Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Elektrik ve Elektronik Mühendisliği Bölümü, Türkiye
Tezin Onay Tarihi: 2014
Öğrenci: MUSTAFA KARAGÖZ
Danışman: OSMAN SEVAİOĞLU
Özet:Virtual router is an essential solution to fulfill the increasing demands of network services. A virtual router, having a single hardware platform, serves several networks concurrently and hence provides cost saving. A virtual router maintains multiple forwarding tables that belong to separate internet service providers (ISPs) and performs IP lookup and forwarding functionality for each ISP in one common platform. IP lookup in a virtual router is performed by inspecting the incoming packets that also carry information about their ISPs. There exist various software and hardware IP lookup solutions in the literature. In software solutions tree or trie based data structures are usually employed which are relatively slower. Hardware solutions use TCAMs or SRAMs and are much faster. Limited on-chip memory is the main bottleneck for hardware implementations. If all ISP forwarding tables of a virtual router are stored separately then a large amount of memory is required and there occurs no benefit for having a virtual router. Therefore tables are usually merged in such a way that the overlapping parts are stored in one common data structure more efficiently. Decreasing the size of the memory and increasing the performance of look up and update tasks are among the primary concerns and challenges in virtual routers. Lookup performance is considered by means of latency and throughput issues. In this thesis, we investigate and propose an efficient trie overlapping approach and aim to decrease the memory requirement while achieving a good IP lookup and update performance. During this study, we examine real life prefix tables that belong to existing routers. We first merge these IPv4/IPv6 core and edge router tables in a simple manner into a single trie and observe that some parts of the trie use a large number of nodes to store a small number of prefixes. This observation has motivated and led us to reduce the size of the trie by truncating some of these low density subtries without destroying the advantageous trie structure too much. 2- 3 tree data structure is then proposed to be used as a secondary storage to keep the deleted prefixes from the trie separately. 2-3 tree is preferred because of its support for incremental updates. Our thesis identifies truncation metrics and we use them to keep most of the prefixes still in the trie. Our approach is evaluated and we have shown that it is possible to achieve around 5% reduction in memory size for IPv4 core routers and around 35% reduction for other type of routers in comparison to existing trie merging solutions. Hence memory efficiency is increased while supporting an update process that is possible using a single write bubble. Our solution achieves one lookup per clock cycle throughput and operates with a latency that is standard among other trie based solutions.