home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!inmos!wraxall.inmos.co.uk!aust.inmos.co.uk!thompson
- Newsgroups: comp.sys.transputer
- From: thompson@aust.inmos.co.uk ()
- Subject: Re: Questions on IMS C104 Buffers
- Message-ID: <1992Jul30.180310.24409@wraxall.inmos.co.uk>
- Reply-To: thompson@aust.inmos.co.uk ()
- Organization: INMOS Architecture Group
- Date: Thu, 30 Jul 92 18:03:09 BST
- Lines: 53
-
- conrad@disuns2.epfl.ch (Christian Conrad) writes:
- : hello,
- :
- : I'm performing a simulation on message transfers between T9000, through an
- : IMS C104. Some questions occurred:
- : 1) It is said in the documentation that if the output link of the IMS C104
- : is not free, then the packet is bufferized, but WHERE ? (input or output link
- : of C104)
-
- I don't think this is exactly what the documentation says.
-
- The important point to bear in mind here is that each connection between C104s
- (and T9000s etc.) is completely flow-controlled at the *byte* level. This
- means that it is not actually necessary for a packet which cannot proceed to
- be buffered entirely within one device. It is possible for a packet to be
- buffered half in the input buffer of the device receiving it and half in the
- output buffer of the device transmitting it. It is even possible for a
- particularly long packet to be buffered over several devices. This means that
- the C104 can route packets of arbitrary length. However the performance of
- networks of C104s is improved by using packets which are sufficiently short
- that they can be buffered in at most two devices, for example those used by
- the T9000 (which are at most 32 bytes plus a header and terminator).
-
- : 2) What happens if two packets from different sources try to reach the same
- : busy destination ? They should both be bufferized, so how many buffers there
- : are ?
-
- If two or more packets are routed to the same output, only one can proceed
- (making the output `busy') and the others are all stalled as described above.
-
- : A possible solution to these problems would be to have 32 buffers of 32
- : bytes for each output link of the IMS C104, is it correct ?
-
- However another possible solution is to use a low-level flow-control technique
- which makes it unnecessary for packets to be buffered by the receiving device.
-
- : Thanks in advance for answers.
- : Christian Conrad
- : LSP, laboratoire de systemes peripheriques, EPFL
- :
-
- You're welcome.
-
- --
- Peter Thompson, INMOS Ltd JANET: thompson@uk.co.inmos
- 1000 Aztec West UUCP: ukc!inmos!thompson
- Bristol BS12 4SQ, U.K. INTERNET: thompson@inmos.com
- +44 454 616616 ROW: thompson@inmos.com OR thompson@inmos.co.uk
- --
- Peter Thompson, INMOS Ltd JANET: thompson@uk.co.inmos
- 1000 Aztec West UUCP: ukc!inmos!thompson
- Bristol BS12 4SQ, U.K. INTERNET: thompson@inmos.com
- +44 454 616616 ROW: thompson@inmos.com OR thompson@inmos.co.uk
-