 
links are.  However, this is more work.  You need to make sure that all 
VoIP packets are tagged as high priority, either in a port-by-port 
basis in the switch or by using 802.1p support in devices.  Just 
because your switch says "QoS" on the box doesn't mean that it'll 
magically know which packets are most important.

Back to your question: the most important feature for VoIP quality in a 
LAN is almost certainly packet loss, followed by latency.  VoIP is much 
more sensitive to lost packets then any other network service that I'm 
used to running.  In a LAN, you *should* only get packet loss when a 
link is nearing capacity--the switch should have a small buffer, but if 
you send too many packets at once, something has to get dropped.  With 
more expensive switches, you'll see latency start increasing before 
packets actually get dropped.  With really cheap switches, there may 
not be enough of a buffer to really matter--I've never tried measuring 
it.

By increasing the speed of the links, you decrease the probability of 
getting a short burst that overwhelms that switch's buffers.  it's just 
statistics.

If you have a decent switch, then you should be able to monitor for 
dropped packets.  With Cisco switches running IOS (29xx, etc), look at 
the 'Output queue' line in 'show interface'.  If it reports 0 drops, 
then you're probably fine.  Other switch types should be similar.


Scott