Storage continues to be a volatile segment of IT. Hot areas trending in the news this month include NVMe over Fibre Channel, which is being hyped heavily now that the Broadcom acquisition of Brocade is a done deal. Another hot segment is the hyper-converged space, complimented by activity in software-defined storage from several vendors.
Flash is now running ahead of enterprise hard drives in the market, contributing to foundry changeovers to 3D NAND to temporarily put upward pressure on SSD pricing. High-performance storage solutions built on COTS platforms have been announced, too, which will create more pressure to reduce appliance prices.
Let’s cover these topics and more in detail:
- NVMe over Fibre-Channel is in full hype mode right now. This solution is a major step away from traditional FC insofar as it no longer encapsulates the SCSI block-IO protocol. Instead, it uses a now-standard direct-memory access approach to reduce overhead and speed up performance significantly.
While FC is well-embedded in the traditional SAN, the new approach seems a Hail Mary play in a market that’s already comfortable with InfiniBand and Ethernet NVMe approaches.
An admin might be asked, “Why not use the same fabric for everything?” It’s not a small change to FC installations. At a minimum, the server adapter cards will need replacing, together with the storage appliances, since this isn’t just a minor software upgrade at the storage end.
It’s too early to figure performance and, anyway, benchmarks published tend to be self-serving and confusing, so I’m guessing it’ll be mid-2017 before anything credible is available.
- The hyper-converged infrastructure space is dominated today by just a few suppliers, Dell Tech, HPE and Cisco. There are two leading software stacks for creating a virtual SAN involved, and the system leaders sell one or both of these on their HCI offerings. With limited hardware options and very tight control of add-ins like drives, HCI prices are somewhat high considering the simple COTS servers underlying the code.
The game is, inevitably, changing. The #2 software stack vendor, Simplivity, is partnering with Huawei, bringing a giant low-priced Chinese player to the game. It’s a fair bet that we’ll see ODMs entering the fray in 2017, following the model of Huawei with Nutanix and Simplivity and also perhaps following SuperMicro, which offers IBM XIV and VMWare vSAN solution platforms.
In the end, there will be a ready supply of unbundled code and a simple enough integration that DIY hyper-converged systems and low-priced boxes will be common.
- SuperMicro announced a major NVMe over Ethernet initiative, which ties back to the two previous trends. They’ve tied NVMe in with Microsoft Storage Spaces on MS Server 16, which is a big play for the performance-oriented Windows user.
SuperMicro is doing well. They have enjoyed good revenue CAGR over the last few years and are now at $2.2B annually, making them a player. They identify their platform offerings as building blocks, taking advantage of COTS interchangeability to make building systems easy.
- Primary Data announced a new solution, Datasphere, which is aimed at building a virtual pool of storage from storage appliances such as RAID arrays, iSCSI units and also cloud storage spaces. The aim is to deliver an NFS 4.2 scale-out NAS. The code is “software-defined storage” running in virtual instances on servers and scalable to meet workload demand.
The Datasphere package does what EMC’s vIPR was aimed at, though the latter has gone on the back burner. The idea of a single unified pool allows a home for legacy storage, while generally simplifying management of storage for a cloud or virtual environment.
This type of approach needs strong support for auto-tiering and the ability to handle QoS and performance or data integrity differences between different pool elements. As I’ve pointed out before, any complex data pool like this also requires a near-real-time storage analytics tool to identify bottlenecks and instabilities.
- HPE has been developing a “monster” new systems architecture for the last two years that involves moving internal communications in a server to a new silicon photonics fabric. This would put DRAM and NVDIMM memory, CPUs/GPUs and accelerator FPGAs all together with connections to drives and communications adapters all serially connected to an in-server switch by very fast links.
HPE just announced they are pulling back from their program. Reading the runes, they are seeing industry efforts to achieve the same solution, such as Gen-Z, now becoming serious. At the same time, all of these efforts now await s signal of direction from Intel, the elephant in the room (at least when CPU architectures are involved).