CrouZ (2) [Avatar] Offline
#1
In section 1.5.2 it is stated that B is zero without synchronization, which I believe is promising too much. I mean, there is still serialization from, for example, the code sequence for each actor. Don't you rather mean that synchronization does not add unnecessary serialization?
roland.kuhn (39) [Avatar] Offline
#2
Re: Amdahl's law
No, what is written is that IF there is no synchronization necessary (meaning purely parallel non-communicating processes) then B=0 and scalability is perfect. The following paragraph then details why that limit cannot be reached, so I don’t think that we’re promising too much.
CrouZ (2) [Avatar] Offline
#3
Re: Amdahl's law
But don't the parallel non-communicating processes themselves have internal serialization, for example, a sequence of statements to execute?
roland.kuhn (39) [Avatar] Offline
#4
Re: Amdahl's law
The question is: if I run five of these completely independent processes, each on one fifth of the work, will they finish in one fifth of the time that it takes a single process to perform the whole work? Executing statements sequentially within one process is always implicitly the case, the only issue concerning scalability is inter-process synchronization.

(I’m using the word process here synonymously for any execution mechanism, e.g. threads or O/S processes)