BAD HARDWARE WEEK

SUNNYVALE, Calif., Mar 24, 2008 (BUSINESS WIRE) -- Yahoo! Inc. (Nasdaq:YHOO), a leading global Internet company, and Computational Research Laboratories (CRL), a wholly owned subsidiary of Tata Sons Limited, today announced an agreement to jointly support cloud computing research. As part of the agreement, CRL will make available to researchers one of the world's top five supercomputers that has substantially more processors than any supercomputer currently available for cloud computing research.

This effort is the first of its kind in terms of the size and scale of the machine, and the first in making available a supercomputer to academic institutions in India. The Yahoo!/CRL effort is intended to leverage CRL's expertise in high performance computing and Yahoo!'s technical leadership in Apache Hadoop, an open source distributed computing project of the Apache Software Foundation, to enable scientists to perform data-intensive computing research on a 14,400 processor supercomputer.

Called the EKA, CRL's supercomputer is ranked the fourth fastest supercomputer in the world - it has 14,400 processors, 28 terabytes of memory, 140 terabytes of disks, a peak performance of 180 trillion calculations per second (180 teraflops), and sustained computation capacity of 120 teraflops for the LINPACK benchmark. Of the top ten supercomputers in the world, EKA is the only supercomputer funded by the private sector and is available for use on commercial terms. EKA is expected to run the latest version of Hadoop and other state-of-the-art, Yahoo!-supported, open-source distributed computing software such as the Pig parallel programming language developed by Yahoo! Research.

BAD HARDWARE: Ok, let see.

HP SFS

HP StorageWorks Scalable File Share delivers up to 100 times more bandwidth when compared to Network File System (NFS), the current de facto standard for sharing files across a network.

Processor Intel EM64T Xeon 53xx (Clovertown) 3000 MHz (12 GFlops) !

System peak is 180 GFlops.

120 TFLOPS, 4th in the world.

• 2.5MW of electricity
• 400ton cooling capacity
• 10km of electrical cabling
• 10km of ethernet cabling
• 15km of infiniband cabling
• 1800 blades

Its strategy against the network congestion is mathematical discipline called projective geometry, keeping faults located in blades, and not blocking the whole supercomputer or requesting application checkpoints for recovery (and system resource hogging too).