Tórur Biskopstø Strøm: Ever since processors were first developed, researcher and engineers have pushed their computing power. This development has usually been in the form of increased parallel capabilities and/or higher clock rates.
Whilst mainframes and high-performance computers
adopted multiple processors/processing-cores early on, this trend didn\u2019t take hold in home
computers until the late 90\u2019s early 00\u2019s, when the issues of ever increasing clock rates
became too prominent.
Today most home computers, laptops and smartphones have at least 2 cores, with some
even containing 8 or more cores. This progress is understandable, as the theoretical
performance increase when going from 1 to n cores is n . However, this is rarely the case
and part of the problem lies in synchronization and communication.
When 2 or more cores share some resource, they must coordinate their access to it so that it
does not become corrupt. To this end processors employ some synchronization mechanism
that controls how, and when, the cores access the resource. The resource is comparable to
a meeting, with attendees being cores and the synchronization mechanism being moderator.
If attendees speak at the same time, information is half heard or lost. It is the moderator's job
to ensure that this does not happen and that all information is properly conveyed.
The use of synchronization mechanisms adds overhead to the resource manipulation. For 2
cores this overhead is typically small, but as the number of cores increases, the overhead
increases as well. In this project the aim is to reduce the synchronization overhead by
creating synchronization and communication mechanisms that are designed for multi-core
processors.
PhD project title: Real-Time Multi-Core Communication and Synchronization
DTU supervisors: Main supervisor is Martin Schoeberl and co-supervisor is Jens Sparsø from the Section for Embedded Systems