Title: inTune: Achieving Performance and Isolation Properties in Virtualized Islands-based Multicore Systems

Priyanka Tembey
School of Computer Science
College of Computing,
Georgia Institute of Technology

Date: April 6th (FRI), 2012
Time: 11:00am - 1:00pm (EST)
Location: KACB 3402

Committee:

• Dr. Karsten Schwan (Advisor, School of Computer Science, Georgia Tech)
• Dr. Ada Gavrilovska (Advisor, School of Computer Science, Georgia Tech)
• Dr. Sudhakar Yalamanchili (School of Electrical and Computer Engineering, Georgia Tech)
• Dr. George Cox (Intel Research Labs, Oregon)
• Dr. Thomas Gross (Department of Computer Science, ETH, Zurich)

Abstract:
Multicore platforms are moving from small numbers of homogeneous cores toward multiple sets or islands of potentially heterogeneous cores residing on a single chip. Beyond core heterogeneity, there may be heterogeneity in memory structures and functional differences in per-island resource managers, an example being managers that supervise communication vs. computational tasks. Hardware structured as ’islands of cores’ may offer improved scalability and functionality, but for applications to realize such opportunities, systems software must meet challenges that include
(a) providing end-to-end performance guarantees for applications operating across multiple islands, (b) mediating use of shared resources and offering isolation guarantees for multiple applications running on consolidated platforms, and
(c) delivering global properties such as platform-level power or utilization caps.

The goals of this thesis addresses (a)-(c) as follows:
(1) by developing system-level mechanisms for addressing challenges (a)-(c), (2) by demonstrating their ability to deliver improved application performance and platform efficiency, and (3) by creating principles and representative methods for realizing the isolation properties sought by applications and the efficiency sought for plat- forms. The concrete realization of these goals is inTune - a framework for inter-island operation, offering APIs and mechanisms that permit resource islands and their resource managers to jointly achieve application performance guarantees and global platform-level properties. Its hypervisor-level implementation provides such properties to arbitrary operating systems and applications, without requiring their complex functionalities to be changed. Per application or platform-level policy objectives are implemented as inTune ’overlays’ that efficiently interact with islands and their resource managers. For multiple consolidated applications (and their overlays), there are additional methods that mon- itor and mediate their use of shared platform resources, while decision control models built into inTune ensure basic properties like progress and stability.

InTune is implemented in the Xen hypervisor for a range of multi-core systems, including a heterogeneous communication-accelerator - the IXP network processor, and for x86 multi-core platforms with multiple NUMA memory nodes. Evaluation with representative parallel, web-based, and real-time applications and application mixes demonstrate the benefits of using inTune to achieve performance and platform policy objectives.