{"456051":{"#nid":"456051","#data":{"type":"event","title":"PhD Defense by Naila Farooqui","body":[{"value":"\u003Cp\u003E\u003Cstrong\u003ETitle:\u0026nbsp;\u0026nbsp;Runtime Specialization for Heterogeneous CPU-GPU Resources\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u003Cstrong\u003ENaila Farooqui\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003ESchool of Computer Science\u003C\/p\u003E\u003Cp\u003ECollege of Computing\u003Cbr \/\u003E Georgia Institute of Technology\u003Cbr \/\u003E \u003Cbr \/\u003E Date: October 19, 2015 (Monday)\u003Cbr \/\u003E Time: 12:00 PM - 2:00\u0026nbsp;PM\u0026nbsp;(ET)\u003Cbr \/\u003E Location: KACB 3100\u003Cbr \/\u003E \u003Cbr \/\u003E\u003Cstrong\u003E Committee:\u003C\/strong\u003E\u003C\/p\u003E\u003Cp\u003E---------------\u003C\/p\u003E\u003Cp\u003EDr. Karsten Schwan (Advisor, School of Computer Science, Georgia Tech)\u003C\/p\u003E\u003Cp\u003EDr. Sudhakar Yalamanchili (School of Electrical and Computer Engineering, Georgia Tech)\u003C\/p\u003E\u003Cp\u003EDr. Ada Gavrilovska (School of Computer Science, Georgia Tech)\u003C\/p\u003E\u003Cp\u003EDr. Richard Vuduc (School of Computational Science and Engineering, Georgia Tech)\u003C\/p\u003E\u003Cp\u003EDr. Vanish Talwar (Research Scientist, PernixData)\u003C\/p\u003E\u003Cp\u003EDr. Rajkishore Barik (Research Scientist, Intel Labs)\u003C\/p\u003E\u003Cp\u003E\u003Cem\u003E\u003Cstrong\u003E\u0026nbsp;\u003C\/strong\u003E\u003C\/em\u003E\u003C\/p\u003E\u003Cp\u003E\u003Cem\u003E\u003Cstrong\u003EAbstract:\u003C\/strong\u003E\u003C\/em\u003E\u003C\/p\u003E\u003Cp\u003E------------\u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003EHeterogeneous parallel architectures like those comprised of CPUs and GPUs are a \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Etantalizing compute fabric for performance-hungry developers. While these platforms \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eenable order-of-magnitude performance increases for many data-parallel application \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Edomains, there remain several open challenges: (i) the distinct execution models \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Einherent in the heterogeneous devices present on such platforms drives the need to \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Edynamically match workload characteristics to the underlying resources, (ii) the complex \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Earchitecture and programming models of such systems require substantial application\u003Cbr \/\u003E\u0026nbsp;knowledge and effort-intensive program tuning to achieve high performance, and (iii)\u0026nbsp;\u003Cbr \/\u003E as such platforms become prevalent, there is a need to extend their utility from running\u0026nbsp;\u003Cbr \/\u003E known regular data-parallel applications to the broader set of input-dependent, irregular\u0026nbsp;\u003Cbr \/\u003E applications common in enterprise settings.\u0026nbsp;\u003Cbr \/\u003E\u0026nbsp;\u003Cbr \/\u003E The key contribution of our research is to enable runtime specialization on such hybrid \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003ECPU-GPU platforms by matching application characteristics to the underlying heterogeneous \u003Cbr \/\u003E \u003Ca name=\u0022section-executive-summary.tex-25\u0022\u003E\u003C\/a\u003Eresources for both regular and irregular workloads. Our approach enables profile-driven \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eresource management and optimizations for such platforms, providing high application \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eperformance and system throughput. Towards this end,\u0026nbsp; this research will: (a) enable dynamic \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Einstrumentation for GPU-based parallel architectures, specifically targeting the complex \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003ESingle-Instruction Multiple-Data (SIMD) execution model, to gain real-time introspection