Newsgroups: comp.sys.transputer
From: davidb@ndl.co.uk (David Boreham)
Subject: Re: Re: LOOKING FOR DISTRIBUTORS...
Organization: Network Designers Limited
Date: Wed, 28 Sep 1994 09:13:48 GMT
Message-ID: <Cwu1n0.79z@ndl.co.uk>

In article <1994Sep26.185719.7512@eso.org> rreiss@hq.eso.org writes:
>Maybe for a PC that's quite ok. We routinely reach about 120kBytes/sec
>from Transputer memory to a Unix disc(!) using iserver routines on a
>Sparc2E and a B016 transputer interface board. We reach about
>900kBytes/sec by directly talking to the B016 'device' (bypassing
>iserver). We reach almost twice that rate using 2 links of the B016
>(has 4 in total) simultaneously...

Ahh, that's my boy...   (to be said in the voice of "Spike" from Tom and Jerry)

>
>Ok, it's unfair to compare a B008 (or compatible) with a B016, but when
>it comes to data rates close to the max. data rate of transputer links
>a PC host is a bad choice.

An interface with B016-like architecture (DP RAM basically) on PCI bus
in a PC running a proper OS (Linux, UNIX, NT ...) would perform just as
well. It's not the PC per say that's the problem but the C012 (spit)
and DOS. I get 4.4Mbytes/s from disk to user process on my PC :)

(Phil, if you're reading this, perhaps you could tell us about performance
over host/C40 interfaces ??)


