Virtual Cloud Strategy

Do You Need WAN Optimization?

An expert weighs in on how to decide that you need more bandwidth.

• By Greg Shields, Don Jones
• 08/04/2014

Surprisingly, the answer to the title of this column might be, "No."

That's the guidance we learned from Steve Schall, a principal product manager in the Citrix CloudBridge product group. Schall has been in the networking business since 1984, and tells us he's seen first-hand how networking has evolved from its earliest days.

We sat down with Steve after hearing him present his "Rules of WAN Optimization." For a guy paid to sell WAN optimization hardware, we were pleasantly surprised at his candor in what might seem to be a sales pitch not to buy his product:

• Rule No. 1: With 20ms of latency and a maximum of 30 percent wire utilization, WAN optimization will make little impact on the UX.
• Rule No. 2: Increase utilization on that same network to 50 percent, and WAN optimization will have a significant impact.
• Rule No. 3: Conversely, with high latency (500ms), low bandwidth, a maximum 30 percent utilization, WAN optimization will have little impact on the UX.

In a world where a WAN link and poor UX are often synonymous, we asked Schall for further clarification:

Schall: These rules are designed to point out a clear problem that latency and congestion go together in terms of affecting the end-user experience at a remote site. Many people look at latency at being the problem when, in fact, it's congestion that's causing the delays, dropped packets and the slowness in response time.

And so people might be seeking WAN optimization to solve the wrong problem?
Schall: An important realization you can't ignore is that WAN optimization cannot defy the laws of physics. The speed of light is the speed of light. If you have a 250 ms delay between the United States and Sydney, Australia, you've got a 250 ms delay. There's nothing you can do to really fix that.

But, if you have a 250 ms delay and a 50 percent utilized WAN segment, that's bad for users at the far end. In this case, TCP is doing everything it can to protect the integrity of the datagrams. That combination of backing off, slow restart, dropped packets and other tricks starts impacting the UX in ways that doesn't need to happen when utilization isn't high.

What's the rule of thumb for the maximum amount of utilization for a WAN connection?
Schall: I've been in the networking business since 1984, and back when we had 200 users on a 10MB shared segment, the minute utilization got above 30 percent things started going wrong. Interestingly enough, that number still holds true today.

The issue here is what's called congestion collapse. One might expect WAN utilization to increase linearly with the amount of traffic, but as the traffic peaks at around 30 percent, the "good data" actually starts to decline. Take this to the extreme, and at 95 percent utilization your amount of good data is less than 5 percent.

Then what are the use cases where WAN acceleration best solves the problem, as opposed to just buying more bandwidth?
Schall: The short answer is, "when the ROI is about six months or less." Look at the costs of increasing bandwidth and then at the costs of adding WAN optimization. If you can't pay for the WAN optimization in six months or less, you're probably better off buying bandwidth.

What leads you to that short a time span?
Schall: It's the way the market lays out. If your WAN connections stay within domestic borders, your payback is typically longer than six months, more like a year to a year-and-a-half.

It seems Steve's rules aren't necessarily an anti-sales pitch, but instead a real rationalization of the decision-making process that should happen before considering WAN optimization. Your answer might not be "no," but it might be something altogether different than you originally considered.