Research Projects

Power and Thermal Management in Storages (2005-Present)

  • Power and Thermal Management in Disk-based Storage Systems

    • Growing power density makes thermal-aware computing crucial to the design of all computer system components ? from micro-processors to peripherals within a computer system and from server units in rack-mounted servers to even entire datacenter machine rooms. Thermal-aware design of disk drives is important because high temperatures can cause reliability problems. Dynamic thermal management (DTM) techniques have been proposed to operate the disk at the average case temperature, rather than at the worst case by modulating the activities to avoid thermal emergencies caused by unexpected events, such as fan-breaks, increased inlet air temperature, etc. A delay-based approach to adjust the disk seek activities is one such DTM solution for disk drives. Even if such a DTM approach could overcome thermal emergencies without stopping disk activity, it suffers from long delays when servicing the requests. In this project, we investigate the possibility of using a multispeed disk drive (called dynamic rotations per minute (DRPM)), which dynamically modulates the rotational speed of the platter for implementing the DTM technique. Using a detailed performance and thermal simulator of a storage system, we evaluate two possible DTM policies?time-based and watermark-based?with a DRPM disk-drive and observe that dynamic RPM modulation is effective in avoiding thermal emergencies. Also we explore how flash memory can help in designing such a solution for rack-mounted hybrid storage servers combining solid-state and hard disk drives.
      (Papers: THETA07, ASME Transactions 2008)

  • Understanding Performance-Thermal Behavior in Disk I/O of Server Workloads

    • Designing and optimizing DTM techniques requires a careful analysis of how different drive activities impact the temperature, using real workloads. For instance, if we have a disk operating at a given RPM, how do the seeks in the workload increase the temperature? How far apart do seeks need to be in order to remain within the thermal envelope? Within a seek, how do the different phases - acceleration, coast, deceleration - impact the temperature? Can we modulate the head scheduling or request service schemes for DTM? Given different DTM alternatives, how do we pick one over another for a given set of workload conditions and disk drive parameters? Such a detailed understanding of the interaction between workload activities and disk drive parameters, and their impact on temperature, requires detailed toolsets that are currently unavailable. Though there are tools such as Disksim which are widely used for performance studies, there is no tool available today to study the temperature of a drive running a real workload. The earlier work in thermal modeling of disk drives has been more intended to study the temperature of drives under steady state conditions for static configurations of different drive parameters, and have not really looked at the temperature during the dynamic execution of a workload. With these motivations, in this project we presents the first integrated performance-thermal simulator to study the temperature of disk drives with real workloads. We profile the thermal behavior of real server workloads and show how the temperature varies during the execution. We also show that the spatial locality (minimizing seek activity) and the temporal separation between the seeks is adequate in these workloads.
      (Papers: TR CSE05-007, HPCA06, IEEE MICRO06)

  • Project Homepage: Not available