Trends in NGDC Discussions

Sorry, yet another acronym -- this one stands for "Next Generation Data Center".

I'm now involved in about a half-dozen ongoing NGDC discussions with different customers.  And out of the haze, clear patterns and trends are starting to emerge, sort of a rough consensus as to what's going to be important, and what's not.

All sorts of caveats apply to what I'm about to share (see below), but I'm finding it interesting that (a) so many large IT shops are having these discussions, and (b) the similarities are starting to be more pronounced.

A Bunch Of Caveats Apply

First, we're talking IT at significant scale here.  Most of my list comprises financial service companies who probably spend billions collectively every year on IT.  And, at that sort of scale, you tend to think of things a bit differently.

Second, we're talking longer timeframes here -- depending on who you're talking to, time horizons stretch from 2012 to 2017 and beyond.  That being said, the future can come at you pretty fast: certain decisions have to be made sooner than later, especially at the scales we're talking about.

Third, there's still very healthy debates in many corners -- when talking about this stuff, no two people share the same opinion.  That makes it fun, doesn't it?  But, at the same time, I can feel some trending starting to happen.

Trend #1 -- The Future Is Most Definitely Virtual

Very little argument about this one -- most people are specifically envisioning a scale-out VMware farm for a majority of their computing landscape.  And there's almost no argument on traditional x64-oriented workloads.

What's up for grabs is to what extent virtual desktops will become prevalent.  That's a major chunk of computing potentially moving from people's desktops to centralized data centers, and if you're thinking in terms of infrastructure and real-estate to support, say, 50,000 virtualized desktops -- well, I'd call that a key issue.

Another worthwhile discussion point is just how far up the food chain will VMware scale in terms of large apps?  I mean, look at the progress over just the past few years.  Extrapolate that out another 3-5 years, and you're talking some very beefy applications indeed.

Now, everyone I'm talking to realizes that mainframes will be a key part of their landscape in these timeframes. 

No, they're not going away.  Ever. 

But -- given the potential scalability of VMware and supporting server technology, there are some serious strategic questions as to what extent "Big UNIX" might play going forward -- will we still need proprietary SMP-style UNIX boxen and their brethren?

The potential is tantalizing: two layers of infrastructure computing (VMware and zOS), both fully virtualized, both very feature rich -- and a good degree of shared infrastructure between the two.  Put differently, if large IT shops can get away with focusing on two primary platforms, rather than three or more, most people see that as a huge win.

There's also widespread agreement that infrastructure -- and operational processes -- will look very different in a fully virtualized environment.  More on that later.

Trend #2 -- The Fabric Will Be Converged

Cisco and other vendors have put their weight behind converging multiple data center protocols to run on a single, channelized (presumably 10Gb or 40Gb) ethernet wire.  The idea is "wire once and walk away".  If it works out, enormous savings results across the landscape: better server designs, reduced network complexity, the basis for integrated management, capex and opex reductions -- the list goes on.

You're starting to see the first wave of it now, with all the interest in FCoE.  But that same ethernet wire can be channelized to support server-to-server communications (e.g. RDMA and other clustering protocols) as well as our old friend TCP/IP for standard server connectivity.

Will all the necessary pieces get here in time for these next-gen build outs?  We'll get a clearer picture during 2009, but I'm reasonably optimistic that we'll be safely be able to plan for many of these build-outs assuming a converged fabric.

And, no, I'm not getting paid by anyone to say this ...

Thread #3 -- Information Continues To Grow, and Grow, and Grow ...

The magic of compound growth, no?  Take that innocuous 60% information growth number we all use as a default, and start looking at how this affects total information growth in outlying years:

• 2007 = X amount of corporation information you have today
• 2008 = 1.6x
• 2009 = 2.5x
• 2010 = 4.1x
• 2011 = 6.55x
• 2012 = 10.5x
• 2013 = 16.8x
• 2014 = 26.8x
• 2015 = 43x
• 2016 = 69x
• 2017 = 110x

So, in ten short years, you might be looking at over 100x times the information you've got to handle today.  OK, maybe less -- but also -- maybe more. 

I wish I could find a few investments that could do half that good ...

And I don't think anyone expects storage media to be 100x more capacious, or 100x more cost-effective -- in that same timeframe.  Managing information growth -- rather than just buying more storage -- will become a serious issue in next-gen data centers, as you can see.

