SLAs are an “important” predictor of future performance: This is one of the secrets of the entire IT services industry. People hate it when I say this, but its true: SLAs are all a scam. Enterprise decision makers especially hate it, because they predicate their reliability decisions on SLAs. Here’s an experiment - ask your provider if they’ll give you a better SLA. They will agree - trust me. Then, ask them what they changed to make that better SLA happen. Guess what? They didn’t change anything.
Don’t despair, however - There is a solution. The best predictor of future performance is not an SLA, its PAST PERFORMANCE. In other words, ask for the last five years of datacenter and network downtime, in minutes, across all facilities. If they say that they don’t record the data, run away - fast.
Design is the the “most” significant determiner of reliability: Don’t get me wrong - design is important. However, there are very few data centers with innovative designs, outside of increased efficiency. The move towards PODs, pre-fab, and containerization has undermined efforts towards innovative designs. At this point, there are exactly two providers with an innovative design focused on reliability.
So, what is the most significant determiner of data center reliability, right now? The data center’s critical facilities engineering and operations groups. Wait, your data center DOES have one of those, doesn’t it? You shouldn’t assume - these functions are often outsourced - poorly. Assuming your provider performs these functions in house, look for the following yardsticks: degreed engineers in management positions, master electricians, navy nuclear-trained technicians, and technicians who have worked for the major vendors (GE, Schneider, Emerson, York, etc) utilized in that facility. Also, ask for preventative maintenance records - if they can’t produce their PM history and schedule, run away.
“Multiple Power Grids” - Some data centers brag that they are on multiple power grids. The problem is that it is never true. In the US, there are only three power grids - East, West, and Texas. Unless your datacenter is in lovely downtown Texarkana, its on one grid. Sometimes, operators mean that they have feeds from multiple substations, but the same grid. This can be factually accurate, but is rarely effective - 90% of the outages that would bring down one substation would bring down an adjacent substation. Data centers must always assume they will lose utility power and operate accordingly. There is a certain level of hardening that is appropriate for connectivity to your power utility, but additional layers of connectivity beyond that point rarely bring additional uptime or reliability. Look for underground conduits and use of primary transmission voltage (34.5 or 69 kVA).