Faculty Members:  Dr. Nirmala Shenoy and Prof. Bruce Hartpence

Cisco Champion: Rafael Mantilla, Montalvo. Ph. D.

ITS Technical Support: Andrew Elble

Period of activity: Winter Quarter 2003.

Activity 1: A mobile station moving from a cellular network to another cellular network during an active data session.

As of now a base model has been completed. The base model consists of two cellular networks with GPRS overlay network with an actively roaming mobile user. As mentioned before, a handoff mechanism as well as a redirection process has been designed for successful transfer of the mobile user data to the new cellular network. Once these mechanisms were implemented, delays were incorporated into the model. Specifically, the delays implemented were processing, database, store/retrieve, and channel allocation delays. These delays were placed at pertinent nodes in the model to achieve a more realistic scenario. Once the delays were implemented, validation of the model was performed. Following the validation, the model was tested for several scenarios. The scenarios were based on varying the processing and database job arrival rates at the various nodes in the networks. The effect of the varying arrival rates was observed on the two overall delays that were being studied. The two overall delays of interest are the delays encountered by the mobile user as a result of handoff and redirection of packets. The base model is intended as a stepping stone to further research on more efficient handoff and redirection mechanisms between the cellular networks. With that in mind, the base model has been modified to create two other models namely, early queue release and bicasting models. The early queue release model is a variation specifically at the redirection process allowing packets destined for the new network to be transferred with lesser delays than the base model. The bicasting model is a variation that uses an IP tethering point to act as a crossover point and transmit identical packets to both cellular networks during the handoff process enabling the mobile user to receive data packets as soon as a handoff has been accomplished. These models will be validated and tested to obtain results.

Activity 2: Roaming from Cellular to WLAN

The 'data session' sequence in a cellular network is very different from that in the WLAN. In a cellular network, the mobile station (MS) will first attach to the network, then activate, and once activated the MS will start the data session. At the end of a data session or when the MS wants to end the data session, the MS first deactivates with the network and then detaches from the network. In the WLAN, the MS first authenticates with the Access Point (AP) and then associates with the AP. Once the MS is associated with the AP, it starts its data session with the WLAN at any time. However, this authentication of the MS with the AP is a layer 2 authentication. When we are considering the roaming of the MS from the cellular network into the WLAN, it is important to consider user authentication of the MS with the home network of the MS. So the concept of "Profile Server" was introduced in the WLAN. A node acting as a "Profile Server" along with an AAA server was used in the WLAN into which the mobile node roamed. So, before the MS is allowed to start a data session with the WLAN, the AAA in the WLAN authenticates the MS with the home network of the MS and downloads its profile. Also, it was important to develop a multi-mode MS that can communicate with a cellular network and a WLAN by switching frequencies and operational modes.

Once the handoff and redirection mechanisms were implemented in the model, delays at the various nodes were introduced. The delays implemented were:

1. Process Delay – The time taken by node to process an incoming packet as it goes through the different protocol layers. This delay is introduced in all the nodes in the model.
2. Database Delay – The time taken by the HLR/ Profile Server to retrieve information about a Mobile Station from its database. This delay is only introduced in the HLR/Profile server (AAA server) before sending the authentication reply for the MS. This gives the effect of checking for information on the MS in the database before sending the authentication reply.
3. Total Delay (= Store Delay + Retrieve Delay) – To handle packet redirection a set of queues were introduced at the various nodes which handle the packet retransmission. The time taken by the nodes to store and retrieve data packets from the queues during data redirection process constitutes the total delay.
4. Encapsulation Delay – The time taken by GGSN to encapsulate data packets traveling from the cellular network into the WLAN through the tunnel.
5. Decapsulation Delay – The time taken by Gateway to retrieve the original data packet from the encapsulated data packets received from the GPRS network. through the tunnel
   The M/G/1 queue was used to model all these delays.

The Encapsulation and Decapsulation delays were assumed to have fixed values.

The two delays of interest for the mobility management studies are the:

1. Handoff Delay – It is calculated from the time BTS initiates Handoff Request on behalf of the mobile the time BTS gets Handoff Complete from MS, before ti switches over to the frequencies of the new network and channel.
2. Redirection Delay – It is calculated from the time BTS queues first data packet to the time MS receives first redirected packet.
   

The handoff delay and the data redirection delay were calculated manually as well as through the simulation results for validation purposes. For the validation, the actual simulation time at every event in the handoff and redirection process was taken and was compared to the manually estimated time according to the delays that were incorporated in the model.
A number of scenarios with varying seed values were used to test the model. In these scenarios, the loads at the various nodes were changed to give the effect of the network being populated with multiple BTSs and MSs.

Activity 3 : MobileIPv4 Test Bed

The following changes were made to the network topology since the last test phase. The Foreign Agent now resides on a server (Sun OS) instead of a router. We have also started to gather statistics with the Windows XP Birdstep client. In this report, we discuss in further detail the status of all issues covered.

