Last month, GCJ editors attended North America's largest IT infrastructure event -- Interop -- to hear what networking professionals are saying and hearing about Grid computing and related emerging technologies. This year, the event's first-ever "Data Center Summit," produced by analyst firm Nemertes Research, included a number of Grid, virtualization and related next-gen datacenter discussions (including a Globus presentation by Carl Kesselman). While there, GCJ had the opportunity to catch up with Andreas Antonopoulos, Sr. Vice President at Nemertes, to hear his take on the confluence of Grid, virtualization, and the various themes that are on networking pro's minds these days.

GCJ: Congratulations on the Data Center Summit that Nemertes organized at Interop. Every session that we sat in, the room was packed.

Antonopoulos: There's no question that data center managers are really looking at how to use emerging technologies, best practices, and new organizational approaches to make their datacenter shine. Grid is one of the new emerging technologies that people were specifically interested to hear about, so we were thrilled that Carl Kesselman could join us to educate them about Globus and Univa.

At the same time, one of the takeaways in the conference was that on the road to Grid, there are many steps that a company can take. A lot of companies -- without necessarily having Grid as their ultimate goal -- are beginning to take steps to make their datacenters more flexible. They're looking at server virtualization, storage virtualization -- and increasingly network and security virtualization -- for better ways to allocate, provision and re-provision resources.

As the typical large enterprise continues to face massive consolidation in the datacenter, more distributed user bases, and other density and planning issues -- datacenter managers are dealing with a much higher demand from the business units, both in terms of the services and the availability of those services. At the same time, they don't have equivalently corresponding budgets. So it's reached a point where managing your datacenter is extremely challenging, and IT managers are actively seeking new approaches.

GCJ: Grid is still a tough sell for the mainstream enterprise audience. There are a lot of different ways to skin a cat -- and as you mentioned, many organizations are leveraging virtualization without necessarily having Grid as an end goal.

Antonopoulos: Yes, but I see the technologies as synergistic.

One aspect of virtualization that doesn't get enough discussion is that while it can make your datacenter more flexible, there are also new issues of design versus run-time optimization. In the past, what we did with the infrastructure was design-time optimization. We'd build the infrastructure for the specific application. If you wanted ERP, you'd build your three-tier or four-tier system. You'd design every component to deliver to the capacity that you'd need. And then you'd turn it on and expect it to meet the demand.

Now -- as we're doing this flexible datacenter with pools of servers and storage -- that optimization happens at run-time. We're deciding how to allocate the resources to achieve the demands at each moment, because now there's nothing really pre-configured to run your application in that infrastructure. It's all shared.

What that means is that we're putting increased pressure on our management platforms. So now we're trying to do provisioning, monitoring, change management, service activation, load-balancing and workflows all from one monolithic management platform. And the result is that many people have found that their management systems are inadequate.

What I've been talking about at the conference is that virtualization alone doesn't bring you any savings. It creates a standardized platform which can then be managed more easily -- which means that if you add automation to it, you can move virtual machines around, you can create configuration copies, pull them out of the freeze-dried configuration you put in your SAN and make changes, etc. If you do all of that with automation, then you get savings. Without the automation, virtualization just increases your complexity.

So we've reached a point where existing management systems -- with or without virtualization -- are proving very difficult to run. And the old sweeping vision you get from the big vendors about an integrated environment where pretty much everything is from the same vendor, or with small variations, is becoming impractical. The problem is that such an environment no longer exists. Or if you do manage to create such an environment, you're one merger away from ruining that nice standardized platform you've built.

So the problem is how do you manage all of this?

And Grid says that instead of a top-down approach, use a bottom-up approach -- a middleware solution that does the allocation and the workload management in that virtual pool, and is really designed for very heterogeneous platforms, which may not even have a single owner. Radically breaking from that top-down mold is what Grid does best, because a lot of the principles are very much peer-to-peer and creating ad hoc networks and machines when you need them to solve specific problems or do specific jobs. It doesn't require the kind of command-and-control architectures that most of the major management systems imply or assume. And because you can't really apply those sorts of command-and-control architectures very nicely to the datacenter, maybe a bottom-up approach is better. So I think there's a lot of potential for Grid in helping solve the complexity crisis.

