1 / 5

3 Questions for Cluster and Grid Use

3 Questions for Cluster and Grid Use. Asheville North Carolina September 11 2006 Geoffrey Fox Computer Science, Informatics, Physics Pervasive Technology Laboratories Indiana University Bloomington IN 47401 http://grids.ucs.indiana.edu/ptliupages/presentations/

abushnell
Download Presentation

3 Questions for Cluster and Grid Use

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 3 Questions for Cluster and Grid Use Asheville North Carolina September 11 2006 Geoffrey Fox Computer Science, Informatics, Physics Pervasive Technology Laboratories Indiana University Bloomington IN 47401 http://grids.ucs.indiana.edu/ptliupages/presentations/ gcf@indiana.eduhttp://www.infomall.org

  2. Is the work and activities of Global Grid Form (GGF/OGF) of any value to CCGSC Community? • Currently OGF is meeting (September 11-14) with • Development of standards • Varied collection of “forums” including workshops – it now has a specific eScience “function” (also Standards and Enterprise functions) • For example, OGF19 (North Carolina Jan 29 – Feb 2) one day workshops • Web 2.0 and the Grid • Federated Identity in Grids and (Virtual) Organizations • Currently very few OGF standards used but many W3C, OASIS standards are also not used • Mashups are composed from JavaScript, AJAX and REST and not BPEL WSDL and SOAP (http://www.programmableweb.com/matrix 270 APIs 1000 Mashups) • GridFTP popular (BitTorrent more popular); BES (Basic Execution Services) with JSDL (Job Specification Language) likely to be popular • “Standardly” available software like GT4Condor and SRB are used • Wide use of standards should help users as allows more sustainable software with multiple interoperating “vendors” • Could OGF activities be tweaked/redirected to be of more value? • Measurement: Number of OGF Standards used in real Grids • Measurement: Number of (eScience) attendees at OGF meetings

  3. Can we survive/make use of innovative multicore programming models I? • Multicore chips will generate programming models that are optimized for use in “broad applications” (aka “Microsoft Word”) and scaling up to eventually to some 128 cores • Fast thread switching operations with microsecond latency • It is unlikely that openMP and MPI will be dominant “broad programming” models as these optimized for different criteria • ParalleX (LSU), Software Transactional Memory STM, Microsoft’s Concurrency and Coordination Runtime CCR, and Functional Languages are approaches that have pretty different concurrency models from MPI/openMP • Don’t know how CCA and HPCS Languages sit with respect to multicore and its application base • One will get far-out “broad programming models” as well as those aimed at evolution over next 5-10 years

  4. Can we survive/make use of innovative multicore programming models II? • Are these new models a distraction or an opportunity for scientific computing? • Can we / Should we produce a new generation of hybrid programming models that say span MPI to CCR? • Could open up applications that didn’t work so well on traditional clusters/programming models • Discrete event simulation with Time Warp etc. is one example • Tree algorithms like Branch&Bound and Computer Chess will run well • Measurements: • Number of new applications enabled by multicore • Number of applications using “new” programming models • Performance of “new” programming models (and their run time)

  5. Do we need more expertise/ education/ consultants in parallel/concurrent computing? • There is a plethora of students with Grid and Internet technology skills • Few computer Science students (at Indiana) take more than an optional introductory parallel computing class • Application scientists are using parallel computers (clusters) but in my experience often use rather inefficient naïve algorithms • e.g. O(N2) not O(NlogN) algorithms • e.g. misuse of MPI collective communications • Probably DoE has plenty of experts • Do we need more students and more classes taught by more faculty? • Do we need more “consultants” • Measurements: • Quality of Code produced by “new” users • Number of Students graduating per year in various parts (CS and applications) of parallel computing

More Related