Abstract

model [2] for Big Graph analytics, where application pro-grammers need no knowledge of parallel or distributed sys-tems. Instead, they just need to “think like a vertex ” and write a few functions that encapsulate the logic for what one graph vertex does. The vertex-oriented programming model has been found to ease the implementation of distributed graph algorithms to a great extent. There are several efforts (e.g., [1, 4]) to build Pregel-like systems from scratch. However, building such a system is complicated because a complete implementation must con-sider network issues, memory management, message deliv-ery, parallel task scheduling, fault-tolerance, vertex storage, and out-of-core support. Moreover, if the system developer wants to support different runtime execution choices, do-ing that in the built-from-scratch world would add still more complexity. 2. THE PREGELIX APPROACH In this work, instead of building a Pregel system from scratch, we explore an architectural alternative — express-ing Pregel’s semantics as database-style dataflows and exe-cuting them on a general-purpose data-parallel engine using classical parallel query evaluation techniques. We have used this approach to build Pregelix — a dataflow-based Pregel implementation for Big Graph analytics. Pregelix examines the value of a “one-size-fits-a-bunch ” philosophy, demon-strating that a general execution engine can also support Big Graph analytics well. Copyright c © Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted with-out fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party compo-nents of this work must be honored. For all other uses, contact the Owner/Author. Copyright is held by the owner/author(s).