|
Thread Index [isp-bgp] Re: Outbound Balancing
Iljitsch van Beijnum wrote: Suggested by the mere nature of the propagation of the attributes. In other words, why use an attribute that propagates to the next adjacent AS for intra-AS tuning?On 28-apr-04, at 12:49, Pete Templin wrote:Folks, use the tools in the manner that they're suggested, rather than a convoluted way, since there's obviously a wide selection of tools in the tool kit.Suggested by whom? If there is a single authority on BGP traffic engineering that fact has managed to escape me over the past years. Weight or local preference are very blunt tools for traffic engineering, as they completely overrule any other considerations, such as the path. If you want to be sure that your traffic to X always flows over Y even though the path over Z is much shorter, this is the way to go. If the problem is that most of your paths are the same length and you want to add a tie breaker to balance traffic when paths are the same length, local pref or weight are not what you need. If you're not thrilled with either one, what do you suggest? In practice, setting MEDs to select a preferred link for outgoing traffic to a single external AS isn't all that common: in many cases you either want to load balance, or the other side will send MEDs and it's usually better to accept those.Right, because it takes a certain size network before they're comfortable having two links to the same provider (i.e. they'll install links to diverse carriers first). I do find using MEDs for balancing outgoing traffic over multiple transit ASes useful, though. Just set "bgp always-compare-med" and apply the desired MEDs using route maps, making sure that all routes have MEDs as the interpretation of no MED can differ. The advantage is that the MED is only considered when weight, local preference and AS path length are all the same, so you get to nudge traffic in the right direction rather than bludgeon in. Also, MEDs are orthogonal to the local preference so you can use both without traffic engineering cosiderations getting in the way of more important settings which the local pref is then used for. And changing MEDs is mostly invisible to external ASes.How can you say MEDs are orthogonal to local preference when you said MED only matters if local preference matches on two path choices? Better to use the same matching algorithms (if AS path = 12345 1234) and twiddle local preference, since "always compare MED" will become a disaster if there's ever a second link to the same network added. pt To unsubscribe via postal mail, please contact us at: Jupitermedia Corp. Attn: Discussion List Management 475 Park Avenue South New York, NY 10016 Please include the email address which you have been contacted with.
Thread Index |
|
