Effective TCP Flow Management Based on Hierarchical Feedback Learning in Complex Data Center Network

Many studies focusing on improving Transmission Control Protocol (TCP) flow control realize a more effective use of bandwidth in data center networks. They are excellent ways to more effectively use the bandwidth between clients and back-end servers. However, these schemes cannot achieve the total optimization of bandwidth use for data center networks as they do not take into account the path design of TCP flows against a hierarchical complex structure of data center networks. To address this issue, this paper proposes a TCP flow management scheme specified a hierarchical complex data center network for effective bandwidth use. The proposed scheme dynamically controls the paths of TCP flows by reinforcement learning based on a hierarchical feedback model, which obtains an optimal TCP flow establishment policy even if both the network topology and link states are more complicated. In evaluation, the proposed scheme achieved more effective bandwidth use and reduced the probability of TCP incast up to 30% than the conventional TCP flow management schemes: Variant Load Balancing (VLB), Equal Cost Multi Path (ECMP), and Intelligent Forwarding Strategy Based on Reinforcement Learning (IFS-RL) in the complex data center network.

PDLB: Path Diversity-aware Load Balancing with Adaptive Granularity in Data Center Network

Modern data center topologies often take the form of a multi-rooted tree with rich parallel paths to provide high bandwidth. However, various path diversities caused by traffic dynamics, link failures and heterogeneous switching equipments widely exist in production data center network. Therefore, the multi-path load balancer in data center should be robust to these diversities. Although prior fine-grained schemes such as RPS and Presto make full use of available paths, they are prone to experi-ence packet reordering problem under asymmetric topology. The coarse-grained solutions such as ECMP and LetFlow effectively avoid packet reordering, but easily lead to under-utilization of multiple paths. To cope with these inefficiencies, we propose a load balancing mechanism called PDLB, which adaptively adjusts flowcell granularity according to path diversity. PDLB increases flowcell granularity to alleviate packet reordering under large degrees of topology asymmetry, while reducing flowcell granularity to obtain high link utilization under small degrees of topology asymmetry. PDLB is only deployed on the sender without any modification on switch. We evaluate PDLB through large-scale NS2 simulations. The experimental results show that PDLB reduces the average flow completion time by up to ∼11-53% over the state-of-the-art load balancing schemes.

ReDCN: A Dynamic Bandwidth Enabled Optical Reconfigurable Data Center Network

A reconfigurable optical data center network is proposed, in which the optical bandwidth can be automatically reconfigured by reallocating time slots based on the real time traffic. Numerical investigations validate that the network performance of packet loss after reconfiguration decreases by 58.5%, and the end-to-end latency decreases by 63.8% with comparison to the network with rigid link interconnections, and thereby increasing the 9.4% of throughput at load of 0.8.