Asynchronous algorithms

Many parallel computations use a synchronous algorithm.

In such an algorithm all parallel processing elements move forward in lock-step synchrony. As the number of processing elements becomes very large (and it could be in the millions for future supercomputers), this synchronisation can become seriously inefficient.

One way to improve the situation might be to move to an asynchronous algorithm: one in which different processing elements can move forward independently and without the need for synchronisation. At EPCC, we are investigating the use of asynchronous algorithms, particularly in the context of sparse linear algebra.

A second important feature of asynchonous algorithms can be their ability to withstand certain types of faults (which are expected to be frequent on a very large machine). Such fault-tolerance is expected to be an important property of future algorithms.


Projects: What We Do

Check out our portfolio of projects and see what keeps our clients coming back for more

View our research papers

A full list of EPCC's research papers and other outputs can be found via the University of Edinburgh's Research Explorer

Upcoming courses and events

We also publish information about forthcoming events on Twitter and in our News feed

Our commercial brochure