NOVEL FAT TREE TOPOLOGIES AND ROUTING SCHEMES FOR DATA CENTER NETWORKS
MennaTallah Adel Ahmed Mazen Hussein;
Abstract
In this thesis, optimized network topologies for enhanced throughput performance in data centers containing tens of thousands of nodes are explored. In particular, the focus will be on the Fat Tree topology, where optimization techniques are proposed along two major directions. First, the performance gains attributed to "extra" inter-pod links creating shortcut paths at the two bottom levels of the Fat Tree are quantified. Using the extra links will relieve the top level core switches, especially for local (between neighboring/adjacent pods) traffic. This entails novel routing algorithms, inspired by two-level table routing, which is a static routing algorithm, to best leverage the extra inter-pod links, while balancing the load across the datacenter network. The proposed topology and associated routing techniques are backward compatible to Ethernet, IP and TCP, and fall back to the classic Fat Tree in the absence of extra links. Results are obtained from simulation to quantify the throughput gain attributed to extra links under a variety of random and extreme traffic patterns. These proposed solutions present low cost add-ons to existing Fat tree networks that require more aggregate bandwidth to support new applications. Second, a fundamental question is posed contrasting inexpensive extra links versus costly core switches. A novel topology is developed in an attempt to answer this question, namely the FitFat Tree. This gives rise to an interesting trade-off which we shed light on in this research work.
The simulation results unveil a valuable insight, that is major investment can be saved for greenfield datacenters, by using the proposed extra links to partially replace some of the core switches in the Fat Tree and still show remarkable improvement in network’s performance for different traffic patterns. We argue that the proposed FitFat Tree is a promising direction for scalable, cost-effective datacenter networks of the future specially if 60 GHz wireless technologies, like IEEE 802.11ad, are leveraged.
The simulation results unveil a valuable insight, that is major investment can be saved for greenfield datacenters, by using the proposed extra links to partially replace some of the core switches in the Fat Tree and still show remarkable improvement in network’s performance for different traffic patterns. We argue that the proposed FitFat Tree is a promising direction for scalable, cost-effective datacenter networks of the future specially if 60 GHz wireless technologies, like IEEE 802.11ad, are leveraged.
Other data
| Title | NOVEL FAT TREE TOPOLOGIES AND ROUTING SCHEMES FOR DATA CENTER NETWORKS | Other Titles | طبولوجيات ومخططات توجيه جديده لشبكات مراكز البيانات ال FAT TREE | Authors | MennaTallah Adel Ahmed Mazen Hussein | Issue Date | 2017 |
Recommend this item
Similar Items from Core Recommender Database
Items in Ain Shams Scholar are protected by copyright, with all rights reserved, unless otherwise indicated.