[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: EOS WG Minutes - Minneapolis

Juergen Schoenwaelder wrote:
> >>>>> David Battle writes:
> >> Jeff's presentation was certainly entertaining. However, I did not
> >> really get his argument. I would like to learn more so that I can
> >> understand whether this "go to jail" phenomenon really exists.
> David> I quizzed him about this after the meeting.  The "starting
> David> over" happens only if the connection is completely dropped.
> David> For example if because of packet loss the operation takes so
> David> long that one of the endpoints crashes or times out due to a
> David> keep-alive not being acked.
> Not sure I understand this either. I think Jeff was talking about a
> sequence of SNMP transactions, e.g. walking a table. If you do that
> over TCP and the connection breaks in the middle, then I think there
> is no requirement to actually throw away the data which you have
> already received and to start from scratch.

Though, of course, in cases where absurdly large object sizes are being
used (TCP makes that more possible, right), if the connection breaks
after 99% of the response has been received, that could force a data
and re-retrieval.

Personally, I believe adding TCP is a good thing, and that
to do otherwise "hints" of practicing a technology religion
as opposed to heeding market needs.  I also think TCP is
extremely useful in fault management applications as that
helps SNMP managers implement event driven alarm models instead
of poll driven models.  I gather poll driven alarm models are
considered next to useless in at least many (most?) telecom arenas.