Papers by Alireza Mortazavi
ACM SIGOPS Operating Systems Review, 2010
Enterprise and scientific data sets double every year, forcing similar growths in storage size an... more Enterprise and scientific data sets double every year, forcing similar growths in storage size and power consumption. As a consequence, current architectures used to build data warehouses are hitting a power consumption wall. We propose a novel alternative architecture comprising a large number of so-called "Amdahl blades" that combine energy-efficient CPUs and GPUs with solid state disks to increase sequential I/O throughput by an order of magnitude while keeping power consumption constant. Offloading throughput-intensive analysis to GPUs integrated in the server's chipset increases the relative performance per Watt. While keeping the power consumption constant, Amdahl blades offer 13 times the throughput of a state-of-the-art computing cluster for data-intensive applications. Finally, using the scaling laws originally postulated by Amdahl, we show that systems for data-intensive computing must maintain a balance between low power consumption and per-server throughput to optimize performance per Watt. Using a mix of SSDs and low-power hard disks we achieve a remarkably balanced system with excellent IO, power consumption and disk capacity.
Uploads
Papers by Alireza Mortazavi