North American Network Operators Group
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

Re: Anycast 101

  • From: Iljitsch van Beijnum
  • Date: Mon Dec 20 07:30:15 2004 
On 18-dec-04, at 22:31, Paul Vixie wrote:


i would be interested in hearing from anybody else who thinks that
turning on pplb in a eyeball-centric isp that has multiple upstream
paths is a reasonable thing to do, even if there were no anycast
services deployed anywhere in the world.



so far, no takers. i've heard from rfc-writers who say pplb was never

meant to be used the way Iljitsch is describing it, and i've heard from

equipment vendors who say their customers don't do that and that if
some

customer did that and asked for support the response would be "don't do

that!", and i've heard from network operators who say they would never
do

that, and i've heard from customers of network operators who did that
with

notable bad effects.




but so far nobody has said "yes, what Iljitsch is describing should
work."



Apparently you also didn't get any pointers to RFCs or other
authoritative sources that say "each and every packet injected into the
internet must be delivered in sequence".



You feel you get to decide what other people should and shouldn't do. I
find that dangerous. As long as there is no standard or law that says
something can't be done, people are free to do it.



Apart from that, I'm not convinced per packet load balancing is as bad
as people keep saying. In the absense of any research that I know of,
my position is that per packet load balancing does have potential
adverse effects, so per destination load balancing is preferred, but if
there is a reason why pdlb doesn't fit the bill, pplb is a reasonable
choice.




let me summarize. Iljitsch says that pplb is incompatible with
anycast,



No. What I'm saying in general is that anycast isn't 100% problem free,
so:



1. There should always be non-anycast alternatives

2. It would be good if we had a way (= BGP community) to make sure that
anycasted routes aren't load balanced across



I don't think either of these is unreasonable.




since a pplb-using access router at the inner edge of an ISP could hear
two different IGP routes to some destination, which ended up taking
different exits from the ISP and thus different BGP paths.




I'm not even sure if I understand this sentence, but it sure doesn't
look like something I said. What I said was, that if you inject packets
towards an anycasted address into two different routers within a
certain AS, there is a very real possibility these two packets will end
up at different anycast instances. I'm on very firm ground here as this
follows directly from the BGP path selection rules. (Although in real
life this wouldn't happen too often because customers tend to connect
to two routers in the same or neighboring pops.)




whereas pplb

would normally only operate on equal-cost paths, the BGP->IGP path
would

hide the variance in BGP paths and make these "paths" eligible for
pplb.


Again: huh?




i've said that pplb is only useful for turning two OC3's into an "OC6"
(or

similar circuit bundling where a pair of routers has multiple
connections

to eachother) and that even in this case, packet reordering is likely
to

occur, which will make tcp-flow performance suffer across this "link".



But would the TCP performance over this "OC6 link" be better than that
over a single OC3 link? That's the real question.




i have also said that turning pplb on across non-parallel links, such
as to

multiple providers or through multiple tunnels or whatever, would
pretty

much guaranty that a word rhyming with "massive suckage" would occur.
and

i've made these claims independent of anycast -- that is, life will be
bad

if you use pplb outside its intended purpose, even if nobody anywhere
was

using anycast.



Your argument is that since it's a bad idea to do this, nobody will, so
making it even worse is ok. My argument is that even though it's a bad
idea, some people will do it we shouldn't unnecessarily make things
worse and/or make a reasonable effort to repair the damage.




loath though i am to treat a "preponderance of assertion" as equivilent

to "proof", i see no alternative on this issue. noone is defending
the use

case Iljitsch is proposing. noone is even saying "i tried that and it
was

OK". lots of people are saying various things like "don't do that!"
and

"are you crazy?"



And we all know that when you tell people not to do something they
don't, and there are no crazy people connected to the net.