(j3.2006) (SC22WG5.4029) [ukfortran] LOCK/UNLOCK question

[Van Snyder]

Malcolm Cohen wrote:
> I might not agree with Van that LOCKs are too low-level to be 
> effectively usable....
It depends upon the definition of "effectively."  JPL + Univac got Exec 
to run for days on a 4-processor 1108 without crashing.  The machine 
typically ran with 50 time-sharing jobs, 10 batch jobs, and an 
occasional real-time job -- with 262144 36-bit words of core and 11 core 
loads of 4ms swap drums.  This is longer than Windows does on a single 
processor with a far simpler load.  But all too frequently we found a 
bug having to do with synchronization (usually a deadlock) or we never 
found it (not revealed on a core dump, and not reproducible).  When we 
started paying more attention to Hoare and Per Brinch-Hansen, and trying 
to emulate their ideas in assembler with test-and-set, things got better.

Test-and-set worked, but if you looked at the labor cost of making it 
work, and reliability that was probably less than might have been 
achieved with more structured tools, you might argue against "effectively."