Re: [TUHS] More Spider

[Paul Ruizendaal <pnr@planet.nl>]

On 26 Jan 2020, at 17:56, Ken Thompson <ken@google.com> wrote:

> the pierce loop had its own protocol  on its own wire.
> that meant it could only be local-area. the PL was
> in operation on a packard-bell 516 when i arrived
> at the labs in june '66. carl christensen was the
> software person for both the loop and the 516.
> i assume that pierce and condon were the hw
> guys, but that was before my time.

Ah, so the Pierce loop was operational several years before the papers were submitted to BSTJ. That explains a lot.

> spider was similar, but was designed to run on
> the standard telephone T1 lines. thus, the whole
> idea was more wide-area. the major draw back
> of spider, and probably the reason it was never
> really used, was that it couldnt make a connection.
> all connections were pre-created at boot time.

The Spider report has a section on the protocol for making/breaking connections dynamically, but this is not used in the surviving programs which hard code a destination (channel). I wondered about that when Noel first found the sources for ’nfs’. Maybe this connection protocol was planned but never implemented.

> a lesser reason was that the controller was a
> tempo computer that no one loved. the system
> software sucked. quickly it became unmaintained.
> i think tempo went out of business. anyway, the
> spider controller was the first and only tempo
> computer that i saw or even heard of.

The video that Sandy Fraser recorded has a segment about that at 25:30. Apparently, just before Christmas Ed David gave him $60K budget to buy a computer to build the Spider switch. The catch was that it had to be delivered before New Year. With the above sentiment in mind it was perhaps no surprise that Tempo had a unit sitting on the shelf that could be shipped in that time frame.

> On Sun, Jan 26, 2020 at 6:05 AM Paul Ruizendaal <pnr@planet.nl> wrote:
> I noted with much pleasure that the main bitsavers site is back up, and that at some point it has added a full set of scans of “Datamation”. The Feb 1975 issue contains an article from Dr. Fraser about Spider and the network setup in Murray Hill early in 1975:
> http://bitsavers.org/pdf/datamation/197502.pdf
> 
> For ease of reference I have also temporarily put the relevant 4 pages of the issue here:
> https://gitlab.com/pnru/spider/blob/master/spider.pdf
> 
> I find the graphic that shows how Spider connected machines and departments the most interesting, as it helps understand how the pro’s and con’s of Arpa Unix might have been perceived at that time.
> 
> The more I read, the more confused I become whether the “Pierce loop” was a precursor to “Spider” or a parallel effort.
> 
> The facts appear to be that John Pierce (https://en.wikipedia.org/wiki/John_R._Pierce) submitted his paper to BSTJ in December 1970, essentially describing a loop network with fixed size short datagrams, suggesting T1 frames. It is quite generic. In February 1971 W.J. Kropfl submits a paper that describes an implementation of the ideas in the Pierce paper with actual line protocols and a TIU. In October 1971 C.H. Coker describes in a 3rd paper how to interact with this TIU from a H516 programming perspective.
> 
> Several Spider papers mention that the project was started in 1969 and that the first Spider link was operational in 1972. The team appears to be entirely different: the h/w is credited to Condon and Weller, and the s/w to Frazer, Jensen and Plaugher. The Spider TIU is much more complex (200 TTL chips vs. 50 in the Kropfl TIU). The main reason for that - at first glance - appears to be that in the Spider network the TIU handled guaranteed in order delivery (i.e managed time outs and retransmissions), whereas in the Kropfl implementation this was left to the hosts.
> 
> It would seem logical that the latter was an evolution of the former, having been developed at the same site at the same time. A 1981 book seems to take that view as well: “Local Computer Network Technologies” by Carl Tropper includes the text "Spider Spider is an experimental data communications network which was built at the Bell Telephone Laboratories (Murray Hill, New Jersey) under the direction of A. G. Fraser. A detailed description of the network is given by Fraser [FRAS74]. This network was built with the notion of investigating Pierce's idea of ...” The chapter is titled “The Pierce loop and its derivatives”. This is a much as Google will give me - if somebody has the book please let me know.
> 
> On the other hand, the Spider papers do not mention the Kropfl network or Pierce’s paper at all. The graphic in Datamation appears to show two Kropfl loops as part of the network setup. Yet, this is described in the accompanying text as "4. Honeywell 5l6: Supports research into comunications techniques and systems. The machine has a serial loop I/O bus threaded through several labs at Murray Hill. Equipment under test is connected either directly to the bus or to a minicomputer which is then connected to the bus. Also avail- able are graphics display terminals and a device that can write read-only memory chips.” Maybe this is a different bus, but if it is the same as the Kropfl loop, to call it a “serial loop I/O bus” suggests it was a parallel effort unrelated to Spider.
> 
> Does anybody on the list recall whether Spider was a parallel effort or a continuation of the earlier work?
>