Improving parallelism in I/O systems
Keok-Kee Lee, Peter Varman · 2002
The problem of improving the performance of external merging in a parallel I/O system using read-ahead prefetching and disk scheduling is studied. A Markov model is developed and numerically solved for a two-disk two-run system. Larger configurations are studied using simulation. A scheduling algorithm to maximize I/O parallelism is also presented.