Weighted Flowtime on Capacitated Machines

It is well-known that SRPT is optimal for minimizing flow time on machines that run one job at a time. However, running one job at a time is a big under-utilization for modern systems where sharing, simultaneous execution, and virtualization-enabled consolidation are a common trend to boost utilization. Such machines, used in modern large data centers and clouds, are powerful enough to run multiple jobs/VMs at a time subject to overall CPU, memory, network, and disk capacity constraints. Motivated by this prominent trend and need, in this work, we give the first scheduling algorithms to minimize weighted flow time on such capacitated machines. To capture the difficulty of the problem, we show that without resource augmentation, no online algorithm can achieve a bounded competitive ratio. We then investigate algorithms with a small resource augmentation in speed and/or capacity. Our first result is a simple (2 + e)-capacity O(1/e)-competitive greedy algorithm. Using only speed augmentation, we then obtain a 1.75-speed O(1)-competitive algorithm. Our main technical result is a near-optimal (1+e)-speed, (1+e)-capacity O(1/e3)-competitive algorithm using a novel combination of knapsacks, densities, job classification into categories, and potential function methods. We show that our results also extend to the multiple unrelated capacitated machines setting.