I was reviewing some documentation the Cisco web site for an architectural white paper and came across this image in some marketing material
What this image clearly shows is that the UCS strategy is really about getting a fancy NIC into a server.
Each blade has a fabric connection card to the UCS 6100 (fabric interconnect), which then connects to a Nexus 5000 (the fabric master).
Why does this matter ?
Simple. Lets consider HP and their C-Class chassis. HP has their “Virtual Connect” strategy for blade servers with the NonStop fabric that connect all the servers via the mezzanine cards to Interconnect modules. Interconnect modules support Fibrechannel, Infiniband and Ethernet and so on. This means that HP can control all the choices that the customer can make for the servers because they “own” the design of the blade server chassis.
When bladeservers were first released, IBM and HP had Cisco develop modules that could be installed. More recently though, IBM has stopped even offering Cisco switches for large parts of its portfolio and HP is quite hostile – in the “HP Virtual Connect for Cisco Network Administrators” they refer to the Cisco switches as “traditional approach to administering the network”. Thats got to smart a bit.
Want to bet that HP didn’t want to play ?
Why wouldn’t HP want Cisco to provide the networking ? My guess is that HP has networking under development and they are freezing Cisco out of the data centre. HP has made substantial inroads at the edge of the network with their ProCurve products. The Data Centre network is an entirely logical next step. Its incremental, extends their server expertise / portfolio and integration and can easily be sold to existing customers through the server teams.
So HP might have shut Cisco out?
So Cisco does servers ?
I think HP did so. HP used the Tandem acquisition to put the NonStop server fabric into their blade servers, then built networking modules that positioned them as competitors to Cisco. This was probably some time in 2007.
Cisco response has been to build a network fabric, functionally similar to the NonStop server fabric, but with much deeper integration into the network offering much greater virtualisation of the storage and network layers. This is consistent with their current marketing message around their network history (although, with video, voice and home products who knows what Cisco is really about these days).
And its all about the message
This gives Cisco a “message” that their product is “different” from HP or IBM. A differentiator like this can be used to initiate new opportunities with customers – to make that into English, you can approach the CIO / CTO and try and flog your new strategy. They had to make some new technologies such as FCoE and all its supporting technology, but that’s just a few broken eggs.
And they had to build the blade servers. Obviously. If Cisco didn’t make their own servers no-one else was going to put Cisco networking in. Easy decision from that perspective, and John Chambers is an ambitious man. Growing the business in the data centre by extending into the server farm makes logical sense. Remember, HP is likely to extend from the server into the network, so meeting them in the market is natural. Especially with the hype about Cloud Computing developing over the last two years.
But the actual servers are not different from anything else in the market. They are a little bit ahead of the curve, but HP/IBM will there with similar products in a few months.
Its all about the network.
So, where is the Special Sauce ?
The HP Virtual Connect system does Ethernet, Fibrechannel and Infiniband as separate elements or mezzanine cards.
Cisco uses a single Ethernet fabric to do all of this. By creating FCoE from nothing, Cisco can combine all storage and network access over a single Layer 1 connection. To tick off the marketecture:
- no fibrechannel = less components = less power & cost
- unified fabric = less configuration = less OpEx
- unified fabric = less equipment = less CapEx
- unified fabric = more flexibility = service automation = cloud computing (BINGO!)
You get the idea, I am sure.
HP and Cisco – Frenemies ?
I wrote something about this here
Conclusion
As a designer, this little graphic made UCS much more comprehensible. The marketing message that UCS is some sort of revolution has obscured my thinking. I was looking for rocket science, and all I got was marketing smoke. And some bendy mirrors.
But I wish they would use nice smelling smoke next time. This one still smells a bit fishy.
Loading ...
About Greg Ferro
You’re an idiot. Go back and read the design docs instead of making a moronic statement after reading a one page press summary. Jags like you should keep on doing it the old way and when what little credibilty you have is in the cr-pper we welcome you to join the rest of the engineering crowd
Mmmm, doesn’t leave much room for discussion. Indeed I have read the design docs, and reviewed much of the marketing material available. However, I am not a true believer in FCoE. In many aspects I am a Ciscotist, but the Data Centre strategy built around FCoE is transient and probably not a happy long term strategy.
I can certainly see the value of the Network Fabric concept espoused in UCS, but currently think it should be based on IP, not Ethernet. Intelligent comment welcomed!
I don’t agree with your thought that FCoE should be based on IP. I think Cisco have it correct, make it as simple as possible. IP is required for crossing networks and getting from end to end. However, FCoE is targeted at SANs and therefore will always be switched within a network.
For your regular network communication, the fabric will support traditional TCP/IP traffic. Also, there is nothing to indicate that FCoE is going to be transient. It will become a standardized protocol and 10GBoC will proliferate.
I think the FibreChannel is a pile of crud, and putting it over ethernet is double crud. The idea of intelligent network fabrics (which is what fibrechannel does) that somehow morph or munge the data flow has never been successful and it’s unlikely to start now.
The rise of iSCSI and is NAS implementations clearly shows that FC/FCoE does not have a long term future and acts only as a transition mechanism for legacy FC-attached storage
Greg:
So, I think the other differentiator for the UCS is the ability to manage it more holistically–more comprehensively and in a way that is more virtualization aware than currently possible with other solutions. I know there is not a whole lot of detail about UCSM right now, but I did a couple of recent blog posts on this and I promise more detail is shortly forthcoming.
On a totally different topic, being the “official FCoE standard bearer”
, I know we have agreed to disagree on the FCoE front, but I am curious–what would you need to see in the industry (beyond final FCoE and DCB standards) or our approach to change your mind about FCoE having some legs.
Regards,
Omar Sultan
Cisco
Well as Omar has responded not sure what point there is having little ole me writing something but what the heck.
I think you have missed quite a bit of the details around what UCS does as well as how it compares to the competitors. To compare them both you are going to have to go a bit deeper than you have gone. However I understand that often in a blog post this is not the purpose.
UCS is a lot more than a fancy NIC, its the far beyond that with abstraction at a large scale. You can do some of this with Virtual Connect but you need extra sauce to do it outside of chassis.
UCS breaks the mold in quite a few other areas, I would suggest you have bit more of a dig. If you are keen, grab a copy of the California book.
Having played with the kit for the last few weeks I can attest that its a lot more than smoke, and it has not caught on fire yet.
Cheers
Rodos
Yep. Read it, reviewed it (somewhere on this site), thought about it. Haven’t changed my mind.
There is nothing revolutionary about UCS, more like a very small step in evolution for virtualization. I do not miss that Cisco needed to do something unusual to penetrate an (otherwise) closed market, specifically, the marketing needed to have a narrative that could transcend the noise from IBM /HP /Dell, and this is the result. But doing some fancy silicon tricks and closed form packaging is not a benefit.
I retain an open mind. Today, the system is unusable except for niche use cases, and the wider market has no interest in cloud computing.
A a result, UCS boils down to server with a fancy NIC. There is no change to Windows/Linux, no change to Intel architecture, no change to anything, but a bit of silicon in the gear box.
YAWN.