The Doctoral Dissertation Award Committee has awarded the 2016 Principles of Distributed Computing Doctoral Dissertation Award to Dr. Hsin-Hao Su and to Dr. Shahar Timnat.
Dr. Hsin-Hao Su completed his dissertation “Algorithms for Fundamental Problems in Computer Networks” in July 2015, under the supervision of Professor Seth Pettie, at the University of Michigan, Ann Arbor.
Hsin-Hao’s thesis provides efficient algorithms for fundamental graph problems that arise in networks, in both sequential and distributed settings. Among the latter, the most prominent are his results concerning graph coloring. He showed that numerous existential results in graph theory can be viewed as distributed algorithms with a tiny probability of success (guaranteed by the Lovasz Local Lemma) and that a fast distributed algorithm for the constructive LLL could be used to amplify the success probability to nearly 1. Hsin-Hao presented a O(log n)- time randomized algorithm for the LLL, and illustrated how it could be applied to graph coloring problems where the existence of the coloring is not obvious. Moser and Tardos observed that any LLL algorithm in their “resampling” framework requires Ω(log n) time, so this result is optimal within a natural design space. Hsin-Hao used his LLL algorithm to establish an O(log n)-time algorithm for (4+o(1))Δ/ln Δ−coloring triangle-free graphs. This result more than any other exhibits the technical virtuosity of Hsin-Hao: he discovered not only a great algorithm, but a new bound on the chromatic number of triangle-free graphs.
Before Hsin-Hao’s work many symmetry-breaking problems appeared to have similar complexity: (Δ+1)−coloring seemed similar to the Maximal Independent Set (MIS) problem and (2Δ−1)−edge coloring seemed similar to Maximal Matching. Hsin-Hao developed new tools for analyzing randomized coloring algorithms in locally sparse graphs, one consequence of which is that (2Δ−1)−edge coloring is provably easier than maximal matching.
Shahar’s dissertation provides an outstanding advance in our understanding of concurrent algorithms, including novel efficient practical algorithms and a theoretical study of their fundamental properties. The literature on highly-concurrent data structures focuses on lock-freedom, which guarantees that some thread will eventually make progress, and wait-freedom, which guarantees that all threads will eventually make progress in spite of failures and delays of other threads. It was believed that the overhead and complexity required to achieve wait-freedom is too high for practical systems. Shahar’s thesis changes this traditional belief by showing that lock-free algorithms can be made wait-free automatically and with a small performance penalty. His construction is realistic and practical.
Shahar provides a practical wait-free iterator, an original construct that no one knew how to do before. Another contribution is a novel and helpful analysis of the common “helping” pattern that is typically used for constructing wait-free algorithms. This analysis shows that there exist circumstances where some form of helping is required. Like many lower bounds, this has practical impact because it spares data structure designers from wasting their time trying on other approaches. Finally, the thesis proposes a simple transactional interface that is well-adapted both to architectures that provide hardware support for transactions, and to those that do not, yielding a way to design data structures that easily can be ported from one platform to another.
The award is sponsored jointly by the ACM Symposium on Principles of Distributed Computing (PODC) and the EATCS Symposium on Distributed Computing (DISC). This award is presented annually, with the presentation taking place alternately at PODC and DISC. The 2016 award will be presented at DISC 2016, to be held in Paris, France.
The 2016 Principles of Distributed Computing Doctoral Dissertation Award Committee:
- Yoram Moses, Technion
- Andrzej Pelc (chair), Université du Québec en Outaouais
- Paul Spirakis, University of Liverpool