Communications of the ACM
Practical Parallelism
A new system dubbed Fractal achieves 88-fold speedups through a parallelism strategy known as speculative execution.
Credit: MIT News
Researchers at the Massachusetts Institute of Technology's (MIT) Computer Science and Artificial Intelligence Laboratory have developed a system called Fractal that enables up to 88-fold acceleration in a parallelism strategy known as speculative execution.
"What these systems do is execute...different chunks [of tasks] in parallel, and if they detect a conflict, they abort and roll back one of them," says MIT professor Daniel Sanchez.
Fractal solves several inefficiencies introduced by large atomic tasks. A programmer adds a line of code to each subroutine within an atomic task that can be performed in parallel, usually enlarging the length of the serial version of a program by a few percent, whereas an implementation that explicitly synchronizes parallel tasks will often increase it by up to 400 percent. Circuits hardwired into the Fractal chip then manage the parallelization.
Fractal also assigns a time stamp to each atomic task, with task order preserved by the subroutine order.
From MIT News
View Full Article
Abstracts Copyright © 2017 Information Inc., Bethesda, Maryland, USA
No entries found