Ph.D. Thesis Proposal: Vishal Gupta

Add to Calendar
May 4, 2012 2:00 pm - 4:00 pm
KACB 3108

Title: Energy-Efficient Resource Management on Heterogeneous Platforms

Vishal Gupta
School of Computer Science
College of Computing
Georgia Institute of Technology

Date: May 4th (Friday), 2012
Time: 2:00PM - 4:00PM (EST)
Location: KACB 3108


  • Dr. Karsten Schwan (Advisor, School of Computer Science, Georgia Tech)
  • Dr. Sudhakar Yalamanchili (School of Electrical and Computer Engineering, Georgia Tech)
  • Dr. George Cox (School of Computer Science, Georgia Tech)
  • Dr. Scott Hahn (Intel Labs, Hillsboro) Dr. Ricardo Bianchini (Department of Computer Science, Rutgers University)

Recognizing that energy-efficiency remains a critical concern for both server systems and mobile devices, heterogeneous platform organizations, consisting of components that are not identical, have been proposed to improve the efficiency of future platforms. However, introduction of such heterogeneous architectural components also requires system software to embrace heterogeneity for efficient operation. To address these issues, this dissertation presents resource management methods to exploit the underlying heterogeneity by matching execution resources to application needs, and thus provide higher performance and also improve energy-efficiency. It also highlights the need for hardware and system software to consider the non-CPU resources on future platforms to obtain such efficiencies.

To this end, this dissertation makes the following specific contributions:

  • A solution for client devices that uses heterogeneous processors to provide a high dynamic power/performance range. It uses 'core groups' as the abstraction that groups a small number of heterogeneous cores to form a single execution unit. Also demonstrated is the growing importance of uncore power and the need for a scalable uncore design to realize the intended gains from heterogeneous cores.
  • An opportunity analysis of the design space of multicore processors exploring datacenter applications with SLAs. Using analytical models, it compares the relative benefits of heterogeneous processors over area-equivalent homogeneous configurations and also highlights associated practical challenges for using these architectures in datacenters.
  • A power management solution for future multicore platforms containing over-provisioned resources. Exploiting over-provisioning using heterogeneity, it presents the design of a resource manager which governs allocation of constrained energy resources to applications in order to meet budgeting constraints, user-preferences, and system-wide policies.
  • A system-level resource management technique for heterogeneous memory platforms containing both on-chip stacked DRAM and traditional off-chip DRAM. By heterogeneity-aware data placement and power shifting techniques, it improves the utilization of the scarce stacked-DRAM resources, resulting in improved performance.