Electrical and Computer Engineering

CALCM Awarded Grants From NSF and Intel

January 15, 2004

The Computer Architecture Lab at Carnegie Mellon (CALCM) has received $1.2 million from NSF and $350,000 from Intel Corporation for research on Total Reliability Using Scalable Servers (TRUSS). CALCM also won an equipment grant from Intel for a $40,000 cluster of 26 high-performance desktop and dual servers for a high-throughput simulation testbed to build future large-scale multiprocessor servers. The testbed will serve the lab, including the TRUSS study and the Proactively Uniform Memory Access Architecture (PUMA²) project.

The TRUSS plan promises to innovate servers that don't stop. Led by Babak Falsafi, Associate Professor of ECE and CS, James Hoe, Assistant Professor of ECE and CS, and Andreas Nowatzyk, Associate Professor of Robotics and ECE, the team will examine novel server architectures that are:

• Non-Stop, i.e., rapidly detect and correct hardware component failure and/or transient error in computation due to future hardware's vulnerability to cosmic rays;
• Fully-redundant, and therefore have no single point of failure in the system, and;
• Scalable both in performance (scaling up to hundreds of nodes) and cost, capitalizing on inexpensive desktops as building blocks for large multiprocessor servers.

Server reliability, availability and serviceability (RAS) is an increasingly critical aspect of computing because information processing and storage are key pillars of modern society's infrastructure. Unfortunately, while RAS is progressively crucial, it is also ever more challenging to design, manufacture and market reliable server platforms.

Falsafi is the Principal Investigator of another computer architecture initiative, PUMA², which proposes Memory Streaming using hardware prediction and speculation to bridge the ever-growing processor/memory performance gap. For more information, please visit http://www.ece.cmu.edu/~puma2.