![[RSS]](http://thefutureofstorage.com/wp-content/themes/dellscsi/images/rss_feed.gif)
Seven Fundamental Reasons Why FCoE Will Fail in the Market
Submitted by Greg Ferro on May 15th, 2008
I have been evaluating FCoE for a while now, and been researching the technology for my latest project. It’s a fine technology, if you believe in Fibrechannel. I believe it has some nice features, and will offer customers some very compelling reasons to purchase.
But if you stand back and look at the market space at a whole, I cannot perceive that FCoE will be successful. Here are my reasons.
There are no standards
The first rounds of FCoE standards process are estimated (yes estimated) to be finished by the end of 2008. All areas of FCoE won’t be finished until 2009/2010. This assumes that none of the vendors start bickering over the details (think of 802.11 wireless standards). If any of participants start bickering you could be looking at another WiMAX / 802.11n debacle.
Have a look at the list of FCoE vendors and decide for yourself if they can deliver: Cisco, Infinera, Ericsson, Qlogic, Emulex, Juniper Networks, Fulcrum Micro, Brocade, IBM, Fujitsu icroelectronics , Broadcom, Intel, HP
The year of 10Gb ethernet won’t be until 2010.
We are seeing some early stage shipments of 10 gigabit ethernet switches from the usual players. However these are very simple ethernet switches that are missing of the features that we expect and want to be there. Sure they have ethernet and some basic layer 3 routing but the rest of the features will not be mainstream until 2010.
And of the 10GB switches today, only the Cisco Nexus 5000 supports pre-standard FCoE. A single vendor does not make a market (no matter how large you are). And make note of the pre-standard – being stuck with pre-standard FCoE would be a very very expensive mistake.
For FCoE to be successful, you must buy new switches that support PFC, ETS, and DCBX Data Centre Ethernet extensions.
These features were submitted to the IEEE in February 2008 as draft submissions. Just when will these be approved, and then implemented by the switch manufacturers? This year? Next year?
DCE capable switches are not here yet, they are going to cost a bomb, and will take a while to be confident of their performance and delivery. It is going to take quite some time before the big customers will purchase these switches. I am sure that there will be a lot of smoke around DCE, but I find it hard to the believe that substantive volumes of equipment will ship before 2010.
You must buy FCoE HBA for servers, and then wait for the drivers to be certified by all the storage vendors
Each FCoE HBA will need design, production and software drivers. But even when this is done, these adapters will need validation from the storage vendors. This process is going to take at least six months for the slowest and least capable FCoE adapters. The high performance and feature rich adapters will take about a year to move through the production cycle.
Why did Cisco buy Nuova Systems so quickly ?
The FCoE is largely developed and driven by Nuova Systems, a Cisco funded startup. Cisco has unexpectedly purchased the remaining interest of Nuova Systems in April 2008. Many people commented at the time that this was quicker than normal and an unusual move by Cisco.
Did Cisco fast track the acquisition of Nuova to head off the iSCSI movement? It’s very possible. The vendors who make the iSCSI HBA will now have to split their resources to develop FCoE adapters as well and this will slow down the iSCSI development.
FCOE is a transition technology
One of the most surprising aspects is that everyone agrees that FCoE is a transition technology; it is not the final destination for Network Attached Storage. You can see this repeatedly in early usergroup and forum postings. And even in the latest submission for DCBX to the IEEE, it clearly shows FCoE is regarded as a temporary measure to transition Fibrechannel systems to an ethernet physical connection and then IP backbone. It also talks about FCoE being for legacy or existing Fibrechannel sites, and greenfield sites are not expected to use FCoE.
Surprised ? Well, yes I am. I don’t want to be investing in short term technology. Living in a niche marketplace is not a happy place and bad for career aspirations.
ISCSI will move into the gaps
By the time these issues have stabilised, iSCSI will have moved in to these gaps and made FCoE obsolete before it achieves significant market penetration. Even though many people call iSCSI an SMB technology, or an unreliable or untrustworthy system, they will also tell you that many of their servers are using iSCSI today. It won’t take long before the price and simplicity advantage will outweigh the supposed benefits of FCoE.
