8 Feb
2013
8 Feb
'13
7:50 p.m.
I wrote about CI, that it's rate was:
I'll agree that CI is slightly more similar to Ethernet than to Token
RIng. Both CI and Ethernet use Manchester encoding and a 32-bit CRC as
a FCS. However, CI does not use CSMA/CD arbitration, which is one of
the defining characteristics of Ethernet (aside from the newer
point-to-point Ethernet links, which have no arbitration). CI has its
own arbitration scheme based on node numbers, and does not use
exponential backoff. Not to mention that CI doesn't support a bus
topology, only a star. And a CI network is limited to a 90m radius,
while Ethernet was designed for a maximum station separation of 2.5Km.
So please remind me again just exactly what is so similar between CI and
Ethernet?
70 Mbps.
Rich Alderson wrote:
Not originally. That was the HSC70 upgrade, which was
VAX only.
The HSC70 upgraded several things, but not the raw data rate, which was
70 Mbps from day one. See, for example, the HSC50/70 Hardware Technical
Manual, EK-HS571-TM-001. The line rate is determined by the "Port Link
Module" (L0100). This same module is used in the HSC50, HSC70, and
CI20. The module is hardwired for 70 Mbps operation.
As far as I've been able to determine, the main performance boost of CI
network traffic with the HSC70 was that it supported longer packets,
reducing overhead. However, in most cases it was probably more
significant that due to the higher processor performance of the HSC70
vs. HSC50, it could handle more disk I/O operations per second.
It's definitely a shame that the HSC70 didn't support 576-byte sector
mode for use with 36-bit systems. This resulted from a deliberate
choice to drop 576-byte mode from the HSC70 software, as there is no
such limitation in the HSC70 hardware.
Although it
clearly was inspired by Ethernet, it is dissimilar enough
that I'm not sure I'd even call it a (lower-case "l") ethernet.
It's similar enough that I would. I think that any documentation from Digital which
suggests that the CI was
intended for anything other than the Jupiter is revisionism in extremis.
So the dates on the schematics in the Field Maintenance Print Set for
the CI20 are revisionist history? Fascinating.
Also the DEC internal memo "Clusters Program Analysis" memo by Tony
Sukiennik, dated June 18, 1982, states on page 13:
The CI is a high speed (70 megabits/second), multdropped, short
distance (90 meter radius) interconnect designed to pass data
and control infomration among intelligent computers.
The computers currently supporting the CI are as follows:
1. VAX-11/780
2. VAX-11/782
3. VAX-11/750
4. 2060
5. 2080 (JUPITER)
6. VENUS
7. HSC-50 (I/O server)
[Note that the presence of the VAX-11/750 on the list contradicts my
earlier suggestion that the CI750 might have been an afterthought.]
If it had really been the case that the CI20 wasn't planned until after
the Jupiter cancellation, there's no way that engineering could have
gotten it ready between then and when it first shipped. Just the
engineering of the three CI20 modules that are installed in the KL's
RH20 backplane (M3001, M3002, and M3003) appears to have taken at least
two years.
Eric