Thursday, May 18, 2006

ADSL for HSDPA backhaul - bad idea

I am just back from a marathon travel binge - lots of conferences, customer meetings and trips to HQ.

I spent last week in Berlin at Transport Networks for Mobile Operators (TNMO) which is a fantastic show put on by IIR. It is a very well-attended conference focused solely on, you guessed it, transport networks for mobile operators. Backhaul and core network transport.

Most 3G operators are launching HSDPA now or very shortly. This poses a big problem in terms of backhaul capacity since most cell sites have 2 or 3 E1s (4-6Mbps) connecting them to the core, but HSDPA will provide 1.8M or 3.6Mbps to each user (depending on how the operator implements it). You do the math - the current backhaul capacity just won't cut the mustard.

So many operators are adding another E1 and plan on contending the service similar to your residential broadband (20:1, 50:1). That could work for the short term. But the problem is only postponed as you get more and more users (up to 300 per cell) demanding bandwidth and HSDPA evolves to higher bit rates. Simply adding more E1s is too expensive and TDM doesn't transport IP traffic efficiently.

So a lot of operators want to split off HSDPA traffic and backhaul it separately. This makes a lot of sense - keep your best effort data traffic separate from your real-time voice and video. Backhaul for HSDPA doesn't need such stringent QoS as that for voice.

I am a big proponent of using Ethernet for this backhaul (I'll post a separate link as to why I think this is good). But at TNMO a LOT of operators were musing on the benefits of using ADSL: It's cheap, it's all IP, it's ubiquitous and mature, it's asymmetric, etc....but mainly the cost element is what's so attractive.

I think ADSL is a bad idea. Here's why:

1. HSDPA will be contended as I already said. Commercial ADSL services are also contended. QoS will suffer big time.

2. HSDPA will be used for streaming video. DSL providers hate streaming. This is why they all put usage limits on their services.

3. You might say, "Hey, I'll just negotiate an SLA with unlimited usage and zero contention." OK, fine. But how much will that cost? The reason ADSL is cheap in the first place is because of contention and usage limits. It minimises the amount of infrastructure the ISP has to build and operate.

4. OK, say you manage to get a good SLA. Then you start rolling out hundreds of HSDPA sites in high traffic areas (which also happen to be high traffic areas for the ISP. This means capacity bottlenecks for them). I know of NO ISP's network that can cope with the kind of high-capacity streaming traffic that HSDPA will generate.

5. How do you manage hundreds/thousands of DSL modems?

6. It's yet another overlay network. Mobile operators already know the hassle of overlays from their 3G Rel. 99 experience.

The only time it might make sense is if the mobile operator builds its own DSL network. Dry copper is available in many places now. The business case seems to be sound but I am not sure of the technical problems of self-build. What about operators with sister ISP companies? I think they'll run into the same issues I've outlined above.

I am just not convinced that you'll be able to get the SLAs you need for carrier class ADSL with good QoS. And the management of the network will be a nightmare.

Ethernet is the way to go. It's also cheap, it's also high capacity, and it's carrier class (well, it will be soon). And oh, by the way, you can use Ethernet to backhaul ALL of your TDM and ATM traffic too using 802.1p/q VLANs. Sure, ubiquity is an issue, but the fixed line guys are catching up.

1 comment:

Martin said...

Hey Kevin,

interesting post and I had to think about it every now and then since I first read it in May. I've finally managed to put my thoughts on the topic on my blog as well: