Back to Research

Simon Fraser University  

Project 1154 - Scalable Synchronization for Dynamic Workloads on Large-Scale Computing Systems

SHARE THIS POST

Running from 2022 to present

Scalable Synchronization for Dynamic Workloads on Large-Scale Computing Systems

Products

Contention arises when applications and runtimes (such as JVM) use multiple threads on many-core CPUs to access shared data. Such contention---no matter high or low---must be correctly and gracefully handled. Yet existing approaches tend to fail to scale under dynamic workloads. For example, centralized spinlocks are easy to implement and use but collapses under high core count and contention; local-spinning locks can scale under high contention, but are slow under low contention. Some solutions exist, but they tend to require more complex interfaces or trade space for efficiency and functionality. Our goal is to devise new locks that gracefully handle both low and high contentions. At the same, we preserve such desirable features as easy-to-user interface found in traditional spinlocks and compact lock storage to maintain backward compatibility as much as possible. Further, we will explore opportunities to co-design locking and runtimes/applications in complex systems for optimal performance.

Learn More about the Research Team.  

Explore the product that harvests this research results  

Research team:

  • IBM Project Lead (RCL): Tobi Ajila, IBM
  • IBM Manager (RCM): Tony Renaud, IBM
  • IBM Sponsor (RCS): Vijay Sundaresan, IBM
  • IBM Contributor (RCC): Babneet Singh, IBM

Institution:

Simon Fraser University   

SHARE THIS POST