[hpc-announce] Call for Papers: First Workshop on Managing Overprovisioned Systems

Carter, Nicholas P nicholas.p.carter at intel.com
Mon Dec 30 17:54:54 CST 2013

                               Call for Papers

           First Workshop on Managing Overprovisioned Systems (W-MOS)
                                March 1, 2014
            In conjunction with the 2014 International Conference on
              Architectural Support for Programming Languages and 
                          Operating Systems (ASPLOS)

Computer systems are at an inflection point where power budgets are
becoming more of a limiting factor on designers than the number of
transistors that can be fabricated on a chip. Transistor density
continues to grow at geometric rates, but power budgets prevent
systems from using all of their transistors simultaneously. As a
result, architects have begun to propose overprovisioned designs that
contain more hardware than they can utilize simultaneously, including
architectures that incorporate fixed-function accelerators, single-
and multi-ISA heterogeneous architectures, and “melt the chip”
architectures in which each subsystem (cores, memory, and network) can
consume the vast majority of the system’s power budget when that
subsystem is the bottleneck on application performance.

A common theme among all of these overprovisioned designs is that they
transform computer architectures from static designs into dynamic
systems that must be continually monitored, reconfigured, and managed
in order to deliver maximum performance without exceeding their power
budgets. Since the system cannot use all of its hardware
simultaneously, some agent (hardware, software, or a combination
thereof) must determine which hardware should be used at any given
time, depending on the application or applications being run. This
will require information from multiple layers of the system stack,
including architecture, runtime/OS, and hardware, creating a complex
co-design problem that must be solved if computer performance and
energy efficiency is to continue to improve at current rates.

The Workshop on Managing Overprovisioned Architectures seeks to bring
together researchers from the OS, architecture, and application
communities to discuss techniques to configure, monitor, and map
applications onto architectures that contain more hardware than they
can utilize simultaneously. We are soliciting papers on topics related
to these areas, including, but not limited to:

- Power management
- System introspection and adaptation
- Self-aware architectures and runtimes
- Using accelerators for general-purpose programs
- Power-aware scheduling
- Techniques to select the best hardware for an application
- Allocating power to resources based on program needs
- Migrating execution to new resources as requirements change
- Cross-layer communication APIs
- Resilience in overprovisioned systems

Nick Carter (Intel), 
Romain Cledat (Intel), 
Radu Teodorescu (The Ohio State University),
Josep Torrellas, (University of Illinois at Urbana-Champaign)

Important dates: 
Paper submission: February 3, 2014
Acceptance notification: February 9, 2014 
Camera-ready due: February 24, 2014 
Workshop date: Saturday, March 1 2014

Authors should submit papers of at most four pages (two-column
format, single-spaced, at least 10-point font) in length to: 
https://www.easychair.org/conferences/?conf=wmos2014.  To make the
best use of the limited space, authors are encouraged not include any
discussion of the technological trends that are leading to
overprovisioned systems unless their ideas depend on a specific aspect
of future fabrication technologies.

Authors of accepted papers will be asked to prepare six-page versions
of their papers, which will be distributed electronically at the
workshop and on the workshop website

More information about the hpc-announce mailing list