{"id":247778,"date":"2016-07-02T22:27:22","date_gmt":"2016-07-03T05:27:22","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-project&p=247778"},"modified":"2022-11-28T13:19:00","modified_gmt":"2022-11-28T21:19:00","slug":"microsofts-wide-area-optical-backbone","status":"publish","type":"msr-project","link":"https:\/\/www.microsoft.com\/en-us\/research\/project\/microsofts-wide-area-optical-backbone\/","title":{"rendered":"Wide-Area Optical Backbone Performance"},"content":{"rendered":"

\"L3\"<\/p>\n

Wide-area backbone networks (WAN) of Internet service providers and cloud providers are the workhorses of Internet traffic delivery. The providers spend millions of dollars toward building access points across the world and interconnecting them using optical links. Improving the availability and efficiency of the WAN is central to their ability to provide services in a reliable, cost-effective manner. Consequently, there has been significant research into measuring and characterizing various aspects of the WAN, such as topology, routing, traffic, and reliability.<\/p>\n

Much of prior work, however, focuses on the IP layer, and little is publicly known about the characteristics of the\u00a0optical layer which forms the physical transmission medium of WANs. While laboratory measurements can inform certain quantities (e.g., expected signal quality), only “in the wild” characterization can shed light on the behavior and the combined impact of many relevant factors in practice.<\/p>\n

Studying optical-layer characteristics of backbone networks is not simply a matter of curiosity. The performance of this layer ultimately determines the network’s effectiveness at carrying traffic.<\/p>\n

Dataset<\/h2>\n

Our dataset is based on 14 months of data, from February 2015 to April 2016, taken from Microsoft\u2019s optical backbone in North America. This backbone has O(50) optical cross-connects, O(100) WAN segments, and O(1000) optical channels. We poll the aggregation devices every 15 minutes for their optical signal q-factor, transmit power (dBm), chromatic dispersion (ps\/nm), and polarization\u00a0mode dispersion. Below we share the data set for random 4000 channels across random 115 optical paths.\u00a0For confidentiality reasons, we have taken out days at which there has been an outage. This data is the first public release of a large-scale optical backbone and provides researchers with a unique opportunity to study temporal behavior of optical channels, the quality of signal, correlation among channels, correlation among segments, and more.<\/p>\n

Sample channels:<\/p>\n