An Enhanced Handoff Mechanism for Cellular IP

Transcription

1 An Enhanced Handoff Mechanism for Cellular IP Jong-deok Kim, 1 Kyung-ah Kim, 1, 2 JaeYoon Park 2, Chong-kwon Kim 1 1 School of Electrical Engineering and Computer Science, SNU, Seoul, Korea 2 Korea Telecom, Seoul, Korea kimjd@popeye.snu.ac.kr Abstract Handoff is one of the most important factors that may degrade the performance of TCP connections in wireless data networks. In this paper, we present a lossless and duplication free handoff scheme called LPM (Last Packet Marking) for improving Cellular IP semisoft handoff. LPM signals the safe handoff cue by sending a specially marked packet to mobile hosts. With this control packet, MH can adapt to network dynamics simply and efficiently. Our performance study shows that LPM achieves lossless packet delivery without duplication and increases TCP throughput significantly. Keywords Micro Mobility, Cellular IP, Hard Handoff, Semi-soft Handoff, Last Packet Marking 1. Introduction Currently, there are many efforts underway to provide Internet services on integrated wireless and wireline networks. Supporting efficient IP mobility is one of the major issues to construct IP-based wireless access networks. Mobile users will expect the same level of service quality as wireline users. Even though access point of mobile user changes, IP connections should be maintained transparently. The Mobile Internet Protocol (Perkins, 1996, Johnson and Perkins, 2000) is the current standard for supporting global IP mobility in simple and scalable manner. But, in processing frequent handoffs in cellular based wireless access networks, Mobile IP has some limitations. After each migration, a new local address called the care-of address must be obtained and registered to a possibly distant home agent. This incurs increasing handoff latency and load on the global Internet, and mobile users suffer service degradation during the handoff period. A number of solutions (Cambell, et al, 2000, Ramjee, et al, 1999, Mankin and Soliman, 2000) have been discussed to overcome these problems. These approaches are extending Mobile IP rather than replacing it. To handle local movement of mobile hosts without interaction with the Mobile-IP-enabled Internet, they adopt a domain-based approach. That is, these intradomain protocols are used for establishing and exchanging the state information inside the wireless access networks, so as to get fast and efficient intra-domain mobility or micromobility control Among these intra-domain mobility protocols, Cellular IP (Cambell, et al, 2000) attracts special attention for it s seamless mobility support in limited geographical areas. Since COA(Care-of-Address) is not changed in local mobility control, it eliminates the load for location update in the global Internet. And by using paging concept, Cellular IP provides simple location tracking scheme called cheap passive connectivity, which imposes neither traffic nor processing load on the global network as long as the host is idle, and preserves the power reserves of mobile nodes.

2 A previous study (Seshan, et al 1997) showed that the performance of TCP degrades significantly due to packet losses in frequent handoffs in small cell wireless networks. Cellular IP introduces semisoft handoff mechanism to reduce the number of lost packets. It establishes a routing path to the new BS (Base Station) before handoff, and temporarily bicasts data to the both old (current) and new BS. The throughput of TCP using semisoft handoff is much better than that of hard handoff. Nonetheless, packet loss or packet duplication still can occur in Cellular IP semisoft handoff, which results in the degradation of TCP performance. In this paper, we propose a simple handoff scheme called LPM (Last Packet Marking) for Cellular IP. LPM gives exact cue to mobile host for safe handoff to remove packet loss or packet duplication. This paper is structured as follows. In section 2, we briefly preview the Cellular IP handoff mechanism. In section 3, we describe last packet marking method for improving Cellular IP semisoft handoff. In section 4, we verify LPM by computer simulations. Finally, we conclude in section Review of Cellular IP In Cellular IP, a group of BSs (Base Stations) forms one access network. Each access network attaches to the Internet via a gateway router. A BS, which provides wireless access service to MHs (Mobile Hosts), is a special-purpose router with mobility-related functions. Cellular IP assumes the tree network topology to simplify routing within an access network. An MH which visits a certain access network uses the IP address of the gateway router as its care-ofaddress. Within the access network, the MH is identified by its home address. Since COA (Care of Address) is not changed in local mobility control, Cellular IP eliminates the load for location update in global Internet. Let us examine Cellular IP handoff schemes briefly. Cellular IP proposes hard and semisoft handoff. The hard handoff scheme, the basic handoff mechanism of Cellular IP, makes a new route to the new BS after real handoff. Hard handoff suffers from severe packet losses and results in performance degradation. To provide adequate performance to both TCP and UDP traffic while maintaining the lightweight nature of the base Cellular IP protocol, Cellular IP introduced a new handoff method called semisoft handoff. Prior to make a real handoff, an MH makes a brief connection to a new BS and sends a semisoft request packet. The semisoft request packet is forwarded toward the gateway router and eventually it reaches a crossover router which is a branching point of the path to the old BS and the path to the new BS. Receiving the semisoft request packet, the crossover node updates its route cache by adding the path to the new BS and starts to buffer packets destined to the MH in a delay device. The MH makes a real handoff after a pre-determined semisoft delay. The MH issues a route update packet to the crossover node after the real handoff. The crossover node updates the route cache and transmits packets buffered in a delay device and packets arrived thereafter. Since optimal semisoft delay cannot be predicted accurately, Cellular IP fixes the semisoft delay proportional to the maximum mobile-to-gateway round-trip delay. On various network topology and network traffic, applying constant semisoft delay may cause an MH to receive duplicated packets or to suffer packet losses. Duplicated packet does not disrupt many applications, but packet loss limits TCP's performance severely. To eliminate the packet loss, Cellular IP temporarily introduces a constant delay along the new path between the crossover node and the new BS using a delay device on the crossover node. So, just after a handoff, most MH suffers packet delay or duplication, which can cause quality degradation of real time traffic or can trigger TCP congestion control.

