[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: Issues 050, 052, DAs in general, and who's DA is it anyway?
Das are a miss-feature, an unnecessary complication, the
cause of too much grief, inherently ugly, the source of too much debate, and begging
to be profiled away. I am having extreme confliction over the role, intention and
value of delivery assertions as they are being discussed in relation to several
issues current and past. The intention of the protocol, as I see it, is merely to
define a mechanism whereby a sender transfers messages to a receiver in such a
manner that the sender understands that responsibility for the message’s
ultimate delivery has been passed from the RMS to the RMD. The transfer of responsibility is one way only, and that a
reliable exchange of request and response requires that each end of an exchange
perform both RMS and RMD roles. I fail to understand why the RMS might care what Das are
offered, since Das effect the behavior of the RMD to AD interaction and only incidentally
the pattern of events on the wire. Assuming the AD and AS may be independently developed and
designed to fulfill their individual purposes, what business does the RMS have
in understanding the AD to RMD contract. The AD may be implemented to
care or not to care about ordering, exactly once, at most once, at least once
and so forth. I assume competence on the part of the authors of AD.
It is quite conceivable, that the authors of DA might employ algorithms insensitive
to order or duplication. I would rather that all of these Das be eliminated entirely,
and that the RMD be specified to operate in one way only. If a vendor chooses
to implement Das, they are a contract between AD and RMD only, and as such may
exist in the API of the offered RMD and thus safely outside of the scope of
this specification and the charter of the WG. I can imagine a TCP analogy to this discussion. Would
we rather have TCP endpoints that optionally reassembled message fragments? We
already had UDP, which amongst other reasons was why TCP was defined. There are well-known protocols without intermediary tolerant
delivery characteristics, those still exist and can be used should Das less
than the full set be required. |
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]