topic Re: is energy a real problem? in General

is energy a real problem?

Dong Li — Tue, 17 May 2005 21:19:50 GMT

Dear all,

Recently there are lots of people talk about energy saving (especially for disks) in large data center and doing trade-off between system performance and energy saving.

I am wondering whether enrgy consumption is really a big problem in data center and other large storage systems, or people just need a topic to write papers.

If it is a problem, what are the common requirements? for example "maxize the energy saving but the performance degradation should be less than 5%"

Thanks a lot and have a good night

Dong

Re: is energy a real problem?

Uwe Zessin — Wed, 18 May 2005 00:29:40 GMT

The problem is that the electrical energy that is fed into the data center costs money and a lot of electrical energy is converted into heat which needs to be removed from the data center again. For that you need additional equipment which, again costs money, so its a good idea to try and save the electrical energy in the first place.

I have heard about attempts to make servers less energy hungry. For example, by turning off parts or a whole CPU when it is not needed, but I have never heard about any such attempts in the storage space. You can try to build a disk drive with less mass, so you can feed it with less energy to keep the moving parts rotating, for example, but I have never heard about strategies to degrade performance to save energy. Do you have some pointers?

We have discussed this in the past and I like to repeat that I don't think it is a good idea to spin down a high performance disk drive multiple times.

Re: is energy a real problem?

Dong Li — Wed, 18 May 2005 20:59:33 GMT

Hi Uwe,

Thanks so much for your reply.
I do remember our discussion about saving disk energy by spinning down disks in July,2004 on this forum.

On single disk environments,like PC or laptop, a common disk energy saving policy is to spin down disk to standby state after a no-request period. The disk will be spun up when a request arriving.

If we can generate such a long idle period for some(maybe not all)server disks with affordable performance degradation, do you think it's an acceptable idea to save disk energy for storage systems?

Dong