Floor space, power, and cooling – Setting records
by Hu Yoshida on Jan 24, 2011
Gartner published a report last November on storage predictions for 2011. In that report, they said that the need to reduce power and capacity costs will become critical. They made the statement that power, cooling, and floor space will be a limiting factor for growth.
These past few weeks I have been travelling in Asia, and see this as a real threat to high growth countries like China and India. The International Energy Outlook report that was recently published in 2010 showed that while worldwide energy consumption will grow by 4.9% per year, energy consumption in Asia will grow by 118%.
This chart shows that the share of energy consumption is quickly shifting to Asian countries like China and India.
While the focus is on these larger countries, consumption in countries like Vietnam with a population of over 80 million also is growing rapidly. Power consumption was top of mind for many of the customers that I visited in Hanoi last week. Last November, the Prime Ministers of Japan and Vietnam signed a strategic agreement to develop nuclear power generators to meet this increasing demand. However, the first nuclear power plants will not be available until 2020, which means that they must maximize the power that they have in the meantime.
Data centers require a lot of power and are key growth engines for emerging economies, so reducing power and maximizing the use of power through tools like virtualization is of high interest to data center managers. Two key tools for reducing power consumption for storage are Small Form Factor 2.5 inch disk drives and dynamic tiering or sub LUN level tiering.
The 2.5 inch disk drive is a no brainer since it consumes about half the power of traditional 3.5 inch disk drives. When combined with point to point SAS connectivity, it can be packaged in dense modules, which can further reduce footprint. Instead of the traditional 19 inch rack storage drawer with 15 drives across the front, the Hitachi AMS 2000 drawer can contain 48 drives, while the VSP storage system’s drive module can contain up to 128 2.5 inch drives.
Dynamic page level tiering or sub LUN level tiering enables consolidation of storage onto a fewer number of drives without sacrificing performance. For instance, a 36 TB storage requirement can be configured with 120 x 300GB drives or a combination of 4 x 200 GB SSD drives, 10 x 300 GB disk drives, and 32 x 1 TB SATA disks. This reduces the disk count from 120 disks to 46 drives. Since the hot pages are in the SSD tier, we can expect to see an improvement in throughput and performance over a single tier of high performance disk storage.
If we assume 10 watts per drive, we reduce the power consumption by 74 drives or 740 watts. We also have fewer racks and fans for additional power savings. The importance of sub LUN level tiering versus LUN level tiering is that LUN level tiering would require enough space in each tier to hold the entire 36 TB LUN. If we had to move this LUN to the SSD tier to get high performance we would need 36 TB of SSD for the whole LUN capacity. LUN level tiering actually requires more spindles than single level tiering while sub LUN level tiering requires less.
This reduction in number of spindles and rack space is extremely important to data centers today. Last week, I met the CIO of a major stock exchange in Asia who is in the process of moving to a new site and consolidating his data centers due to the high cost of data center space. He was very interested in the reduction of floor space that can be achieved by the use of 2.5 inch drives and page level tiering.
If you could reduce your footprint for storage from 11 frames that stretch across 11 x 76.7 cm or 8.4 m with 6 frames that only spans, 6 x 61 cm or 3.7 m, and also reduce power consumption by 40% with the use of 2.5 inch drives, wouldn’t you be interested too?
Reducing foot print and power consumption is a better way to set records than jumping over a long span of storage arrays. For a fun perspective on this, see Martin Glassborow’s post on Storagebod.
Comments (3 )
I agree with your valuable points. But in Vmax, we use SATA drives which can be powered off if not in use (Spin Down). It can accomodate more flash drives which consumes less power than native drives. There by it helps reducing the power consumption.
Hitachi takes a very conservative approach to spin down of disk drives. When a drive is not spinning we cannot monitor it and do not know the state of the drive until we power it up. We support spin down on the AMS storage system, but it is intended for use with temporary data like backup. For instance, we can spin up a drive and check its health before we copy data to it for backup. Once the backup is completed from the copy, the drive can be powered down again until the next copy/backup cycle. We do not recommend it for production data storage.
As far as SSD drives are concerned, most venders support 256 drives. Additional drives are supported by RPQ. While SSDs consume less power, they consume more bandwidth. This can be especially disruptive if the SSD drives share a FC loop with non SSD drives. Since the VSD uses point to point SAS connections, this contention is avoided.
Dear Maestro Hu,
Can we say that VSP does supports disk spin down but depending on the types of drives?
For SATA,SAS and SSD’s we could do it manually (depending on OS?) but it’s not a common features on a VSP as it were on the AMS2000 line ups.
Kindly correct my understanding.