Grant: Discovery Grants Program - Individual
University of British Columbia
$40,000.00
Reconsidering distributed storage system design to harness massively multi-core commodity hardware
FY 2012-13
Summary
The rate of progress in a number of scientific fields (e.g., bioinformatics), as well as the productivity of highly-trained professionals that interactively use data-intensive computing systems (e.g., radiologists), are limited by the peace at which computing systems are able to manage and process large volumes of data.
To address this issue this is proposal is set to explore two hypotheses. First, we believe that active storage techniques can be efficiently and transparently integrated with today's massively multi-core processors (e.g., Graphics Processing Units or GPUs) to accelerate the execution of data-intensive applications in fields as diverse as bioinformatics and medical image analysis. The second hypothesis is that the order-of-magnitude reduction in computation costs massively multi core devices offer can be harnessed not only to accelerate specialized applications, but also to accelerate, enhance, or reduce the cost of generic storage systems. Inexpensive computation can be used to support key techniques that impact storage systems reliability, security, or performance by using erasure coding, data similarity detection, encryption, or on-the-fly compression.
We plan to: (i) study the viability of offloading data-processing intensive operations to the GPU; in addition to synthetic workloads we will use medical imaging storage and processing systems to anchor this project with realistic workloads, demonstrate feasibility, and asses performance gains (ii) develop accurate performance models that will inform when outsourcing to the GPU is advantageous and will guide the performance optimization efforts; (iii) improve development and runtime support (e.g., GPU-specific memory management, automated choice of parallelism levels and thread mapping) that will reduce the cost of application migration while extracting maximum performance; and, finally, (iv) explore solutions to virtualize the GPU and share it between multiple applications. Progress along each of the above directions would represent a valuable contribution to the body of theoretical/practical knowledge related to storage system design.
Program
Discovery Grants Program - Individual
Recipient
Location
British Columbia, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
327287-2009
Name nom
Ripeanu, Matei
Department département
Electrical and Computer Engineering
Organization id
2
Institution établissement
University of British Columbia
Province en
British Columbia
Province fr
Colombie-Britannique
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2012
Competition year année de concours
2009
Award amount
40000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
330
Committee name en
Computing and Information Sciences - A
Committee name fr
Informatique - A
Area of application code
802
Area of application group en
Information and communication services
Area of application group fr
Services d'information et de télécommunications
Area of application en
Information systems and technology
Area of application fr
Systèmes d'information et technologie de l'information
Research subject code
2704
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Distributed and parallel processing
Research subject fr
Traitement réparti et simultané
Partie
4
Nb partie
5
Application title
Reconsidering distributed storage system design to harness massively multi-core commodity hardware
Keyword
NSERC
Application summary
The rate of progress in a number of scientific fields (e.g., bioinformatics), as well as the productivity of highly-trained professionals that interactively use data-intensive computing systems (e.g., radiologists), are limited by the peace at which computing systems are able to manage and process large volumes of data.
To address this issue this is proposal is set to explore two hypotheses. First, we believe that active storage techniques can be efficiently and transparently integrated with today's massively multi-core processors (e.g., Graphics Processing Units or GPUs) to accelerate the execution of data-intensive applications in fields as diverse as bioinformatics and medical image analysis. The second hypothesis is that the order-of-magnitude reduction in computation costs massively multi core devices offer can be harnessed not only to accelerate specialized applications, but also to accelerate, enhance, or reduce the cost of generic storage systems. Inexpensive computation can be used to support key techniques that impact storage systems reliability, security, or performance by using erasure coding, data similarity detection, encryption, or on-the-fly compression.
We plan to: (i) study the viability of offloading data-processing intensive operations to the GPU; in addition to synthetic workloads we will use medical imaging storage and processing systems to anchor this project with realistic workloads, demonstrate feasibility, and asses performance gains (ii) develop accurate performance models that will inform when outsourcing to the GPU is advantageous and will guide the performance optimization efforts; (iii) improve development and runtime support (e.g., GPU-specific memory management, automated choice of parallelism levels and thread mapping) that will reduce the cost of application migration while extracting maximum performance; and, finally, (iv) explore solutions to virtualize the GPU and share it between multiple applications. Progress along each of the above directions would represent a valuable contribution to the body of theoretical/practical knowledge related to storage system design.