Utility Optimal Thread Assignment and Resource Allocation in Multi-Server Systems

Achieving high performance in many multi-server systems (e.g., web hosting center, cloud) requires finding a good assignment of worker threads to servers and also effectively allocating each server’s resources its assigned threads. The allocation components this problem have been studied extensively but largely separately the literature. In paper, we introduce assign allocate (AA) problem, which seeks simultaneously find an that maximizes total utility Assigning together can result substantially better overall than performing steps separately, as is traditionally done. We model thread by function giving resources. first prove AA NP-hard. then present $2 (\sqrt {2}-1) > 0.828$ factor approximation algorithm for concave functions, runs notation="LaTeX">$O(mn^{2} + n (\log mC)^{2})$ time notation="LaTeX">$n$ notation="LaTeX">$m$ with notation="LaTeX">$C$ amount each. give faster same ratio notation="LaTeX">$O(n complexity. extend two more general settings. First, consider nonconcave 1/2 algorithm. Next, using multiple types resources, show achieves empirical performance. conduct extensive experiments test our algorithms on both synthetic realistic they achieve over 92% optimal average. compare number practical heuristics, up 9 times higher utility.