Seamless 3GPP/WLAN offload: simulation approach Nikolay Amelichev∗
Kirill Krinkin
Open Source & Linux Lab Russia Email:
[email protected]
Open Source & Linux Lab Russia Email:
[email protected] Abstract
In this paper, offload in heterogeneous mobile ad-hoc networks (MANETs) containing of 3GPP and WLAN nodes, is studied. In such a network, there are several WAN-connected mobile access gateways (MAGs) providing connectivity. For a moving mobile node (MN) to sustain connectivity, offload from one MAG to another must periodically occur. It is natural to search for techniques maximizing mobile node connectivity (or, equivalently, minimizing offload time), while not incurring high costs on network bandwidth and base stations’ loads. Several mobility protocols have been proposed for offload. Currently, all third generation mobile user equipment (UE) should implement 3GPP TS 23.402 [2], which supports using MIP (Mobile IP), MIPv6 (Mobile IPv6) and PMIPv6 (Proxy Mobile IPv6) protocols for offload. We evaluate performance of PMIPv6 protocol in a simple simulation. Simulation is done using discrete event simulator ns-2 [1]. UDP client-server scenario with WLAN-based MAGs and a mobile node, measuring throughput of MN. Our preliminary conclusion is that PMIPv6 protocol reduces interruption period length from 2..12 s to 0.1..0.3 s. Area of our further research is modeling of PMIPv6 with both 3G and WLAN MAGs, and comparing performance and bandwidth usage of PMIPv6 versus MIPv6.
I. I NTRODUCTION A. MIPv6 Mobile IPv6 [6] is a subset of IPv6 developed to support mobile connections. Most importantly, it solves the problem with device changing IP address when connecting to another MAG (effectively, moving to another subnet). Each device in MIPv6 is identified by its unchanging home address, although it may be connecting through a foreign network. This is done by having a home agent (HA), which receives and stores data about MN’s location, intercepts packets intended for it, and tunnels them to MN’s current location. Since its introduction, several more effective protocols, such as HMIPv6 (Hierarchical MIPv6) [9] and PMIPv6 (Proxy MIPv6) [5] have been proposed. B. PMIPv6 PMIPv6 [5] has been proposed in 2008 as a new mobility protocol. It expands the ideas of MIPv6, introducing the concept of local mobility anchor (LMA). LMA is the home agent for the mobile host in a PMIP domain. MAGs send updates to LMA about current location of MN. This avoids MN participation in signaling, which conserves network bandwidth. As explained in [4], PMIPv6 performs handover faster than MIPv6 primarily because the MNs do not participate in signaling, but also because it uses additional link layer attachment information and micro mobility characteristic.
II. E XISTING METHODS Neither MIPv6 nor PMIPv6 can prevent packet loss in the interruption period (i.e., before MN is attached to new MAG). Several techniques to alleviate the problem have been proposed, such as using Smart Buffering with PMIPv6 [4], which effectively caches packets sent in the interruption period on the old MAG, and forwards them in-order to new MAG when MN has established connection with it. III. P URPOSE OF THIS RESEARCH In this paper we investigate how well PMIPv6 performs compared to no handover protocol at all (that is, effectively, dynamic route discovery using ad-hoc on-demand distance vector protocol (AODV) [8]). Our comparison is in terms of minimal interruption period time, and minimal additional bandwidth utilization. • We estimate interruption period length by measuring the time MN throughput is effectively zero. • We also measure bandwidth utilization of PMIPv6 by measuring MN thoughput with PMIPv6 on and without it running. IV. S IMULATION A. Overview Our simulation model consists of a mobile node repeatedly moving horizontally from one edge of rectangular map to the other, and backwards. There are two MAGs, and switch between them occurs when MN is situated approximately at the horizontal center of the map. B. Model Characteristics 1) General Characteristics: a) Model Type (Transitory or Steady-State).: Our model is transitory, that is, we don’t wait for it to reach a steady state (unchanging node positions, stable routing tables, etc.). We believe that transitory model better stresses weaknesses and strengths of handoff protocols, specifically the time of transition from one MAG to another. b) Node movement.: Model includes one node, repeatedly moving from west to east of the map and backwards. c) Timing.: Simulation lasts 100 seconds, with handoff at approximately 50.0 s and 63.0 s. d) Sampling.: We take samples of MN link throughput (Kbits successfully sent) every 0.1 s. e) Randomization.: Currently we don’t use randomization in our model, but we consider generating randomized handoff scenarios in our further research of PMIPv6 protocol for WLAN, 3G, and mixed WLAN-3G MANETs. 2) Topology: We use a rectangular 150 m × 150 m grid as topology. No nodes are elevated above the earth level (all Z = 0.0 m). Radio visibility distance (distCST parameter of modeling) is 85.3 m, with receiving and sending thresholds (RXThresh and CSThresh respectively) calculated by propagation.cc program from ns-2 [1] standard distribution.
g
CN (100, 120)
LMA (55, 85) g
MAG1 (20, 30) f
MAG2 (100, 30) g
MN (0, 0) v = 5 m/s e : Fig. 1.
