Talk:Speedup
From Wikipedia, the free encyclopedia
Surely there is a conflict here!
"It is not uncommon to observe more than N speedup when using N processors in parallel computing, which is called super linear speedup. Super linear speedup rarely happens...."
not uncommon = happens moderately often
rarely happens = happens hardly ever
This needs tidying up. David Martland 11:32, 7 July 2006 (UTC)
[edit] parallel computing?
OK the definition currently goes like "In parallel computing, speedup refers to how much a parallel algorithm is faster than a corresponding sequential algorithm.", but the way I learned about speedup in uni is that it's a much more general thing in computer architecture and doesn't have to do with parallel computing. You could talk about speedup about any technology that makes processing faster. For example, you could talk about the speedup of using MMX-optimized code versus non-MMX code. This whole article focuses around parallelisation and the number of processors, and I think all of this needs rewriting, or maybe merging with amdahl's law. Someone please give your input on this. Alex.g (talk) 15:01, 23 January 2008 (UTC)
[edit] Explanation of parallel efficiency
Where does the $ \frac{1}{\log p} $ for "many difficult-to-parallelize algorithms" come from? Could someone please provide an example of an algorithm with such poor parallel efficieny? --Rolf (talk) 21:59, 22 March 2008 (UTC)
[edit] discussion of superlinear speedup misleading
It seems to me that superlinear speedup as defined is heavily dependent on the sequential algorithm. If the sequential algorithm is suboptimal, then the speedup can be anything and superlinear speedup will not be surprising. I suggest that for the purposes of discussing how big speedup can be as a function of the number of processors, it is better to assume that the sequential algorithm is optimal. For example, if the multiprocesor algorithm can have a thread that stops execution of a different thread, then why can't the sequential algorithm have the same threads? In addition, there is ambiguity about the problem size in regards to the multiprocessor algorithm. Do we assume that we are solving exactly the same problem with P processors as with 1 processor? If so, then we have to deal with the size of the memory when there are P processors. If the multiprocessor is allowed to have more memory then speedup may be due to the size of the memory rather than the number of processors. In the extreme case, the problem might be intractible with a single processor because of lack of memory. Speedup would then be infinite. Ec23nav (talk) 04:08, 16 April 2008 (UTC)
