ADRIAN@vax.oxford.ac.uk (11/24/88)
And a comment on long links. >Subj: Any ideas on fast disk access? >Date: 21-NOV-1988 10:12:20 GMT 00:00 >From: MJP@uk.ac.roe.vaxb >To: TRANSPUTER@uk.ac.oxford.prg Just a couple of points that may be of interest: 1) We have been running a transputer link over a complete reel of 90 ohm coax for a couple of years, and have never seen an error. If I remember, the total length is 100 metres. It is driven with TTL levels, and is properly terminated. It is 10 Mbit/sec, using old C003 chips. I don't think they overlap acknowledge, and we have done no proper measurment of thoughput. But I see no reason why one can't extend this approach to longer lengths. Eventually propagation delay will start to reduce thoughput a little, and at some stage losses or even dispersion might mess things up, but I should think that one might get away with at least 1 km. The only potential problem at 20 Mbits/sec might be controlling skew in the drivers: we actually used parallel F244s (if I remember correctly), but would have used single F3037, 3038, 30244s (etc), which can drive 30ohms to TTL levels. These latter chips are available now, but weren't around when I designed the interface. 2) I will be designing a NETBUS board with 6 M212s (or 3 M212s and 3 T2s). This is intended to address the disc bottleneck. The M212s will have their own hard discs (=disks if you are in USA!): the T2s will have their own private SCSI buses. Surely the best approach is simply to go parallel, and have a disc farm? If at some stage you need to interface to a sequential dinosaur, you can always squirt the reconstitued data down it's throat subsequently. This is the approach that we will be using for handling our real time image data. Each disc can be simple, cheap relatively low performance. We just replicate the card and racks of discs until we obtain the required throughput. And of course we can record the data twice on different discs for security at run time with little extra effort. There are obviously many ways of distributing the data across the discs... If you used the new erasable exchangable optical discs (Maxtor?) which store (if I remember correctly) 1Gbyte, have 28ms track-to-track access and cost at the premium introductory price of (I think) around #(pounds)6K, then your data size would be no problem. Just a suggestion. Adrian Lawrence ---------------------------------------------------------------------------- ADRIAN @ UK.AC.OXFORD.VAX Microprocessor Unit, Oxford University, 13,Banbury Road, Oxford, Oxon. OX2 6NN. UK. ---------------------------------------------------------------------------- EARN/Bitnet: ADRIAN%UK.AC.OXFORD.VAX@AC.UK {EARN node UKACRL} ARPA: ADRIAN%VAX.OXFORD.AC.UK ACSnet: adrian@vax.oxford.ac.uk@au.oz.munnari [ean.ac.uk%nummari.cs.mu] -----------------------------------------------------------------------------