[TUHS] Alternative Implementation Proposal for Unix/370 (BTL, 1979)

[arnold at skeeve.com]

Thanks Noel.

Those reasons are quite compelling. One gets the sense that they wanted
to get UNIX going on the 370 as quickly as posible.

jnc at mercury.lcs.mit.edu (Noel Chiappa) wrote:

>     > From: Tom Lyon
>
>     > there were a few icustomer nstallations. Bell Labs Indian Hill was one
>     > - so that's why TSS was the base of their UNIX port.
>
> "A UNIX System Implementation for System/370" (by W. A. Felton, G. L. Miller,
> and J. M. Milner):
>
>   https://www.bell-labs.com/usr/dmr/www/otherports/ibm.html
>
> says "code to support System/370 I/O, paging, error recording and recovery,
> and multiprocessing already existed in several available operating systems,
> we investigated the possibility of using an existing operating system, or at
> least the machine-interface parts of one, as a base to provide these
> functions for the System/370 implementation ... Of the available systems,
> TSS/370 came the closest to meeting our needs and was thus chosen as the base
> for our UNIX system implementation". Alas, it doesn't say which other systems
> were also considered.
>
>
>     >> On May 6, 2022, at 09:39, arnold at skeeve.com wrote:
>
>     >> So, why, given the letter from these folks, including DMR, did they go
>     >> ahead and use the TSS solution anyway?
>
> That paper says: "We initially thought about porting the UNIX operating
> system directly to System/370 with minimal changes. Unfortunately, there are
> a number of System/370 characteristics that, in the light of our objectives
> and resources, made such a direct port unattractive. The Input/Output (I/O)
> architecture of System/370 is rather complex; in a large configuration, the
> operating system must deal with a bewildering number of channels,
> controllers, and devices, many of which may be interconnected through
> multiple paths. Recovery from hardware errors is both complex and
> model-dependent. For hardware diagnosis and tracking, customer engineers
> expect the operating system to provide error logs in a specific format;
> software to support this logging and reporting would have to be written. ...
> Finally, several models of System/370 machines provide multiprocessing, with
> two (or more) processors operating with shared memory; the UNIX system did
> not support multiprocessing."
>
> Presumably these factors outweighed the factors listed in the
> Haley/London/Maranzaro/Ritchie letter.
>
> 	Noel