![[RSS]](http://thefutureofstorage.com/wp-content/themes/dellscsi/images/rss_feed.gif)
It’s A Long Road
Submitted by marc farley on April 23rd, 2008
The Ethernet\TCP\IP stack owns the networking world and sooner or later all other networking technologies succumb. Many believe Fibre Channel (FC) will be the next one to disappear due to the economy-of-scale advantages that 10Gb Ethernet has. If Fibre Channel were more flexible, its chances for survival would be better, but its reputation as a management beast stacks the odds against it.
However, as John Cleese once paraphrased, FC isn’t actually dead yet and some think it is already on the road to recovery. The difference for FC is that it is getting a second chance as Fibre Channel over Ethernet (FCoE), which is an encapsulation of FC over a new variation of Ethernet that Cisco is developing, called Data Center Ethernet (DCE). DCE supposedly will solve certain data center networking problems as well as being the transport for FCoE. High priority traffic (like storage) will be transported at lower latencies with better reliability than vanilla Ethernet.
To the best of my knowledge, FCoE is the first major application for DCE and therefore depends on the success of DCE. DCE’s success depends on the requirement for customers to overhaul their core networks. In other words there are a lot of chores to do before dinner goes on the table. There is no doubt that there are lots of FC customers today who would like to avail themselves of FCoE – but I suspect the path to getting there is going to look much steeper and longer than it does today. In the meantime, iSCSI storage technology will continue to develop and broaden it applicability through standard compatibility with vanilla Ethernet and vanilla 10Gb Ethernet.
So the question I have for people here is when they (or customers they work with) will want to install FCoE. What’s the timeline? Also, what requirements do you see for DCE, independent of FCoE. In other words, if you weren’t thinking about FCoE, would you implement DCE?
![[Ars Technica]](http://thefutureofstorage.com/wp-content/themes/dellscsi/images/icon_ars.gif){kind=icon}
April 23rd, 2008 at 9:16 pm
Marc, anyone’s who spent any time in a large corporate data center wouldn’t be so glib on FC’s “demise”.
Last time I checked with IDC, more FC networked storage was sold than all flavors of IP storage, and that trend shows no sign of abating.
The real question for me is: will iSCSI ever be taken seriously by data center architects?
So far, the answer is “hell no”, regardless of whether it’s 1Gb or 1Tb.
But keep trying, though — it’s a good read!
April 24th, 2008 at 2:09 am
I’m not an DCE expert but I feel it is mostly addressing problems of sharing storage and LAN traffic. The simplest solution for these problems is switch separation however.
Most iSCSI installations I meet actually share both sorts of traffic with no troubles. The separation is done by VLANs only. If they fell in troubles with such design, the first they would do is purchase some new switches to separate traffic. In short term I don’t believe those companies would replace their whole infrastructure for DCE enabled.
April 24th, 2008 at 8:59 am
Hi Chuck, nice to hear from you. It’s true that sometimes I run down futuristic rat holes like this and get myself in trouble. BTW, a lot of the people I work with don’t agree with me either. :)
There’s no way anybody with a sizeable FC installation is going to go anywhere under the common sense law of – if it ain’t broke, don’t fix it. And duh! of course FC has outsold iSCSI – that’s a no brainer too. iSCSI’s growth rates are, of course, much larger. I’m thinking IDC projections are not going to be accurate – they’ve been wrong before.
Roadmaps matter and FC has a bit of a roadmap crisis. It would almost be better if the whole FCoE/DCE thing weren’t hanging out there like a stop sign for native FC, but it is.
April 24th, 2008 at 9:14 am
Hi Lucas, thanks for posting here. There seems to be a bit of a divide about the value of segregating traffic and integrating (or unifying) networks. As you point out, they don’t need to be different with virtual networking technologies.
