Comments on: Deleting Your Hard Drives – Entering a Green Data Center Future of SSDs

By: johnsavageau

johnsavageau — Thu, 22 Oct 2009 02:32:14 +0000

Sam – in the old days in locations such as Los Angeles, the going rate for power was at the breakered amp, and we charged (I come from the data center community) $10/amp. At 120v, at 20 amps (even considering the NEC limitation) that would run $200/month. 208 or 208Y would have an additional charge.

20*120= 2400 wats, or 2.4kwH. The going rate adds a premium to accomodate the cost of air-conditioning, UPS, and other associated mechanical/electrical equipment management and support for the individual breaker.

Now most data centers have moved from breakered amps to usage-based billing. There is a bit of premium added due to the loss of static revenue generated from the breakered amp.

Many users would get a 20 amp breaker, pay for the entire breaker, and only use a small fraction of the breaker.

For an industrial user in LA you will get around $.09/kw as a raw price.

Every city is of course different, with Washington (state) and Virginia being amoung the least expensive locations for power.

Hawaii ($.29/kwH), New York, Washington DC, and Chicago tend to run much higher.

So in short, you are not only paying fort the power used, but the entire mechanical/electrical infrastructure asupporting the power.

Hope that helps

John

By: Sam Leach

Sam Leach — Wed, 21 Oct 2009 17:55:22 +0000

“Data centers can charge anywhere from $120~$225/kw”. These number sound very wrong. The average cost for per kilowatt-hour of electricity in the US is $0.10. That would be a 1700x markup. Throw in cooling, etc. and MAYBE you get 10x the average cost … “$120~$225/kw” needs clarification.

Also, note a disc drive uses about 10-15 watts per hour (that is watts not kilowatts). 12.5 watts/drive is only about 9 kwh/month/drive or ~$1/month/drive.

A typical high end server uses about 15x that for the processor, memory, etc. So the storage is a small part of the overall carbon foot print and electrical costs.

By: johnsavageau

johnsavageau — Wed, 21 Oct 2009 04:48:01 +0000

The only numbers I had to go on were the numbers in their press release. However my calculations on the power, the cost they would normally pay for disk systems, the cost savings in data center cabinet space required to house the smaller form factor – all still looks pretty attractive to me.

Here is a link to the original article that piqued my interest:
http://tech.yahoo.com/blogs/null/153442

Most comments I have received still mention the cost as a limiting item when considering SSD implementations. I personally would have considered quality and potentailly failure rates as a bigger concern – but that just goes to show it is good to listen!

Thanks for the comment

By: vburke

vburke — Wed, 21 Oct 2009 01:49:43 +0000

The numbers I’ve seen for the MySpace 320GB PCIe SSDs is $7K each, 2 in each server for a total of $14K per server. If they’re gaining $20K per month, they’re going to have a hard time paying that back before the servers are ready for refresh.

Vern

By: johnsavageau

johnsavageau — Tue, 20 Oct 2009 21:40:16 +0000

I think the pricing issue is still pretty real, except that if you are able to reduce the eletrical draw to the point indicated in the MySpace articles and vendor documentation it might pay for itself quicker than anticipated.

Also looking at how quickly SSDs are scaling, and the complete integration of SSD i/o into what used to be disk-based SANs and NAS, it is becoming obvious that spinning disks may have some competition.

I also have this sick vision of SSDs being scaled in not a linear form, but in a 4 dimensional form – assuming there is no inherent I/O limitation or reason why SSDs cannot interconnect at more than one dimension.

In my mid-50ish semi-fossilized mind there is also a thought limitation on being able to separate in thick black lines the difference between processors, system memory, SSDs, and other levels of interconnection.

When will we see Brocade come out with a card that fits into the layer 2 switch which makes fiber channel over Ethernet obsolete? Just plug a 10TB SSD card into the switch, and configure it as cloud storage capacity. Make that part of an Internet exchange point, and then CDNs will no longer need to support disk systems in data centers.

I wish I was 40 years younger so I could see what is going to happen in 20 years…..

By: Jameson

Jameson — Tue, 20 Oct 2009 19:26:35 +0000

I personally think we are at least 24 months, or more away from seeing SSD prevalance in the data center. We need to keep in mind that traditional spinning platters have seen a bump in performance (due to increased platter density and higher rotational speeds and more intelligent firmware) and the fact is that there is still margin on spinning disks, that can be shed (somewhat) to fight off the invading SSDs. The most compelling argument for traditional hdds is that capacity will be far, far more inexpensive per GB than SSDs.

I would anticpate (what IMHO is) a natural segmentation whereby SSDs will be used in high IO applications such as DB servers, etc. and HDDs will be used primarily in large storage arrays. The move to 32 and ultimately 22nm processes at the fabs will be interesting in packing the densities higher…..but I think the long term solution for SSDs will be to add more chips and not increase the density. Doing so will necessitate intelligent wear-leveling at the controller level as well as advanced stuff like TRIM.

Exciting tech for sure, but short of faster boot times and scorching DB performance….I don’t expect to see too many High Def video files of ‘the office’ sitting on SSDs…for a while anyway.