Need help on one QOS related issue.
When network utilization is more than 90 % traffic on Wan Link, D1 class applications is facing slow response even though it has been assigned 60% of total bandwidth.
Queue limit is 64 as of now.
If i increase queue limit then will it make any difference ??
Increasing queue depth increases latency although it might avoid drops. I.e. if your D1 class application has slow response to lost packets, and you see queue drops, increasing queue depth may help. If your D1 class application has slow response due to queuing delay, increasing queue depth may not help at all.
You could also try increasing bandwidth allocation to your D1 class applications (to dequeue such traffic over other traffic). However, if that traffic actually needs more than about 2/3s of your link capacity, you might not have enough bandwidth to meet that application's service needs.
(NB: for maximum TCP performance, interfaces need the queue depth to support about half the BDP. 64 packets is often a default queue depth useful for "typical" WAN of T1/E1 bandwidths.)
On some/many platforms, software queuing only "engages" when the interface FIFO queue overflows. Sometimes you also need to decrease the interface tx-ring-limit to get the bandwidth guarantees you've defined.
Thanks for the reply.
WAN link is having bandwidth of 5 Mb, so queue limit can be increased from 64 to something else ??
Here issue is with the overutilization of the WAN link. With a usage above 90% it will certainly impact the performance. Increasing queue limit probably to something like 512 packets will only help you with drops but still you will see the slowness. It is a better option to either reduce traffic load or divert traffic else opt for a higher bandwidth.
Thanks for inputs.
But customer and my concern is that if at high utilization (Not 100%) we are not able to send D1 class traffic well over WAN link then what is use of implementing QOS.
We need to show customer that at high utilization also priority traffic like D1 is not getting
Can you please share some fine tune which we can implement.
Once again we faced same issue.
Customer reported slow response for D1 applications even through there is no drop in D1 class.
Only bandwidth utilization was around 90 %
can you please let me know how i can calculate tx-ring-limit
Cisco suggests not setting it so small that interface hardware waits for packet to transmit. I.e. ideally, you want to be able to keep the interface hardware running at 100%. However, don't recall seeing documentation how to determine that.
In the past, I've just set it to minimal value or perhaps minimal value plus 1.
BTW, what kind of WAN link is this? You're not dealing with any kind of WAN "cloud" technology where there might be delay within the provider's cloud?
We are using metro Ethernet but i checked cache flow
packet is showing TOS value of A8 ??
How can i convert it into equivalent COS parameter.
I need to know whether A8 is equivalent to CS1 or not