I’ve always thought that separate networks for SANs and LANs were best because it simplifies troubleshooting and change management. Of course the downside is that there are multiple network infrastructures to manage. I’m not convinced that virtual networking isolates the problems of change management – in fact it can amplify it if switch ports are consumed more rapidly. When you run out of ports, the physical layer has to change, and things like an expansion of multipathing for new servers in the SAN can consume a lot of ports in a hurry. With 10Gb Ethernet, the need for additional ports in the core should be reduced, but that’s not necessarily a slam dunk.
If you have segregated SANs and LANs, you can still run into the same problem – running out of posrts and integrating new switches. I just think the planning for the change and the implementation of the change is simpler if the physical networks are separated and isolated. Do you agree?
April 24th, 2008 at 10:28 am
Hi Marc,
simple solution usually wins the battle (understand it a vote for segregation).
Those “broad pipe” technologies however – be it 10GbE, Infiniband or anything else – may call for the need of a sharing formula again. Having invested in 10gig infrastructure, some companies will force their engineers to get the most out of it. Instead of separating “expensive” switches they might end up sharing the traffic again.
Still I don’t believe this effect will kill the today’s protocols in favor of DCE.
April 24th, 2008 at 11:21 am
Lukas,
I strongly agree that the simple solution wins the battle. In my earlier insight (”A Skeptic’s View” on April 21st) – that was my main point: it’s the easiest way that rules the day.
What hasn’t seemed to be discussed here as much is that relative simplicity or ease is not necessarily a technical question, but often a political one. How much power are the SAN/FC guys ready to relinquish? I think this may be one of the more significant obstacles to adoption/growth of FCoE.
Jeff
April 24th, 2008 at 12:36 pm
Wow, good point Jeff! I made the mistake of assuming they would be the same people, but of course that doesn’t make sense. Doh!
April 24th, 2008 at 12:56 pm
Marc, I agree with Jeff’s “simple wins” maxim, but then you need to define “simple”. Is “simple” having two totally different networks for storage and communication? So far, that’s been the case, and anti-FCoE but pro-FC folks would argue that it will remain. But consolidated and virtualized servers are already combining multiple virtual NICs and HBAs over a single wire. I contend that converged I/O using Ethernet is the real simple future solution, leaving us with a question of where iSCSI and FCoE (and AoE and NAS and the rest) fit in.
As for Jeff’s other point, I remember writing a Storage magazine article about that way back in 2005. The answer then was “don’t count the storage networking guys out”, but that answer is changing in today’s virtualized world.
April 24th, 2008 at 1:31 pm
Hi Stephen, No question that “consolidated and virtualized servers are already combining multiple virtual NICs and HBAs over a single wire”. I see it too. At the current early phase of virtualization its clear that lots of people find it “simplest” to virtualize SAN ports (with software initiators) over Ethernet ports. But part of the reason for this is to get multipathing in VMware environments.
I question that this will continue to be perceived as “simplest”. I have to think that network management for converged traffic will be extremely difficult – made simpler by segregating storage I/O from LAN traffic.
April 26th, 2008 at 4:13 am
Jeff, regarding the people. From a longer time perspective I believe the people can’t fight against a simple and reliable technology. In most companies the rational arguments usually wins.
On the other site, I don’t worry about the SAN peoples’ jobs. I’m convinced most of them started their career in networking, server management or other sorts of IT positions. Similarly they can move when SAN administration doesn’t need them for full time more.
April 28th, 2008 at 10:02 am
Stephen & Marc, as far as the “simplest wins” – people in our position can not define it. The maxim in these cases is that simple is what the customer defines it to be. And typically that goes against implementation of new standards, for obvious reasons. The value prop has to be *very* large to keep them from doing the easy “business as usual” route. Does FCoE really have that value prop?
Lukas, I agree with your standpoint on the longer time perspective and where SAN people could go – my point was more along the lines of “what battles might you need to face in implementing a new technology?” Politics often factors in importantly when an IT shop looks toward a new platform and, rather than looking at long term potential, people sometimes fight what seems like an immediate threat. And though virtualization is changing things (well put, Stephen) adoption of new protocols or technologies will (almost) always face some sort of political battle -