🚀 go-pugleaf

In article <1ht2p2INNnn1@iskut.ucs.ubc.ca> morrison@ppc.ubc.ca (Rick Morrison -  
ppc sysadm) writes:
> He then goes on to attribute some network problems to the device
> when used in a "high-performance" application.
> 
> I would say that we have been using the device in a fairly
> "high-performance" application: 4 Sparc 2s chained off of the device.
> The length of the thin-wire segment was, however, quite short.
> The thin wire segment has now been replaced with twisted pair. 
> There has been no important change in the error 
> rates observed on machines that migrated from the
> thin wire to the twisted pair.
> 
> We have been satisfied users of the product.
> 
> I guess Richard has a more technical definition of repeater in mind
> than I intended - i.e. a device that joins two ethernet segments
> at the physical layer, providing signal amplification. I gather
> it is a pulse retiming function that Richard feels is missing
> to merit this term. By, the way, what ARE the cheaper competing
> products that provide retiming? The CTP100T is pretty cheap.
> 
> Since this _is_ bc.general I suppose there ought to be some
> B.C. general content. So: cold enuff for ya?

A Sparc 2 is not high performance in the Ethernet department, it doesn't
push the Ethernet specifications anywhere near the limits. (The Sparc 2
has a rather low I/O bandwidth.)  We have Sparc 2s, Sparc 470s, SGI 3ds,
an RS/6000, an Auspex NS5000, and many NeXTs. It's the latter two machines
that *really* push Ethernet to its limits, since they are both designed
with 'smart' Ethernet controllers.  They are capable of issuing packets
that have minimal (but within specification) delay between successive packets.

An 'Ethernet repeater' is a device that retimes the signals as they pass
through. The common term for a device that changes from one Ethernet medium
to another is without retiming is 'media converter'. A CTP100T does not
contain the retiming circuitry, and as a result, cannot live up to the full
Ethernet standard. We have had several discussions with Cabletron about this
deficiency, and their response has been that they understand that this
problem exists, but that the market for CTP100Ts is small enough that it
isn't worth their trouble to redesign, and wouldn't you rather just buy
one of our MR2000's instead (for a larger chunk of money)? (:-) (I say this
latter with tounge-in-cheek, since the Cabletron people said this mostly in
jest, admitting that they didn't really have the product that we were looking
for in this case. Most of the Ethernets in our campus network are Cabletron
MMAC based, we were hoping that we would not need to go to another supplier for
this annoying little piece of hardware.)

The NeXT on my desk was originally connected to our network through a CTP100T.
NFS performance was abysmal, due to the mangled packets. (*Lots* of timeouts
due to retransmission.) Eliminating the CTP100T eliminated the problem. Our
Computing Science Department discovered the same thing with a number of their
machines, and it was Eric Kolotyluk (eric@cs.sfu.ca) who discovered the 
alternate that contains a repeater. I don't remember the manufacturer of the
units that he bought (maybe Micom or Xylogics?). Drop Eric a quick note, and
I'm sure that he'll respond with an address or phone number for the supplier.
I do remember that they were cheaper than Cabletron by $50-100, depending on
quantity and your Cabletron discount. (The last time that I saw the price
for a CTP100T it was in the range of $450.)

I don't actually have one of these devices on my own workstation (I have a
direct Thinwire to a Cabletron MMAC now), but Eric has been quite pleased with
the performance of these units compared to the CTP100T.

- Richard Chycoski
  Academic Computing Services
  Simon Fraser University
  richard@sfu.ca (NeXT Mail OK.)