A novel bandwidth allocation scheme for OTSS-enabled flex-grid intra-datacenter networks
A novel bandwidth allocation scheme for OTSS-enabled flex-grid intra-datacenter networks
复制标题
DOI：
10.1007/s11107-021-00933-z
复制
发表时间：
2021-08
影响因子：
1.7
通讯作者：
Lin Wang;Xinbo Wang;M. Tornatore;Kwangjoon Kim;B. Mukherjee
中科院分区：
文献类型：
--
作者：
Lin Wang;Xinbo Wang;M. Tornatore;Kwangjoon Kim;B. Mukherjee
光路交换网络被认为是一种很有前途的数据中心间网络解决方案。然而，对于数据中心内网络，由于其相对粗粒度的信道分配和特殊的数据中心内流量模式，它们可能无法有效地提供流量请求。光时间片交换（OTSS）是近年来提出的一种光交换技术，它通过将柔性网格交换的优点扩展到时域，从而实现更精细的粒度，从而提供灵活透明的光路。由于OTSS需要昂贵的纳秒速度光交换机，因此对整个数据中心进行一次性升级在经济上可能不可行。因此，我们预计细粒度的OTSS使能和粗粒度的灵活网格使能的光交换技术将在可预见的未来共存。在这项研究中，我们研究了一个OTSS启用的灵活网格（OTSS-FG）架构的数据中心内的网络。对于给定流量的情况，我们开发了一个混合线性规划来研究OTSS-FG架构中的最优带宽分配方案。当业务流是真实的时产生的，通过利用机器学习技术检测流的类型，我们提出了一种流感知的带宽分配（FABA）方案和一种动态的FABA方案，称为“D-FABA”方案。数值仿真结果表明，该带宽分配方案在平均延迟和阻塞概率方面优于基准方案。
Optical circuit switching networks have been recognized as a promising solution forinter-datacenter networks. However, forintra-datacenter networks, they may fall short in efficiently provisioning traffic requests due to their relatively coarse-grained channel assignment and special intra-datacenter traffic patterns. Optical time slice switching (OTSS) has been recently proposed as an optical-switching technique that can provide flexible and transparent optical circuits by extending the merit of flex-grid switching to the time domain, thus achieving much finer granularity. As OTSS requires nanosecond speed optical switches which are expensive, it might not be economically viable to make a one-time upgrade for the entire datacenter. Thus, we expect fine-grained OTSS-enabled and coarse-grained flex-grid-enabled optical switching techniques to co-exist in the foreseeable future. In this study, we investigate an OTSS-enabled flex-grid (OTSS-FG) architecture for intra-datacenter networks. For scenarios where traffic flows are given, we develop a Mixed Integer Linear Program to study the optimal bandwidth allocation scheme in an OTSS-FG architecture. When traffic flows are generated in real time, by leveraging machine-learning techniques to detect flow types, we propose a flow-aware bandwidth allocation (FABA) scheme and a dynamic version of FABA, called “D-FABA” scheme. Numerical simulations show that proposed bandwidth allocation scheme can outperform benchmark schemes in terms of average delay and blocking probability.