Today, Cisco announced their long-rumored entry into the heart of the data center market: the UCS, or unified computing system.
Do not make the mistake of dismissing this as "just another x64 blade server".
Regardless of what you may think about the product, the technology or the strategy, there is no denying that UCS represents serious new thinking and innovation in our industry.
And things won't likely be the same around here anymore ...
The Background
Many of us at EMC have been involved with Cisco on this for quite a while. We've tried to contain our enthusiasm; fortunately we don't have to keep quiet any longer.
We think UCS is way cool. There, I've said it.
There's a serious dose of very fresh thinking to appreciate here, backed by serious technology and a serious vendor consortium.
Here's what jumps out at me ...
First, there's the notion of a unified data center fabric. The entire architecture is based on data center ethernet, meaning a single wire supports converged server/storage, server/server as well as server/user connections.
As a result, the I/O subsystem is so clean and elegant, it makes you just want to cry. No gaggle of I/O adapters, just a couple of clean converged connections that will find their way onto the motherboard before too long.
Not to mention all that good work that Cisco and VMware have done to ensure that network state follows virtual machines around as they move.
Second, it's obvious that this architecture is designed for virtualization at very serious scale.
UCS is fundamentally a very large array of stateless server blades, each with the massive memory and processing power that can push server and desktop virtualization to entirely new levels. And do so more efficiently than any other competing architecture.
And all operating as a single giant computer leveraging the power of VMware.
Third, UCS reflects the management model of the future -- network-centric. The ideal way to manage the UCS is to think of everything as a provisioned network service -- including virtual machines that provide application or infrastructure services.
I think that this particular aspect of UCS is certainly going to cause some controversy in certain corners of the industry -- and it's a healthy discussion that's way overdue.
To next-gen data center geeks like me, this sort of design philosophy will end up being the server architecture of choice for building the next generation of private clouds in enterprises and service providers.
The bigger your environment, the more appealing UCS will be. And there are some very big environments out there ...
And There Are Serious Implications
First, I'm sure that traditional server vendors like HP, IBM and Sun will have something close to a severe allergic reaction to all of this.
I'm betting that the server blogosphere quickly gets as heated as our happy little storage blogosphere as a result. And I'm guessing that HP will lead the charge for the opposition, since they have the most to lose here.
Just for the record, EMC has historically worked with all server architectures -- large and small, mainframes, etc. -- and just about every OS and database out there.
As a result, we've ended up having a unique perspective on the server wars. And since we don't sell any of them, we're pretty open about our opinions.
Here's the headline: although we're not 100% impartial, we don't think the traditional x64 or Big Unix vendors should be dismissing what Cisco has done, or just how attractive UCS is going to be to large enterprise shops going forward.
And all of you enterprise shops running scads of Linux virtual machines on zSeries? You'll eventually have much more than a passing interest in UCS.
Second, the ecosystem of partners that Cisco has assembled around UCS is essentially the best-of-the-best technology that isn't hopelessly tethered to some server vendor's portfolio.
That sort of vendor and system integrator ecosystem will be refeshing as well to many larger IT shops who might feel locked into a particular server vendor's offering in one area or another.
And I think that EMC is an important part of this ecosystem.
The EMC Angle? I'm So Glad You Asked ...
Bottom line: I believe EMC is extremely well positioned to help customers exploit UCS to its fullest potential.
Here's why ...
Serious server architectures usually require serious storage architectures, and EMC is one of the very few vendors who do that well.
Not to mention EMC's extended capabilities with backup and advanced business continuity at serious scale.
Of course, EMC will be doing all the usual industrial-strength eLab quals and performance characterizations that we usually do. And all of that work will help customers deploy UCS with confidence.
I'm not sure too many people realize this, but our EMC Smarts resource management portfolio (not to forget Voyence and Infra) is also extremely well positioned for this next-gen network-centric model that UCS facilitates -- not to mention working extremely well in legacy environments.
Going a bit farther, EMC's RSA portfolio of security products will provide the architectural foundation for next-gen security frameworks that will be very attractive for many of these implementations -- a new world where security policy has to move with the virtualized workload.
And then there's all of EMC's extensive integration and qualification work with VMware at serious scale -- very useful work, now that you mention it ...
EMC teams are also busily working today to prove out very large use cases of UCS technology, using our significant expertise gained from the largest data center environments.
We think this sort of solutioneering work will accelerate time-to-value for customers who want to exploit what UCS can do.
On an even more customer-centric note, EMC and Cisco have been working closely together for several years in large data center environments.
We've learned how to partner with Cisco and others -- before, during and after large-scale implementations.
A Few Final Thoughts
It's amazing what can be done with a clear vision, a clean sheet of paper and a serious R+D budget, no?
Cisco doesn't have the decades-long legacy of historical server architectures and traditional operating systems holding it back.
Think about it.
No HP-UX and Itanium (Itanic?) base to feverishly defend.
No Solaris/SPARC business to position.
And please don't get me started about IBM's plethora of different architectures.
Cisco had the freedom and the insight to design for what will be -- and not what was.
Cisco also had the freedom to invest in what really mattered -- an entirely modern computing architecture -- and to partner freely with the rest of the industry for the technologies and competencies to complete the customer offering.
