Re: Fragments and Interface Speeds

James B. VanBokkelen (JBVB@AI.AI.MIT.EDU)
Sun, 29 Nov 87 22:13:01 EST

>It's hard to believe that in this age of utterly cheap dense RAM,
>otherwise sane people are proposing inserting artificial delays between
>Ethernet packets because a lowball vendor wouldn't put, say, TWO
>buffers on their card!
>[I admit I'm prejudiced, since I worked on Suns, which currently seem
>to have the highest Ethernet thruput, but they were built out of
>standard Ethernet chips and DRAMs available to everyone. You too can
>handle infinite back to back packets, if you just design with that in
>mind as Sun did.]

        But whhhat about all those old, old Suns, the ones without back to
        back capacity. What about the tens of thousands of NI5010's or
        3C501's. People bought them, have them, are working with them
        daily even though no double buffering is available. Its not just
        a querstion of being a lowball vendor, reality is the fact that
        old outmode hardware is out there and used! ...

                                                Larry Backman

One point that may not be well known is that it is not just the single-
buffered PC interfaces that drop packets. I have observed a DEQNA dropping
packets that arrived too close together (or when it had a resource crunch,
or something), sent from a PC. There is also considerable variation in
the performance of different device drivers for the same interface. I've
observed a VMS TCP/IP sending packets (with DEC's driver) much closer together
than Ultrix manages to.

Another point is that even the newest PC interface cards apparently can't
always handle back-to-back packets (I haven't yet met the LAN controller
chip that is as good as its salesmen imply). Presumably other controllers
based on similar chips will have some susceptibility to overrun, also.

So, don't just dismiss the problem: The closer two IP fragments are, the
more likely it is that any arbitrary receiver will drop the second, requiring
a high-level retransmit. Fragments are expensive to deal with in most
environments, and almost everyone does their best to avoid them entirely.
The exceptions I know of (NFS, routing protocols on big networks) that
send fragments deliberately are very much special cases. I wasn't asking
that the gateway people go to enormous lengths to insert delays, just to
do so when convenient, in order to make the best of an already bad situation.


This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:39:57 GMT