Parallel and Distributed Systems Lab (PDSL)

Home Contents

Research

 

News
People
Research
Publications
Facilities
Support
Links
Private

Research within the lab fluctuates over time following the evolving interests of the lab members (both faculty and students). Currently the primary research themes being investigated in the PDSL are:

Effective and Efficient Communication in Computational Clusters

As the use of distributed memory compute clusters increases to meet cost and scalability demands, the importance of providing effective and efficient communication between cluster nodes is growing. Within the PDSL we seek to assess and improve the efficiency of various low-latency interconnects using the "Crash And Burn (CAB)" research cluster as an evaluation tool. We also  try to simplify the process of programming clusters through unique applications of DSM (Distributed Shared Memory) techniques.

Lock-free Concurrent Data Structures

In shared memory parallel computing environments (including small scale multiprocessor and multi-core architectures) the use of traditional lock based synchronization methods introduces unnecessary overhead and undesirable scheduling anomalies. In the PDSL we are designing and implementing key algorithms for complex data structures (most notably balanced trees) using lock-free techniques in an attempt to address these issues.

Mobile Computing

Mobile computing with such devices as PDAs, cell phones and laptops is becoming increasingly common. A large number of capabilities we take for granted in the wired network world are missing or poorly supported in mobile scenarios, especially those based on Ad-Hoc networks. Our interest in the PDSL is to address such fundamental problems as dynamicity, reliability and power management in an effort to build underlying services for mobile computing that are useful to all.

Pervasive Computing

The ubiquity of computing devices is continually increasing with the number of embedded computers being deployed now outnumbering the number of "visible" machines by multiple orders of magnitude. Integrating all of these devices together so that they can interoperate to the advantage of human users is an immense challenge. Of particular interest in the PDSL are issues of interoperability, context sensitivity, automation of device discovery and interaction and mechanisms for the automated addition of new services.

Wide Scale Distributed Computing (Grids)

The sharing of computing resources has grown from LAN-based sharing of disks and printers through institution-wide sharing of computers to wide-area/Grid-based sharing of machines, datasets, software and high-end visualization facilities. The key challenges in making such systems work and be useful to end users who are typically not computer scientists/engineers are the scale, autonomy and inherent unreliability of such systems. This is often compounded by the very long running times of many jobs. In the PDSL we are interested in adding dynamicity and reliability into grids with a focus on resource discovery and management.

 

Send mail to pgraham@cs.umanitoba.ca with questions or comments about this web site.
Last modified: 03/23/06