6

I'm working on this project to develop a system with centralized information regarding emergencies delivered via open Wi-Fi on a small city.

I'm from Chile, so we thought of this system to work especially when an earthquake strikes the city (just last year we had a 8.8 Richter). My area of expertise is Networking, but I've been working for a while with servers. The thing that I'm new at is server hardware...

Do you know what kind of hardware can hold up better: tower or rack? (tower obiously being attached to the ground somehow)

Also, in your experience, is it better to have redundant internal hardware (redundant hard drives and power supplies) or to duplicate the entire hardware (clustered servers, 2 routers, 2 switches, etc.)?

Thanks

Journeyman Geek
  • 6,977
  • 3
  • 32
  • 50

3 Answers3

4

As mtinberg pointed out, there are standards for earthquake-hardening datacenters. Look them up and follow them.

In terms of "what kind of hardware", most server-class hardware is rack mounted. "Tower Servers" are generally not a bargain in terms of serviceability or room density.
You should expect to be forking out for redundant components (power, disks) and redundant systems - either paired up in the same location, or better as a mirror some distance away. A major earthquake could easily seriously damage or destroy your datacenter, and being able to bring up a secondary site to continue providing vital information to the pubilc is just good sense.

voretaq7
  • 79,879
  • 17
  • 130
  • 214
  • This is correct. You earthquake-proof your DC as a building, not the individual servers inside the DC. – Mark Henderson Aug 31 '11 at 22:25
  • Thanks, but I've been searching for a while now and still can't find the standards you mentioned... –  Sep 01 '11 at 06:24
  • I don't know that they are "formal" standards -- like Mark said you build an earthquake-resistant building (local architects & engineers can tell you what's involved there), and then you fill it with stuff like http://www.rackmountsolutions.net/Seismic_NEBS_Zone_4_Server_Rack.aspto protect the equipment (such racks come with instructions on how to safely install them) – voretaq7 Sep 02 '11 at 05:25
2

I would suggest rack mount servers and have the rack bolted to the floor, bolted to the wall, etc. There are standards for earthquake hardening for datacenter equipment that you should search for which have more details.

mtinberg
  • 1,833
  • 11
  • 9
1

I'm not sure what you mean exactly by a "tower"... but I'm assuming you mean some sort of shelving system... rather than a traditional enclosed server-rack.

A rack can hold up better assuming you bolt it to the floor, but unless the building it is in has some sort of shock-dampening the servers are still going to feel the full effect. (very bad for disk platters) Simply placing your servers on some sort of shelving system, means that they'll be loose and will more-likely to slide around and fall over which can cause even more risk to the equipment. Rack servers have some sort of sliding rail system that "locks" into the server rack and will prevent them from sliding at all.

In my experience, there is no better way to maintain "up-time" than to have at least two of everything in at least two completely geographically separate data centers. If your data center in one location gets hit bad enough that you have to replace equipment... it's significantly less likely that your other data center will be similarly hit. Of course... two of everything means twice the cost/expense. You should always plan for redundancy in everything you do. Disk drives should be run in some sort of parity based raid. (raid 1, 5, 6, 10 etc...) Depending on the quality/grade of routers & switches, you may feel comfortable relying on 1 router and switch rather than two. You may want to also consider using 1 switch for "users" and two switches for fail-over for your servers. Investing in 2 links for every workstation is pretty unreasonable for 99.9% of offices... but depending on up-time requirements and maintenance cycles... you may want to have 2 or more connections for each server... so you can update firmwares & configurations without having to have a disconnected server at any-time.

As this is in a "small-city"... you can probably afford some down-time periodically, and so it really comes down to a matter of cost. 99.9% up-time is pretty easy to maintain on a fairly low budget... but adding additional 9s to that usually drives the price up exponentially.

TheCompWiz
  • 7,409
  • 17
  • 23
  • I would also suggest thinking deeply about what purpose the redundancy is serving and whether it is increasing your reliability or introducing more ways to have a critical failure. Or whether the redundancy is totally defeated by other single-points-of-failure. – mtinberg Aug 31 '11 at 21:45