Model topology
3) Nodes: Our model (fig. 1) consists of 5 mobile nodes: • MN (Mobile node) periodically moving from west to east and back with speed v = 5 m/s. Constant bit-rate protocol (CBR) client is run on this node. • CN (Corresponding node), on which CBR server is run. • MAG1 (Mobile gateway), situated on the west of the map. • MAG2, situated on the east of the map. • LMA (Local Mobility Anchor), situated above the MAGs, within the reach of both of them, but not within the reach of MN. This node establishes connection from CN to MN, and also keeps information about current MAG which MN is attached to. Note. We are using flat routing, that is, no base stations; all addresses are in the same domain. This is mainly for simplification of the model. C. UDP client-server scenario CN runs a constant-bitrate protocol (CBR) server, that serves 1 KB-sized packets. Bitrate is 1 MBit/s. This protocol works over UDP transport. MN runs a constant-bitrate protocol (CBR) client, receiving packets sent to CN. MN periodically (T = 0.1 s) polls the number of bytes successfully sent, calculates throughput and writes it to the simulation throughput file, to be read by xgraph utility from ns-2 distribution [1].
Fig. 2.
Visualization of our model at t = 8.0 s in iNSpect
D. Simulator configuration We used ns-2 discrete event network simulator [1], version 2.33. To add support for PMIPv6 protocol, we ported ns-2.29 patch from [3] to ns-2.33. Patched version of ns-2.33 is capable of simulating flat routed networks with PMIPv6, both wired and wireless. Base stations are not currently supported in our patch, primarily because ns-2.29 NIST implementation of 802.11 MAC layer differs much from ns-2.33 802.11 MAC implementation. We would likely support hierarchical routing in the future. For visualizing simulator traces, we used iNSpect 4.0 beta 3 [7] from Toilers research group at the Colorado School of Mines. Example of visualization at t = 8.0 s is shown in fig. 2. For visualizing MN throughput, we used xgraph utility coming from ns-2 distribution [1]. V. S IMULATION R ESULTS We obtained the throughput graphs for model with PMIPv6 (fig. 3) and without it (fig. 4). These graphs show that throughput of MN in connected state is essentially the same for both configurations. For the handoff times one could notice that PMIPv6 reduces interruption time more than tenfold, from approximately 12 s to 0.1..0.2 s. VI. C ONCLUSION Our simple model shows that performance of PMIPv6 in an all-WLAN network is rather good. It has a slight (0.1..0.3 s) delay while changing MAGs, which may be undesirable for some sensitive applications, though. Another advantage of PMIPv6, that being constantness of mobile node’s IP address while changing subnetworks, is not shown here, because we decided not to have hierarchical routing. This is to be addressed in a later work.
Fig. 3.
Fig. 4.
MN throughput with PMIPv6 enabled
MN throughput without PMIPv6. Notice the much longer handoff time
We observed that peak throughput of MN with PMIPv6 and without it is essentially the same, perhaps because our scenario is oversimplified. PMIPv6 MAC level optimizations may be also resposible for almost no visible bandwidth usage by the protocol. Comparison of bandwidth utilization by PMIPv6 versus MIPv6 in an environment with more MNs and MAGs would definitely be interesting. It would be also useful to research in PMIPv6 handover procedures outlined in 3GPP TS 23.402 [2] to see if they’re more effective than plain PMIPv6 used on WLAN MN and WLAN MAGs. We consider simulating handoff in mixed 3G-WLAN network in both TCP and UDP client-server scenarios, with more nodes, to get more concrete results on PMIPv6 performance. R EFERENCES [1] ns-2. the network simulator, version 2.33. http:// www.isi.edu/ nsnam/ ns/ , 2009. [2] 3rd Generation Partnership Project. Technical Specification Group Services and System Aspects; Architecture enhancements for non-3GPP accesses (Release 9), December 2009. [3] HyonYoung Choi. Proxy mobile ipv6 in ns-2.29. http:// commani.net/ pmip6ns, 8 2009. [4] HyonYoung Choi, KwangRyoul Kim, HyoBeom Lee, SungGi Min, and Youn-Hee Han. Seamless handover scheme for proxy mobile ipv6 using smart buffering. In The International Conference on Mobile Technology, Applications and Systems (Mobility Conference). Association for Computer Machinery, September 2008. [5] S. Gundavelli, K. Leung, V. Devarapalli, K. Chowdhury, and B. Patil. Proxy Mobile IPv6. The Internet Society, August 2008. [6] D. Johnson, C. Perkins, and J. Arkko. RFC 3775. Mobility support in IPv6. The Internet Society, June 2004.
[7] Stuart Kurkowski, Tracy Camp, Neil Mushell, and Michael Colagrosso. A visualization and analysis tool for ns-2 wireless simulations: inspect. In MASCOTS ’05: Proceedings of the 13th IEEE International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunication Systems, pages 503–506, Washington, DC, USA, 2005. IEEE Computer Society. [8] C. Perkins, E. Belding-Royer, and S. Das. RFC 3561. Ad hoc On-Demand Distance Vector (AODV) Routing. The Internet Society, July 2003. [9] H. Soliman, C. Castelluccia, K. El Malki, and L. Bellier. RFC 4140. Hierarchical Mobile IPv6 Mobility Management (HMIPv6). The Internet Society, August 2005.
