SEARCH
ISTC-CC NEWSLETTER
RESEARCH HIGHLIGHTS
Ling Liu's SC13 paper "Large Graph Processing Without the Overhead" featured by HPCwire.
ISTC-CC provides a listing of useful benchmarks for cloud computing.
Another list highlighting Open Source Software Releases.
Second GraphLab workshop should be even bigger than the first! GraphLab is a new programming framework for graph-style data analytics.
ISTC-CC Abstract
Just Join for Parallel Ordered Sets
SPAA 2016. 28th ACM Symposium on Parallelism in Algorithms and Architectures. Jul 11, 2016 - Jul 13, 2016. Asilomar State Beach, California, USA.
Guy Blelloch, Daniel Ferizovic*, Yihan Sun
Carnegie Mellon University
*Karlsruhe Institute of Technology
Ordered sets (and maps when data is associated with each key) are one of the most important and useful data types. The set-set functions union, intersection and difference are particularly useful in certain applications. Brown and Tarjan first described an algorithm for these functions, based on 2-3 trees, that meet the optimal Θ(m log (n/m + 1)) time bounds in the comparison model (n and m ≤ n are the input sizes). Later Adams showed very elegant algorithms for the functions, and others, based on weight-balanced trees. They only require a single function that is specific to the balancing scheme—a function that joins two balanced trees—and hence can be applied to other balancing schemes. Furthermore the algorithms are naturally parallel. However, in the twenty-four years since, no one has shown that the algorithms, sequential or parallel are asymptotically work optimal.
In this paper we show that Adams’ algorithms are both work efficient and highly parallel (polylog span) across four different balancing schemes—AVL trees, red-black trees, weight balanced trees and treaps. To do this we use careful, but simple, algorithms for JOIN that maintain certain invariants, and our proof is (mostly) generic across the schemes.
To understand how the algorithms perform in practice we have also implemented them (all code except JOIN is generic across the balancing schemes). Interestingly the implementations on all four balancing schemes and three set functions perform similarly in time and speedup (more than 45x on 64 cores). We also compare the performance of our implementation to other existing libraries and algorithms.
FULL PAPER: pdf
