We were Hybrid before Hybrid was Hip

The storage array Industry’s product landscape has, generally, three backend media configurations:

  • All-Flash – SSD only
  • Hybrid – Combination of SSD and HDD
  • All-HDD – HDD only


Each serve a given performance or capacity-cost requirement.  On the extreme ends of the bell curve, there are the workloads and business requirements that gravitate towards SSD or HDD only configurations. Highly random, latency sensitive workloads benefit from SSD while sequential workloads or that requiring large capacity benefit from HDD. Datacenters usually have the need to serve both workloads. In such a scenario, covering the majority, a Hybrid solution prevails. 

Over 20 years ago, when the NetApp founders Dave Hitz and James Lau wrote the original WAFL (write anywhere file layout), it was optimized and integrated for RAID4.  The benefit of RAID4 was the ability to add capacity, on the fly, to a Volume without having to recalculate RAID.  Though RAID4 uses a single disk for parity opposed to spreading parity among each of the data disks, like RAID5, this causes the parity disk to run hot and become a performance bottleneck. 

To negate this performance bottleneck, while embracing the flexibility of growing volumes on the fly and getting many other auxiliary benefits, Dave and James built WAFL to have journaled block write integrated with the raid controller. This facilitated incoming writes to go directly to DRAM and logged to NVRAM (a fast microsecond process) then acknowledged, back to the source. DataONTAP coalesced the writes in memory and flushes to disk in large sequential stripes across all of the available storage devices maximizing the performance capabilities of the devices, and limiting disk contention.  Removing mechanical disks (and RAID4) from the Acknowledged write path dramatically reduced write latencies essentially enabling writes to solid state memory.  This is, by my definition, is a Hybrid storage array.

Additional benefits realized over 20 years ago by this initial Data ONTAP release:

  • Optimized RAID striping with data being laid out evenly across all disks. This created an almost linear sum of I/O performance per disk spindle.
  • Random writes were written sequentially.
  • Writes took place with minimal disk head movement.
  • Block changes were calculated while the acknowledged data was in NVRAM negating the need for the disk heads to seek to the original block location on disk. Talk about Virtualized block system!
  • Because all writes were being acknowledged when logged to NVRAM, this freed up the disk heads to service reads.
  • In the instance of an unclean shutdown ONTAP was ready to immediately serve data without having to do a whole system block check. This is because the acknowledged writes in the NVRAM get immediately committed to disk concluding the CP (Consistency Point) of any uncompleted writes.

I can’t help to marvel over how the original WAFL journaled write mechanism of 20 years ago has proven to be such a pillar and core DNA, helping enable many of NetApp’s innovative technologies and solutions today.  Technologies such as RAID-DP which protects against double disk failure but matches the same space efficiency as our competitors arrays’ yet they have inferior single disk protection.  Rapid and space efficient data copy function found in FlexClone is also born out of this same technology. The list goes on.

It would be difficult to predict 20 years ago the vast capabilities of what Clustered Data ONTAP has today.  These capabilities born out of that same DNA is testimony to the core technical focus which makes NetApp’s products and solutions so great.  It’s also what enables our independence as a company; to have a business model which is all embracing of both VAR, SP and SI partners as well as attaching to and adding value to alliance partners and even potential competitive offerings.  NetApp’s business model is unique and will drive the company’s success in the future; we have our core technology to thank for that and our Enterprise customers would agree.