Below is the current status of the matrix. The highlighted portion will be discussed in more detail later in the report. We’ve gathered a variety of statistics for Experiments 1, 13, 25, to include total download time, redirection times, Mobile IP delays, various WLAN statistics, retransmissions, etc. As for the other experiments, we would change the topology to match the situation and then attempt a file transfer to test mobility success or failure. The matrix depicts this. We have been unable to complete experiments 11, 23, and 36 because of the inability to capture the TCP traffic due to WEP encryption. We currently have a request for information with LinkFerret support. Because we are nearing the end of the quarter, I suggest that we can move forward by taking statistics from the wired side and forgo the delays involved with the Layer 2 side. We are finding that these delays are minimal compared to Mobile IP registration and Redirection.

Experiments 1, 13, 25 – Best Case Scenario – Open Authentication, No WEP, Same SSID and Channel.

Experiments 11, 23 – Probable and Worst Case for Linux – Open Authentication, Same WEP, Different SSID and Channel.

Experiment 36 – Probable and Worst Case for Windows – Shared Authentication, Different SSID, WEP and Channel.

Results and Observation

Below are results of Experiments 13 and 25. We conducted five tests of each, and then analyzed the data.

Total – Total download time, in seconds, form SYN to FIN.
MIP Lag – The time from Mobile IP registration (tunnel erected) to the first ACK from the mobile node on the foreign network.
Redirection – The time difference between the same ACK, mention above, and its corresponding data packet on the home network sent from the server.
MIP Reg – Mobile IP Registration, the time difference between a Mobile IP Request, from the mobile mode, and a Mobile IP Reply.

Experiment 13 Observations. Test 1 and 5 had long MIP Registration times. We believe this was caused by a lost WLAN packet. The request was made by the mobile node, although there was no matching packet on the wired side. This will affect the total delay greatly, because the Linux node waits until it hears another advertisement by the FA, every seven seconds in this experiment. Plus this happened twice out of five tests. Some other observations, total download time seems to not have a great effect on redirection time.

Experiment 25 Observations. Birdstep Mobile IP registration time is minimal. Birdstep is more consistent, but still has a large variance. Again, total download time does not have a great effect on redirection time.

We would like to conduct the same experiment with Birdstep under ideal conditions, tweaked. Maximize Birdstep’s settings and increase the advertisements on the FA should give a greatly reduced Redirection Delay and Mobile IP Lag.

Activity 4: Delay and handoff mechanisms study at Layer 2 and Layer 3 when using MobileIPv4

With mobileIPv4 set up in the network topology as shown in the figure above, work was done in testing all the scenarios under to study layer 2 mobility. We narrowed our study on most probable cases and observed their characteristics. The most probable cases were

1. same SSID, same channel, no wep key
2. different SSID, different channels and wep key

We mainly studied 3 scenarios of raoming from one subnet to another and they were

1. Birdstep client without tuning the parameters for efficient roaming
2. Linux client
3. Birdstep tuned to handoff faster

Under this part, we find that layer 2 delay accounts for major part of the overall delay which can be seen in the hand off delay between two access points with different SSIDs. Moreover with the untuned birdstep client the throughput was almost half that for the linux client.

Set up: Different subnet, windows as mobile IP node running Birdstep mobile IP client in its default configuration. Different SSID, same channel with no encryption enabled.

Wired Side capture for birdstep client indicated a redirection delay of 50.716170 seconds.

On the wireless side: Total time delay seen for the same scenario and captured concurrently was 50.723 seconds. After the registration of the mobile client in the foreign network, it took less than 5 seconds for the data transfer to resume.

Major portion of the delay was seen to be occurring on the layer 2 side of the handoff. Therefore the handoff was studied again in layer 2 to verify this.

Set up: Under the same experimental setting as before but with both APs in one subnet.

WindowsXP client was used and the capture as seen on the wired Side indicated a Time of 46 seconds which confirms the fact that delay were more significant in layer 2 than in layer 3. To study this further.

     Aanak Patwa

    Vishal Mankotia

Actvity 5: Set up an IPV6 test bed using Linux routers to study the mechanisms at layer 2 and layer 3 during handoff using MobileIPv6

Developed an IPv6 test bed using Linux based routers from IPV6 freeware.

On the Linux routers, we plan to install the home router module provided by Lancaster University, UK. The Lancaster University also provides the “Correspondent Node” software.

Windows XP Professional has been used primarily to test Windows Support.
Redhat Version 9.0(kernel 2.2-4) is running on all Linux hosts/routers.
Already an FTP server (vsftpd) has been set up. This server that supports IPV6. FTP client (tnftp) which supports Ipv6 is used.

Cisco routers are currently being upgraded with the beta version of IOS 12.3T that has support for Mobile IPv6.

Actvity 6: QoS support in WLAN

Modern 3G and wireless networks have evolved to the point where they are now able to provide QoS to mobile nodes. Current mobile users however find themselves confronted with a dilemma when attempting to maintain connectivity while roaming with mobile IP. The ability to map QoS capabilities of the wireless networks to 3G QoS systems is critical to enable seamless roaming across Next Generation Wireless Networks. Current research will be focused on the DiffServ model and in particular on Per Hop Behaviors. Specifically initial research will examine the combination of Expedited forwarding and Assured Forwarding contained within the current DiffServ framework. The combination of these current mechanisms will be mapped to the four 3G QoS traffic classes: conversational, streaming, interactive, and background. Combinations of subscript labeling within AF may also be applied per Behavior Aggregate.
Details of Qos support in Cisco APs were obtained. These will be tested.

John Sheftic