Non-volatile memory: It's nothing to SNIA at - NetApp
'Storage class memory' could soon outclass NAND Flash
A non-volatile memory technical workgroup has been set up by the Storage Networking Industry Association (SNIA) and NetApp's canned quote on the move sheds a bit more light on the Sunnyvale storage giant's flash thinking. From the wording, it would seem that the storage firm is preparing for its arrays to move fast data storage into external solid state cards and boxes.
The SNIA has set up a cross-member workgroup and forum whose task is to develop system standards and communicate the need for operating system changes to support non-volatile memory (NVM) hardware. Wayne Adams, SNIA chairman, said: “The classic architecture of operating systems, servers and storage will rapidly evolve with NVM to meet the continued computational needs using larger data sets and driving new uses of information for business and science.”
Adams singled out storage class memory – the use of flash as a virtualised adjunct to main memory – as a significant application performance enhancer and said that operating systems need to support that.
The SNIA release said: "The TWG (technical workgroup) will develop a programming model specification for in-kernel NVM interfaces supporting functionality as requested by TWG members. Additionally, the TWG will develop a specification defining the interaction between applications and the kernel for applications directly using NVM storage (such as database software)."
Most of the big storage vendors are involved: Dell, EMC, Fujitsu, HP, IBM, Intel, NetApp, Oracle, QLogic and Symantec. They all issued supporting quotes and NetApp's said:
It has become clear that solid-state storage is having a profound impact on computer and storage system architectures. The next wave of disruption will be the advent of ‘storage class memory’ – the successor to today’s NAND Flash, offering up to 10 times the performance and a fraction of the latency. Standards enabling applications to take direct advantage of this technology will be essential, and this proposed new SNIA Non-Volatile Memory Technology Working Group is a good first step, and NetApp intends to actively participate in this effort.
This came from NetApp's chief scientist, SVP Steve Kleiman. The company already has its Virtual Storage Tiering concept in which flash storage is used as a cache inside a storage pool to accelerate I/O for the most accessed data. What it aims to do now is to extend VST outside a NetApp FAS array and into server-located solid state caches and storage memory and also (we're told) into networked solid state storage arrays.
NetApp is telling its customers that the storage array future is for disk drive-based arrays to become focused on capacity, with big and fat SATA drives, retaining their existing data management features. VST will be used to feed solid state caches, storage memory and external flash arrays. These will be focused on performance and use deduplication, thin provisioning and cloning to minimise the footprint of their stored data.
We understand NetApp does not intend to produce its own PCIe flash card hardware for servers, instead relying on third parties to do that, whcih means it will need an open standard interface to simply move data to and from such caches. It seems likely it will take the same tack with storage memory.
However that approach would not be appropriate for solid state arrays which come with their own operating system and controller software. It is El Reg's belief that NetApp is thinking about producing its own networked solid state array: flash at first and then, down the road, phase-change memory or whatever other follow-on technology to NAND is chosen by the storage and server industry.
It is unlikely that a NetApp flash array will run WAFL, the Write Anywhere File Layout technology at the heart of NetApp's Data ONTAP operating system, as WAFL is focused on optimising I/O to and from arrays of disk drives.
Will NetApp buy in flash array technology? That would speed development and there must be a reasonable chance that NetApp will take that route so it can flash forward and not fumble around with its own ground-up development. ®
Sponsored: Hyper-scale data management