• October 2009
    M T W T F S S

Parallel course

MIT News (10/23/09) Hardesty, Larry

Researchers at the Massachusetts Institute of Technology’s (MIT’s) Computer Science and Artificial Intelligence Lab are helping make programmers’ move to parallel programming less onerous as computer chip manufacturers produce multicore technology to upgrade performance. “Just writing anything parallel doesn’t mean that it’s going to run fast,” says MIT professor Saman Amarasinghe. “A lot of parallel programs will actually run slower, because you parallelize the wrong place.” Amarasinghe also thinks that computers are capable of automatically determining when to parallelize as well as which cores to assign which jobs. His group’s multicore computing effort is split along two lines–tools to ease programmers’ switch to parallel programming and tools to optimize programs’ performance once that switch has been accomplished. Amarasinghe and two graduate students have designed a system to increase the predictability of multicore programs by assigning a core attempting to access a shared resource a priority not according to the time of its request but according to the number of tasks it has performed. Amarasinghe’s lab has several projects focusing on parallel program optimization, one of which helps programs adjust to changing conditions on the spur of the moment. His group has devised a language that asks the developer to specify different techniques for executing a given computational job. When the program is operational, the computer automatically identifies the method with maximum efficiency.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: