🚀 go-pugleaf

In article <1992Aug18.111938.19628@hubcap.clemson.edu>, Ian Heavens <ian@spider.co.uk> writes:

|> This issue becomes important because multiple-reader/single-writer
|> locks are much more convenient to use, and reduce contention, at
|> a cost of increasing the overhead (three times that of a mutex lock?).
|> 
|> Are their cache consistency or other reasons to inflate the overhead
|> of testing and setting a memory location?  What have I missed?
|> 
|> Can someone suggest a rough equivalent for a spin lock acquisition in
|> terms of (some arbitrary mix of) instructions execution, where both 
|> instruction and data fetch are cache hits?

Try running some industrial CFD applications with writable global data 
(like the Barnes-Hut algorithm) on 100's of processors.  You will literally
see the machine comes to a halt...

Our experience on the Stanford DASH machine is that, when simulating
iPSC/860 functions such as those global gadd or gXXX functions, the 
contention can get so bad that a single instruction on a vector machine
can turns into 20,000+ instructions, all because of lock-spinning.
We end up having to redesign the algorithm.

It all comes down to how "real" your application is.

Regards,
  Bernard Lee, Stanford CS Dept.