|
The Scalable Systems Lab is a research laboratory in the Computer Science Department at the University of New Mexico. The Laboratory enjoys a close affiliation with the UNM Center for High Performance Computing, the Scalable Computing Systems Department at Sandia National Laboratories, the Computer and Computational Sciences Division at Los Alamos National Laboratory, and research groups at Sun, Intel, and IBM Research.
The goal of the Scalable Systems Laboratory is to develop approaches for the design and implementation of large-scale, high-performance computing systems for resource constrained (Grand Challenge) applications. The approaches considered will be driven by the measurable needs of resource constrained applications.
Our goal is to provide an execution environment for resource constrained applications.
We define success by our ability to deliver 90% of the available resources to the application (e.g., 90% of the physical memory, 90% of the available communication bandwidth)
Our current focus is on building very large computing systems from commodity components.
To be able to take advantage of the delivered resources, applications must see easily predictable performance (e.g., message passing latencies) so they can be tuned for the environment.
Our approach is to minimize the services that are required and to make as many services optional as possible. That is, whenever a service can be moved to an application level library, without a significant performance or security penalty, we will prefer that approach. An alternative is to consider dynamic services that are migrated into and out of the lower levels as needed (this is tricky, because it may introduce unpredictable behavior into the application).
The challenge associated with using commodity components is the conflict between the intended use of the component and our planned use of the component.
Kurt Ferreira just had his paper entitled “Characterizing Application Sensitivity to OS Interference Using Kernel-Level Noise Injection accepted for publication at Supercomputing 2008. In addition, the paper has also been nominated for best paper and best student paper awards at the conference. Great job, Kurt!
~~DISCUSSION~~
Rolf Riesen, Ron Brightwell, Patrick Bridges, Trammel Hudson, Barney Maccabe, Patrick Widener, and Kurt Ferreira just had their paper entitled “Designing and Implementing Lightweight Kernels for Capability Computing” accepted for publication in Concurrency and Computation: Practice and Experience.
~~DISCUSSION~~
Wenbin, Patrick, and Barney have had their paper entitled “Lightweight Online Monitoring and Tuning with Embedded Gossip” accepted for publication in IEEE Transactions on Parallel and Distributed Systems.
Wenbin Zhu successfully defended her Ph.D. dissertation: “Lightweight Online Performance Monitoring and Tuning with Embedded Gossip”.
James Horey presented his paper, Kensho: A Dynamic Tasking Architecture for Sensor Networks, in April at the Workshop for Wireless Sensor Network Architectures at IPSN 2007 (Boston).
|
-
Cache Injection
In this project we investigate techniques for reducing memory bandwdith utilization due to incoming network messages in multiprocessor multi-core architectures.
ConfigOS
We are developing a framework for configuring and building operating and runtime systems using composable micro-services that are combined to provide only the features needed for a given ultrascale environment.
-
-
-
-
Message-centric Monitoring
A light-weight approach for performance monitoring of large-scale, long-running applications and systems.
Network Testbed
A testbed for investigating real-application performance of NIC stack manipulations.
OS Noise
This project investigates the sensitivity of HPC applications to various types of OS Noise.
-
Sensor Networks
We are interested in creating a complete software stack that addresses next-generation, multi-purpose sensor networks.
Virtualization in HPC
This project investigates the benefits of virtualized system environments in High-Performance Computing
|
