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

[ih] The origin of variable length packets

When we wrote the proposal to do the IMP development job, we wrote the inner loop of the program and calculated how many packets per second an IMP could process, e.g., service a packet-arrived interrupt, lookup the next IMP on the route, queue the packet for output, and do the output instructions and the time for the hardware to take the packet out of memory, etc.  I am sure we did this for short as well as for full length packets.  I suspect these performance predictions were noted in the bid evaluation process at ARPA.

On Feb 28, 2011, at 9:23 PM, jnc at mercury.lcs.mit.edu (Noel Chiappa) wrote:

>> From: Dave Walden <dave.walden.family at gmail.com>
>> I can't look up what ARPA's call for bids (RFQ) to develop the ARPANET
>> IMP's specified.
>> ...
>> .. the packet-subnet of IMPs broke these messages into about 1,000 bit
>> packets with the last packet in a message being possibly being shorter
>> than a full 1,000 bits.
> I'm also too lazy to go check the RFQ or the BBN proposal, but I did look at
> the Heart et al paper, and although it doesn't _explicitly_ say that the IMPs
> used shorter packets, and give the details on how, there are a lot of things
> that implicitly say so, e.g.:
> "a line fully loaded with short packets will require more computation than a
> line with all long packets ... a line will typically carr a variety of
> different length packets" (pg. 564)
>    Noel