Cisco – TX-ring-limit Values

ciscoqosrouterswitch

So, I am getting deep into my CCIE study and I just read about the TX-ring-limit interface command. I have used QoS a bit to oversubscribe software buffers to prevent frame loss on switches and whatnot, can you use the "TX-ring-limit" in a similar manner to decrease the likelihood of frame loss during bursty traffic on a LAN?

Additionally I understand the following:
1) TX-ring-limit should be set to either a value of 1 or 2 for a low speed interface such as a T1 serial
2) Higher values can be configured on higher speed interfaces
3) Setting a high value can introduce jitter on a link because the software queue of a IOS device will send the packets to the hardware queue where they are simply FIFO

Has anyone experimented with setting this to a higher value on ethernet LAN interfaces to allow the hardware queue to "hold on to frames" during a bursty traffic period, rather than drop them?

Best Answer

The TX-Ring is the hardware interface queue and is FIFO. QOS does NOT begin to operate in a router until the TX-Ring is full. The tradeoff is a small TX-Ring causes QoS to cut in quicker (good when you want to use LLQ for VoIP) at the cost of CPU, a large TX-Ring uses less cpu (don't need to service the queue as much) but will increase jitter (remember that the queue is Fifo) for voip applications. This is an important VoIP tuning knob on the older cisco routers for the above reasons.

It does not impact drops because TX-Ring only exists in routers and when the tx ring is full traffic is queued in router memory.