Skip to main content
. Author manuscript; available in PMC: 2013 Feb 6.
Published in final edited form as: Proc IPDPS (Conf). 2011;2011:435–443. doi: 10.1109/ipdps.2011.184

Figure 4.

Figure 4

Plots for the relative speedups for different optimizations. Note that each axis is logged, so a straight line would show optimal speedup. Splitting the reads across multiple nodes achieves linear speedup. Splitting the memory across multiple nodes can obtain nearly log-linear speedup. Because of the high cost of thread synchronization, linear speedup is not possible beyond 8 threads.