NetApp has no plans to join the fray in the all-flash array market. That’s according to NetApp CTO Jay Kidd, who spoke to Computer Weekly at NetApp’s Insight partner event in Dublin this week.
Jay Kidd says NetApp does not think the all-flash array market is big enough to get into and there are better ways of providing the boosted IOPS that flash solid state storage can provide to high performance applications, especially by providing flash acceleration at the server.
The NetApp view runs counter to storage industry trends. In recent times, market leader EMC acquired all-flash array vendor XtremIO; Hitachi Data Systems (HDS) added an all-flash component to its VSP storage arrays; and a host of startups have started offering very fast (circa 1 million IOPS) all-flash arrays such as Violin Memory and Pure Storage.
Jay Kidd’s argument boils down to two main points: that the all-flash market is currently too small for NetApp to be urgently interested in it; and that the array is not the right place for flash to reside anyway.
Kidd said: “We see the all-flash array market as an immature market, a start-up market. All-flash array vendors probably have less than 1% of array revenues currently. And even if an organisation uses, say, a Violin array, the rest of it will be held on array like ours, so it’s not a market we want to compete in right now.”
Kidd added, however, that NetApp, finds the all-flash market interesting.
But is the company in a position to develop an all-flash product if it decided it needed to?
All-flash array products cannot use controller/storage operating systems (OS) that were developed for spinning disk, because flash operates at much higher speeds and requires management of read, write and erase operations that differ radically from those of conventional hard drives.
Kidd agreed such capabilities do not exist in its NetApp’s Data ONTAP OS, but said NetApp possesses the expertise to develop them. He would not be drawn, however, on whether or when such development would bear fruit.
Instead of the flash array, NetApp puts forward a range of approaches to flash: pools of flash as cache in the array and controller plus flash on PCIe cards in the server host. It was the latter of these that Kidd was most keen to emphasise for high-performance operations.
“The enterprise hard drive was built on the assumption that hard drive access times would be ten times the latency of the network. Now we’re looking at flash storage with latency 100 times less than the network. So, the most logical place to put flash is in the host, and we’re seeing things like SAP’s Hana in-memory databases being loaded into host flash as a working set.”
Here NetApp – with its Flash Accel server flash – is mirroring a storage industry-wide trend pioneered by start-ups such as Fusion-io, but also adopted by giants such as EMC with its VFCache.
The main drawback of server-side flash is that hardware cards reside as potentially siloed chunks of data in individual servers that, if they suffer an outage, could take out an entire unprotected set of data.
Kidd said that by treating these as cache you potentially lessen the impact of such an outage but that there are still questions that need to be answered regarding data protection of scattered flash cache instances.
“You need centralised support, feeding and managing the host flash. Flash Accel is our first step but we still have to figure it all out,” Kidd said.