Online Demand Scheduling with Failovers

TL;DR

This paper addresses the problem of optimally deploying hardware in cloud data centers under power and robustness constraints, introducing the Online Demand Scheduling with Failover problem and proposing algorithms for worst-case and stochastic demand models.

Contribution

It formulates a new online assignment problem with failover constraints and provides algorithms with provable guarantees for both worst-case and stochastic demand scenarios.

Findings

Deterministic algorithm achieves approximately 50% competitiveness in worst-case.

Stochastic model algorithms attain sub-linear regret, approaching optimal utilization as demands grow.

New techniques include a configuration LP and an online monotone matching procedure.

Abstract

Motivated by cloud computing applications, we study the problem of how to optimally deploy new hardware subject to both power and robustness constraints. To model the situation observed in large-scale data centers, we introduce the Online Demand Scheduling with Failover problem. There are $m$ identical devices with capacity constraints. Demands come one-by-one and, to be robust against a device failure, need to be assigned to a pair of devices. When a device fails (in a failover scenario), each demand assigned to it is rerouted to its paired device (which may now run at increased capacity). The goal is to assign demands to the devices to maximize the total utilization subject to both the normal capacity constraints as well as these novel failover constraints. These latter constraints introduce new decision tradeoffs not present in classic assignment problems such as the Multiple Knapsack problem and AdWords. In the worst-case model, we design a deterministic $\approx \frac{1}{2}$-competitive algorithm, and show this is essentially tight. To circumvent this constant-factor loss, which in the context of big cloud providers represents substantial capital losses, we consider the stochastic arrival model, where all demands come i.i.d. from an unknown distribution. In this model we design an algorithm that achieves a sub-linear additive regret (i.e. as OPT or $m$ increases, the multiplicative competitive ratio goes to $1$). This requires a combination of different techniques, including a configuration LP with a non-trivial post-processing step and an online monotone matching procedure introduced by Rhee and Talagrand.