Providing fairness and system efficiency are important, often conflicting, requirements when allocating shared resources. In a hybrid storage system the problem is complicated by the high variability in request service times, caused by speed differences between heterogeneous devices and workload-specific variations in access time within a device.

Providing QoS guarantees for hybrid storage systems made up of both solid-state drives (SSDs) and hard disks (HDs) is a challenging problem. Since HDs and SSDs have widely different IOPS capacities, it is not sensible to treat the storage system as a monolithic black box, instead a useful QoS model must necessarily differentiate the IOs made to different device types. Traditional storage resource allocation models have largely been designed to provide QoS for a single resource type, and result in poor utilization and fairness when applied to multiple coupled resources. In this paper, we present a new resource allocation model for hybrid storage systems using a multi-resource framework. The model supports reservations and shares for clients sharing the storage system. Reservations specify the minimum throughput (IOPS) that a client must receive, while shares reflect its weight relative to other clients that are bottlenecked on the same device. We present a formal multi-resource allocation model to allocate IOPS to clients, together with an IO scheduling algorithm to maximize system throughput. The model and algorithms are validated with empirical results.

The paper examines the problem of fair bandwidth allocation in heterogeneous storage systems in the framework of multi-resource allocation. We first extend the Bottleneck Aware Allocation model recently proposed by the authors to directly compute the maximum allocation satisfyinglocal fairness, envy freedom and sharing incentive. Next, we broaden the solution space to all allocations that satisfy envy freedom and sharing incentive even if they do not satisfy local fairness. We present an efficient algorithm to maximize the system utilization in the more general model.

Data center computing is gaining popularity due to the cost efficiencies of consolidation, centralized management, and high reliability. The unpredictable bursty nature of typical workloads where the instantaneous arrival rates can significantly exceed the average long-term rate, requires the server to significantly over provision resources in order to meet response-time service level agreements (SLAs), resulting in low resource utilization and higher costs. In this paper we consider a statistical on-off model for bursty workloads to explore the relationship between burst size, frequency, capacity, and response time distribution. We analyze the performance of a workload shaping method to reduce capacity requirements by providing a graduated two-level SLA. Statistics of the request arrival to the overflow queue are characterized in terms of the underlying Markov chain passage times, and used to estimate its capacity.