On the Throughput Optimization in Large-Scale Batch-Processing Systems

We analyze a data-processing system with n clients producing jobs which are processed in batches by m parallel servers; the throughput critically depends on batch size and corresponding sub-additive speedup function that arises due to overhead amortization. In practice, optimization relies numerical searches for optimal is computationally cumbersome. this paper, we model terms of closed queueing network assuming certain forms service speedup; standard Markovian analysis yields w n4 time. Our main contribution mean-field has unique, globally attractive stationary point, derivable form. This point characterizes asymptotic as can be calculated O(1) Numerical settings from large commercial reveal optimum accurate practical finite regimes.