IDC’s Worldwide Hard Disk Drive Forecast for 2015 to 2019, published in May 2015, predicts that “Slow HDD areal density (capacity per disk) growth means that a steadily increasing number of components per drive will be needed on average to reach higher capacity points, particularly for the enterprise segment. This dynamic will push the overall blended average HDD ASP higher each year over the forecast period….”
IDC’s prediction that a steadily increasing number of components will be required to reach higher capacities is borne out by the new 6 TB capacity disk drives from Seagate and HGST. HDDs, up to the 4 TB HDDs, all contained 5 disk platters, and were essentially the same no matter who made them. Now we are seeing major differences between the 6 TB HDDs from Seagate and HGST/Western Digital. HGST seal their Head Disk Assembly (HDA) in Helium and insert 7 disks to achieve 6 TB. Seagate has a higher bit density and is able to achieve 6 TB by adding a separator plate to improve airflow and reduce head flutter and vibration to enable them to insert 6 platters. Attached is a table that I found on the web that compares the HGST and Seagate 6 TB disks. Notice that there are differences in performance, MBTF, weight, and power.
Seagate has recently announced an enterprise 8 TB HDD and HGST announced an enterprise 10 TB disk. They achieved these densities by adding shingled magnetic recording (SMR) to their 6 TB HDDs. Writing a magnetic track takes a lot of energy, so the width of the track is wide. However, reading the track takes less energy, so the full width of the track is not required when reading. Shingled magnetic recording (SMR) takes advantage of this to increase track densities by over writing a previously written track, leaving only enough of the width to read. As you can imagine this is not suited for update write or random write activity, since all of the tracks that were over written must be read into a buffer before the target track can be updated, and then all the overlapping track must be written back. To limit the number of tracks that must be read to update previous tracks, a number of tracks are grouped together separated by a guard band. As you can imagine update writes can be very slow as a read/modify/write must be done. Seagate and HGST are positioning SMR HDDs as archive HDDs with Enterprise availability. So while capacities can still be increased, it will be at the sacrifice of performance. I am not sure what this does to RAID rebuild times. With 8 and 10 TB HDDs, RAID 1 or erasure coding may be the only practical approaches to recovering from HDD failures.
Another point that is made in the IDC report is that the mix of enterprise HDDs versus commodity HDDs has been shifting with the enterprise share growing faster. I believe this is more about flash replacing commodity HDDs in commodity markets like PCs. This has a further impact on HDD ASP since the volumes in the commodity market that used to help reduce the overall costs of development and manufacturing of HDDs is dwindling.
In my last post I suggested that, based on the rising HDD ASP, it was time to separate the acquisition of media like HDDs, from the acquisition of storage controllers. This would allow us to lower costs by extending the capitalization period for HDDs from 3 to 5 years, to 5 to 7 or even 10 years, while we refresh the controllers more frequently to remain current with the latest storage enhancements. We do this with our G Series block storage controllers through storage virtualization, our high performance HNAS heads, and our Hitachi Content Platform for object storage.
In subsequent posts we will discuss other ways to address the rising cost of HDD’s.