Michael Vizard over at eWEEK makes an interesting prediction about the future of application acceleration: "Some day the whole concept of application acceleration will be baked into the core routers and switches we have in place."

I disagree. Routers and switches are packet-based. They focus on getting a single packet from here to there based on layer 2/3 information. Application acceleration solutions require action higher in the stack, usually layer 4 through 7; they are flow or connection based, and are often specific to the application (think CIFS, SAMBA, HTTP, etc..). The information necessary for application acceleration solutions to improve the performance of applications is just not available at the layers of the stack at which routers and switches operate.

While in the case of HTTP it is generally possible that some application acceleration features can be employed, such as manipulation of cache-oriented headers, simply because they are contained within a single packet, routers and switches don't have the awareness of that layer to be acting upon it.

But, you say, Michael is postulating that they will, someday.

Again, I'm going to disagree with that prediction. We've already tried to apply application performance-oriented functionality in routers and switches. The TOS (Terms of Service) bits (which have been redefined in recent years as Differentiated Services Code Point (DSCP)) in an ethernet header can easily carry basic information regarding the quality of service desired for a particular packet. This was initially used by routers (and still is in many implementations of WAN optimization solutions) to determine prioritization and handling of individual packets.

"Coloring TOS bits" was a euphemism used nearly ten years ago to describe the common scenario of prioritizing packets based on customer service level (gold, silver, bronze) by flipping TOS bits in the ethernet header. Problem was that most routers and switches didn't honor TOS bits, and there was no way to enforce the quality of assurance indicated by the TOS bits once the packet hit the Internet. Some routers simply didn't interpret TOS bits and ignored them.

An entire market of products arose out of the need for QoS solutions that actually worked. Bandwidth management products like Sitara and Packeteer rose from obscurity and quickly saw the need for more application specific prioritization and immediately "'moved up the stack" into layer 7 in order to provide that functionality. TOS bits were discarded by the market as a whole for being unable to provide the level of specificity required to manage applications. TOS bits have made a bit of a comeback with the advent of WAN optimization as a hot technology, but I suspect that it, too, will go the same route (sorry, no pun intended) as its forerunners earlier this century in application acceleration.

Application acceleration in a router or switch would require that the router or switch become layer 7 aware, and flow-based. If they were to attempt to apply application acceleration to XML-based data, they would have to go one step further and become a full proxy. They would have to become an application switch rather than an L2/3 device.

This is a bad idea for several reasons. First, there already exist a plethora of application switches that more than ably perform the task of application acceleration. Second, the core packet-processing hardware and software of switches and routers is optimized to pass packets, and would need to be completely reworked in order to provide application acceleration functionality. This would drive the cost of core infrastructure higher and unnecessarily add cost for enterprises who certainly only need application functionality at specific points in their architecture, not in every wiring closet on their campus. Third, the process of packet aggregation required to act upon application data rather than individual packets necessarily adds latency. In many cases it's barely noticeable at any given hop, but when it happens at every point device through which a packet must travel it adds up and will certainly be noticeable to the end-user.

Routers and switches are the core of the network, of the infrastructure that makes networks and the Internet "just work". Adding unnecessary functionality that requires a dramatic change in the architecture of these devices is simply a bad idea, financially and functionally. Routers should route, switches should switch, and application acceleration should be left to solutions specifically engineered to provide such functionality.

 

Follow me on Twitter View Lori's profile on SlideShare AddThis Feed Button Bookmark and Share