Article Info
Big Bandwidth in the Big Apple
I spent the day shuttling around Manhattan. My travels took me near 111 8th Avenue, one of the big carrier hotels in Manhattan. Unlike previous encounters with data centers, I was unable to go inside. This building is a brick behemoth and literally spans an entire city block. Google has an entire floor to themselves.
This sort of arrangement strikes me as a little silly. Clearly it would be better to locate in lower cost real estate areas. Not everyone needs to be 500 microseconds from Wall St. If bandwidth were truly cheap people wouldn’t still be operating data centers in Manhattan. They would all be in caves in South Dakota.
Price of bandwidth is not the only factor in choosing the location for a Datacenter.
Web sites (not to mention picky applications) perform better when latency between client and server is lowest. NYC happens to be geographically near to a lot of eyeballs as well logically in the Internet as a result of US and European ISP transit and peering connections that terminate in the area.
It is no accident that the most popular places to host web applications is in the vicinity of the original NAP’s and MAE’s (DC, NYC, San Francisco, and to a lesser extent, Chicago). Hosting your applications in these areas will generally guarantee the best web performance to the entire globe.
Remember too that the mid-sized buyers of datacenter space (the clients I serve with the facility I run in Seattle) tend to want to have their machinery close at hand.
While it is a very sexy concept of having a server grid of commodity hardware that routes around component failure, very few companies beyond Google have actually pulled it off. Most are still using the standard web/mail/database server layouts that have existed since HTTP was a minority protocol.
Having your servers close by eases the mind of IT guys who trade increased cost for shorter (panic!) drives when the inevitable happens.
–chuck
111 Eighth is a multi-tenant “carrier hotel” where many of the tenants interconnect with one another’s networks. While the cave in South Dakota would be cheaper for a single-tenant data center, it doesn’t work for carriers or high-traffic sites that use interconnections to extend their network or use peering to reduce costs. The aggregation of bandwidth is really what draws new customers to places like 111 Eighth or 60 Hudson.
Google has a big location at 111 Eighth, but oddly enough it’s a sales location rather than a data center.
This reminds me a lot of the way the Interstate system developed in the US. Originally all of the roads went right through downtown urban areas. Likewise, all of the big data centers are in urban areas because the fiber was laid that way.
Eventually, people built beltways that circumnavigated big urban areas. It would appear to me that the next big opportunity are datacenters in this beltway configuration, with lower operating costs.
Likewise we see Google and Yahoo and MSFT building datacenters in the middle of nowhere, just like FedEx has their hub in Memphis (sorry Memphis readers)and not at ORD or DFW.