I'm upgrading our cabinet at our colocation for A/B power and wanted input on which of our options are good or bad.
First, a bit of an overview of our setup: We currently have a single circuit feed into our cabinet. We have a few servers that are mission critical with dual power supplies and all the other mission critical servers are upgradable to dual power supplies. Our options as I see it are this:
- Wire the servers with dual supplies to A/B power and then divide the other servers evenly across A/B sources.
- Get an extra power supply for all mission critical servers and wire all servers to A/B power. (with the exception of our firewall and switches which cannot accommodate dual power – those would require manually plugging into the B source if A went down.)
- Get a PDU that automatically switches to B power when A goes down (like here: http://www.tripplite.com/en/products/model.cfm?txtModelID=4809)
To me #3 sounds like the best option since we could just plug in our servers to that PDU and if A goes down it will automatically switch to B. The main possible drawback is that it does contain an ATS which could potentially fail and then we'd be in serious trouble. For #3 to work we'd need to know that a unit like in the link above was reliable. Otherwise we'd opt for #1 or #2.
I'd be interested it hear your thoughts on if #3 is a good approach to increase our power redundancy or if we should take another route.
Best Answer
No matter how much redundancy is claimed to be inside a device, I always prefer to have redundant devices if at all possible. Therefore I wouldn't be too thrilled myself about #3. I don't really have any science to back that up -- it is more that I don't trust all of the internals to be perfectly engineered with redundancy.
Your #1 is a good choice. With redundant power it is easy not to keep track of your total utilization, and then what can happen is that you end up overloading the other power feed during a failover situation when all power gets switched to the redundant feed.
Another option which you didn't list. With the mission critical servers, if you can live with only some of them, wire some to A and some to B (and maybe some to A/B). So, for example, what I did with our web servers is to break them into groups of 3 servers. Server A is hooked into power A, server B is hooked into power B, server C is A/B. That way if I lose power on a single feed I am operating with 2/3rds of my web servers which should be fine.
Regarding your switches, you might consider buying secondary switches and then teaming your NICs in a failover mode that can span multiple switches.
Above all, get PDUs that display the power you are drawing and make sure you don't go too close to the limit.