Ethernet, Fibre Channel sales boom in Q4
We don't need no stinkin' FCoE
Networking and telecom market watcher Dell'Oro Group says that shipments of 10 Gigabit Ethernet and 8 Gb/sec Fibre Channel adapters are booming. And, somewhat surprisingly, they're not eating into each other, despite the convergence of server and storage fabrics with Fibre Channel over Ethernet (FCoE) support on mny 10 Gigabit Ethernet switches.
Dell'Oro makes its living selling data about the networking markets, so it didn't provide a lot of detail about what went on in the fourth quarter of 2010. But the company does put out some teaser information every once in a while, and the most recent figures show that sales of 10 Gigabit Ethernet controllers and adapters rose 57 per cent in the final quarter of last year, to just under $100m.
"What really took us by surprise was the strength in network adapters rather than LAN on Motherboard, or LOM," explained Tam Dell'Oro, president of the company, in a statement. "Intel and Solarflare just about doubled their adapter shipments quarter-over-quarter, while Emulex's were up over 60 per cent, and all this is on the older generation servers that support 50 Gb/sec of in/out throughput. What do you suppose will happen when the next generation servers with over 100 Gbps of throughput start shipping?"
Dell'Oro tracks the top nine makers of 10 Gigabit Ethernet controllers and adapters, and in this latest report one big shift that effected the numbers is Hewlett-Packard using silicon from Emulex instead of from Broadcom for its BladeSystem blade servers.
The advent of FcoE was supposed to nuke Fibre Channel adapter sales, at least according to those like Cisco Systems and HP that are pitching converged networks very hard. But it looks like FC port shipments are continuing to grow.
In its latest quarterly SAN connectivity report, Dell'Oro says that over 900,000 FC ports shipped in the fourth quarter of 2010, a record high level of shipments for any speed of Fibre Channel adapters in the history of the SAN market and representing a 15 per cent boost in port count compared to Q4 2009.
"The Fibre Channel market appears to have returned to end-of-year strength which is the typical seasonal pattern in a healthy economy," said Dell'Oro. "Many server manufacturers have announced very strong quarters, and Fibre Channel adapter sales generally pattern those of servers. What really took us by surprise was the magnitude of Fibre Channel strength. We expected more users to shift to Fibre Channel over Ethernet (FCoE), particularly in blade server chassis, but it didn't happen."
What happened instead is that a lot of customers decided to move to 8 Gb/sec Fibre Channel adapters and keep separate things separate on their networks. Without FCoE having been an option for the part two years, it is logical to assume that FC port counts might have been even higher in Q4 2010. Dell'Oro did not divulge how many FCoE ports shipped in the quarter, but that would be an interesting number to see. ®
This is easy to explain
Here's my 2 cents.
Firstly, FCOE made promises about convergence - simplifying the data centre.
You can get simplification from iSCSI, firstly at 1Gbps and now at 10Gbps. And unless you're really looking for very high deterministic performance, iSCSI can hang with the big guys. iSCSI continues to grow - rapidly - due to its lower cost, simplicity and good enough performance for the majority of applications (consolidation, virtualization, near-line archive, backup to disk, etc).
So why Fibre-Channel's continued popularity? Several reasons.
Firstly, the FCOE protocols have just been settled, but the Ciscos, Junipers and Brocades are all arguing about 10Gbpe lossless Ethernet connections, with little common ground.
Secondly, converged network adapters (CNAs) only started to appear on the market in late 2010, which slowed growth. The article mentions blade servers, but if you recall, rack server introduction of CNAs came first and only in late Q4, early Q1 were the blade form factor CNAs introduced for a couple of the major server vendors.
Thirdly, fibre-channel introduction of FC8 protocols offered deterministic connectivity at speeds (really capacity) similar to 10Gbps FCOE..
Fourthly, a storage architect who has built his reputation on 12 years of fibre-channel is going to be highly skeptical of ripping and replacing for something so new.
Fifthly, has anyone seen the cost of the Cisco Nexus gear that you are expected to roll out should you go this direction? in the current climate, that's sobering enough to make you extend the investment you've already made for another business cycle...
Not really. Most people that build a SAN infrastructure did it for performance or to manage a lot of storage. Say you toss in iSCSI or FCoE, those that care about performance will try and segregate it for performance reasons. If you're going to maintain another infrastructure, why not just stick with FC? It performs better, and I don't see a cost benefit. Granted, we use Cisco which is pricey -- there are cheaper 10GbE vendors.
And that's the infrastructure. There aren't many FCoE arrays. And if you look at arrays that support iSCSI , most would argue if they're as reliable or perform as well as a FC array. Discussion in itself, but I'd say it's a factor in why FC is doing well and not going away anytime soon.
Agree with AC in post 2
In short as far as I see it... fiber channel is designed for use with storage devices; it is designed for low latency, high speed, and to ensure data integrity. However, the controllers and equipment are very expensive, the few FC bits I've played with the drivers were difficult. Ethernet has relatively inexpensive hardware, but can have packet losses and retransmissions; for SAN use, there's iSCSI, (perhaps) ATA over Ethernet, or the likes of NFS if you don't need block-level access. FCoE? You have the packet losses and retransmissions of ethernet, while still needing expensive, specialized cards and tricky drivers like fiber channel. It's like the worst of both worlds.