Ph.D. Defense of Dissertation Announcement

Title: Informed Storage Management for Mobile Platforms

Hyojun Kim
School of Computer Science

College of Computing

Georgia Institute of Technology

Date: July 12 (THU), 2012
Time: 2:00pm - 4:00pm (in EDT)
Location: KACB 1116 East

Committee:

• Dr. Umakishore Ramachandran (Advisor, CoC, GATECH)
• Dr. Ling Liu (CoC, GATECH)
• Dr. Marilyn Wolf (ECE/CoC, GATECH)
• Dr. Moinuddin K. Qureshi (ECE, GATECH)
• Dr. Remzi H. Arpaci-Dusseau (CS, University of Wisconsin–Madison)

Abstract:
Smartphones are not just ordinary phones; they are being used as mobile platforms for serving the computing needs of a significant segment of the user community.  We believe that storage in smartphones is an Achilles' heel to achieving the full performance potential of smartphones for meeting the future computing needs of the user community that is starting to become more heavily dependent on mobile platforms.  The poor performance of storage in smartphones can be attributed to the fact that smartphones typically employ flash storage, and the OS storage stack is not optimized to deal with the performance quirks of such storage.

I have proposed multiple storage software solutions to handle the durability issue and to improve the performance of Flash storage. There are always design tradeoffs involved in building software systems. RAM based write buffering can enhance storage performance, but sacrifices reliability. Redundancy (via replication) can improve data availability, but it has implications for data consistency. Log-structured design to combat the “small write” problem solves the random write issue in Flash storages, but introduces overhead for maintaining mapping information between logical and physical blocks.

In this proposed research, entitled Informed Storage Management (ISM), I aim at providing a dynamic decision-making framework for system design, specifically targeted to storage systems on mobile platforms. The goal of ISM is maximizing the performance benefits while minimizing the side effects of the design choice. As a concrete example of ISM, we provide mechanisms along three axes, namely, type, temporal, and spatial for selectively supporting write-back buffering, which can be used judiciously by the upper layers of the operating system.  We implement and evaluate our solution on a real Android smartphone, and demonstrate significant performance gains for everyday apps on such platforms.