WirelessMoves

While doing some research on how capacity will grow in fixed line and wireless networks in the future I stumbled over the following thing right in my neighborhood: Future bandwidth increases on the last
mile to the subscriber come with an additional cost in comparison with today’s
standard ADSL or ADSL2+ deployments because of extra hardware that has to be installed close the the subscriber.

For ADSL2+ the DSLAM is usually installed
in the telephone exchange and the cable length to the subscriber can be up to 8
km for a 1 MBit/s service. For VDSL, which offers data rates of 50 MBit/s
in downlink, the cable length must not exceed 500m. Thus, DSLAMs can no longer
be only installed in central telephone exchanges but equipment has to be
installed in cabinets on the street. The cabinets themselves are quite large (about half the size of a GSM or UMTS base station),
require power, active cooling, and create noise. For the installation
of the cabinets earthworks are necessary to lay the additional fiber
and power
cables required to backhaul the data traffic. The picture on the left shows a VDSL DSLAM cabinet that
has been installed alongside a ‘legacy’ small telecom cabinet as part of the current VDSL build out in my region.

To connect a
new subscriber a technician is required to manually rewire the customer’s line
to one of the ports. Different sources currently
specify the maximum capacity of such cabinets from about 50 to 120 VDSL
ports. To support 500 VDSL connections per km², several cabinets are thus
required. I wonder what happens when 5 different companies put such DSL ‘base stations’ in place!?

I often wondered in the past what ‘Secondary PDP Contexts’ are good for in UMTS networks. I had a vague idea but never had the time to look up the details. These days I had and here’s a short explanation:

‘Secondary PDP contexts’ can be used to separate the real time data traffic from background or signaling traffic into different streams on the air interface while keeping a single IP address on the mobile device. This is done by an application providing the network with a list of IP addresses in a Traffic Flow Template. The mobile device and gateway router (GGSN) in the network will then screen all incoming packets and handle packets with the specified IP addresses differently, like not repeating them on the RLC layer after an air interface transmission error. This is transparent to the IP stack and the applications on both ends of the connection.

External providers of speech services such as Skype, however, do not have access to this functionality. A big advantage for operator controlled IMS services when things get wild on the air interface!?

Resources:

• Secondary PDP Context Activation: 3GPP TS 23.060, Chapter 9.2.2.1.1 (Rel 6)
• Traffic Flow Template Description: 3GPP TS 24.008, Chapter 10.5.6.12 (Rel 6)

For those interested in getting a feeling of how Wifi works on the physical layer, Metageek’s Wi-Spy is the ideal tool. I’ve already reported about my first experiences here and here. Wifi has become very popular in Paris due to DSL being quite cheap. So it should thus probably not be surprising that I can see 13 access points in my Paris apartment. As there are only three non-overlapping Wifi channels on the 2.4 GHz ISM band the question is what kind of impact such a high number of access points has on throughput.

Matters are made worse by the fact that some of the networks I can see are used for TV and video streaming. This is quite popular in Paris as this is offered by most DSL ISP’s and one even offers video streaming over Wifi to a remote set top box. The picture on the left shows a trace taken with Wi-Spy under these conditions. The lower graph in the figure shows the frequency range of the ISM band between 2400 and 2480 MHz. Instead of showing the frequency in MHz the x-axis labs show the 13 available Wifi channels. On the y-axis the amplitude of the signal received over the band is shown. The color of the peak depends on the intensity of the signal received during 60 minutes. Bright color indicates high activity. The graph shows five partially overlapping networks with their center frequency on channel 1 (only little traffic so the arch is not very well visible), channel 3, 5,6 and 11.The most activity can be observed in the wireless network that is centered around channel 11.

The upper graph shows a time graph over the frequency range. On the y-axis I’ve chosen a resolution of 60 minutes to show the activity in the ISM band in the course of one hour. The Wifi networks on channel 5 and 11 were most likely used from streaming as there is uninterrupted activity throughout the test period. The Wifi networks on 3 and 6 were also used for streaming. Streaming was stopped on the Wifi network on channel 6 after about 12 minutes while streaming was started on the Wifi network on channel 3 about 40 minutes into the trace.

To see what the impact of that streaming has on throughput in my network I used two notebooks, one connected via Ethernet, the other via Wifi and Iperf, a UDP and TCP throughput measurement tool. With a fully overlapping Wifi network which is used for TV streaming, capacity of the Wifi network under test was reduced to 72%. Partial overlapping entails an even bigger speed penalty and performance was reduced to 59%.

Here are the absolute values:

• No interference: 22.5 MBit/s
• Full Overlapping: 16.3 MBit/s
• Partial Overlapping: 13.4 MBit/s

Looks like it is time equipment manufacturers are taking the 5 GHz band a bit more seriously…

For more traces take a look at my previous traces or head over to Metageek where you can download the software and check out some sample traces yourself.

Back in mid September I reported on using my Linksys WRT54 Access Point in "Access Point Client Mode" to create a wireless link to another access point for a number of notebooks which are connected via Ethernet to the Linksys. The traces which I took on the Linksys and on the notebooks indicated that the Linksys replaces the MAC addresses of the notebooks with its own before it sends the packets over the wireless link. Equally it replaced its own MAC address in incoming packets with the MAC address of the real recipient. This is neither layer 2 bridging nor layer 3 IP switching but something in between. I couldn’t quite believe it.

In the meantime thanks to the suggestions I received I made some further tests and I can now confirm that the Linksys really does replace the MAC addresses. Take a look on the picture on the left which shows the ARP table of a PC connected wirelessly to the real access point. The notebooks connected to the Linksys Client AP both have the same MAC address. The MAC address is that of the access point! Quite sophisticated! (Note: All devices in the network are in the same IP subnet)

I am not sure how this feature should be called. It’s not really ‘Layer 3 switching’ which is already a highly overloaded term anyway. I’d prefer the term ‘MAC masquerading’ although the term is also already used for something else as well.

Thanks to all who sent their comments and suggestions!

A fierce competition is raging in Austria between DSL and 3G operators positioning 3G data cards as an alternative for DSL connectivity. Prices are interesting too, so many people are going wireless these days. Which leaves the question of how much capacity mobile networks could have compared to DSL.

Certainly not an easy question to answer so let’s take a couple of assumptions:

Austria has 4 HSDPA networks today. Let’s say in a city like Vienna the average cell inter distance is 1km. Usage is still in it’s early stages so only a single 5 MHz channel is used in a 3 sector cell. Per sector throughput is assumed to be 2.5 MBit/s. Since the cell covers an area of 1 km², the capacity in that area per operator is thus 2.5 * 3 = 7.5 MBit/s. All 4 operators together would thus create a capacity per km² of 30 MBit/s.

On the fixed line side I would say that DSL today offers a speed of on average of 4 MBit/s to housholds in cities like Vienna. Vienna has a a population density of 4000 inhabitants per km². Let’s say the average household has 3 people and DSL penetration is 40%. Thus there are (4000 / 3) * 0.4 = 533 DSL lines per km². With an average speed of 4 MBit/s per DSL line that would be 2.113 GBit/s. Sounds like a lot more than what the 3G calculation results in above. But wait, there’s a catch. The 4 MBit/s are only valid between a subscriber and the DSLAM (DSL Access Multiplexer). The connection to the core network is usually much smaller. I’ve heard the ‘oversubscription’ is anywhere between 1:20 and 1:50. Let’s assume the oversubscription is 1:30. As a result, the DSL capacity per km² would be 71 MBit/s.

30 MBit/s wireless vs. 71 MBit/s via DSL

The example stands or falls with the DSL oversubscription ratio. If you have more details on this please let me know!