4 ms·

I don't think these kinds of facilities swap hard drives. If one fails, you have others. By the time a lot of them fail, the whole container is probably obsolet

by spolsky 18y ago

I don't think these kinds of facilities swap hard drives. If one fails, you have others. By the time a lot of them fail, the whole container is probably obsolete anyway... you swap it out for a new one shipped from wherever they're manufactured. Why would you get increased failure rate without the roof? Shipping containers are designed to be weatherproof. One of the the biggest advantages with the containers is that you can manufacture entire containers of servers wherever it's cheapest and ship them anywhere in the world for about $3000. Think how much you save on the labor of racking up servers if you can do it in China. The server never has to see the inside of a cardboard box, either.

Retric 18y ago
You could be right, but I don't think you can completely separate your CPU, Motherboard, and RAM. If they are each 50$ and one fails then replacing a 50$ component will rescue two 50$ components which is a net gain if it cost's less than 100$ in manpower to do the swap. Without temperature control the failure rate will increase some. I guess they could work on a 3 month cycle and move the container to a nearby repair location. Thinking about it there is probably some interesting trade offs depending on the local labor supply etc. But giving up the ability to repair PC's is a significant cost. In the US 400,000 square feet of warehouse is around $19/square foot and you could avoid weatherproof all those containers and electric cables. You could also use larger more efficient fan's to move all that hot air. And you could use the space for other reasons if you don't want to use that land just for a data center. Unlike this custom building which is going to be hard to use for other things. Why would you get increased failure rate without the roof? Bugs, animals, and weather, combined with a defective container. Then again if they only need to last 3 years the failure rate is probably going to be low. PS: I think of a rack as 100k+ worth of equipment that lasts for 3 years that has a vary small footprint and uses a lot of energy. But, when you start building data centers like MS or Google I guess the buildings start becoming expensive on their own.
- bemmu 18y ago
  Wouldn't bugs and weather also be a problem if you had a defective building?
  - Retric 18y ago
    I think this is a defense in depth issue, with large walls to cut down on wind you can use really simple roof over what you want to protect and you end up with a strong defense vs rain, wind, and snow. When the outside is 1 foot from your equipment a 1 in 1000 fault becomes a problem, but you can use two 1 in 50 solutions to end up with better protection if the failures don't cascade. (1/50 * 1/50 = 1/2500.) PS: A friend built some really cheep rooms using plastic and plywood inside an old barn that had large gaps in the walls but an OK roof to protect his record collection on the cheep.
lacker 18y ago
I do think these kinds of facilities swap hard drives. It's much more convenient for the developer when you can count on every functional machine having the same setup. You don't want to just give every machine more hard drives than you need because that's a lot of added cost.

by spolsky 18y ago

Retric 18y ago

You could be right, but I don't think you can completely separate your CPU, Motherboard, and RAM. If they are each 50$ and one fails then replacing a 50$ component will rescue two 50$ components which is a net gain if it cost's less than 100$ in manpower to do the swap. Without temperature control the failure rate will increase some. I guess they could work on a 3 month cycle and move the container to a nearby repair location. Thinking about it there is probably some interesting trade offs depending on the local labor supply etc. But giving up the ability to repair PC's is a significant cost. In the US 400,000 square feet of warehouse is around $19/square foot and you could avoid weatherproof all those containers and electric cables. You could also use larger more efficient fan's to move all that hot air. And you could use the space for other reasons if you don't want to use that land just for a data center. Unlike this custom building which is going to be hard to use for other things. Why would you get increased failure rate without the roof? Bugs, animals, and weather, combined with a defective container. Then again if they only need to last 3 years the failure rate is probably going to be low. PS: I think of a rack as 100k+ worth of equipment that lasts for 3 years that has a vary small footprint and uses a lot of energy. But, when you start building data centers like MS or Google I guess the buildings start becoming expensive on their own.

bemmu 18y ago
Wouldn't bugs and weather also be a problem if you had a defective building?
- Retric 18y ago
  I think this is a defense in depth issue, with large walls to cut down on wind you can use really simple roof over what you want to protect and you end up with a strong defense vs rain, wind, and snow. When the outside is 1 foot from your equipment a 1 in 1000 fault becomes a problem, but you can use two 1 in 50 solutions to end up with better protection if the failures don't cascade. (1/50 * 1/50 = 1/2500.) PS: A friend built some really cheep rooms using plastic and plywood inside an old barn that had large gaps in the walls but an OK roof to protect his record collection on the cheep.

lacker 18y ago

I do think these kinds of facilities swap hard drives. It's much more convenient for the developer when you can count on every functional machine having the same setup. You don't want to just give every machine more hard drives than you need because that's a lot of added cost.