Using The Mobile WAN for apps? Look Out For Latency!

SnailThe mobile phone operators promise us blinding speeds, especially with their newest 4G and LTE networks, but will this actually be any use to business — or will deploying your modern apps via smartphone without WAN acceleration make your users feel like they have stepped back into the Dark Age?

For the mobile networks, the elephant in the room is latency: how long it takes for a data packet to traverse the connection. There’s a reason why the networks prefer to talk about download speeds and how fast you can pull stuff off iTunes and YouTube, rather than anything more business-useful, and it’s that their latency is too often rubbish. In fact, the latency on current 3G networks can be about the same as the latency involved in sending data halfway around the world.

Worse, it can vary massively, doubling or halving seemingly at random. As WAN managers should already know, with business applications this means unpredictable response times and therefore frustrated users.

Don’t believe me? Try it for yourself. I recently spoke with someone who did exactly that, Greg Howett of JetNexus. His company does application delivery controllers — a form of accelerator that often complements the work of WAN optimizers such as Silver Peak. He had been using apps via his mobile phone and finding them sluggish, so he put on his engineer’s hat and went investigating.

What he found astonished him: ping times on all five (officially four, as T-Mobile and Orange have merged, but they still seem to have a fair amount of separate infrastructure) of the UK’s 3G networks were 200-300ms. That is slower than pinging India from the UK, and about the same as pinging Australia or Singapore, and it means that if you access an app via your smartphone, you can expect about the same latency as if you had hosted the app on the other side of the world.

Of course, in the grand scheme of things this is still pretty good. Those of us who cut our mobile data teeth on 1200bit/s analogue cellular modems were delighted when GSM came along, offering digital connections at 9600bit/s, and with sub-second latency.

Dark Age mobile data

That really was the Dark Age of mobile data, however, and for modern web apps, especially for line-of-business tools, 300ms latency is just not good enough. Sure, it is fine for streamed audio or video, but a web app can make a dozen or more round trips to assemble a single page, which could add up to several seconds per page, never mind that you have ‘unlimited data’.

Why is it so bad? After all, 3G — and especially the HSPA (high speed packet access) upgrade that is sometimes referred to as 3.5G — was supposed to remove the latency issues that we had with 2G GSM/PCS data and its 2.5G upgrades EDGE and GPRS. We were told that 3G would offer latency in the 50-250ms region, but now we discover it is much nearer the latter than the former. (With the benefit of hindsight, that was an absurd promise anyway as it includes an order of magnitude’s variability.)

Part of the problem is the complexity of the network engineering and the many trade-offs and network elements involved, both wired and radio. For instance, research by students at the University of Edinburgh a few years ago pointed a finger at performance-enhancing proxies (PEPs). Installed as part of the HSPA upgrade process, these gave faster downloads but also appeared to break the end-to-end TCP connection and increase latency.

Greg is sanguine that the problem is rather more relevant to the WAN optimizers than it is to his company — although he says there are still some application acceleration techniques that could usefully be applied, such as in-lining small images and merging all the JavaScript on a page into one file. At root though, he says it is a WAN problem — and one that could come as a nasty surprise.

“I think that enterprise service delivery architects and network managers are all very familiar with the latency issues of delivering interactive apps over the WAN, but I suspect they are not so prepared for the impact of delivering them over the mobile,” he explains. “They may assume — as did I —  that the performance is fast and consistent. Given the growth of mobile as the new business terminal they will start to get application latency issues closer to home.

“I think the other problem is the variation in end-user experience due to this unpredictable latency. Users can deal with reliably slow apps — they moan a lot, but they get used to them — but what really makes people unhappy is what we call the performance delta. One day its nice and fast, setting high expectations, and the next it is running like a dog.”

The answer is, of course, that we need to design and deliver applications that have a performance profile that is reliable, and is not so sensitive to latency. Fortunately, all the skills and technologies to do this already exist, such as WAN accelerators and ADCs, although some of the skills date back to those data Dark Ages, so they might need a bit of dusting.

And telecoms buyers need to tell the network operators to stop always talking about bandwidth only, and keep pushing them for better latency. They can do it if they try!

Image credit: tamburix (flickr) – CC-BY-SA