Backup, Virtualization, and the Next Big Thing: Large Data Centers
In our final post on next generation data center trends, we thought we’d simply show you as of mid-2014 estimates of the number of servers in some of the largest and best known data centers – and discuss trends associated with more servers with fewer IT staff.
Server counts are a closely held secret at many of the largest data centers – so you end up having to piece together and extrapolate the number of servers within these data centers. Google has an estimated one to two million servers operating; Microsoft has an estimated million servers; and Amazon has an estimated 500,000 servers. You can find Akamai, Facebook, Intel, Yahoo, Godaddy, Rackspace, Ebay, and others as well on this chart. Google has stated publicly that they are planning their infrastructure to support ten million servers.
The most important thing to note here is that the number of servers, while interesting, is relatively meaningless. The generation of server, and thus the core count, memory, I/O backplane, and other attributes of the servers means that comparing a server built in 2010 versus a server built in 2014 isn’t a very constructive exercise.
What’s more interesting is the sheer number of servers per operations personnel. Facebook has confirmed as of 2013 that each of its operation personnel manages on average 20,000 servers – and some personnel manage as many as 26,000 servers. This is handled through the automation of most data center tasks. Facebook has developed its own software, which they call CYBORG, which detects server problems and attempts to resolve those problems. If automated repair attempts fail, CYBORG sends a ticket so that a human being can become involved.
In SMB (Small and Medium Business), where the average number of assets supported per IT staff is 30:1, the number of servers being supported per operations person seems daunting. But the answer is relatively straightforward: automation.
That’s all we’re going to talk about in this series regarding trends; in the next set of posts in this series we’ll begin discussing strategies for building next generation data centers today.
This is part 5 of an on-going series. Part 1: Virtualization isn’t the next big thing (NBT) because it was the last big thing (LBT); part 2: data center IP traffic growth; part 3: data center IP traffic sources; part 4: cloud workloads.
