6 Aug 1986 07:31-EDT
I don't think I could reconstruct all the paths we took in examining
different ACK schemes, but I do recall that we examined the selective
ack idea in considerable detail. One of the "criteria" which caused
difficulty had to do with disorderly arrival and with the complexity
of maintaining a kind of scrolling status of which bytes had been
received and which had not since the accounting (sequence numbering)
was byte oriented. There were many reasons for that decision but
chief among them was uncertainty as to maximum packet sizes and the
need for an indeterminate number of fragmentation steps - the
hierarchical numbering plans suffered with complexity compared to the
simplicity of sequential, byte-oriented numbering.
As the accounting unit gets larger, selective acking becomes easier
to account for and, indeed, you find it in various link level or
even transport level protocols which are block rather than byte
It is certain that lack of congestion control and integrity enforcement
produces very serious performance deficits under some conditions-
all the flow control in the world won't help if packets are discarded
for some other reason than the source's transmission rate.
I agree with you that the performance issues deserve very careful attention;
we need to learn from the mistakes we may have made or, at the least,
treat all our experiences in complex internetting as the results of
endless experiments intended to guide us in developing better protocols.
This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:36:35 GMT