What are the components and building blocks for a femtocell gateway? I've suggested that it's likely there will be relatively few of these beasts, co-located with major regional internet connection points, aggregating a lot of voice and data traffic. Can they scale up to meet the full capacity required and what are the bottlenecks?
What is the femtocell gateway capacity required?
Some analyst’s figures are for very high takeup of femtocells. This will depend on how attractive the pricing, benefits and packaging are for the end users, combined with how much is invested in marketing promotion. Let's pick a figure of 3% takeup within 6 months of launch (just a wild guestimate on my part). You might believe its half that, or much larger - just bear with me...
For a reasonable sized operator with 10 million subscribers, that’s 300,000 femtocells.
It could just as easily be 600,000 or more.
Where's the femtocell traffic going?
In terms of volume, data traffic has overtaken voice in terms of traffic mix on mobile networks - i.e. there are more data bits than voice bits being sent over mobile networks today.
Femtocell data traffic will be routed in from the internet through the femtocell gateway then onto the SGSN and GGSN and back out to the internet. (CDMA networks route their data traffic through a PSDN instead of SGSN/GGSN).
Femtocell voice traffic will be routed to an MSC probably in the same building and from there onto its final destination which could be anywhere - another mobile phone on the same network, to another mobile phone network, or to another telephone network elsewhere. Network operators using the latest IMS architecture for call handling are likely to have a similar topology.
So consolidation is the key
Large femtocell gateways with capacities of 300,000 to 1 Million femtocell access points are going to be needed, and probably from a fairly early stage in the launch program.
If the launch is limited, say focussing on the enterprise market first, then lower capacities would be needed - perhaps some 10 to 30,000 femtocells - but each femtocell access point could be handling up to 8, 16 or more simultaneous calls.
With long distance telecommunications links between major switching centres now being comparatively cheap, it's often more efficient and cheaper to co-locate ever larger nodes at fewer data centers, which are managed by a small number of staff.
Perhaps a sensible sizing for a femtocell gateway would be around 100,000 simultaneous calls/sessions, depending on the geographic nature of internet exchange connection points and customer locations around the country.
Time to market versus capacity and efficiency
When 3G systems were first rolled out, time to market was key. This meant that some of the equipment designs really weren't very efficient. I recall one large vendor requiring some 17 cabinets for the first generation of their RNC (Radio Network Controller), taking up huge amounts of space and power consumption.
Similar inefficiencies have been seen with some femtocell gateways, which weren't originally architected to handle large numbers of internet connections. There are several critical components within a femtocell gateway:
- A Controller (computer server controlling the sessions), communicating with all the femtocells, handling registration and session setup/teardown using signalling messages (not the bearer traffic).
- An IPsec Security Gateway which terminates all the secure IP tunnels, decoding the encrypted traffic in real time.
- A media gateway which terminates the voice calls (bearer traffic only).
- A separate TR-069 provisioning gateway which also maintains, diagnoses and upgrades the femtocells.
There would also be a firewall and other IP routing components to protect the network.
Some early femto gateways sit behind Cisco or Juniper routers, and leave them to handle the IPsec tunnel decryption. But plugging an IPsec card into (say) a Cisco 7600 router (a fairly common model) can reduce its performance by 70%. Standalone dedicated IPsec hardware which is designed for this purpose can achieve the same function at a cost advantage in some cases as high as 10:1.
Learning from UMA
Lessons can be learnt from UMA which has a similar architecture, feeding individual IP sessions from each UMA connected phone through to the UNC (which performs a similar function to a femtocell gateway). A typical UNC has a capacity in the region of 100,000 to 300,000 sessions - these relate to individual UMA phones, not the number of Wi-Fi access points which the phones can use. These can also scale up as required and all mainstream models use a dedicated box for IPsec processing.
Scalable IPsec hardware will be an essential part of the long term femtocell gateway solution.
Rapid go-to-market solutions are helpful to facilitate trials and initial product launches. They don't need to be super-efficient, just reliable, proven and easy to manage and maintain.
But as the size of live deployments grows rapidly, the cost and efficiency of the femto gateway will gain importance.
Whilst the cost of the gateway is a very much smaller piece of the overall femtocell solution (the femtocell itself is the critical cost item because there are so many of them), as volumes ramp up we can expect to see more attention paid to the efficiency of the gateways and their scalability.