The trend?  First, acknowledgment that, yes, this is an issue.  But the best response seems to be multi-faceted.  I'm arguing for improved focus on information governance (minimize what you have to keep), aggressive tiering with newer technologies (insert long list here), increased investment in tools that move information to the right service level (or delete) automatically, increased use of data reduction technologies (single instancing, dedupe, compression, et. al.) -- and closed-loop tools that measure what's actually happening on the ground, so to speak.

Most of these techniques and technologies are in the market today, albeit in limited form.  And getting started on this particular problem can happen well in advance of any specific technology -- or data center build out.

Thread #4 -- Beyond Disaster Tolerant Computing To Active/Active

Anyone in this club needs to keep their IT running regardless of what might happen.  And, since most of the IT processes are interwoven, thinking that only Important Applications need to be discussed, the view has shifted that pretty much all of IT is a candidate -- including desktops, if you're thinking VDI.

The default recovery architecture seems to be a redundant local data center (think speed-of-light latencies), and a longer remote one -- the farther the better.

But no one seems enamored with the idea of all that computing resource sitting around idle, waiting for a problem to happen.  Just about everyone (technology vendors included) believes we can get to a more widespread use of active/active models between data centers; one where workloads can be regularly moved as needed, whether to load balance, or recover something in a big hurry.

If my first assumption is correct (e.g. we'll see a predominantly two-tier computing architecture, VMware and zOS), both vendor's likely roadmaps lead us to this scenario to make this a serious architectural contender.

Thread #5 -- Beyond Management To Orchestration

It's also pretty clear that we won't be able to manage IT the way we largely do today.  Functional stovepipes with serial processes will give way to integrated, parallel processes.  Responses to new requirements will be more dynamic and more autonomous.  IT will get measured on speed and flexibility, in addition to service levels and efficiency.

Not only will the tools have to change (think underlying use of model-based technology), but the roles and responsibilities of IT operations will change as well. 

And, at least for me, the challenge of introducing new technology pales in comparison to the challenge of introducing new behaviors and skills into large IT workforces.

Thread #6 -- Security Built In, Not Bolted On

OK, this one isn't exactly a consensus yet (I'm hopeful), but I think there's an opportunity to redefine the fundamental security model in these next-gen data centers -- evolving from a perimeter-centric model (keep the bad guys out) to an information-centric model (keep tabs on your information).

Like deciding you want CATV in every bedroom in your new house, the costs of doing this during construction are quite minimal, but deciding you want it after the house is built, well -- that's a bit pricey.

I'll keep evangelizing on this one ...

Thread #7 -- Energy Really, Really Matters

I guess $4 gas here in the US has finally got everyone's attention ;-)

No, seriously, there's no predicting the future, but it's safe to say that energy costs (and related social concerns) aren't going to abate any time soon. 

Part of this discussion leads to what you put in the data center: servers, storage, etc.  Another part leads to the construction of the data center: design, cooling, power distribution, UPS, etc.  And, ultimately, it might lead to a discussion of where you put the data center itself ...

I think there's yet another layer of the discussion that hasn't really been explored.  In a world of multiple data centers, and the ability to dynamically reconfigure and relocate workloads -- one  could imagine an extension of modeling technology that incorporated real-time energy costs and needs into the overall dynamic optimization equation.

Actually, I know a few people who are looking at that one right now.  And, if energy costs (and variability) continue to increase, their work may be very popular indeed ...

Thread #8 -- Improve, Then Move?

In most of these discussions, the team is imagining an entirely new facility somewhere.  Which brings up the question of actually moving to the new facility, doesn't it?

When you start to think of what's becoming possible in virtualization and related storage technologies, a case could be made that the data-center ought to be next-genned *before* it's moved.

Why?  Simply because the next-gen technologies make moving all the more easier.  Relocating workloads in an evolved VMware SRM environment is gonna be a whole lot easier than doing it with tapes, as an example.

Thread #9 -- Will Applications Change?

During these discussions, I try to find out a bit about the envisioned application environment.  I'm not having much luck.

Will we see more modular, decomposed SOA-ish applications?  Or will the landscape be dominated by more monolithic application entities?  Or, perhaps, it's a more fine-grained world of web services and ad-hoc mashups?

