Quality of Service on today’s network is almost a must, whether you have VoIP or not.
During recent visit to a client of mine (they were having some major voice issues), I discovered some minor QoS configuration issues. I also found hardware QoS issues.
One of the problems came down to a 10/100 blade in a Cat6500. It was one of the original 10/100 blades created, and only had 2 output queues, with 2 thresholds each. After reviewing the system, I noted that the queue designated for voice was seeing drops.
Normally, most places could run in this situation without issue. Here however, they have hard core data users that truly use their network. This created numerous situations of high bursts in traffic. These high bursts were causing the output buffers to fill almost completely on the switch. This in turn caused voice problems, since their voice gateways are also their WAN routers.
The eventual solution was to move the routers to a blade that has a dedicated priority queue, where voice traffic essentially cannot be dropped (it’s possible, you just need a LOT of traffic to hit that queue). This appeared to smooth things out quite a bit.
In Cisco lingo, the old blade had 2q2t (two queues, two thresholds per queue). Almost all newer blades have 1p2q2t (1 proiority queue + 2q2t) or 1p1q8t.
Moving forward, it’s a good thing to remember that if you’re seeing voice problems, it CAN be the network, no matter how much it is over-built.
QoS is critical. In my opinion, more companies should be building at least basic QoS now, if not for voice, for their critical applications.