To:
<ietf-provreg@cafax.se>
From:
Sheer El-Showk <sheer@saraf.com>
Date:
Thu, 22 Mar 2001 16:06:50 -0500 (EST)
Sender:
owner-ietf-provreg@cafax.se
Subject:
UDP pro's and con's (fwd)
I'm going to repost this one more time since it didn't seem to make it to the list the last two (so if you see it two more times pleae be forgiving). ---------- Forwarded message ---------- Date: Wed, 21 Mar 2001 17:55:44 +0000 (WET) From: sheer@saraf.com To: ietf-provreg@cafax.se Subject: UDP pro's and con's (fwd) I've noticed the mailing list sometimes ignores/delays my mails, particularly when they're CCed to it so here's a repost, sorry if it shows up again in a little while: It strikes me as slightly incongruous to even be considering UDP in a system where we will _definately_ have to verify transactions (ie a registrar _must_ know if a domain registration went through or not) and thus need to develop/adapt some intermediate transaction layer (which TCP provides) while at the same time using XML as the protocol language and rejecting out of hand any binary conversion. This is not a disguised rallying cry for a XML-binary conversion, just a call for some perspective. I think we've already made some implicit performance descisions in adopting XML for the protocol and we should be consistent in the level of performance we're demainding. In addition I think TCP overhead is going to be very light compared to the encryption systems we're considering (TLS, BEEP) and TCP (or the gaurantees it provides) will be necassary for any transport-level encryption that I know of (though of course we could take the encryption into the protocol and ignore the transport layer). If UDP has a place I thinks its in the "Check Domain" command that we descided we might have to consider as a special case since its requirements are vastly different from the rest of the protocol's. Regards, Sheer On Wed, 21 Mar 2001, Brian W. Spolarich wrote: > > | Is there an example of a UDP-based transport protocol that might > | be applicable? > > Rx, which was part of the CMU AFS work, was a UDP-based RPC > protocol with some of the characteristics that are required > here. > > Do folks really think the TCP overhead is that big of a > problem? Memory and CPU is much cheaper these days, and I haven't > seen a requirement stated that would necessitate going down > the UDP road. > > -bws >