Newsgroups: comp.sys.transputer
From: andyr@wizzy.com (Andy Rabagliati)
Subject: Re: Transputer link speeds
Organization: W.Z.I.
Date: Tue, 12 Apr 1994 06:04:56 GMT
Message-ID: <Co4u88.136@wizzy.com>

Carew Oladapo <co@buck.ac.uk> wrote:

>
>I have re-checked the speed settings for data transfer over transputer 
>links for the IMS T800. It shows that data transfer using the bidirectional
>link is faster than using the unidirectional link.

Thank you for going back to the book - I do not have one handy.

Let us just be clear here - the figures quoted are the sum of the data
transfers in both directions. If what you want to do is move a lot of
data from one transputer to the next, it is usually quicker to do it
without any traffic in the opposite direction. I do not think Carew was
making that assumption, but I would not like you, gentle reader, to be
misled.

So, in the table (well, we are only interested in 20Mbits, right ??)

> ---------------------------------------------------
> |  Link   |  Linkn  |  Mbits/sec |  Kbytes/sec    |
> | Special | Special |            |  Uni   |     Bi|
> |---------|---------|------------|--------|-------|
> |  1      |  1      |    20      | 1740   | 2350  |
> ---------------------------------------------------

That means that T800 to T800 traffic runs at 1740 with no traffic in
reverse, and 1175 in each direction when fully loaded both ways.

The loss of bandwidth, as noted earlier by Carew, is because the acks
are being interleaved in the data stream.

To clarify my earlier post, the overlapped acks in the T800 show the
greatest benefit in unidirectional transfer. The interesting figures to
compare these to would be the same figures for a T414.

(Still) not having a book handy (I am sure Carew could oblige) I shall
guess the following without proof, just to illustrate my point :-

Unidirectional, 20 Mbit, T414,    1 Mbyte/sec
Bidirectional,  20 Mbit, T414,    1 Mbyte/sec (per direction)
Bidirectional,  20 Mbit, T414,    2 Mbyte/sec (aggregate)

Since the data bytes on a non-overlapped-ack link have (at least)
ack-sized spaces between them, adding traffic in the opposite direction
has little (no ?) impact on throughput.

By the way, link speeds are measured by ispy/check - the Mb column.
This is the input speed of the boot link, identified by the Bt column.
It is a data transfer into internal RAM, and is (of course) measuring
the unidirectional speed. You may see the difference in transfer rates
by looking at a T414 booting through a T800, and vice versa. The
receiver determines the overlapped-ness of the acks.

Of course, all mention of T800 above applies equally well to any
transputer implementing overlapped acks, such as the T425 or T225.

Cheers,      Andy.


--
The birds have vanished into the sky,	     Andy Rabagliati  andyr@wizzy.com 
   and now the last cloud drains away.			    W.Z.I. Consulting
      We sit together, the mountain and me,			(719)635-6099
         until only the mountain remains.   -- Li Po (701-762)

