A certain algorithm may be parallelized onto a cluster of N nodes such that each node in the cluster operates on 1/Nth the total data, but must also exchange its results with the (N-1) other nodes in the network. An unusual characteristic of this code is that it is completely parallelizable (there is no serial portion of the code). Based on this characteristic, what would the expected speedup be for 2, 8, and 32 nodes (show your work)?

Since there no serial code, the speedup will linearly with processor nodes (2, 8, 32)

Based on this analysis, does this algorithm appear to be a good candidate for parallelization (justify your answer)?