Thanks again to those who commented on my "WFQ on high speed (16 Mbps HSSI) serial interfaces" discussion. I was concerned that this might be a burden on the processor and leave the queue depth too shallow. I was finally able to get my hands on the routers yesterday. Changing the interface back to FIFO shaved 3 or 4 % off the processor load. It would likely have been more, but the router had just the one flow to manage. Since the traffic wasn't the least bit bursty, the shallow queue (64 message congestive discard) didn't seem to matter much either.
The throughput issue seems to be one of "bandwidth-delay product." This was an issue in my satcom days where the delay in getting acknowledgements back caused the transmitter to sit and wait. That also now can be the case with modern day wideband pipes (over 10 Mbps sustained tcp throughput in our case) combined with moderate terrestrial delays (about 55 msec in our case). We were able to support our mathematical model that showed we had just a few too many kbytes on the wire for the max tcp window size by splitting the load across two different sockets. Viola! Throughput full speed ahead. Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=35558&t=35558 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
