Evaluation of Object Placement Techniques in a Policy-Managed Storage System

Storage management cost is a significant fraction of the total cost of ownership of large, enterprise storage systems. Consequently, software automation of common storage management tasks so as to reduce the total cost of ownership is an active area of research. In this paper, we consider a policy-managed storage system—a system that automates various management tasks—and focus on the problem of the storage allocation techniques that should be employed in such a system. We study two fundamentally different storage allocation techniques for policy-managed systems: narrow and wide striping. Whereas wide striping techniques need very little workload information for making placement decisions, narrow striping techniques employ detailed information about the workload to optimize the placement and can achieve better performance. We systematically evaluate this trade-off between “information” and performance using a combination of simulations and experiments on a storage system testbed. Our results show that an idealized narrow striped system can outperform a comparable wide-striped system at low levels of utilization and for small requests. However, wide striping outperforms narrow striped systems when the system utilization is high or in the presence of workload skews that occur in real I/O workloads. Our experiments show that the additional workload information needed by narrow placement techniques may not necessarily translate to better performance. We identify the stripe unit size to be a critical parameter in the performance of wide striped systems, and based on our experimental results, recommend that (i) policy-managed systems use wide striping for object placement, and (ii) sufficient information be specified at storage allocation time to enable appropriate selection of the stripe unit size.