Re: Parallelizing (WAS: Death by pointers.)

>[I once read an IBM tech report that claimed that 10 slow processors give you
>considerably less throughput than one 10X processor in most database
>applications because of increased lock contention on "hot spot" records.
>But I'm not sure that's a compiler issue. -John]


This is a general result from Queueing theory. If you have the choice
between one server that can operate at speed n, or m servers that
operate at speed n/m, then you will be better off to take the one
server at speed n. The average time in the system, and the average
wait time will be much lower.


A good reference for this can be found in Sheldon Ross' _Introduction
to Probability Models_. Third Adition. Academic Press, Inc. ISBN
0-12-598463-4. Orlando, 1985. In this edition, Chapter 8 presents
this result. As this book is intended for undergraduate studies (or
first year graduate studies), it limits its results to the M/M/1 and
M/M/k systems (although M/G/1 systems are treated briefly). M/M/1
means that the system's arrival time deltas have a Poison
distribution, as well as its servicing time having a Poison
distribution. This simple model covers a wide variety of systems.


Warner


--

Return to the comp.compilers page.
Search the comp.compilers archives again.