Chen-Mou adds: “When using Amazon EC2 as a metric, the parallelizability of the algorithm or the parallelization of the implementation is explicitly taken into account, as opposed to being assumed or unspecified. The Amazon EC2 metric is thus practical and easy to use.”
The group now uses Hadoop Streaming in their architecture, and runs their programs with Amazon Elastic MapReduce and Cluster GPU Instances for Amazon EC2.
Chen-Mou notes, “Our purpose is to break the record of solving the shortest vector problem (SVP) in Euclidean lattices. The problem plays an important role in the field of information science. We estimated that we would need 1,000 cg1.4xlarge instance-hours. We ended up using 50 cg1.4xlarge instances for about 10 hours to solve our problem. Now, the vectors we found are considered the hardest SVP anyone has solved so far.” He adds: “We only spent $2,300 for using the 100 Tesla M2050 for ten hours, which is quite a good deal.”
Since switching to AWS, the group indicates that the machine maintenance costs have been reduced, and they have experienced more stable and scalable computational power. The group’s favorite component of AWS is Amazon CloudWatch, which it uses to watch computer utilities while also improving their program.
Looking into the future, says Chen-Mou, “We want to increase our GPU Cluster quote and solve a higher dimension SVP. We are also considering renting an AWS machine for setting up an SVN server.”
To learn more, visit http://www.ntu.edu.tw/.
Added May 18, 2011