The other key issue is the problem with the way we're managing the datacenters today -- you have all of these small, interdependent components at the low-level which create these emergent systems. So you have these high-level business services that depend on dozens or even hundreds of underlying components -- all of which may provide multiple services, and also may depend on multiple other services. And all of these complex interdependent interactions create highly unstable systems. They create systems where a single problem down in your infrastructure could have a cascading affect, and could topple several of your business services. And the difficulty here is that tying the top-level business services to the elements in your infrastructure is enormously complex. And you've got an environment that's forever changing.

So part of the issue there is that you can't really make assumptions about what your infrastructure will look like, or what the demand for business services will look like. So the top-down approaches are beginning to reach the limits of their usefulness, and people are beginning to turn to other approaches.

GCJ: So do you see Grid eventually starting to displace some of the big, monolithic management platforms?

Antonopoulos: Perhaps.

As we look at the research that we've done with IT executives -- we find that even with people who have adopted some of the large management platforms -- they typically have adopted only a very small part of them. They might use the network monitoring piece, for example. Or they might use the server provisioning piece.

But these types of systems are not very commonly adopted whole-hog. Part of the reason for that is the complexity of getting such a system up and running. The complexity of configuring it and fine-tuning it -- and doing this while everything around you changes, both in terms of demands, and in terms of your infrastructure - is too much to overcome for the vast majority of organizations. By the time you actually finish the project, it's no longer corresponding to your strategic goals, and you have to start again.

Grid, on the other hand, assumes from the get-go that these resources will not be available all the time. It assumes they will be available on different platforms with different capabilities, and perhaps owned by different organizations, with different security policies. Grid may be more suitable to dealing with the future needs of enterprises. So I see a very powerful reason to look at Grid, even within the other virtualization activities that you're doing within the datacenter.

And beyond HPC, many companies today are dabbling in accumulations of large amounts of data -- whether that's RFID-fueled, or medical imaging data, or whatever. And data mining and business intelligence on that data is a key application that we see many companies trying to deal with. Of course, very computationally-intensive and network-intensive and storage-intensive requirements ... those types of jobs certainly have a lot of similarities to Grid applications in HPC and science. And therefore, we should start to see Grid get more enterprise traction in these data-heavy areas.

GCJ: Speaking of application development - there are still not many enterprise Grid applications to point to. Where do you see the application development trends heading? And how will this tie back to Grid?

Antonopoulos: The truth of the matter is that we've effectively rediscovered the maxim that "interoperability is cheaper than integration." This comes directly from Peter Yared's (ActiveGrid CEO) speech at my session. What we see is that we can do a lot of things with interoperability in the consumer world that we can't even imagine doing with our enterprise applications, such as ERP and CRM. And that's a pretty glaring difference that shouldn't be there.

What I'm seeing in the industry is that all of the big software vendors are beginning to take those monolithic applications which internally are modular in some fashion -- and making them radically modular, and at the same time implementing either the internal web services interfaces or SOA as their development model. We see this across many of the large enterprise software vendors. What that trend does is enable vendor interoperability and software re-use.

As Grid has adopted web service technologies within the Globus Alliance and GGF, I think you're starting to see the convergence of software development practices on one hand, and the need to manage the software lifecycle and the provisioning of resources on the other. Because the more fine-grained you get at the low-level, the more complex it becomes to do top-down management.

If instead of one 4-tier web app for ERP you're running in fact a collection of 150,000 web services that create a composite ERP application, you have enormous flexibility to move those services, re-use those services, manage and back-up those services separately -- and really smooth out demand across those services and make your applications much more responsive and agile from a business perspective. All of that is great, but if you try to do that with a top-down management approach, you're either going to manage just the elements (which gives you no visibility into business service management), or you're going to try to tie the elements to the business elements (which creates enormous complexity).

I see an alignment of the fine-grained, loosely-balanced, software-as-a-service model that we're seeing emerging out of SOA. Even though enterprises are not deploying that whole-hog now, their software vendors are internally at least, and will gradually start externalizing it. So there's this very gradual confluence of that and the direction that Globus is taking -- where Globus will emerge as one of the best platforms to manage these enterprise applications when they are SOA.

If you look at SOA, it has a layer in there for orchestration and service discovery and delivery -- which is not the service delivery itself, it's a meta layer of how you manage and deliver that service and how you orchestrate all of the small services to create that composite app. That layer is where I envision Grid fitting in -- it fits in very nicely to the architecture model.

close window