wiki:public/20121012

Project: Hard real-time data stream processing on heterogeneous CPU/GPU systems with multiple GPU accelerators

Team: Prof. Avi Mendelson (Technion – Israel Institute of Technology), Uri Verner (HPI research school, Technion – Israel Institute of Technology)

Research institution: Israel Institute of Technology

Abstract:

FirstPicture

"The ability to process large volumes of streaming data in real-time is required in many application domains. This project aims to provide a framework for processing data streams with real-time latency constraints utilizing CPUs and GPU accelerators. Take for example a server that provides a service of secure voice and video calls over an IP network. Each side of the call produces a continuous stream of data packets that need to be processed by the server within a limited time in order to avoid noticeable latency. Moreover, calls may differ in communication latency, which imposes the effective deadlines for processing data to be individual to each stream. Processing operations on such data include decryption/encryption, transcoding, recording, and even analyzing to improve quality or check for illegal content. Given a description of a workload of hard real-time data streams our system finds a distribution of the streams to CPUs and GPUs in the system, for which hard real-time dealine compliance is guaranteed. The system takes into account the properties of each processor and stream characteristics -- data rate and relative deadline -- in order to trim most of the possible distributions, and only consider a polynomial subset of distributions that have a potential to result in a deadline-compliant distribution."

Second Picture

Last modified 7 years ago Last modified on Jul 31, 2012 9:13:20 AM

Attachments (2)

Download all attachments as: .zip