Execution time in minutes, required to complete various computational problems, using the optimization techniques discussed.

<p>Panels A and B show the execution time as a function of problem size for 10,000 bootstrap resamples conducted on datasets varying in size (A) and time required to run a stochastic population model against the number of time steps (B). "Naive" R code, in which no optimizations are applied, uses most computing resources (solid lines in A and B). Optimized R code, with use of efficient functions and optimal data structures pre-allocated in memory (dashed lines in A and B), is faster. In both panels A and B, the largest speed-ups are obtained by using optimal R code (black lines). Subsequent use of parallelism causes further improvement (dot-dashed green line) in A. In panel B, using R's byte compiler improved execution time further above optimal R code (dotted lines in green) while the smallest execution times were achieved by refactoring code in C (red dot-dashed lines). Panels C and D give the computing time (in minutes) needed to conduct the calculations from (C) Merow et al. [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.ref010" target="_blank">10</a>] and (D) the calculations represented by Fig. 3 in Visser et al. [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.ref011" target="_blank">11</a>]. Bars in panel C represent the original unaltered code from [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.ref010" target="_blank">10</a>] (I), the unaltered code run in parallel (II), the revised R code where we replaced a single data.frame with a matrix (III) and the revised code run in parallel (IV). Bars in panel (D) represent the original unaltered code [<a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.ref011" target="_blank">11</a>] (I), original run in parallel (II), optimized R code (III), optimized R code using R's byte compiler (IV), optimized R code run in parallel (V), optimized R code using byte compiler run parallel (VI), code with key components refactored in C (VII), and parallel execution of refactored code (VIII). All parallel computations were run on 4 cores, and code is provided in <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.s001" target="_blank">S1 Text</a>, section 3, <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.s002" target="_blank">S2</a> and <a href="http://www.ploscompbiol.org/article/info:doi/10.1371/journal.pcbi.1004140#pcbi.1004140.s003" target="_blank">S3</a> Texts.