When Did Hard Drives Get Workload Rate Limits? (Updated)

When Did Hard Drives Get Workload Rate Limits? (Updated)

Chris EvansData Practice: Data Storage, Storage Media

A recent article on The Register reminded me that we’ve started to see HDD vendors introducing warranty limits on their hard drives for the amount of data that can be transferred to/from them in any one year.  Solid state drives have always had this restriction, because we know that flash is damaged by being written to, but does the same now apply to HDDs?

Note: This post was originally published in April 2016. It has been updated to reflect the changing endurance levels of the latest 30+TB drives.

As a starting point, I had a quick look at the HDDs from the usual suspects.  Here’s what I found:

  • Seagate Enterprise Capacity 3.5″ drive – 10TB (spec) – 550TB/year, or 0.15 DWPD
  • Seagate Archive 3.5″ drive (5TB – 8TB) (spec) – 180TB/year or 0.06 – 0.1 DWPD
  • Seagate Enterprise NAS drive (2TB – 8TB) (spec) – 300TB/year or 0.1 – 0.4 DWPD
  • Kinetic HDD – 4TB (spec) – 180TB/year or 0.12 DWPD
  • WD Gold (4TB – 8TB) – 550TB/year or 0.2 – 0.4 DWPD
  • WD Re (250GB  – 1TB) (spec) – 550TB/year or 1.5 – 6 DWPD
  • WD Re (1TB – 6TB) (spec) – 550TB/year or 0.25 – 1.5 DWPD

The data above was from 2016. Today, the latest drives have similar profiles:

  • Seagate Exos 24 – 24TB (product manual) – 550TB/year or 0.06 DWPD
  • Western Digital Gold – 24TB (spec) – 550TB or 0.06 DWPD

It’s notable that the workload capacities (the TB written or read per year) aren’t easy to find in the specification.  There’s typically no row in the tables; some vendors hide the data in the footnotes, others have it only on their website in the product descriptions.  In addition, WD Gold and some of the WD Re devices are rated at 550TB/year, but their MTBF and AFR figures are calculated based on transferring only 219TB/year (Gold) or 240/300TB/year (Re) – half to two-thirds the rated capacity.  Does this mean in reality if you pushed these drives to their warrantied limits, then they would be much less reliable?

What is a Warranty?

Now at the beginning of this post I did say “warranty limits”, however there’s no direct information in the spec sheets to say drives are warrantied for data written/read.  In fact, terms like “designed for” are used more often, so where do we stand with warranty?

I need to check some of the detailed product sheets, however it would be interesting to find out whether these drives now report back the amount of data transferred, as flash drives do.

So how do these figures compare to SSDs?  Here’s some data I found:

  • Intel SSD DC S3710 (200GB – 1.2TB) – 10 DWPD (spec)
  • Intel SSD DCP3500 (400GB – 2TB) – 1095 TBW (assuming 5 year lifetime, 0.3 – 1.5 DWPD) (spec)
  • SanDisk Lightning (200GB to 1.6TB) – between 3-25 DWPD depending on model (specs)
  • SanDisk CloudSpeed Ultra Gen II (400GB – 1.6TB) – 1.8 DWPD (spec)
  • SanDisk CloudSpeed Eco Gen II (480GB – 1.92TB) – 0.6 DWPD (spec)

The data above was collected in 2016. Here are the data for some more recent drives:

  • Kioxia FL6 Series – up to 3.2TB capacity (high write focused) – 60 DWPD (spec)
  • Kioxia CM7-V Series – up to 12.8TB capacity (mixed workload) – 3 DWPD (spec)
  • Kioxia CM7-R Series – up to 31TB capacity (read intensive) – 1 DWPD (spec)
  • Samsung QVO SATA 2.5″ 8TB – 0.62 DWPD (spec)
  • Samsung P893 up to 8TB – 1.0 DWPD (spec)
  • Samsung PM897 up to 4TB – 3.0 DWPD (spec)

Depending on the models you use as a reference, SSDs could be seen as more reliable than HDDs, however we need to look a little closer.  All of the HDDs I’ve referenced are the capacity models.  I could find nothing indicating that top end 15K drives were in any way rated for endurance (update, these drives are no longer marketed).  Obviously we see “duty cycle” referenced a lot, but these drives are expected to have duty cycles of 24×7, whereas SATA (especially consumer) drives are typically rated at 8×5.

The Architect’s View®

HDD vendors are pushing limits on some of their technology, such as drives built around SMR.  Instead of writing individual sectors/blocks, SMR drives have to re-write large amounts of data for each track written due to the overlapping track design.  This occurs because the write head size in the drives are currently at limits of miniaturisation; track sizes can be reduced but the write head size can’t and SMR is the workaround to fix this (note read heads aren’t affected because they can be made smaller).  As a result, drives are reading and rewriting much more data than is being physically stored by the host (what is generally called write amplification).  This must be being seen as a potential endurance issue and drives are being warrantied with limited DPWD figures as a result. This is our working theory, without additional evidence, as non-SMR drives now also seem to have workload limits.

Unless we some some magical breakthrough, it looks like increasing HDD capacity is hitting problems.  I do wonder with the release of 16TB flash drives like the Samsung PM1633a, whether there’s a appetite to invest millions of dollars into trying to extend HDD capacities further – especially considering the time it would take to RAID rebuild an entire 10TB drive from scratch (update: drives have increased well past the 10TB mark).

What do you think?  Is 10TB now the limit for HDDs?  Do we need bigger drives when large capacity flash is just around the corner?

Hard drive capacity has just reached 30TB with the announcement of Seagate Mozaic 3+ drives. These translate to a DWPD of just 0.06. This is way under the capabilities of even the read-intensive QLC media we see on the market. Solidigm has already demonstrated penta-level cell (PLC) technology, which will offer additional capacity benefits over QLC of around 25%. Endurance can expect to drop an order of magnitude compared to QLC, so PLC devices could be rated at 0.1 DWPD or lower.

However, this level of endurance is on a par with high capacity HDDs, where the endurance is for both read and write I/O. SSDs only suffer endurance issues with write activity. It is possible that PLC NAND could directly challenge high-capacity drives, with a narrowing of the price differential between the two.

Could the imposition of workload limits on HDDs be the restriction that finally kills off the hard drive?

Further Reading

Copyright (c) 2009-2024 – Post #80F9 – Brookend Limited, first published on https://www.architecting.it/blog, do not reproduce without permission.