Bookmark and Share
Title: A New Approach to Design of Massively Parallel Systems
Go to 0:03
Go to 0:27
Go to 0:33
Go to 1:57
Go to 3:28
Go to 4:24
Go to 5:05
Go to 6:10
Go to 7:09
Go to 9:10
Go to 9:33
Go to 10:26
Go to 11:24
Go to 12:11
Go to 12:52
Go to 13:42
Go to 14:28
Go to 14:53
Go to 15:17
Go to 16:48
Go to 17:10
Go to 18:09
Go to 20:18
Go to 20:44
Go to 21:33
Go to 22:23
Go to 24:32
Go to 26:06
Go to 27:04
Go to 27:19
Go to 27:56
Go to 29:17
Go to 30:34
Go to 30:57
Go to 31:04
Go to 31:24
Go to 31:30
Go to 32:46
Go to 33:01
Go to 33:35
Go to 33:49
Go to 34:54
Go to 36:05
Go to 36:20
Go to 36:46
Go to 38:41
Go to 39:51
Go to 40:30
Go to 40:45
Go to 40:58
Go to 42:00
Go to 42:44
Go to 44:48
Go to 45:13
Go to 47:42
Go to 52:01
Go to 52:05
You need Flash player 8+ and JavaScript enabled to view this video.
Link:
Authors: Narendra Karmarkar
Published: 2008-08-15
Channel: googletechtalks
NEW: Explore related talks
Abstract: Google Tech Talks
August 13, 2008

ABSTRACT

Based on insights derived from theory of global optimization, we have made a major breakthrough in physical design of several massively parallel systems. These include:

New design of parallel supercomputers based on massively parallel quantum tunneling, electron optics and finite projective geometry. This approach overcomes the bandwidth and latency obstacle faced by other contemporary designs. Other advantages include large improvement in teraflops per kilowatt, significant reduction in programming complexity and broad applicability to many domains, including those requiring a lot of pointer chasing. Our design can be implemented using known fabrication techniques.

New design of low latency, multi-ported secondary storage based on magneto-optics that implements shared memory directly at physical level. This feature is highly valuable to business data bases as well as applications requiring frequent shared access to massive amounts of common data such as google earth.

New design of high bandwidth switches required for building next generation internet infrastructure.

Speaker:
Narendra Karmarkar
Dr. Karmarkar received his B.Tech at the IIT Bombay in 1978. Later, he received his M.S. at the California Institute of Technology, and his Ph.D. at the Institute of Computer Science at the University of California, Berkeley.

He published his famous result in 1984 while he was working for Bell Laboratories in New Jersey. Karmarkar was a professor at the Tata Institute of Fundamental Research in Bombay.

Karmarkar received a number of awards for his algorithm, including the Fulkerson Prize and the Lanchester Prize.