First, I would like to thank Douglas Gourlay for responding to my post at http://www.networkworld.com/community/node/24657/177203#comment-177203. You can find his home page at Cisco http://blogs.cisco.com/datacenter/ in his role as Senior Director of Marketing and Product Development for Cisco Data Centre.
All that Fibrechannel and FCOE adds to a data centre is more complexity, and more organisational pain. 'Operational Orchestration' does not occur. Fibrechannel enables the Server and Networking teams to balkanize and does not create synergy. Worse, it creates an opportunity to have a dialog about my FC is better than your IP.
I do not fondly remember the same debates when we had FDDI and Token Ring. And just change the word Token Ring for Fibrechannel and we are having exactly the same debate from 15 years ago. Have we learned so little ?
In data centers that use iSCSI as a primary method, and Fibrechannel as the "if we have to" storage works just fine.
I can understand the Fibrechannel over Ethernet (FCoE) is stateless, and IF THE UNDERLYING TRANSPORT is lossless you get a BIT more performance for some people. But FCoE is not an enabler for "Synchronous replication, storage media encryption, data migration, etc." thats a software feature that occurs above the transport layer, and can also be done using iSCSI and even FCIP.
If the iSCSI cost point is higher, lets compare that against the financial cost of implementing Fibrechannel at all. Fibrechannel anything costs a squillion bucks to buy, and two squillion bucks to support. So what if a bit of performance is lost with iSCSI, it costs a quarter of squillion to do.
You can certainly throw some of the cash you saved by NOT using Fibrechannel to applying 'buffers' or anything else that fix the problem such as buying more TOE cards. Even buying more servers is viable because of the monstrous power consumption of Fibrechannel chips and equipment.
On the latency 'thing'. If you need real time lossless connectivity, then just engineer the IP network to do that. Its easy enough to do, just because IP CAN drop packets, doesn't mean that it always does ? Can't software drivers be developed that stop buffering ?
(Aside: I remember a few years back that Cisco had the latent, store and forward, lossy 1.2Gb/s shared backplane Catalyst 5000 switch competing against the nonblocking Bay Networks 28000. Customers voted by buying the one that was cheaper)
Even now most servers cannot use the performance of fibrechannel anyway, and are not likely lift performance in the next five years to use it either. The X86 CPU/Memory bus architecture doesn't let data transfer that fast - witness the failure of 4GB fibrechannel except as a Fibrechannel switch interconnect technology.
(Note: bus architectures for other platforms exist, but, who cares ? X86 defines the data center market)
Lets face it, Fibrechannel is a poor mans Infiniband, and Infiniband is the real deal. Lets get over Fibrechannel and move on to the end game.
PS: The Cisco Nexus 7000 looks like a good ethernet switch for my data centre. I will be putting them into my strategy and it stops Foundry and Woven Systems from bothering me.
Latest software headlines from Network World:
At 10, Google reiterates commitment to CIOs
As Google turns 10, enterprise success in question
Zoho adds Google Docs-like file management
|
Does Verizon's Voyager stack up to the iPhone? |
|
|
5 IT skills that won't boost your salary
[1,407]
Women 4 times more likely than men to cough up personal info
[589]
Japan's 10 funniest tech-related commercials [Videos]
[407]
Throwing away a promo CD is "unauthorized distribution"?
[1,265]
Adults too quick to dismiss educational video games
[682]
Attack of the iPhone clones [Slideshow]
[578]
10 things IT needs to know about AJAX
[1,258]
This Year's 25 Geekiest 25th Anniversaries [Slideshow]
[409]
|
|
Fibre Channel is not Token Ring
As much as I see some similarities to Token Ring, the Fibre Channel market is actually growing and most hosts are attached at 4Gb today (not just ISLs) and are going to 8Gb next year, whether they need it or not.
In a perfect world, iSCSI would be the right choice to converge storage and data traffic but the reality is that the Fibre Channel market is much too big and entrenched to change anytime soon.
FCoE accomplishes many things. It offers a lower-cost, higher-performance option that will appeal to the Fiber Channel customer much more than iSCSI ever can.
I want a perfect world - and nothing less!
FCoE is NOT lower cost, it means replicating name resolution, flow control, connection types and more. The support cost is enormous as people have to learn fibrechannel and its mystical properties. Why bother ?
FibreChannel at 4GB/s is pointless, current bus architectures can't move data at more 600-800 megabits.
Fibrechannel is growing becuase the storage vendors are marketing it heavily. But SMB is buying iSCSI and its already on the way into big data centers.
Fibrechannel accomplishes nothing more than any other solution. It appears to a marketing technique so storage vendors can hold the revenue for so-called 'storage switches'.
iSCSI is the right choice, lets start on it today.
=====================================================
If your only tool is a hammer, every problem looks like a nail - Abraham Maslow
You have to deal with the world the way it is...
Of course, if we were just beginning the convergence process, iSCSI would be very attractive.
But if you ask an existing FC customer, who has invested millions in their SAN over the last few years, iSCSI doesn't seem so attractive any more. They've already invested in "name resolution, flow conrol, connection types and more" so FCoE is a much simpler migration than moving to iSCSI.
For customers who haven't invested in FC like many SMBs, I agree with you, iSCSI is certainly a simpler alternative.
and thats what they said about token ring
And no one was happy with it either.
If your only tool is a hammer, every problem looks like a nail - Abraham Maslow