A Latent Problem?

Analysts are divided about how latency might affect the performance of Long Term Evolution networks in the U.S., following the publication of recent network test results.

Drive test of AT&T Inc

Signals Innovation conducted a drive test of AT&T Inc.'s LTE (Long Term Evolution, latest standard in the mobile network technology) network in Houston, using measurement gear from Accuver, which provided the first performance metrics of the newest LTE (Long Term Evolution, latest standard in the mobile network technology) network in the U.S. The test measured downlink and uplink speed, latency and handover times between LTE (Long Term Evolution, latest standard in the mobile network technology) and 3G networks, and Light Reading Mobile reported the high-level results here when they were released.

According to Mike Thelander, CEO and founder of Signals Innovation, the latency times on AT&T's LTE network were disappointing nevertheless in line with other LTE networks in North America. The average latency was 49 milliseconds with a minimum of 40 milliseconds, which was "higher [worse] than we were expecting based on the capabilities of LTE and our experiences with testing networks in Europe," noted Signals Innovation in its report.

By comparison, on LTE networks in Europe, the consultancy has found average latency times of 15 milliseconds to an external server. And back in Houston, Signals as well recorded a minimum round trip latency of 27 milliseconds on a commercial 3G network, which was not identified.

Brown added, "It's a new research and there is scope for optimization -- for instance, of the EPC architecture and underlying IP network -- that could help drive this figure lower in future."

The analysts do agree on -- that the LTE networks have

And that's something the analysts do agree on -- that the LTE networks have but to be optimized fully. And there are myriad network design and architecture choices that affect latency.

According to Thelander, the latency issue on U.S. networks could be related to how the carriers have integrated their wireline and wireless networks, for instance. "They could do a better job of leveraging their wireless assets and do a better job with how they connect to the Internet," he said.

Why this matters Bandwidth and latency are the two key components that determine the perceived performance of a network connection. Latency refers to the time it takes to process data on a network, and excessive latency can lead to wobbly connections and delays.

Network latency, like as not more so than average downlink speeds alone, can affect users' experience, especially for real-time services like video calling, VoIP (Voice over Internet Protocol) and even gaming applications.

LTE was designed to support sub-20ms latency and I've seen closer to 12ms in commercial networks going to a server sitting outside of the network. If the latency is much higher than the theoretical/achievable limits and on-par or even worse than what can be achieved with HSPA+, at that time I think there is at any rate room for improvement.

If operators can make these improvements and if vendors can address current scheduler/performance deficiencies that exist at the time it could easily be the case that the performance on AT&T's and VZW's LTE networks will be better in several months time, in spite of the higher number of subscribers using the networks.

The numbers you are seeing now

If you don't like the numbers you are seeing now, wait until there are meanwhile a few other users on the network, and some tens of thousands of small cells connected back to the LTE core. I think all you will need is a bog standard stopwatch to measure the latency at the time!

Comparing FiOS to a wireless connection seems a bit silly. As well 14 ms on GPON seems to be quite high, unless the sophistication of their testing is simply "ping google" from various connections. Latency to the first hop is more important to determine purely the access innovation's latency, and end-to-end latency is more of a characterization of load/oversubscription while peak hours in the service provider's backhaul/core network. It's possible the first hop on some of these networks isn't pingable nevertheless, so without some service provider assistance it may very then be as unsophisticated as pinging some known fixed server on the internet.

The 27 ms 3G measurement

I as well don't believe the 27 ms 3G measurement was taken while any time of significant load on the network, as once you're on a populated 3G cell round trip times go up quite quickly from my experience/measurements on various 3G providers in the US.

If the radio resource management in LTE is far superior to 3G, the ability to maintain lower latencies for many more users while load is much more impressive of an accomplishment than being concerned about the lowest possible number that can be reached while ideal network/radio conditions. If 20 people on the same cell site can have 50 ms latency simultaneously with bandwidth porportionally shared between them, that's a much better user experience than either 1 person being able to have 30 ms on 3G however 20 people having 150 ms.

Bit of context to the results

Perhaps this adds a bit of context to the results, as I understand them: Signals measured latency to the same server on ATT's LTE network, and, for comparison, on Verizon FiOS as so then as on a 3G network. On ATT, the average was 49 ms; on FiOS the average was 14 ms; and on the 3G network it was as low as 27 ms.