How Low Can We Go?
by Hu Yoshida on Jul 21, 2006
David Merril posted an excellent blog on perpendicular recording in which he raised the warning that Moore’s Law, in regards to the declining cost of magnetic disk capacity, may begin to slow if manufacturers have to retool to deliver new technologies or the rate of density increases start to stall. Fortunately, the manufacturing process for perpendicular recording is not a major change from current horizontal recording so little retooling will be required. However, future technologies like HAMR, Heat Assisted magnetic recording, where a laser is used to heat the recording surface may require retooling or packaging which will affect the price that we will pay for disk media.
The industry has been used to a yearly 30 to 35% decline in storage capacity prices for some time. What happens if that decline starts to level off at 20% to 25 % or less? That could have major ramifications in an increasingly storage hungry world.
David goes on to point out that disk prices are not the only cost component in storing data. In fact his following post architecture enumerates 30 different cost factors.
I like to point out that the separation of the control unit from the storage media, an approach which defines and differentiates HDS storage products, will help when this occurs. The disaggregation of storage systems into intelligent control units and commodity disk media will enable us to differentiate the costs and capitalization rates of these components.
Today most companies will capitalize their storage hardware over 3 or 5 years. Even though the disks are built to last much longer than 3 to 5 years, storage systems are replaced in this time frame because the control unit functionality will change dramatically and up to now, the yearly capacity increases in disk media, enabled a price erosion of 30% to 35% which justified the replacement of the disk in three to 5 years.
However, as the price erosion of disk capacity starts to slow down, the disks may need to be kept longer to off set this slow down. The reliability of magnetic disks will support this extended life. We have disk subsystems that are over 10 years old, still running in production. However if the control units are bundled with the storage disks, the storage system may still need to be replaced in order to take advantage of the new technology in the control unit.
Remember when open systems servers came with internal storage? The same justification was used to desegregate the storage from the server.
With the separation of Control unit from storage media, each could be purchased and capitalized at different rates. Perhaps the control unit continues to be capitalized over 3 years but the storage media is capitalized over 5 to 7 years. Of course there will have to be changes in accounting practices, and in warrenties for control units versus media.
As the capacities of these disks increase, it also would make sense to do this migration less frequently, since the time to move the data to the next generation will get longer and longer.
The question for future storage capacity costs may not be “how low can we go in price, but how long can we keep it?.
Traditionally we have replaced storage arrays after the “free” maintenance period expires because the cost to maintain the array is greater over a 3-5 year period than it would be to purchase a new array with 3 years maintenance included.
Your challenge in the future, should what you describe here come to fruition, will be to lower the maintenance costs for keeping the arrays longer. While we would understand why we’d have to pay more for the change in disk technology, we would expect to pay less on some other line item(s) in our ROI study to make up for the price difference.
The long and the short of it is, I will expect HDS (or any vendor I do business with) to make up for the increase in disk cost (or slowing of the declination) by maintaining the normal declination of pricing by adjusting the price of something else.