There is a tendency to describe every device on a network as simply “the network” regardless of whether that device is dedicated to security, or application delivery (layer 4-7), or actual network (layer 2-3) functionality. It’s an artifact of aging data center architecture models that there exists an artificial line of demarcation between web and application servers and everything else. We used to depict “everything else” as a cloud, but with the emergence of The Cloud doing so simply complicates discussions even further because the “network” necessary to support a dynamic, on-demand operational model of computing like “cloud” is more complex. Load balancing and application switching functionality is no longer a “nice to have”, it’s a “must have” in order to properly abstract applications such that users can remain blissfully ignorant of the inner workings of the cloud. But it’s rarely broken out that way; everything that’s not a server is evidently part of “the network” regardless of where its true demesne within the architecture may lie.
The problem with the “tossed salad” approach of discussing “the network” is that there are very real differences between the network (layer 2-3) and the application network (layer 4-7) that make such a mixing of solutions problematic. Certainly every device on the network is capable of performing some layer 2-3 functions else it could not reside “on the network” in the first place, but that does necessarily mean that every device on the network is focused on providing additional value above and beyond what is necessary to forward packets in the most expedient manner possible.
A network is primarily concerned with the routing of packets from one host to another through a series of connections. It may apply some logic to that routing in order to comply with business goals related to response time or availability, but in general the “network” doesn’t really care whether the application is SAP or SharePoint or a custom application. It only cares about the packets and how best to route them to the next hop. The only time a network really cares about the application is when it needs to decide which packets to forward first when there is limited bandwidth and then it only cares about the application in so far as it can distinguish between SharePoint and SAP so it can make a routing decision.
An application delivery network, or just application network, is less concerned with the routing of individual packets (it does it, but that’s not its primary concern) and is more focused on the entire application stream. It must know what application is being processed because its decisions about how to treat an application stream are almost always derived from a policy that is based on the application. An application network must be intimately familiar with all the particular nuances and quirks of an application and its associated protocols – from layer 4 to layer 7 to the application itself – because much of its value is in being able to provide application specific optimizations, acceleration, and security. The behavior of an SAP application, for example, is much different than that of SharePoint. The usage patterns are different, the interaction between client and server is different, and the very way in which these applications exchange data is different.
What an application network does is provide an overlay of intermediaries (brokers, proxies, whatever you’d like to call them) that essentially mediate between the client and the server and the application and the network. It’s a layer of functionality over the network that understands both sides of the equation – application and network – and uses information from both to make decisions regarding the best way to ensure an application is available, fast, and secure.
LET THEM USE THE NETWORK
Now usually what happens in discussions is that some folks just blithely say “that functionality belongs in the network”. But what most people think of when the term “the network” is used is routing and switching and thus reasonable folks will immediately dismiss the notion of putting X functionality “in the network” because it doesn’t make sense. Web application security, for example, doesn’t belong in a router or switch because those devices are packet-based and generally unaware of the nuances involved with inspecting web application data for SQL injection or XSS attacks. Similarly, load balancing “in the network” is associated with port trunking/link aggregation or assuring availability through the use of two or more ISP links, not application-specific switching and routing.
The use of “the network” as an umbrella term to describe a varied range of functions and responsibilities across the networking stack is as confusing as the use of “the cloud” to refer to the myriad data center architectural models emerging today. The lack of qualification of “the cloud” in discussions causes as much confusion as does the lack of qualification around “the network.” Consider as an example the differences between network acceleration and application acceleration as an incarnation of this particular problem. The two are very different technologies and yet they are often lumped together as just “acceleration” without any care or concern for not only the technological differences but the distinct business value each adds.
The reason this is so very frustrating is that as soon as an application network device is described as “in the network” the very people who might benefit most from much of its functionality – developers and application architects – immediately dismiss the solution as something outside their demesne to be relegated to the network operations folks for further exploration. That’s a problem because it means that the application half of an application network is often summarily ignored in favor of the network half of the device because that’s what the network operations team understands and has responsibility for.
EVOLUTION ISN’T JUST FOR TECHNOLOGY
The real issue here is that as technology has evolved and resulted in new solutions the terminology we use to describe it has not. Because we continue to use the same terms to describe completely different foci it is nearly impossible to have a meaningful discussion without first coming to an understanding of what we’re talking about. A “session”, after all, has many different meanings depending on what layer of the network stack you are focusing. The lack of qualification wastes time that many people today just don’t have and allows for the manipulation of terminology to the detriment of the end-user/customer.
The “network” is not the “application network” – and vice versa. The benefits, roles, and responsibilities relevant to the delivery of applications are very different at layer 2-3 than they are at layer 4-7. Lumping the two together makes it difficult to fully realize the benefits of the latter because it ends up treated as something it is not because we all have preconceived notions about what a “network” can and cannot do. And it also makes it difficult to leverage layer 2-3 to its fullest potential because either we expect it to be more intelligent about things outside its domain (like application specific functionality) or we treat it like a dumb pipe, neither of which is a good way to view the “network.”
Maybe I’m just being hypercritical (wouldn’t be the first time) but it makes things extremely difficult when two groups of people have a completely different understanding of what “the network” means when we’re trying to talk about an application delivery network. One of the goals of SOA was to provide a terminology such that business stakeholders and IT could start from a common understanding and collaborate on solutions because it was recognized that all of IT and the business owners needed to work together to leverage the concepts of a service-enabled architecture. We need to do the same and not only share terminology with the business but also within the disparate silos that have formed over time in IT.
There is a difference between the “network” and the “application network”. It’s not all that dissimilar from the difference between the tiers in an application architecture. We don’t just say “server” because that could mean “web server”, or “application server” or “database server”. We spell it out. The same is true on the “network” side of the house: there are tiers within the infrastructure architecture that need to be recognized if we’re going to understand how to fully leverage a dynamic infrastructure and on-demand computing in the future.
So mean what you say, and say what you mean.
Note: Yes, I know that layer 5 & 6 in the OSI model above are “missing” and essentially incorporated into layer 7. That’s because we really don’t use those layers individually, they’re treated as layer 7 in what most folks refer to as the TCP/IP stack or model, which has only the 5 layers. Consider this view of the stack a hybrid of reality and theor.