Overview

TritonSort is case study in building balanced computing systems, specifically data-intensive scalable computing (DISC) systems.  TritonSort is a staged, pipeline-oriented sorting system designed to sort data with a very high degree of per-node efficiency.

Project members:

Project Alumni:

Publications

Alexander Rasmussen, George Porter, Michael Conley, Harsha Madhyastha, Radhika Niranjan Mysore, Alexander Pucher, Amin Vahdat. TritonSort: A Balanced Large-Scale Sorting System. Proceedings of the 8th USENIX Symposium on Networked Systems Design and Implementation (NSDI 2011), Boston, MA, Mar 30-Apr 1, 2011. [slides]

World Records

TritonSort and its successor, Themis, hold the 2011 world records in the "Indy" and "Daytona" variants of 100TB GraySort as well as the "Indy" variant of MinuteSort and the "Indy" and "Daytona" variants of 100TB JouleSort. 

We'd like to thank the National Science Foundation (grants #CSR-1116079 and #CNS-0923523), Cisco Systems, and NetApp for their generous support of this project.