Planning a Small Server Room

An anonymous reader writes: "Our company is planning to build a small server room. Initial requirements are for two or three enclosed server cabinets in which various servers and network gear will be installed. The cabinets are planned to hold between 15 to 20 servers of various types and sizes, switches, routers, four dial-in modems for after hours use by staff who do not have ISPs and a KVM switch. We would expect for a small desk as a work area, a book case, storage for some spare parts as well as server documentation and records. We know that we need some power protection in the way of a UPS and a generator. We also expect that this room will get quite warm in the summer months so it will need more air conditioning than the rest of the office. What should we expect for power and cooling needs? Are there any 'rules of thumb' when it comes to building a server room. Good suggestions and help would be appreciated."

Comments

Cooling Theory

[PhysicsGenius]

You shouldn't need to use too much cooling. Yes, the CPU produces heat but keep in mind that a server room is a closed environment--no energy (e.g. thermal energy) is actually created. The heat produced is given off by the entropy reversal of information being created. When that information is destroyed, in practical terms just deleting a file, some of that heat is sucked back up and it cools the room back down.

Of course, if you intend to send large amount of data out over the internet the environment is no longer entropically closed and you will experience heat buildup. In fact, Josh Bell proved in 1999 that data transmitted over a CAT5 cable is mathematically isomorphic to heat transferred backward over that same cable.

Since you are probably intending to have a net link, make sure you insulate your T1 connection well to keep this heat gain to a mimimum.

Re:Cooling Theory

[Goldsmith]

You're assuming that deleting a file is the same thing thermodynamically as removing information. This might work for some mathmatical models, but I know of no disk structure that allows it's magnetic domains to return to thermal equilibrium for every bit of encoded data.

On top of this assumption, you are assuming that the computer acts as a better-than-ideal engine. The amount of heat put off, even if deleting a file did cool a computer, would still be extreme because work is done by the computer, and I'm sure that code optimization does not mean thermodynamic optimization.

There are many more problems with this argument, even the internal clock in the computer is going to create this type of entropy generated heat.

What you absolutely cannot get around is the heat generated by the current in the wires. Even in superconducting wires, current generates heat. Factor in resistance and you have a source of heat far greater than any heat generated by entropy reversal.

These are fine theoretical assumptions, but in practice, computers generate a lot of heat.