Howard has mentioned this a couple of times - the situation where route caching causes one link out of multiple parallel links to be over-utilised while another is under-utilised. I was thinking about this (don't worry, a couple of aspirin will cure the after-effects :-) , and came up with a few questions. For those who can't recall the previous discussions, the setup is (for example) two parallel links, of equal cost/bandwidth/metric, between two routers. If route caching is not used, packets will be distributed evenly between the two links. If route caching is used, packets will be distributed between the two links based on destination or flow, depending on what sort of route caching is in use. The routing process has no way of knowing what volume a flow will be, so flows are distributed between the links more or less evenly by number of flows - about half the flows on each link (correct me if I'm wrong here). If there are a couple of high-volume flows and lots of low-volume flows, then if the high-volume flows end up on the same link, that link could end up congested while the other link is twiddling its metaphorical thumbs. OK, so given that scenario... can this situation be self-correcting? I'm thinking that if one link gets too congested, packets will be dropped, timeouts will occur... and if sessions are dropped, they're just as likely to restablish on the other link (yes I know, meanwhile your users are cursing). Unless it's a high-volume flow that drops out, though, it may not make much difference. Or, in practice, does the cache change fast enough (new flows etc) that there is no time for self-correction to occur (OK, I guess that one is going to be different for each network)? When assigning a flow to the cache, is it simple round-robin, or does the algorithm take note of how many flows are already assigned to that link? Does pinhole congestion tend to be a problem in production networks? I'd be interested to know if anybody has any practical experience with this, or knows of any research. Thanks, JMcL Message Posted at: http://www.groupstudy.com/form/read.php?f=7&i=193&t=193 -------------------------------------------------- FAQ, list archives, and subscription info: http://www.groupstudy.com/list/cisco.html Report misconduct and Nondisclosure violations to [EMAIL PROTECTED]
