On 28. aug. 2013, at 17:52, Scott Brim wrote: > On Wed, Aug 28, 2013 at 11:34 AM, Brian Trammell <trammell@tik.ee.ethz.ch> wrote: > My only concern here -- and I say this as someone who's a big fan of the general approach of endpoints being smart in how they work around deficiency in the network -- is that SCTP and TCP provide different services. ... > > So even if the OS handles the fallback (which it should), the application still has to determine whether the SCTP features it wants to use are available > > Brian, this has been the case since (at least) IBM access method assembler macros. Somehow the apps that want to do adjust, if you give them the hooks to do so. > > We don't need to make it simple for everyone, just simple for those who want to use it. Agree 100%. But I'll also say that in fact, streams should be hidden, and so should multihoming (all an application wants is to efficiently communicate to the other host). As for the actual application services provided by SCTP, such as unordered delivery and partial reliability, it seems to me that they're semantically consistent with TCP's ordered total reliable transfer, so the fall-back should just make things less efficient, but still correct for what the application expects. Either way, we are, at this point, very much leaving the field that Mike would like to stay in when he says "Theory is nothing - execution is everything." Being an academic, as much as I do value execution, I just can't agree that theory is nothing :-) Cheers, Michael
Received on Thursday, 29 August 2013 07:36:31 UTC