Free-ebooks-library - Bahar Ali Khan

Figure 22-4. Chunk versus range partitioning
calculator might perform poorly with range partitioning. An example of when range
partitioning would do well is in calculating the sum of the square roots of the first
10 million integers:
ParallelEnumerable.Range (1, 10000000).Sum (i => Math.Sqrt (i))
ParallelEnumerable.Range returns a ParallelQuery, so you don’t need to subsequently
call AsParallel.
Range partitioning doesn’t necessarily allocate element ranges
in contiguous blocks—it might instead choose a “striping” strategy.
For instance, if there are two workers, one worker might
process odd-numbered elements while the other processes
even-numbered elements. The TakeWhile operator is almost certain
to trigger a striping strategy to avoid unnecessarily processing
elements later in the sequence.
Optimizing custom aggregations
PLINQ parallelizes the Sum, Average, Min, and Max operators efficiently without additional
intervention. The Aggregate operator, though, presents special challenges
for PLINQ. As described in Chapter 9, Aggregate performs custom aggregations. For
example, the following sums a sequence of numbers, mimicking the Sum operator:
PLINQ | 889
Parallel
Programming