How much equipment integration is enough?

One of the most interesting trends to watch over the coming year will be the integration of network functions into a single platform.  And the question that you’ll have to answer for yourself is: “How much integration is the right amount?”

Over the years, a plethora of network edge devices have hit the scene.  Routers, “dumb” DSU/CSUs, intelligent DSU/CSUs, firewalls, traffic-shaping appliances, network performance monitors, Secure Sockets Layer accelerators, and compression devices are just a few that have benefited the WAN.

As time marches on, the functions in these devices are growing increasingly similar.  For instance, a firewall that permits differential amounts of traffic – rather than only binary permit/deny decisions – could be viewed as a traffic-shaping appliance.  Conversely, a traffic-shaping appliance that blocks certain traffic is performing a firewall-like function.  And in some cases, these functions are integrated right into your WAN access router, which has its own core set of basic transport duties to perform.

So what is the optimal number of edge products?  From a performance perspective, there can be advantages of integrating certain functions into a single box.  In fact, we highlighted some test results from both Packeteer and Redline back in December.

At the same time, there’s the obvious risk of having too many of your proverbial eggs in a single basket.  Many network managers like having the separate pieces of equipment so each can be controlled and diagnosed as a discrete functional unit – and to minimize the single-point-of-failure threat.

Having too much integration also diminishes the extent to which an independent view of the network is possible.  Back when Steve was involved in network operations for the University of North Carolina, his boss described the datascope that sat between the network equipment and the WAN as the “de-liar.”  This device provided an absolutely independent view of the network with no bias as to how a given piece of equipment was performing.

Further, since the analysis equipment wasn’t borrowing resources as an embedded function, there was no question as to whether the act of observing the network was having an impact on performance. 

So we’d like to hear from you.  What do you see as the optimal amount of integration?  Are you ready for an omniscient “god-box” at the edge of your network?  Or do you prefer separate devices with specific functions?