Improving Storage Densities When Areal Density Slows Down
by Hu Yoshida on Jun 6, 2012
A recent paper in the IEEE Transactions on Magnetics Vol 48 May 2012 shows that the roadmap for areal density increases in magnetic hard disk drives (HDD) has slowed down to 20% per year.
Areal density is measured in bits per square inch and has the most impact in driving down the costs of storage capacity, power, cooling and floor space. This is related to the capacity of HDDs. However, an HDD’s capacity can vary depending on other factors like sector size, track size, disk size and the number of disks within an HDD. A 3.5 inch HDD can hold more capacity than a 2.5 inch HDD, but since the 2.5 inch HDD has less mass to spin and less distance to traverse to seek and search, it has better performance characteristics and requires less power. The smaller size also enables us to package 24 of these in a 2U drawer versus 12 of the 3.5-inch HDDs in our new Hitachi Unified Storage (HUS) system, providing more access arms per drawer.
The IEEE areal density chart shows areal densities for HDDs crossing the 1TB per square inch barrier this year. However, this goes beyond the capabilities of current PMR (Perpendicular Magnetic Recording) technology. Seagate announced this breakthrough in March of this year with the introduction of HAMR (Heat Assisted Magnetic Recording) technology, where a laser is used in the read/write head to enable the use of harder recording materials for greater densities. With this technology Seagate announced that they would be able to double the capacity of 3.5 inch HDDs from 3TB to 6TB by the end of the decade. I mentioned this in a previous post: Seagate’s HAMR Reaches the 1TB Per-Square-Inch Barrier and HGST Announces 4TB Disk
In that same post I reported on a 4TB HDD that was announced by HGST (now Western Digital). HGST accomplished this with the current perpendicular recording technology by adding another platter and increasing the number of disks in a disk module to 5 disks, which adds two more recording surfaces with additional recording heads and arms.
One other change that I failed to mention is that this disk has increased the sector size from 512 bytes to 4K bytes, which will be the new standard sector size for all disk vendors. Each 512 byte sector has an additional 80 bytes for ECC and sync bytes. By consolidating 8 of these sectors into a 4K sector, these 8 x 80 bytes can be consolidated into one with a greater number of ECC bytes, which would not only increase capacity by 7 to 11% but also increase error correction by 50% according to Western Digital.
As you can imagine, increasing the sector size has implications for operating systems and servers as well as storage, in terms of sector alignment. This will require a transition period where these disks will support 512 byte emulation mode for compatibility to legacy operating systems, servers and storage systems. More information will become available as these new drives begin to roll out. You can expect to see other storage vendors roll out this new sector size not only as a way to increase capacities but also to improve ECC with longer correction codes.
Another way to increase densities with the current PMR technology is to use Shingled Magnetic Recording. This technology is based on the fact that it takes more energy to write a magnetic bit than it takes to read it. As a result a track is very wide when it is written, but it does not require the full width of the track to be able to read it. Therefore once a track is written, you can write the next track over part of the old track as long as you leave enough of the old track to be seen by the read head. In this way you “shingle” the writes like you would shingle the tiles on a roof. This enables you to pack more tracks on the surface of a disk. However the tracks that are partially overwritten are write-once-read-many and recording on this would be like recording sequentially on tape but with the ability to randomly read the recorded data. The last track in a shingled group of tracks would not be overwritten, so that track could still be randomly updated. This type of recording would require more data management, but could be viable if the data is not frequently updated.
The IEEE chart also shows that areal densities for NAND devices are on a similar slope as HDDs. However, NAND has other possibilities such as SLC (Single Level Cell), MLC (Multi Level Cell), and TLC (Three Level Cell). The challenge with NAND is the ability to manage durability as the durability decreases with the increase in number of cell levels. While areal densities for tape are much lower, they are continuing on a steeper slope. There may still be a place for tape in the future.
So even though the increase in areal density for magnetic disks has declined to 20%, there are other ways to increase HDD densities through new technologies, repackaging and new form factors. But all of these can add new cost considerations, such as retooling for manufacturing of HAMR or additional data management and systems costs. The simple days of 30% price erosion, year after year, for the past 50 years are over. We need to put more focus on operational savings through virtualization, thin provisioning, dynamic tiering and active archive if we want to contain our storage costs. With dynamic tiering, NAND and even tape may play a bigger part in this equation, but I believe the management of HDD storage will still be the major cost consideration for storage for the next 4 to 5 years.
Comments (3 )
Great Explaination .
thanks a ton
An enlightening point of view.
Most storage users (at least most that I know) have a “there will always be more” mindset and use medieval-like competition techniques between vendors to lower prices year after year after year. It has been easy to have the vendors make-up for their untamed growth in storage needs.
Seems now like that strategy needs a second look.
Very good blog. Thanks a lot.
[...] years now, storage demand has been growing faster than hard drive density. Meeting our voracious appetite for data storage [...]