Something the traditional server guys just weren't able to do.
On a practical note, it'll take a while for all the UCS components to mature and prove themselves. It'll also take a while for data center architects to fully appreciate what Cisco has done here.
Either way, before too long, it looks like the game will have changed in this part of the IT marketplace, and changed for the good.
Bravo, Cisco — you’ve got our respect and support!
-------------- updates --------------
Full EMC press release can be found here.
Full VMware press release can be found here.
Full Cisco press release can be found here.
Registration page for Cisco webcast can be found here.
I look forward to seeing one of these systems up and running, but unfortunately it looks like it'll be several years before something like this becomes available to companies with less than $1 million per year IT budgets.
Indeed, I wonder if it will ever be deployed by anyone outside the super-large enterprises - will everyone else go directly to the Amazon style cloud environments?
Posted by: Ewan | March 16, 2009 at 11:14 AM
Ewan, then I suspect you'll see something before you think, only it won't be coming from Cisco and will be based on real industry standard parts. 10gb ethernet is much more than just Cisco, and while the press loves a battle, technologists really like solving problems, and there are plenty of people and companies out there solving problems that work with and without Cisco.
Posted by: Mark Cathcart | March 16, 2009 at 11:26 AM
Hi Mark (of Dell?)
No argument -- really.
First, it's not just about 10Gb enet -- there's far more to this picture if you go look closelet.
Second, you're right, there are a few "secret sauce" parts in this architecture, but I see them as "pre-standard" rather than the basis of the business model.
Someone's gotta innovate out there to get the party started -- and that's what I see from the Cisco UCS announcement.
Thanks for writing!
Posted by: Chuck Hollis | March 16, 2009 at 11:53 AM
Yes Chuck, agreed, 10gb ethernet is the entry technology that enables this though, the low latency and no packet drop, are the start point for some very interesting changes in direction. However there are other issues that need to be solved.
For example, at 40gb ethernet, why put memory in servers where it is isolated and can't be shared, but still sucks power? Why not make memory a "server" on its own? Sure you'll always need high speed memory in the core or close by, but a 40gb, close is relative. Rather than think flash drive inside the server, think shared memory cache on the data center network, and on. Time to rethink the server?
Theres much more to the future than just the fact that Cisco are moving into the server, and much more to be lost in the industry if we silo the technology. So, like you, I can only hope these are pre-standard.
And yes, I used my work email address, read back through recent blog posts, you'll get the picture!
Posted by: Mark Cathcart | March 16, 2009 at 12:33 PM
Server memory as a global shared resource? Moving beyond classic SMP? Now that's worth considering ...
Although -- regardless of the technology and the economics -- convincing people that processor and memory are better off seperated, well, there's some serious evangelism work to consider.
Kind of the same evangelism work EMC and I did back in the mid 1990s to convince people that servers and storage should be separated :-)
-- Chuck
Posted by: Chuck Hollis | March 16, 2009 at 12:53 PM
I agree with you Chuck; we're seeing "open" system computing finally catching up to the mainframe world, at least in hardware terms. Now, if open systems would just get their processes nailed down and in place, then they'd really have something...
Posted by: Jerry Thornton | March 16, 2009 at 04:09 PM
Datacenter ethernet is available through a number of vendors, and plugging that into a backplane like HP's VirtualConnect or Flex-10 modules is essentially the same thing. Having virtualized ethernet and storage adapters is very cool, but not unique to Cisco.
It also remains unclear if Cisco is positioned well to be a tier 1 server vendor. Cisco has really been hyping California as something new and revolutionary, but it seems like it's just an entry point into the blade market for them with some new(ish) technology.
Maybe the devil is in the details... would be nice to see some.
Posted by: Dominic Rivera | March 16, 2009 at 04:41 PM
Hi Dominic -- I don't think that Cisco was able to provide full detail with the announcment today. Many of us at EMC have a better view of the bigger picture, so I'd encourage you to wait a bit until some of the other pieces fall into place ...
Posted by: Chuck Hollis | March 16, 2009 at 05:09 PM
Cisco might be looking to buy a storage vendor next! ;)
Posted by: Jaipuk Sedarak | March 16, 2009 at 05:42 PM
Couldn't agree more about RSA stuff too, Chuck. We're pretty excited too, and think it has a big impact in the security world
http://www.rsa.com/blog/blog_entry.aspx?id=1433
Posted by: Paul Stamp | March 17, 2009 at 10:27 AM
Cisco's offering looks still like FrankenComputingSystem to me...
Posted by: Brainy | March 17, 2009 at 03:06 PM
Everyone's welcome to their opinion, which is good.
To me, it's the single most elegant large-scale computing architecture I've seen in more than a decade.
We'll see what people think in a year or so, no?
Posted by: Chuck Hollis | March 17, 2009 at 05:52 PM
Does Cisco's "Unified Computing System " -
mean that Cisco feels that routing (and telecomm in general) products market got saturated and there is only way to expand further - is to go into applications/computing ?
Would it be the same market saturation soon for data storage applications ;-) ?
Posted by: Alexander R. Povolotsky | March 19, 2009 at 01:15 PM