As a research community, we have seen transactional memory and other forms of speculative execution ride the wave of the Gartner hype cycle from "next big thing" to the "trough of disillusionment". While these technologies appear to be slowly climbing the "slope of enlightenment", through this workshop, we hope to give them a jolt of encouragement. If best-effort hardware support for speculation and rollback is given (as exemplified by Azul/Transmeta/Sun Rock/AMD ASF), what alternative uses might it have? What new possibilities would emerge if systems supported no-overhead/infinite speculative footprints?

Given such a substrate, what new microarchitectural optimizations are possible? How can various layers of the system stack use such features? What programming abstractions are possible to the developer, beyond transactions? What impact would unbounded speculation have on operating system services, virtualization, debuggers, profilers, I/O, and accelerators?

The workshop will consist of a set of presentations/positions on all things speculative: authors should submit a 2-4 page abstract (details below) summarizing their idea, position, vision, or dream. Supporting these ideas through quantitative evidence is acceptable but not required. In this spirit, we hope to capture the enthusiasm of successful "Wild and Crazy" idea sessions of years past, while keeping the program sane enough that all presentations are subject to serious thought and reflection.

Submission Topics Include

Submission Guidelines

Authors should submit their ideas in the form of 2-4 page papers, excluding bibliography. To ensure submissions are accessible to a broad audience of researchers in Languages, Compilers, Operating Systems, Architectures, and Run-Time Systems, we ask authors to include a brief background section. Papers will be reviewed for their novelty, ability to cut across areas, and ability to generate discussion.

All paper submissions should be made electronically. The link for submitting papers will appear on this page within a few weeks. Note that there will not be a printed proceedings, nor will accepted abstracts appear in any electronic archival form.

Important Dates

Submission Deadline: August 29, 2011 9:00 AM EDT (GMT - 5:00)
Notifications: September 9, 2011
Final Copy Due: September 26, 2011
Workshop: October 10, 2011

Program

1330 - 1340 Opening Remarks
1340 - 1440 Keynote: Hardware Support for Transactional Memory and Thread-Level Speculation
in the IBM Blue Gene/Q System

Martin Ohmacht
The IBM Blue Gene/Q Team
Keynote Slides
1440 - 1500 Hints based Speculative Execution for Exploiting Probabilistic Parallelism
AndrĂ¡s Vajda and Per Stenström
Ericsson SW Research & Chalmers University of Technology
pdf
1500 - 1530Break
1530 - 1550 Speculating on top of an unmodified Java VM
Ivo Anjo
INESC-ID Lisboa/Instituto Superior Tecnico/Universidade Tecnica de Lisboa
pdf
1550 - 1610 FaulTM-multi: Fault Tolerance for Multithreaded Applications Running on
Transactional Memory Hardware

Gulay Yalcin, Osman Unsal, Adrian Cristal and Mateo Valero
Barcelona Supercomputing Center
pdf
1610 - 1655 Panel Session (Topic: "Hot or Not: What key topics have the research community
overlooked, and what topics have been overstudied?")
 
The following positions were introduced and debated during the panel:
  • We pay too much attention to debugging
  • Supporting legacy languages is a waste of time
  • Progress guarantees are underrated
  • All speculation instrumentation should be explicit
  • We need to pay more attention to profiling
  • The only good transaction is a small transaction
  • Nested speculation is essential
1655 - 1700 Concluding Remarks

Submission Website

You may register your submission to WANDS here.

Program Committee

Organizing Committee