To:
"Hollenbeck, Scott" <shollenbeck@verisign.com>, "'Liu, Hong'" <Hong.Liu@neustar.biz>
Cc:
"'ietf-provreg@cafax.se'" <ietf-provreg@cafax.se>
From:
"Gould, James" <JGould@verisign.com>
Date:
Wed, 24 Jul 2002 09:43:06 -0400
Sender:
owner-ietf-provreg@cafax.se
Subject:
RE: Proposed Document Changes
Scott, I really don't see the value in the <status> command. In its proposed form, it doesn't provide any value to the client, since the client will most likely have a pool of sessions and finding the status of the last command processed on any one of the sessions doesn't provide any valuable information. There is even more of an issue of supporting the <status> command for non-transform commands, where determining if a <check> was successfully processed is pointless and represents huge overhead for the server. Non-transform commands should simply be re-transmitted. Transform commands can also be re-transmitted or the client can use the <info> command to determine the current state of the object. Jim -----Original Message----- From: Hollenbeck, Scott [mailto:shollenbeck@verisign.com] Sent: Wednesday, July 24, 2002 7:14 AM To: 'Liu, Hong' Cc: 'ietf-provreg@cafax.se' Subject: RE: Proposed Document Changes > I assume you resend this to see if there are more comments. I > don't mean to > beat up the dead horse, but I would like you to clarify the > two paragraphs > quoted below. I didn't resend it. Take a look at the message headers -- it looks like Jaap might have manually approved it for distribution to the list, or something like that. > The first quoted paragraph, could you explain a bit why "congestion > control/ack" need to be defined if the proposed change is not > made? Do you > mean application level framing for flow control (windowing > mechanism)? I > just want to understand it. This gets back to the "delayed execution" discussion. If commands don't get immediate responses, we'd have to have some mechanism in place to let the client know that commands have been received, but not processed. > The second quoted paragraph, could you clarify: > (1) What is the last command processed? Is it the last > command per registrar > (client) or the last command per session of the client? I > assume that a > client is allowed to established multiple sessions with the server. Last command completed, which might not have anything to do with a session. Remember, the session concept tends to be specific to connection-oriented transports, and we shouldn't have transport dependencies on generic commands if they can be avoided. > (2) Why is <status> command ever needed if its use is so > restricted? The > client can simply resend the command if it does not hear from > the server > after a timeout. EPP ensures idempotency for each command, so > there is no > harm to resend it. To be perfectly honest I don't think the <status> command is needed at all. It got added rather late in the review process when someone asked for the ability to check on the completion status of a previously executed command [1]. No one objected when it was discussed on this list back in September of last year. I would very much like to ditch the command for the very reasons you cited. This is a good time to agree or disagree, folks. Is the <status> command really needed or not? -Scott- [1] The thread: http://www.cafax.se/ietf-provreg/maillist/2001-09/msg00071.html http://www.cafax.se/ietf-provreg/maillist/2001-09/msg00072.html http://www.cafax.se/ietf-provreg/maillist/2001-09/msg00074.html Plus other messages from September 2001, all visible in the archives.