into \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eapplication behavior; (b) leverage such dynamic performance data to support novel online \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eresource management methods that improve application performance and system throughput, \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eparticularly for irregular, input-dependent applications; (c) automate some of the programmer \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eeffort required to exercise specialized architectural features of such platforms via \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Einstrumentation-driven dynamic code optimizations; and (d) propose a specialized, affinity-aware \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Ework-stealing scheduler for integrated CPU-GPU processors that efficiently distributes work at \u003C\/p\u003E\u003Cp\u003E \u003C\/p\u003E\u003Cp\u003Eruntime across all CPU and GPU cores for improved load balance, taking into account\u003Cbr \/\u003E\u0026nbsp;both application characteristics and architectural differences of the underlying devices.\u003C\/p\u003E\u003Cp\u003E \u003Cbr \/\u003E\u0026nbsp;\u003Ca name=\u0022section-executive-summary.tex-25\u0022\u003E\u003C\/a\u003Eresources for both regular and irregular workloads. Our approach enables profile-driven \u003C\/p\u003E\u003Cp\u003Eresource management and optimizations for such platforms, providing high application \u003C\/p\u003E\u003Cp\u003Eperformance and system throughput. Towards this end,\u0026nbsp; this research will: (a) enable dynamic \u003C\/p\u003E\u003Cp\u003Einstrumentation for GPU-based parallel architectures, specifically targeting the complex \u003C\/p\u003E\u003Cp\u003ESingle-Instruction Multiple-Data (SIMD) execution model, to gain real-time introspection into \u003C\/p\u003E\u003Cp\u003Eapplication behavior; (b) leverage such dynamic performance data to support novel online \u003C\/p\u003E\u003Cp\u003Eresource management methods that improve application performance and system throughput, \u003C\/p\u003E\u003Cp\u003Eparticularly for irregular, input-dependent applications; (c) automate some of the programmer \u003C\/p\u003E\u003Cp\u003Eeffort required to exercise specialized architectural features of such platforms via \u003C\/p\u003E\u003Cp\u003Einstrumentation-driven dynamic code optimizations; and (d) propose a specialized, affinity-aware \u003C\/p\u003E\u003Cp\u003Ework-stealing scheduler for integrated CPU-GPU processors that efficiently distributes work at \u003C\/p\u003E\u003Cp\u003Eruntime across all CPU and GPU cores for improved load balance, taking into account\u003Cbr \/\u003E \u003Ca name=\u0022section-executive-summary.tex-36\u0022\u003E\u003C\/a\u003Eboth application characteristics and architectural differences of the underlying devices.\u003Ca name=\u0022section-executive-summary.tex-37\u0022\u003E\u003C\/a\u003E\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E","summary":null,"format":"limited_html"}],"field_subtitle":"","field_summary":"","field_summary_sentence":[{"value":"Runtime Specialization for Heterogeneous CPU-GPU Resources"}],"uid":"27707","created_gmt":"2015-10-06 09:21:39","changed_gmt":"2016-10-08 02:14:14","author":"Tatianna Richardson","boilerplate_text":"","field_publication":"","field_article_url":"","field_event_time":{"event_time_start":"2015-10-19T13:00:00-04:00","event_time_end":"2015-10-19T15:00:00-04:00","event_time_end_last":"2015-10-19T15:00:00-04:00","gmt_time_start":"2015-10-19 17:00:00","gmt_time_end":"2015-10-19 19:00:00","gmt_time_end_last":"2015-10-19 19:00:00","rrule":null,"timezone":"America\/New_York"},"extras":[],"groups":[{"id":"221981","name":"Graduate Studies"}],"categories":[],"keywords":[{"id":"100811","name":"Phd Defense"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[{"id":"1788","name":"Other\/Miscellaneous"}],"invited_audience":[{"id":"78771","name":"Public"}],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[],"email":[],"slides":[],"orientation":[],"userdata":""}}}