Earlier this year, WAN optimization developer Silver Peak Systems Inc. joined EMC Corp. and network solutions maker Brocade in Boston to talk about distributed storage systems, disaster recovery
Silver Peak took part in the meeting of the network minds because it has built a business around developing enterprise-class WAN performance solutions that reduce such things as packet loss and latency. The company is also fanatical about generic WAN optimization solutions (as opposed to application-specific) and real-time forecasting tools that help spot small problems before they get bigger.
Long after the dessert plates were cleared, SearchEnterpriseWAN.com site editor Tim Scannell spent some time with Silver Peak president & CEO Rick Tinsley to talk about generic WAN optimization approaches, virtualization and file acceleration techniques. What follows is an excerpt of that conversation.
What exactly is wrong with an application-specific approach, especially if a company is just looking at optimizing key applications?
Rick Tinsley: The problem with application-specific techniques is they tend to be very brittle. They may solve today's problems and perhaps may do so very, very well. But chances are that application is going to evolve, and the next release of that application may not be as compatible. And certainly we know that no matter how many engineers we have, we couldn't build a library of application-specific shims or application-specific modules for the myriad of applications that are out in the industry today. And we certainly couldn't keep up on that treadmill as those applications constantly change.
So we think the notion of using application-specific techniques is wrong and will never actually be successful in volume for networking equipment.
The term "WAN optimization" has evolved to include a range of technologies and is not just focused on increasing speed on the network. What are your views on what is included under the umbrella of WAN optimization?
The problem with application-specific techniques is they tend to be very brittle. They may solve today's problems..but chances are that application is going to evolve.
President & CEOSilver Peak Systems
Tinsley: It's a good point. As an industry, we've all struggled for the right label for the business that we are in. Optimization means a lot of different things to different applications. Different applications create different challenges, and different wide-area network conditions create different challenges. So our goal as an enterprise-class WAN optimization vendor is to build generic capabilities and techniques that work across a large variety of applications and network conditions.
In a large global enterprise, for example, you will have a very large and dynamic mix of apps and different network conditions, and both of those will change over time. You'd better have a generic technology that can handle problems today and tomorrow.
How difficult is this from a solutions standpoint? Does this mean each WAN optimization deployment requires a high degree of customization?
Tinsley: Customizing application-specific shims (coding that transparently intercepts an API to maintain compatibility- Ed.) is absolutely anathema to us. That's absolutely the wrong way to go, and we know we are at odds with some of our competitors in that regard. But a lot of our competitors started out with an application-specific product, which may have been a shim to make email run faster or a Web cache that would help with http object caching.
We build plumbing, and we can handle all of the traffic that runs through it and can optimize most of it. We may not be able to optimize every single application in every single scenario, but in no case are we incompatible with anything that runs on IP.
In an effort to reduce capital expenditures, many companies are using virtualized WAN optimization appliances, especially in branch offices and at remote sites. Obviously, this has an impact on the WAN appliance business. What is your view on this trend?
Tinsley: Virtualization is interesting but has to be the most overused buzzword in the industry today. There is server virtualization, which everyone in the industry has had experience with, and the ROI is pretty straightforward. When you get into desktop virtualization, however, you're going to find that your mileage will vary and that it will vary tremendously. Desktop virtualization, along with data center consolidation, is one of the things that is increasingly creating opportunities for us.
When you consolidate data centers, and you go to virtual protocols, the average user is usually farther away from the source of the data and where the data is stored, and that creates challenges. When this happens, the network limitations become far more onerous and far more apparent. So, investing into the network to make those virtual services more effective, more palatable and more acceptable to the users is a huge part of our business going forward.
In terms of virtualizing the network element, this is where the marketing people tend to get a little bit ahead of themselves. We went through this a couple of years ago when some of the vendors were talking about having server blades in their boxes. If you ask people who run networks, most do not want a Windows server on their router. When we went through our own internal server virtualization process, we found that some apps lend themselves very well to virtualization and that you can truly get better server utility and better ROI from these applications.
Virtualizing network elements—like routers and switches and WAN accelerators—is one of those things that makes for a good PowerPoint and good marketing, but I'm not sure where it's going to go in terms of actual deployment.
FTP acceleration is an option for some users, as opposed to full WAN optimization. Are you sensing more competition from this area, especially in this economy?
Tinsley: There are a lot of small companies that can make FTP transfers or TCP run faster. The industry has been through enough experiments—either application-specific accelerators or protocol-specific accelerators—that haven't really stood the test of time. In the late 1990s, for example, Web caches were all the rage. With the proliferation of the World Wide Web, we thought we'd need caches everywhere to store objects. As it turns out, Web content is not that static, it's dynamic—especially if you think about real-time ad insertion and everything else.
In the mid-2000s, we went through an experiment with file caching, but all those companies were acquired by larger companies, and those products were pretty much phased out. What enterprises learned is that while these things are pretty interesting at a demo level, if you actually deploy them on a large scale, you run into synchronization problems and multiple users added to the same file—and bad things can happen if you are both trying to store at the same time.
The technology was very alluring, but it proved not to be an effective enterprise-class solution. These products were also relatively expensive and relatively difficult to configure, and the hardware of this technology was essentially like another server.
So the point is, if you are going to deploy an object cache, why not just deploy servers in the first place? Deploying a caching-based technique or a proxy-based technique sort of flies in the face of server consolidation.
What do you view as the biggest mistake enterprise users make today in terms of the network and network acceleration?
Tinsley: They have an application problem over the wide-area network, which they assume is a bandwidth problem, so they go buy more bandwidth and a bigger pipe. But their problem doesn't improve because maybe it's packet loss, maybe it's latency, or maybe it's something else. They really don't know the underlying limitations of the network. With the right tools and visibility, people can make far more informed decisions in terms of capacity.
Where do you see the biggest opportunities for WAN optimization over the next three to four years?
Tinsley: The biggest opportunity for us (relative to EMC and Brocade) is enabling customers to perform SAN replication on their corporate network. Most large enterprises today buy a separate dedicated network or private-line bandwidth just for SAN replication. For most enterprises, it is probably the last remaining application-specific network in the whole company. We can really enable people to collapse those two networks down to one network.
It's the same kind of convergence we've seen with voice and data—which used to be two separate networks until they collapsed over time—and today most people run voice over their data network. The opportunity is to allow that same sort of network convergence for SAN replication.