[e2e] "PMTUD using options" draft
Jon.Crowcroft at cl.cam.ac.uk
Thu Feb 12 23:32:12 PST 2004
so it depends whether your option needs processing on every hop - thats why ipv6 was done the way it was:-)
note also it depends whetehr options are stateful or stateless whether you care - most stateful options (e.g.
things that pin paths) dont get used on every packet and arent performance critical on the first - most other
options one could think of "sensibly" can obviously be dealt with by some level of programmable packet processor
logic - and the only question in terms of systems archtecure is :
what is the tradeoff between "sensible" optional stateless per packet functionality at a network node, and number
of gates in packet processor logic in router linecards..... routers that send anything to a "central" processor are
going to have a backplane or switch hotspot/bottleneck problem as well as a cpu problem and are out of the ark
designs/dinosaurs for sure....
In missive <DAC3FCB50E31C54987CD10797DA511BA07700473 at WIN-MSG-10.wingroup.windeploy.ntdev.microsoft.com>, "Christian Huitema"
>>> > It is widely believed that all routers process IP packets
>>> > with options in the "slow path", as you claim in your web
>>> > site above. This should be re-filed in the basket of
>>> > "widely believed fallacy". :-)
>>> > In fact, this is no longer true (if it ever was).
>>> Okay, so you say that this '"fallacy"' is not true. And suggest that
>>> it may never have been true.
>>I have no reason to not believe Ran, and to accept that some routers can
>>process some options without any performance penalty. I am not sure that
>>these routers can process every option that can possibly be invented.
>>But as Lloyd points out, there are quite a few routers out there that
>>cannot process the options in real time. Maybe they are using inferior
>>and obsolete technology, but there is a good chance that there will be
>>one such router on any given path.
>>If you use an IP option and there is an ancient router on the path, that
>>router will become your bottleneck, and your performances will be
>>terrible. Thus it is only safe to use options when all routers have been
>>replaced. In short, we have a classic "boil the ocean" scenario. We have
>>to boil the entire ocean, i.e. replace all the routers, before we can
>>cook a single noodle, i.e. use IP options.
>>-- Christian Huitema
More information about the end2end-interest