CICS Professor Emery Berger Granted ASPLOS 2019 Most Influential Paper Award

Professor Emery Berger, co-director of the College of Information and Computer Sciences (CICS) PLASMA Research Lab, has won the Most Influential Paper Award at the 24th annual ACM International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS) for his 2000 paper, "Hoard: A Scalable Memory Allocator for Multithreaded Applications." The paper was co-authored by former CICS faculty member Kathryn McKinley, and researchers Robert D. Blumofe and Paul R. Wilson.

The ASPLOS Most Influential Paper Award recognizes one paper per year that has had a major influence on the field. The conference's program committee nominates papers from any ASPLOS conference that occurred ten or more conferences ago.

In this year's winning paper, Berger and his collaborators presented Hoard, the first memory allocator, also known as memory management software, that simultaneously addressed issues posed by parallel, multithreaded C and C++ programs like database managers and web servers.

At the time the paper was published, multiprocessor memory allocators often suffered from problems that included poor performance and scalability, heap organizations that introduced false sharing (further degrading performance), and a dramatic increase in memory consumption when confronted with the memory allocation and freeing demands of multi-threaded programs.

In response, Berger and his collaborators developed Hoard, an allocator that attacks all of these problems simultaneously. Hoard first manages objects in large groups called "superblocks", anticipating future memory consumption needs and reducing the number of times the application needs to request memory from the operating system. Because each thread gets its own pool of superblocks, which it can access without interference from any other thread, threads can run at full speed. This approach also reduces the risk of false sharing problem by ensuring that memory from each processor is kept separate, avoiding potential conflicts. Finally, to keep space consumption low, Hoard tracks the amount of empty space accumulating in any thread's pool; if this crosses a threshold, it moves an empty, or nearly-empty, superblock to a common shared pool. Together, these approaches make Hoard the first multiprocessor allocator that scales up to large numbers of threads while remaining memory efficient.

In the nineteen years since the paper's publication, Hoard has been integrated into technologies used by a wide range of companies, including British Telecom, Cisco, Credit Suisse, Reuters, Royal Bank of Canada, SAP, and Tata. The Hoard algorithm also serves as the basis for the Mac OS X Memory Manager.

Berger received a PhD in Computer Science from the University of Texas at Austin in 2002. His honors include receiving a Microsoft Research Fellowship, an NSF CAREER Award, and a Lilly Teaching Fellowship, among others. He was named an ACM Senior Member and an ACM Distinguished Member in 2010 and 2018, respectively. He previously received Test of Time Awards at OOPSLA 2012 and PLDI 2016.