[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Re: [ws-rx] NEW ISSUE: semantics of "at most once" delivery assurance
Christopher B Ferris wrote: >Charlton wrote: > > my comment is inline Tom Rutt > > >>It makes the most sense to me to take this approach. Although more >> >> >choice > > >>is usually good, it seems onerous to me to require support of AtMostOnce >> >> > > > >>semantics on the wire. Unless I see something that proves otherwise, I >>feel we should say that our protocol is AtLeastOnce. >> >> > >Actually, more choice typically leads to lower interop:-) However, I >concur >with Anish's #1 that we should stick with the way that the *protocol* is >designed >(AtLeastOnce) > > If this is the decision of the group (it seems to simplify the spec a lot), then we should remove reference to "at most once" DA in the spec. However, is there is an agreed use case for "increasing at most once" , we might specify a simple protocol, which only requires state in the rmd to know the last message delivered, and which filters any messages with smaller ID. The existing Ack behaviour for such a simple protocol would not be required for such a "increasing at most once" DA. >As for DA, again, that is a function of the contract between the RMD and >AD (IMO) >and not at all related to the wire protocol. > >Cheers, > >Christopher Ferris >STSM, Emerging e-business Industry Architecture >email: chrisfer@us.ibm.com >blog: http://webpages.charter.net/chrisfer/blog.html >phone: +1 508 377 9295 > >Charlton Barreto <cbarreto@adobe.com> wrote on 07/28/2005 04:31:48 PM: > > > >>On Thu, 28 Jul 2005 13:06:50 -0700, Anish Karmarkar >><Anish.Karmarkar@oracle.com> wrote: >> >> >> >>>The protocol, as is defined today, is AtLeastOnce (as Chris keeps >>>reminding us), between the RMS and RMD. >>> >>>It seems to me that using this protocol for AtMostOnce is >>>wasteful/inappropriate/expensive. I.e., if AtMostOnce semantics is >>>desired why would one use WS-RM and pay for AtLeastOnce semantics >>> >>> >(whose > > >>>costs can be significantly higher). >>> >>>The two sensible options that I see are either: >>> >>>1) say that our protocol is AtLeastOnce and we don't say anything >>> >>> >about > > >>>AtMostOnce at all. If someone wants to use AtLeastOnce at the protocol >>> >>> > > > >>>level and implement AtMostOnce at the application level they can >>>certainly do it. We don't say that 'Unreliable' DA is supported, but >>>someone can use WS-RM to do unreliable messaging (and pay the higher >>>cost) -- although I don't know why someone would do that. >>> >>> >>It makes the most sense to me to take this approach. Although more >> >> >choice > > >>is usually good, it seems onerous to me to require support of AtMostOnce >> >> > > > >>semantics on the wire. Unless I see something that proves otherwise, I >>feel we should say that our protocol is AtLeastOnce. >> >> > >Actually, more choice typically leads to lower interop:-) > > > >>>or >>> >>>2) say that AtMostOnce semantics are important in the world of >>> >>> >reliable > > >>>messaging and Tom/Gil have cited use cases for it, and change our >>>protocol on the wire to accommodate this. I.e., the protocol on the >>> >>> >wire > > >>>will not always be AtLeastOnce protocol. This would allow the >>>sender/receiver to pay the price for the QoS that is desired (and >>>nothing higher). >>> >>>Comments? >>> >>>-Anish >>>-- >>> >>>Dan Leshchiner wrote: >>> >>> >>>>if AS makes RMS aware that AtMostOnce is used by AD/RMD, then RMS can >>>> >>>> > > > >>>>assume that no retransmissions will be necessary and, consequently, >>>>does not need to take up its resources to make retransmissions >>>>available, right? >>>> Christopher B Ferris wrote: >>>> >>>> >>>> >>>>>I think you meant AtMostOnce mode. I suppose you could do that, but >>>>>from the >>>>>perspective of the RMS, it is still retransmitting until it receives >>>>> >>>>> > > > >>>>>an ack. >>>>> >>>>>Cheers, >>>>> >>>>>Christopher Ferris >>>>>STSM, Emerging e-business Industry Architecture >>>>>email: chrisfer@us.ibm.com >>>>>blog: http://webpages.charter.net/chrisfer/blog.html >>>>>phone: +1 508 377 9295 >>>>> >>>>>Dan Leshchiner <dleshc@tibco.com> wrote on 07/27/2005 06:36:24 PM: >>>>> >>>>> >>>>> >>>>> >>>>>>Christopher B Ferris wrote: >>>>>> >>>>>> >>>>>> >>>>>> >>>>>>>However, the protocol as specified would never be able to complete >>>>>>> >>>>>>> >a > > >>>>>>>sequence >>>>>>>if there are lost messages from the RMD perspective and the RMS is >>>>>>> >>>>>>> > > > >>>>>>>not >>>>>>>retransmitting them. >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>why so? as RMD, if i see a gap or AckRequested for a sequence >>>>>> >>>>>> >number > > >>>>>>i have not received and i am operating in "at least once" mode with >>>>>> >>>>>> > > > >>>>>>my AD, >>>>>> >>>>>> >>>>> >>>>> >>>>>>why cant i just send an Ack? if i did so, wouldn't that enable us >>>>>> >>>>>> >to > > >>>>>>complete the sequence? >>>>>> >>>>>>thanks, >>>>>>dan >>>>>> >>>>>> >>>>>> >>>>> >>>>> >> >>-- >>Using Opera's revolutionary e-mail client: http://www.opera.com/mail/ >> >> > > > -- ---------------------------------------------------- Tom Rutt email: tom@coastin.com; trutt@us.fujitsu.com Tel: +1 732 801 5744 Fax: +1 732 774 5133
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]