Distributed big data in scientific computing presents a major I/O performance bottleneck
when exploiting data paral-lelism. Consumer and producer compute nodes are often …