Grant: Discovery Launch Supplement
University of Toronto
$12,500.00
Networking Data Center Accelerators for Hyperscale Data Management
EF 2024-25
Résumé
The largest data management workloads are hosted in cloud data centers, which have ballooned to the hyperscale level over the years, encompassing billions to trillions of data items and hundreds to thousands of servers. Managing data at such scale requires efficient designs for all of the components involved, including hardware, data center networks, and database management systems (DBMSs). Unfortunately, every component is facing fundamental challenges for achieving high efficiency at hyperscale.
For hardware, the end of Dennard scaling and the slowdown of Moore's law have meant that maintaining the performance advances of general-purpose processors is becoming extremely challenging. For networks, hyperscale workloads are generating traffic and requesting performance that is beyond the capacities of data center networks. For DBMSs, the mismatch between the data growth and hardware speed progress makes it increasingly difficult for DBMSs to achieve expected performance and cost efficiency.
I propose Universal Data Management Acceleration (UDMA) to address hyperscale data management challenges, by rethinking the hardware, networking, DBMSs, and their interactions. First, rather than stick to general-purpose processors, data management workloads can benefit from hardware accelerators in many ways. Given the variety of data center accelerators, unlocking their full benefits requires a unified abstraction that facilitates accelerator selection and DBMS offloading. Second, there need be a network connecting accelerators for query execution, because data movement across workers is guaranteed to occur at hyperscale. The network must respect the heterogeneity of accelerators and provide an easy-to-use interface to DBMSs. Finally, on top of the infrastructure are DBMSs that can leverage the support to optimize the execution of data management workloads.
The long-term vision of UDMA is that the optimal performance and cost efficiency for hyperscale data management comes from universal acceleration where DBMSs take full advantage of data center accelerators. My group will focus on the following thrusts to realize the vision:
1. a unified abstraction of hardware acceleration for DBMSs,
2. a data center networking stack for scaling out data management acceleration,
3. an acceleration-centric DBMS query optimizer for hyperscale analytics,
4. an accelerated hyperscale transaction processing engine, and
5. cost-optimized hyperscale data management.
Programme
Discovery Launch Supplement
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
DGECR-2023-00308
Name nom
Zhang, Qizhen
Department département
Computer Science
Organization id
31
Institution établissement
University of Toronto
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
12500
Program id
DGECR
Program name en
Discovery Launch Supplement
Program name fr
Supplément Tremplin vers la découverte
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
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
2711
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Database management
Research subject fr
Gestion de bases de données
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Networking Data Center Accelerators for Hyperscale Data Management
Keywords
data management; databases; query optimization; computer systems; cloud computing; data center networking; hardware accelerators; heterogeneous hardware; compute offloading; application-infrastructure codesign
Application summary
The largest data management workloads are hosted in cloud data centers, which have ballooned to the hyperscale level over the years, encompassing billions to trillions of data items and hundreds to thousands of servers. Managing data at such scale requires efficient designs for all of the components involved, including hardware, data center networks, and database management systems (DBMSs). Unfortunately, every component is facing fundamental challenges for achieving high efficiency at hyperscale.
For hardware, the end of Dennard scaling and the slowdown of Moore's law have meant that maintaining the performance advances of general-purpose processors is becoming extremely challenging. For networks, hyperscale workloads are generating traffic and requesting performance that is beyond the capacities of data center networks. For DBMSs, the mismatch between the data growth and hardware speed progress makes it increasingly difficult for DBMSs to achieve expected performance and cost efficiency.
I propose Universal Data Management Acceleration (UDMA) to address hyperscale data management challenges, by rethinking the hardware, networking, DBMSs, and their interactions. First, rather than stick to general-purpose processors, data management workloads can benefit from hardware accelerators in many ways. Given the variety of data center accelerators, unlocking their full benefits requires a unified abstraction that facilitates accelerator selection and DBMS offloading. Second, there need be a network connecting accelerators for query execution, because data movement across workers is guaranteed to occur at hyperscale. The network must respect the heterogeneity of accelerators and provide an easy-to-use interface to DBMSs. Finally, on top of the infrastructure are DBMSs that can leverage the support to optimize the execution of data management workloads.
The long-term vision of UDMA is that the optimal performance and cost efficiency for hyperscale data management comes from universal acceleration where DBMSs take full advantage of data center accelerators. My group will focus on the following thrusts to realize the vision:
1. a unified abstraction of hardware acceleration for DBMSs,
2. a data center networking stack for scaling out data management acceleration,
3. an acceleration-centric DBMS query optimizer for hyperscale analytics,
4. an accelerated hyperscale transaction processing engine, and
5. cost-optimized hyperscale data management.