TY - GEN
T1 - FFTree
T2 - 25th IEEE/ACM International Symposium on Quality of Service, IWQoS 2017
AU - Duan, Jun
AU - Yang, Yuanyuan
N1 - Publisher Copyright: © 2017 IEEE.
PY - 2017/7/5
Y1 - 2017/7/5
N2 - In this paper, we propose a novel data center network architecture named FFTree. Compared to widely adopted industrial solutions including fat-tree, this architecture can provide even higher aggregate bandwidth and richer path availability. Most importantly, FFTree provisions unique flexibility so that its topology and performance can be tailored with great freedom before its deployment. The fine granularity and wide dynamic range of its configurations allow FFTree to exactly fit and closely follow the time-variant demands of its applications. Also, this flexibility facilitates the reuse of network devices, so that upgrades could be achieved by rearranging existing switches, avoiding the expenditure on new generations of hardware in each upgrade cycle. Furthermore, this flexibility allows FFTree to be elaborately tuned into cost efficient zones in the sense that it provides higher bandwidth using less network devices, which leads to significant cost reduction. We design the network topologies of FFTree, propose a set of routing algorithms, and demonstrate its aforementioned features. We also analyze the performance and compare it with its predecessors. Technically, this architecture can be seen as a generalization of fat-tree, and is backward-compatible to fat-tree: it consists of commercial off-the-shelf network devices, and the whole system can cooperate with the hosts running Ethernet and IP. We believe FFTree is ready to be deployed in industry, as a perfect replacement of the classic fat-tree.
AB - In this paper, we propose a novel data center network architecture named FFTree. Compared to widely adopted industrial solutions including fat-tree, this architecture can provide even higher aggregate bandwidth and richer path availability. Most importantly, FFTree provisions unique flexibility so that its topology and performance can be tailored with great freedom before its deployment. The fine granularity and wide dynamic range of its configurations allow FFTree to exactly fit and closely follow the time-variant demands of its applications. Also, this flexibility facilitates the reuse of network devices, so that upgrades could be achieved by rearranging existing switches, avoiding the expenditure on new generations of hardware in each upgrade cycle. Furthermore, this flexibility allows FFTree to be elaborately tuned into cost efficient zones in the sense that it provides higher bandwidth using less network devices, which leads to significant cost reduction. We design the network topologies of FFTree, propose a set of routing algorithms, and demonstrate its aforementioned features. We also analyze the performance and compare it with its predecessors. Technically, this architecture can be seen as a generalization of fat-tree, and is backward-compatible to fat-tree: it consists of commercial off-the-shelf network devices, and the whole system can cooperate with the hosts running Ethernet and IP. We believe FFTree is ready to be deployed in industry, as a perfect replacement of the classic fat-tree.
KW - Configurability
KW - Cost efficiency
KW - Data center networks
KW - Network topology
KW - Routing
UR - https://www.scopus.com/pages/publications/85027873854
U2 - 10.1109/IWQoS.2017.7969131
DO - 10.1109/IWQoS.2017.7969131
M3 - Conference contribution
T3 - 2017 IEEE/ACM 25th International Symposium on Quality of Service, IWQoS 2017
BT - 2017 IEEE/ACM 25th International Symposium on Quality of Service, IWQoS 2017
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 14 June 2017 through 16 June 2017
ER -