Tuesday, August 23, 2016

Optimizing Bulk Transfers with Software-Defined Optical WAN

Optimizing Bulk Transfers with Software-Defined Optical WAN

Xin Jin (Princeton)
Yiran Li (Tsinghua)
Da Wei (Tsinghua)
Siming Li (Stony Brook)
Jie Gao (Stony Brook)
Lei Xu (Sodero Networks)
Guangzhi Li (AT&T Labs)
Wei Xu (Tsinghua)
Jennifer Rexford (Princeton)

Wide area data transfer are major part of today's globally-distributed applications. These transfers account for large part of WANs. Network operators are required to carefully schedule these transfers as they can impact not only their business success but also application's performance (e.g. search result quality with outdated index).

In Modern WAN, network layer is built over an intelligent optical layer. Operators can dynamically change their network topology by re-configuring the optical devices. In this paper, authors present Owan, a traffic management system that optimizes wide-area bulk transfers with centralized joint control of the optical and network layers. They use SDN along with ROADM devices (Reconfigurable Optical Add-Drop Multiplexer, a device that connects a WAN router to an optical switch. Operators can change which two router ports are connected by changing optical layer circuits, Modern ROADM devices allow fast remote re-configurations e.g. provisioning a circuit in tens to hundreds of milliseconds instead of minutes). Owan orchestrates bulk transfers in a centralized manner. It computes and implements the optical circuit configuration (the optical circuits that implement the network-layer topology) and the routing configuration (the paths and rate allocation for each transfer) to optimize the transfer completion time or the number of transfers that meet their deadlines.

They build Owen prototype using commodity hardware and have nine sites with Internet 2 topology emulation. They conduct extensive evaluations through both testbed experiments and large scale simulations with data from an ISP network and an inter-DC network. Their results show that Owan improves the transfer completion time by up to 4.45x on average and 3.84x at the 95th percentile, as compared to prior methods that only control the network layer. Furthermore, Owan allows up to 1.36x more transfers to meet their deadlines and up to 2.03x more bytes to finish before their deadlines.

Q1) You give Software defined ability to optical networks and configure optical circuits. How do you regenerate a optical network and what about load balancing the traffic?

A) We build regeneration graph that resembles regular consumption and transforms the problem to a shortest path finding problem. This helps us regenerating the topology.