The Sinking Fibre Channel SAN

As predicted, the weakening market for SAN-based storage arrays is eroding the once-powerful flagship interface of storage. Sales of Fibre Channel SANs are essentially flat lining, Since chipmaker Avago Technologies bought out Emulex, one of the two host bus adapter (HBA) providers, earlier this year, the only exciting thing in SANs seems to be the all-flash array that offers enterprises a cheap and much-needed alternative. Even 16 Gbps and 32 Gbps Fibre Channel don’t seem to create much buzz.

The bottom line is that Ethernet is stealing Fibre Channel’s thunder. Already available in 10 and 40 Gigabit Ethernet  links, next year will see a step up in technology to 25 GbE and 100 GbE. With Ethernet RDMA starting to take share from InfiniBand, too, it’s clear Ethernet has won the horse race.

But there are more than speeds and feeds involved here. The cloud is the major driver for the Ethernet tsunami. A single fabric that can be virtualized, orchestrated and, in a couple of years, turned into software-defined networking is very desirable. Ethernet meets that goal very well, while Fibre Channel isn’t even close.

With limited resources that can be deployed to engineer new solutions, the remaining major HBA companies -- QLogic, Atto and Emulex -- are not in a position to keep up. They’ve recognized that and both have hedged their bets to include Ethernet solutions in their portfolio. These aren’t garnering huge support, though, probably because of the vast competition in the Ethernet space. It’s also hard to be differentiated in Ethernet, which is highly standards driven, as Mellanox is finding with its Ethernet RDMA technology RoCE, which falters over a need for specialized switches for performance.

There are two major Fibre Channel switch companies left, with Brocade and Cisco neck-and-neck on market share. With such a small vendor base, the Fibre Channel SAN industry is vulnerable to technology or market stumbles and a lack of investment cash. Even so, Fibre Channel leverages some Ethernet technology development. For example, the PHY technology for driving the electronics of communications is used by both Ethernet and FC, and switching chips are closely related.

However, SANs are much more than just Fibre Channel. For a couple of decades at least, the SAN has been synonymous with RAID arrays: big, expensive boxes stuffed full of hard drives. The advent of solid-state drives has caused a radical rethink of the use cases for hard drives, with most in the process of relegation to secondary storage or the scrap heap over the next few years. The user benefits of compression and deduplication, now a standard feature in all-flash arrays (AFAs), means that fewer hard-drive arrays are needed and this, together with much larger hard drive capacities, is nailing the coffin shut on hard-drive SANs.

With the rise of AFAs and a decline in the growth of box counts, sales of faster SANs have been slower than expected, Financials for the vendors reflect that, with revenue essentially flat in FC and declining markedly in hard-disk drive RAID array boxes.

The RAID decline is systemic. SSD performance is so much better that a shift to smaller boxes with sets of say 10 drives is beginning, providing better matching to data rates and IOPS, and more economic modularity and packaging --  vendors claim 12 drives can reach 192 TB.

These new boxes have a smart ARM or x64 front end and provide redundancy at the appliance rather than the drive level. The appliance integrity approach means new software and this has led to “unified” storage products based on open source code such as Ceph as well as proprietary technologies. Invariably, these unified solutions use Ethernet for NAS and object modes, which logically push iSCSI as the block access method.

Besieged on many fronts, it’s clear that FC SANs are heading into decline, and will follow SCSI into history. This is going to be a slow process; I’d bet that air traffic control is still SAN-based somewhere in the world in 30 years! In the meantime, it will be a cash cow for the remaining vendors, priced high and falling behind in technology. It was good while it lasted!