![[Ars Technica]](http://thefutureofstorage.com/wp-content/themes/dellscsi/images/icon_ars.gif){kind=icon}
May 16th, 2008 at 7:12 pm
I was directed to your blog from “Another day, Another Skeptic” by Cisco’s Deepak Munjal latest post on their data center blog: http://blogs.cisco.com/datacenter/2008/05/another_day_another_skeptic.html
I think he made some fair points against every one of your seven claims above. I probably won’t agree all of Deepak’s statements but you certainly cannot convince me that the FCoE will fail.
Forget 1G iSCSI. Everyone knows the iSCSI won’t truely take off without 10G support. 10G card is just starting deployment into enterprise server land so it looks to me both iSCSI and FCoE are at the same starting line.
May 17th, 2008 at 4:39 pm
My feeling is that for most people, perhaps that vast majority, iSCSI is a better choice than FCOE. This will lead to market dominance over time and if FCOE does not come out strongly, iSCSI might strangle it before it can penetrate the market sufficiently.
So the points I raise are topical concerns about where FCoE might fail to achieve that momentum. The response by Deepak has a lot of forward looking statements that should be carefully evaluated. Consider the following two:
“The FCoE standard will be completed this year. Most of the hard work such as frame format and addressing schemes has been completed and there shouldn’t be any roadblocks left.”
May 17th, 2008 at 4:51 pm
….posting problem…
…”shouldn’t be” does not mean there won’t be. I like his positive outlook, but want to see the finished standard before I would consider dropping hard cash on it.
“4. “You must buy FCoE HBA for servers, and then wait for the drivers to be certified by all the storage vendors”
The FCoE adaptors (CNA) have been announced by two different vendors and they use the exact same ASIC technology and driver as their current 4G HBA so certification should be relatively swift.”
So far, my point is valid, no adapters, no certfication, but it “should be swift”.
On your statements on 10G support, I believe that you are mostly correct. What I see from my reading, is that a lot of smaller/medium companies are already using software iSCSI on 1GB Ethernet, and they believe in it now.
I postulate that with more iSCSI HBA, 10GBE and Data Centre Ethernet that will become the storage standard and this might extend into larger companies.
After all, this is how Ethernet beat Token Ring.
Thanks for posting.
Greg
May 20th, 2008 at 2:45 am
[...] article I wrote on concerns about FCoE adoption has been posted at The Future of Storage and reviews seven reasons why FCoE might not achieve critical mass in the [...]
June 3rd, 2008 at 8:57 am
Before 40GE/100GE are widely used as switch-to-switch links, it does not make sense to use 10GE as switch-server connections. In a server farm, access-layer switch is better have an up-link:down-link bandwidth ratio of 1:2 (or 1:1). Without 40GE/100GE PHY, 10GE server-switch connections will suffer up-like bottleneck.
thanks
xuping
June 3rd, 2008 at 6:36 pm
Agree – it’s doomed to a slow death. iSCSI is simpler and in many cases about as fast since the bottlenecks are with disk I/O. But most importantly it’s a lot cheaper, and money talks.
June 4th, 2008 at 5:09 pm
No disagreements about iSCSI being simpler and having better capabilities than even Fibre Channel as a storage protocol.
However, what do you suppose we do with the $50B installed base of FC? Token Ring never had that.
FCoE is the best way to get that infrastructure converged onto a unified fabric.
June 5th, 2008 at 10:02 am
FCoE is the best way to get that infrastructure converged onto a unified fabric only if it works on 1GE cat5 Ethernet. Otherwise “what do you suppose we do with the Nx$50B installed base of 1G Ethernet?”
10G Ethernet is a completely new network in terms of physical layer. From an IT managers point of view, it has no compatibility with current infrastructure. Bypass 1GE may become a big mistake of FCoE vendors.
Lack of performance on 1GE? Plenty of room for improvement on south-bridge, bus, TOE, CPU and even OS. 1GE is not the only one should bear the blame.
10GE NIC or CNA for server
June 6th, 2008 at 7:32 pm
Most servers already have multiple 1GbE adaptors along with two FC adaptors. We are already dealing with several copper Cat5 and Fiber cables. A consolidation to a single pair of low-cost SFP+ twinax cables that have better latences, lower power budgets, and better bend radius than Cat5 or Cat6 is the way to go.
Add up the total amount of bandwidth currently being used on a server. Please include interfaces that currently support storage, data, management, backup, clustering, and even Vmotion. This is certainly more than 1GbE and therefore, 10GbE is the best place to start.