Will there be more use of video in the enterprise?  Will there be more social computing, both inside the firewall, and outside the firewall?  Will your workforce become mostly mobile, and demand smartphones for their primary "desktop"?

If you think about it, each has the potential to impact how we think of data center architectures in the future, if only in a subtle manner.

Thread #10 -- And, Of Course, The Cloud

Sorry, can't get out of this one without at least some acknowledgment that the emergence of cloud service providers will have a direct impact on next-gen data center discussions.

If, for example, a significant portion of IT "housekeeping" that doesn't differentiate moves to external cloud service providers, the remainder that sits in the data center might be significantly smaller, and -- perhaps -- significantly more specialized.

Or, if cheap compute and storage becomes widely available as an IT "ingredient", you might be thinking about a much, much smaller data center.

And, finally, if there's some aspect of your business that you see turning into a cloud service for others, that's an entirely different data center discussion, since these cloud-oriented data centers are designed to do a few things extremely well, rather than the generic workloads we see in traditional IT.

And, In Conclusion

There's a lot to think about going forward, to be sure.  And, please keep in mind, I don't have definitive answers to any of this.  I don't think anyone really does.

And that's what makes it such an interesting topic ;-)

Comments

[Bill Bonin]

Regarding "Thread #4 -- Beyond Disaster Tolerant Computing To Active/Active":

Active/active has been popular for awhile with some network devices, so there is some real life data.

#1 The most important attribute of active/active is that lets the IT person sleep well at night. With traditional DR, when do we learn whether the DR will really and truly work? ANSWER: After a catastrophic failure. With active/active when will you know whether DR will really and truly work? ANSWER: All the time because the failover node is constantly processing data.

#2 The other really important attribute of active/active is that it's cost-effective. No one is going to build two fully redundant enterprise-scale data centers to support DR. But you might split processing between two data centers, with each capable of being an emergency failover site for the other. That way if your New Jersey data center gets whacked by a freak hurricane, your Utah data center can take over. That's effective DR without buying double hardware, double real estate, etc.

[Mike Dutch]

RE: Energy Really, Really Matters

Does EMC (or even better, the IT industry) have any lobbying efforts in Washington?

"Greeen" is good as far as it goes but tends to take our eye off the ball... growth takes energy and if our industry is going to grow, Energy SUPPLY Really, Really Matters.

[Owen Garrett]

Active-Active is by no means a panacea to failover problems. You still need to over-allocate capacity; in the event of a failover, the remaining datacenter or server needs to process all traffic, so you end up needing as much capacity as you would have in an active-passive scenario.

For many applications where state must be shared, active-active is also more complicated to implement and manage than active-passive.

The cloud computing / utility / service provider model changes the scene completely. Whether you go active-active or active-passive, you need just pay for what you use and resources can be brought on line dynamically. Some cloud providers do this already - expect to see automated migration between datacenters based on capacity, availability and cost in the near future too.

Tomorrow's college kids are going to marvel that organizations used to build entire redundant buildings to house their applications!

[Chuck Hollis]

Hi Mike

As someone with an economics background, I agree with your basic premise: energy is an economic input, just like capital or talent, and inherently more is better.

My impression is that -- in aggregate -- there is more than enough energy available globally. Like other important commodities (e.g. water, food) it's a bit unevenly distributed.

What's hurting are the global side effects of unbridled energy consumption, especially of the carbon-producing variety.

I remember a time when hydroelectric and nuclear were considered "bad", and coal was considered "good".

Yes, I believe EMC has a small lobbying capability in Washington, but I don't think it's enough to move the needle on this particular issue.

[Chuck Hollis]

Hi, thanks for writing. You make some interesting points, but I'm finding myself disagreeing a bit.

First, while you're right that some capacity might want to be overprovisioned (presumably in reserve for a bad day), the discussion usually goes in a different direction.

In many shops, there seems to be a rather long list of workloads that could be temporarily deferred if the capacity was needed for more important tasks: things like test and dev, reporting runs, administrative grooming tasks. Add that to the "buffer capacity" most shops run with, and we're definitely not talking 2x as you suggest.

Your second point discusses potential complexity issues. Many of the people I work with strongly believe that encapsulating workloads in virtual machines allows a consistent approach to orchestrating load balancing and failover as part of the same mechanism.

Certainly, trying to do this application-by-application would be impractical.

Thanks for writing!