Welcome to TRANSACT 2007, the second ACM SIGPLAN Workshop on Transactional Computing. Long the keystone of the database world, transactions have, over the past few years, become one of the hottest topics in shared-memory computing, driven largely by the need to tame the coming generation of many-core machines. TRANSACT 1, held in conjunction with PLDI in Ottawa last June, was an outstanding success, with three times the attendance originally expected. We hope to duplicate that success in this second incarnation, and plans are already underway for TRANSACT 3, to be held in conjunction with PPoPP 2008 a mere six months from now.
From a strong pool of 23 submissions, the program committee selected 11 papers for presentation at the workshop. Almost all the reviewing was performed by members of the program committee: each PC member read at least 6 papers in detail and skimmed the rest; each paper received at least 4 detailed reviews. A preliminary ranking based on initial reviews was refined through several rounds of email, and finalized during a conference call involving most of the program committee.
The Call for Papers expressly invited submissions from the full range of transactional systems, and the committee attempted not to constrain decisions based on area or level of development. Accepted papers span the range from theory to practice; C to Haskell; and language, library, and hardware-level techniques.
Though all authors have elected to make their papers available on-line, copyright vests with them or their employers, and there will be no formal proceedings. The expressed intent of the organizers has been to avoid any obstacle to subsequent publication in other refereed forums. At the workshop itself, the program is intended to be as interactive as possible, with 15 minute presentations, 5 minutes of immediate questions, and a lengthy period of general discussion at the end of every session.
A theme emerging from several submissions, and from work appearing elsewhere, has been an evolution in hardware design from special-purpose transactional mechanisms to more general-purpose support for reference tracking; conflict detection; buffering, versioning and rollback; contention management and scheduling; etc. — most of which may potentially be used for purposes other than transactions. At the invitation of the program committee, Prof. Mike Swift of the University of Wisconsin–Madison agreed to organize a panel on this subject, to conclude the day’s activities.
Many people have contributed to making the workshop a success. I would first like to thank the authors, including those whose submissions could not be included in the program. My thanks as well to the members of the program and steering committees, and to Rachid Guerraoui, our general chair. Gil Neiger, Costas Busch, and Srikanta Tirthapura, local arrangements and workshops chairs for PODC, did a wonderful job on logistical details, for which I am enormously grateful.
Michael L. Scott
University of Rochester
It may seem strange that PC Chair Michael Scott is co-author of a paper in this workshop. Michael agreed to be PC Chair with the understanding that he would be spared the “reward” of being forbidden from submitting. He asked me to act as PC Chair for his paper. To this end, we used a process that paralleled the TRANSACT 2007 reviewing process, but was done by email outside of the reviewing software at Rochester. I selected four PC reviewers, gathered their reviews, revealed all four reviews to each, and led a discussion that led to a decision. We agreed to hold the paper to a slightly higher standard (e.g., persistent disagreement would sink it), and to decide before the virtual PC meeting, but not to inform Michael until after the rest of the program was set. As you can see from the final program, we accepted Michael’s paper.