This paper, from a major webscale provider, describes how the Microsoft optical network is operated at scale with low margin. The paper does a good job of motivating the data-center-interconnect (DCI) and long-haul architectures that Microsoft has deployed, along with their decision to employ open line systems. The authors provide a wealth of information directly related to their network’s design and operation, including some of the key enablers of low-margin operation. The paper includes numerous network statistics, including: distributions of span distance, span loss, fiber type, route length; Q-factors measured over 50 days; and statistics based on measurements of thousands of in-service transceivers, e.g., temperature, failure rate. Of particular interest is how Microsoft puts into practice two seemingly simple yet powerful principles: trade availability for lower margins at the optical layer and let the upper layers deal with availability; and use simple "minimalist" hardware and software platforms.
You must log in
to add comments.