May 29, 2013
SEATTLE, Wash., May 29 -- Meeting the performance, energy-efficiency, and resilience requirements of systems and applications at exascale will require rapid, accurate, and dynamic evaluation of tradeoffs. To provide these capabilities, significant advances in predictive modeling and simulation methods are needed. Models are key tools in the area of application/system co-design. As applications and systems evolve, models must be able to track ongoing complex changes and predict the impact of developments in both software and hardware design. While today's methods tend to focus on application performance as the metric of concern, modeling methods must evolve to consider performance, power consumption, and reliability in concert. It is critical to develop tools and techniques that will allow modeling capability to spread into the larger computational science community where it will have the greatest possible impact. Simulation and emulation capabilities must also expand along multiple directions, including scalability improvements, interoperability, support for exascale system design, and interfaces with modeling tools.
As part of the process, we solicit community input, in the form of position papers that describe novel research approaches for performance modeling and simulation at extreme scales. Position papers should address one or more of the following areas.
1. Development and generation of machine and application abstractions. Application developers need new abstract machine models that expose expected hardware features (e.g., SIMD, lightweight cores, or specialized functional units) and performance targets in terms of hierarchical parallelism, scale, data movement, computational intensity, bandwidths, latencies, and storage capacities of possible exascale configurations, as well as non-performance objectives such as power and resilience. Meanwhile, system architects need architecture-independent models of important applications that they can use to identify, understand, and evaluate application behavior on these proposed architectures. These abstractions must exist across the modeling, simulation, and measurement infrastructures.
2. Dynamic and Actionable Modeling. Modeling and simulation techniques must provide predictive and optimization capabilities at runtime. This will require the development of rapid evaluation techniques as well as interfaces for incorporating models into the layers within the software stack. In addition to dynamic model creation, models must be actionable. That is, the models must enable appropriate levels of the software stack to react to the instantaneous state of the system, both hardware and software, and provide the necessary information that can be used to
guide and optimize system operation.
3. Standards, Integration, and Interoperability of ModSim Methodologies and Tools. ModSim components will be integrated in different layers of the exascale software stack and must be able to interoperate and leverage common infrastructure including aspects such as time progression mechanisms (e.g. in discrete-event generation), parameter sharing, component querying for internal data, and interfaces to dynamic models.
The Organizing Committee will review these position papers and invite selected contributors to participate in the workshop to be held on September 18-19, 2013, in Seattle, WA. Responsive submissions will be made public via the workshop website and selected papers will be included in a post-workshop proceedings.
Necessary background information, including previous workshop reports and submission instructions, can be found at http://hpc.pnl.gov/modsim/2013/, and submissions can be made with EasyChair at http://j.mp/modsim2013.
Position papers (up to 2 pages) should describe a fundamental computer science research approach in the development of performance modeling and simulation approaches to address the key challenges associated with exascale computing systems. This description should be followed by a brief summary of related work and an assessment of the approach based on the following dimensions.
1. Challenges addressed: Which exascale modeling and/or simulation challenges does this approach address?
2. Maturity: What are the indicators that this approach will address the identified challenges?
3. Uniqueness: To what extent is the proposed approach unique to exascale systems? Could it be addressed by other research programs?
4. Novelty: How is this approach different from existing solutions?
5. Applicability: To what extent will the proposed approach, if successful, be applicable to other areas?
6. Effort: How much effort is needed to effectively explore this approach?
The paper may include any number of authors, but must provide contact information for a single "contact" author. There is no limit to the number of position papers that an individual or group can submit. Authors are strongly encouraged to follow the structure presented above. Authors of selected position papers will be invited to participate in a workshop based on the overall quality of the position paper(s) and our expectation that their inclusion in the workshop will stimulate constructive discussion by workshop participants. Unique positions that are well presented and emphasize transformative approaches will be given preference.
1. Length: Up to 2 pages (a list of cited references does not count against this limit).
2. Due Date: SUBMISSIONS DUE ON JUNE 17 (extended from JUNE 3). Submission site: http://j.mp/modsim2013
3. Notification of Selection for Workshop Presentation: July 12
4. ModSim Workshop, September 18-19, 2013, Seattle, WA
Adolfy Hoisie, PNNL
Darren Kerbyson, PNNL
Bob Lucas, USC
David M. Nicol, UIUC
Boyana Norris, ANL
Arun Rodrigues, SNL
John Shalf, NERSC
Douglas Thain, University of Notre Dame
Jeffrey Vetter, ORNL and Georgia Tech
Sudhakar Yalamanchili, Georgia Tech
Source: Oak Ridge National Laboratory
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?