Fast and Scalable Startup of MPI Programs in InfiniBand Clusters

摘要

One of the major challenges in parallel computing over large scale clusters is fast and scalable process startup, which typically can be divided into two phases: process initiation and connection setup. In this paper, we characterize the startup of MPI programs in InfiniBand clusters and identify two startup scalability issues: serialized process initiation in the initiation phase and high communication overhead in the connection setup phase. To reduce the connection setup time, we have developed one approach with data reassembly to reduce data volume, and another with a bootstrap channel to parallelize the communication. Furthermore, a process management framework, Multi-Purpose Daemons (MPD) system is exploited to speed up process initiation. Our experimental results show that job startup time has been improved by more than 4 times for 128-process jobs, and the improvement can be more than two orders of magnitude for 2048-process jobs as suggested by our analytical models.