How eBay Created Its Own Transaction Metric for Data Centers

Online auction giant eBay has developed an application-specific metric for evaluating the efficiency of its server infrastructure, which it refers to as Digital Service Efficiency, or DSE.

This metric was applied to eBay’s 2012 efficiency performance, establishing a baseline of 45,914 transactions per kilowatt-hour. The company plans on using that number as a foundation of sorts, with a goal of increasing transactions per kilowatt-hour in 2013—reducing the cost of transactions, as well as carbon produced per transaction, in the process.

A number of metrics have emerged to evaluate the efficiency of a data center. The most popular public metric is Power Usage Effectiveness (PUE), which measures how much power is actually used by the data center for computing, and how much is wasted on cooling and other functions. Other metrics measure the water used, or how much carbon dioxide ends up produced.

For its part, eBay touts DSE as “essentially a ‘miles per gallon’ measurement for technical infrastructure that makes an end-to-end connection between what customers do and the fundamental business metrics they influence,” the company wrote in a technical brief (PDF). The numbers will allow eBay to make better business decisions.

In real-world terms, eBay simply broke down its auction Marketplace into its component parts: buying and selling real-world goods via online requests, known as transactions. From its own internal use, it knew that a buy or sell request progresses through a specific number of URLs in order to deliver the service; the same for its PayPal payments business, which relies on similar methods to pay and transfer funds. From there, it established “currencies” for its other systems: for example, search engines, databases, data warehouses and other shared infrastructure used APIs and SQL queries as currencies. Each transaction consumed some amount of the “currency” used by each system; once abstracted, eBay began slicing the data into metrics transactions per server, dollars per transaction, CO2 per transaction, and so on.

From there, eBay began “turning the knobs” that would adjust aspects of the company’s business, trying to isolate the variables that would reduce cost without impacting performance. In one case, eBay slightly trimmed the amount of memory used by an application across a pool of servers, resulting in being able to trim 400 servers from the pool, and saving about $2 million in capital costs alone. In another, eBay eliminated high levels of redundancy and fault tolerance from its data centers, simple because it found it wasn’t needed. The result (eBay claims) was a doubling of data center power capacity, a four-fold increase in rack density, and a 53 percent reduction in cost per megawatt delivered over the last four years.

DSE, eBay said, represents an opportunity for exes to make informed business decisions on how to spend computing resources. “…The choices that reduce carbon footprint won’t be made simply to achieve a badge of honor,” eBay said in its brief. “They will fundamentally be business decisions.”


Image: eBay