The processing of high-frequency data streams is a challenging problem of high commercial interest in the financial sector. High-frequency streaming requires high-performance computing systems that balance computational power with high-throughput memory systems and interconnects. Applications have stringent performance constraints and often require real-time response.
The state of the art in high-performance computing adopts heterogeneous parallel architectures integrating multi-core CPUs with accelerators, such as GPUs, at the node level. These platforms, although computationally powerful, are challenged by the bandwidth requirements of high-frequency streaming. Emerging heterogeneous architectures that integrate CPUs and accelerators at the chip level may provide the necessary bandwidth for real-time processing of in-memory data, but with reduced computational power per core.
Streaming applications also require effective programming abstractions for high-productivity code development, along with efficient runtime systems and sufficient operating system support. System design for high-frequency streaming should balance stream processing between CPUs and accelerators, use effectively the memory hierarchy for bulk data transfers while exploiting locality, use low-latency mechanisms for synchronization and coherence between the memory address spaces of CPUs and accelerators, and perform efficient allocation to multiple continuous streams.
A number of PhD projects are available in this area. These projects are outlined in the objectives section.
Programming abstractions and runtime support for real-time data streaming on heterogeneous architectures: This project aims to investigate programming abstractions, such as language constructs or program annotations, for high-frequency streaming applications, as well as the necessary compiler and runtime support for translating these abstractions into efficient code for different architecture targets (CPUs, hybrid CPU-GPU nodes, CPU-GPU multiprocessors, and clusters of heterogeneous compute nodes).
Resource allocation and management for streaming on heterogeneous architectures: This project aims to investigate new methods for provisioning and managing heterogeneous resources for high-frequency streaming. The project will develop new methods for allocation of computational, memory and interconnect resources to stream processing kernels and the associated data transfers, data partitioning and data consistency management between address spaces of heterogeneous devices, and dynamic load balancing between heterogeneous cores.
Virtualization: This project aims to investigate real-time stream processing in data centers with virtualized computing environments. The associated topics include the study of the impact of virtualization on stream processing throughput, resource provisioning to concurrent virtualized data streaming applications towards multi-objective optimization (preserving SLAs, maximizing resource utilization, maximizing energy-efficiency) and virtual machine monitor design, implementation, and customization for streaming applications.
A minimum of a 2.1 honours degree or equivalent in Computer Science or Electrical and Electronic Engineering, or relevant degree is required.
This 3 year PhD studentship, funded by the Department for Employment and Learning (DEL), covers approved tuition fees and a maintenance grant of approximately £13,590. The PhD studentship normally commences on 1 October 2012 however, this could be available immediately to suitable candidates.
Applicants should apply electronically through the Queen's online application portal at: https://dap.qub.ac.uk/portal/
Further information available at: http://www.qub.ac.uk/schools/eeecs/Scholarships/PostgraduateResearchScholarships/
|Supervisor Name:||Prof Dimitrios S. Nikolopoulos|
|Address:||School of Electronics,
and Computer Science
Bernard Crossland Building,
|Tel:||+44 28(0) 9097 1809|