I responded to a question on Quora a few weeks ago that asked how enterprises separated premium data centers from some of the “bargain” options on the market. In short: experienced data center buyers take the time to understand a service providers’ performance track record.
Most performance issues can be lumped into three silos: reliability and uptime, power and connectivity availability, and support and communication. In terms of reliability, larger companies tend to look for providers that have N+1 (what is needed + 1 spare) architecture designs across power, cooling and connectivity. In addition, they ask that the data center be “concurrently maintainable,” essentially meaning that the data center can perform major maintenance across critical components without having to take the facility offline. They also frequently ask the company or data center manager to provide uptime and maintenance records and procedures.
For power, large companies look for lots of headroom incase they want to increase cab density from two or three kilowatts to 12 kW somewhere down the line. This gives them the ability to “upgrade in place” in the future without moving to a new footprint or an entirely new data center.
Connectivity density is also important. Some legacy telco data centers or “mom and pops” will simply have one backbone choice which really hurts if that carrier happens to goes down (either at the loop or closer to the core). The best-case scenario is a good mix of transit or peering options and fiber providers with route optimization capabilities.
Finally, larger companies look for excellent support. A provider with multiple Network Operations Centers (NOCs) in-house is preferable. When they call the NOC with an issue they also want to be sure they are speaking with a certified engineer instead of a ticket-taker. They look for facilities with 24/7 on-site engineers (imagine this scenario: something goes wrong at the data center, the offsite engineer gets a call from the NOC [maybe], has to drive 30 minutes to get to the facility, and THEN tries to fix the problem). And last, but certainly not least, they want to be notified of any network events or outages. Some providers say “100% SLA” and then never tell you there was a problem. Larger companies are going to be on the lookout for proactive notification AND crediting.
From there, the choice depends on the size of the deployment, location and a specific use case. For more than 500kW of power, a wholesale provider (Digital Realty, Vantage, Dupont Fabros) might provide a more economical solution. Otherwise a deployment of one to 100 racks (~<500kW) in a retail colocation facility like Internap is likely a better option because all the support is taken care of, including environmental monitoring, security and maintenance from the shell all the way to the rack and remote hands. If more flexibility is needed (with more opex/less capex) managed hosting, dedicated hosting, and cloud are all great options.
