External sorting on a reconfigurable message-passing multicomputer: experimental results and analysis

An actual implementation of the external sorting problem on a multicomputer, with careful attention paid to the overlap between computations and I/O in order to minimize total execution time, is reported. The problem is divided into two steps. The first involves creation of multiple sorted runs. The second involves merging the runs. Step 1 is accomplished using pipelined sort; step 2 is implemented on a tree of processors. An analytical model for step 1 is presented. The execution time predicted by the proposed analytical model is compared with the experimental results