This new release represents EMC's first break from the Clariion FLARE codebase that it acquired with Data General back in 1999. (Data General itself introduced the first Clariion way back in 1992.) There are a couple of critical problems with FLARE. The first is that the core RAID and data placement functions were originally written for single-core processors and weren't multithreaded.
As multicore processors made their way into Clariion and VNX product lines, EMC assigned functions like cache management and replication to separate cores, but the limiting factor in system performance was always the core that ran the RAID.
The other problem with FLARE was its old-school approach to RAID and data layout, which affected SSD performance. I've said many times that efficient use of flash takes a different data layout. Flash and disk drives are so different that your data placement algorithms need to have significantly different goals.
A disk system does everything it can to minimize random I/O, because every time you move the disk heads, you add 5ms or more latency. Flash handles random I/O as well as sequential I/O, but writing to the flash adds latency and also wears out the flash. So rather than avoid random I/O, a flash-optimized array minimizes writes to the flash.
[Storage startup Skyera released the highest-density SSD appliance on the market. Get details in "Skyera Soars With 250TB SSD skyEagle Array."]
The new software bundle runs on six new VNX hardware models (VNX5200, VNX5400, VNX5600, VNX5800, VNX7600, VNX8000 and VNX-F). They use the latest Xeon processors ranging from a total of four cores per controller in the low-end VNX5200, to dual eight-core processors in each of the high-end 8000's controllers, which EMC calls SPs, for storage processors.
EMC also used PCIe switching technology to expand the I/O bandwidth to up to 160 PCIe 3.0 lanes with up to 11 PCIe slots per SP. This provides plenty of connectivity for Fibre Channel, 10-Gbps Ethernet and, of course, SAS ports to the back end, where you will eventually be able to hang 1500 drives of the VNX 8000's controllers. Up to 22 lanes of PCIe are used for the inter-SP communications channel to support active-active operations and maintain cache coherency.
The basic hardware design hasn't changed; VNX is still a dual-controller block storage system. With the software redesign, calling it a Clariion is a stretch, with up to eight file system X-blades doing the SMB/NFS heavy lifting as a front end. Some sticklers, especially those from NetApp, will argue about whether a common management platform makes that truly unified, but as long as I don't have to manage block and file separately, or pay a lot extra, it's unified enough for me.
Faster hardware and cleaner software means better performance. EMC SVP Eric Herzog tells me customers can move down the product line a model or two and still get better performance than their old VNXs can deliver. The new VNX7600-F all-flash version can deliver 500K IOPS and up to 400 Tbytes of capacity, which puts it well into the middle of the all-flash-array market. This is a step up from the previous all-flash VNX, which struggled to deliver the performance its SSDs promised.
I have in the past expressed some skepticism that a venerable platform like VNX could be updated to use flash as efficiently as a system designed for flash from the outset. All-flash startups still have an advantage, but EMC has given VNX a new lease on life by rewriting the core of its software.
Some competitors have pointed out that this big a rewrite means many thousands of lines of code that haven't been acid tested in the real world; that's true, but I'm confident EMC's testing found most of the big crunchy bugs. In any case, the new VNXes are a big step forward for EMC.
[Get insights into the pros and cons of flash storage at Interop New York with Howard Marks' session "SSDs In The Data Center."]