Author :
Publisher :
ISBN 13 :
Total Pages : 124 pages
Book Rating : 4.:/5 (865 download)
Book Synopsis Building Efficient Packet Classification Systems Through Software-centric Approaches by :
Download or read book Building Efficient Packet Classification Systems Through Software-centric Approaches written by and published by . This book was released on 2013 with total page 124 pages. Available in PDF, EPUB and Kindle. Book excerpt: Building high-performance and extensible networking devices is one of the hottest topics in computer networking. However, several factors make this task challenging. First, the performance of software routers is limited, although they are easily extensible and programmable. Second, the latest commercial routers are the standard for high performance but they rely on specialized hardware and software that is difficult to extend. This thesis focuses on designing and implementing practical and extensible software-centric solutions for building high-performance routers. We improved the performance of packet classification systems in multiple ways, including speed, power consumption and storage requirements. Our work has had a broad impact on various packet classification systems. 1) Our software router work exploited the parallelism in state-of-art multi-core desktop platforms, and it achieved a packet classification speed of more than 10 Gbps on a single 8-core Intel Xeon processor desktop platform. 2) We introduced the concept of the pre-classifier, which reduced the dimensionality of the multi-dimensional packet classification. We then applied the pre-classifier to a TCAM-based power-efficient packet classification system and a software-based memory-efficient packet classification system. 3) In light of new hardware features in TCAMs, including parallel search and partial address support, our TCAM design, which includes a pre-classifier, reduced power consumption by more than 80% when used for multi-dimensional packet classification. 4) We observed that RAM-based algorithmic solutions for multi-dimensional packet classification either require a large memory size to load the rules, or they require a large number of memory accesses for a packet classification. Therefore, we proposed a memory-efficient packet classification system that includes a pre-classifier and achieves a large memory size reduction at lower cost of memory accesses. For future work, we plan to apply our packet classification techniques to systems with limited resources, such as systems with limited TCAM size. We also plan to apply our techniques to the emerging problems in the domain of Software-Defined Networking.
