Hello......
Please download here the free book about USL methodology from Baron
Schwartz:
https://www.vividcortex.com/resources/universal-scalability-law/
You can read clearly inside the book this:
--
"To forecast scalability beyond what’s observable, be pessimistic and
treat the USL as a best-case scenario that won’t really happen. Use
Equation 4 to forecast the maximum possible throughput, but don’t
forecast too far out. Use Equation 6 to forecast response time."
--
As you have noticed he is not saying "don't forecast", he
is saying clearly: "don’t forecast too far out", that means
that you can forecast not too far out. And that's illogical,
simply because, in a parallel program, a locked region using
locks can escape contention at fewer cores and fewer threads
and can than hit the contention at more cores and more threads..
so you can not use nonlinear regression to predict scalability,
so why Baron Schwartz is saying: "don’t forecast too far out",
so i mean why then forecast, since you can not predict scalability ?
that's like illogical from Baron Schwartz ! so what is the purpose
of the USL methodology that uses nonlinear regression? its purpose
is to permit us to test and see how much there is contention and how
much there is crosstalk(cache misses etc..), that's all, it's not a tool
to predict scalability.
Thank you,
Amine Moulay Ramdane.
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)