Improving energy efficiency and performance in storage server systems

A wide variety of networked services such as web hosting, e-business and out-sourced storage drive the need for deploying large scale server storage systems. In server storage systems, the performance has been a long-term research topic, and will be continuously important with the advent of new storage architectures such as SAN and iSCSI. iSCSI leverages existing Ethernet infrastructure and well-understood TCP/IP protocol suite by encapsulating SCSI command set and block-level data transfer over IP, pushing current networked storage systems evolving from expensive FC (Fibre Channel) SAN to economic SAN built over the iSCSI. Playing a core role in the economic SAN of the back-end storage systems, the most challenging problem for iSCSI-based IP storage servers is the performance when comparing with FC SAN. In addition to the performance issue, the increasing cost of energy consumption is attracting considerable research on techniques for energy conservation in server storage systems. Earlier energy saving solutions for mobile devices usually have significant performance degradation when applying in the server-side environment. Many DRPM disk-based solutions have been proposed to solve the performance concern at server side. But the manufacturing challenges and costs have so far prevented mass commercial deployment of multi-speed disks, which have largely been confined to experimental laboratory studies. Reducing energy consumption without compromising performance is still an open question in today\u27s high performance conventional disk-based storage array architectures. With a focus on both energy-efficiency and high performance aspects of modern server-side storage systems, this dissertation develops novel techniques to address the aforementioned challenges by: (1) constructing an effective NIC caching (DCA) to reduce local-interconnect traffic and thus boosting iSCSI-based storage server performance; (2) exploiting redundancy in the context of hierarchical cache architecture (RIMAC) to save energy and simultaneously improve performance for parity-based storage systems; and (3) flexibly trading off performance and energy saving with Power Aware Load Balancing (PALB) and novel layouts (GLL) in the replication-based storage systems