- From: <noah_mendelsohn@us.ibm.com>
- Date: Fri, 13 Jun 2003 22:02:53 -0400
- To: "Jean-Jacques Moreau" <jean-jacques.moreau@crf.canon.fr>
- Cc: David Fallside <fallside@us.ibm.com>, Hugo Haas <hugo@w3.org>, Martin Chapman <martin.chapman@oracle.com>, XMLP Public <xml-dist-app@w3.org>
Actually, I think we can handle several flavors of multicast with what we have in SOAP 1.2, independent of the "the/an" distinction. I think the key is to realize that for many (not all) types of multicast, it's better to model the MEP as multiple outbound messages, possibly directly to the multicast target, or possibly to nodes that assist with the fanout. If you want responses at all (as opposed to fire-n-forget), it's the MEP that should define the rule for folding the responses. For example, what do you do if an outbound message is sent to 10 nodes, 7 report success and 3 fault? Well, there are flavors of multicast involving voting which might report that as success. There are others in which you would want one or more of the faults reported, or perhaps an array of 10 response statuses. If you want to use a spanning tree to reach a thousand nodes, the MEP can describe the generation of messages by each of the tree nodes, and the folding of responses as approriate. I don't think the exercise is primarily to map our existing request/response MEP to multicast. I think our design enables the specification of MEPs that will model any of the above. The specification for the MEP would describe the routing of the messages, the generation of one message dependent on another (e.g. if you used an explicit tree flood algorithm to fan out the outbound messages and fold in the responses), or whatever. I strongly urge that we NOT design the details here. I'm only making the case that there are many useful flavors of multicast that I believe can be well-handled by specifying the suitable MEPs, and I think that's more than enough to let us move SOAP 1.2 to recommendation. I would also point out that there is a completely different flavor of multicast that I believe we can also handle. Consider a fault tolerant system, perhaps like the space shuttle, where 3 machines all run the same software more or less in lockstep. Thus all requests go to all three, which (in the absence of errors) produce identical responses. Let's assume that a hardware or software mechanism in or below the SOAP binding causes the multiple message copies to be sent, and handles any necessary voting on the responses. I believe we are quite clear that SOAP says nothing about whether the implementation of a SOAP node such as the endpoint is distributed or replicated. In this case, the same chapter 2 soap logic is being executed consistently by a combination of 3 machines which collectively conspire to implement a highly fault tolerant instance of what to SOAP is a single node. In this case, I think you can use our ordinary Request/Response or perhaps even Response-only MEPs. Again, I think our design is more than adequate to claim reasonable 80/20 support for multicast and to move on to recommendation status. I would also point out that during our CR period we heard, as far as I'm aware, no concerns expressed by any suppliers of multicast systems. I hope this is helpful. Thank you. ------------------------------------------------------------------ Noah Mendelsohn Voice: 1-617-693-4036 IBM Corporation Fax: 1-617-693-8676 One Rogers Street Cambridge, MA 02142 ------------------------------------------------------------------ "Jean-Jacques Moreau" <jean-jacques.moreau@crf.canon.fr> Sent by: xml-dist-app-request@w3.org 06/13/2003 04:37 AM To: Dear XMLP Comments <xmlp-comments@w3.org> cc: XMLP Public <xml-dist-app@w3.org>, David Fallside <fallside@us.ibm.com>, Martin Chapman <martin.chapman@oracle.com>, Hugo Haas <hugo@w3.org>, (bcc: Noah Mendelsohn/Cambridge/IBM) Subject: New PR issue: one or more ultimate receiver? The following issue has been raised (e.g. [1]) on ws-arch: is there one and only one ultimate receiver, or can there be several ultimate receivers for the same message? The issue is that multicast bindings, for example, may be prohibited if there is only one single ultimate receiver. Currently, Part 1 specifies there can only be ONE ultimate receiver (THE ultimate receiver). An earlier version of Part 1 used to allow multiple receivers (AN ultimate receiver), as per the resolution to issue 103 [2]. It appears that when the resolution to issue 103 was implemented, not all occurences of "THE" were changes to "AN", and that an (unfortunate) editorial sanity check later replaced all instances of "AN" to "THE", instead of the contrary. We have two options at this stage: 1) Go with whatever is in Part 1 today, considering that we are too late in the Recommendation process; or 2) Reimplement the resolution to 103 (i.e. s/THE/AN/). I have a preference for option 2) above and consider that this is an editorial change only. However, I think we should first investigate whether this change is likely to (severely) impact current implementations. I don't think so, but at the same time I don't want to take the risk of delaying publication. I apologize for raising this issue so late in the Recommendation process. Pls remove xmlp-comments of any follow-up discussions. Jean-Jacques. [1] http://lists.w3.org/Archives/Public/www-ws-arch/2003Jun/0118.html [2] http://www.w3.org/2000/xp/Group/xmlp-issues#x103
Received on Friday, 13 June 2003 22:04:07 UTC