3 3. Last Packet Marking We develop a new handoff scheme called LPM which can signal exact handoff initiation time to an MH. Overall LPM procedure is similar to the semisoft handoff in many aspects. As in Cellular IP, an MH sends a semisoft request packet to the new BS before making a real handoff. Receiving the semisoft request packet, a crossover node creates a mapping for the new path and forwards it to the gateway. LPM makes a distinction in that the crossover node sends a special control message called the semisoft reply to the old BS before multicasting data packets to both the new BS and the old BS. As the semisoft reply assures the MH that it has received all the packets that can be received only through the BS and it can receive following data packets from the new BS. So the semisoft reply can be used as a trigger for safe handoff by the MH. Therefore, receiving the semisoft reply through the old BS, MH initiates real handoff immediately rather than waiting for the expiration of constant semisoft delay as in Cellular IP. Cellular IP introduces the crossover delay to synchronize the delay difference between the new path and the old path from the crossover node in case the new path is shorter than the old path. However this may introduce undesirable additional delay in many situations. LPM makes another distinction in that the crossover node sends data packets to the new BS immediately instead of buffering packets in its delay device for the constant crossover delay. In case packets sent to the new path arrive the new BS before the handoff, the new BS buffer them for fast delivery after handoff. This approach eliminates undesirable packet delay and packet duplication in many cases. Cellular IP does not place any restriction on delaying a real handoff for the constant semisoft delay. However delaying a handoff would be restricted in real situations. Therefore there would be situations that an MH should make a real handoff even if it does not receive the semisoft reply, or it could make a real handoff only after it has received several data packets after the semisoft reply. The former case may result in the packet loss and the latter case may result in the packet duplication. To overcome these, we propose packet forwarding to the new BS by the old BS for the former case, and packet duplicate elimination at the new BS for the latter case. The old BS uses the semisoft reply as a delimiter for the packet forwarding to the new BS. That is, packets arrived at the old BS after the last packet mark will not be forwarded, which alleviates the forwarding overhead compared to the general forwarding mechanism proposed. As there is no sequence number in IP Header, the new BS uses a hash function to identify and eliminate the duplicated packets from its buffer. Compared to Cellular IP, LPM only introduces a simple handoff signal, semisoft reply, that enables MH to adapt to the network topology and dynamics instead of using fixed parameters like semisoft, crossover delays, which results in no packet loss and duplication. In case delaying a handoff is constrained, which is not considered in Cellular IP, LPM makes use of efficient packet forwarding and duplicate elimination mechanisms which help to recover from packet loss and duplication. 4. Simulation Results We used computer simulation for performance analysis. 2 network topologies are considered. One is shown in Figure 1-(a) where all possible MH-to-Gateway delays are same, and possible MH-to-Crossover node delays are also same before/after handoff. Determining proper parameters for semisoft delay and crossover delay would not be difficult in this situation. Actually, this topology is used for the performance study of Cellular IP (Cambell, et

4 al, 2000). Another topology is shown in Figure 1-(b) where neither possible MH-to-Gateway delays nor possible MH-to-Crossover node delays before/after handoff are not same. Considering dynamic network delay due to short burst characteristics of data traffic and competing backgrounds traffics, we assert that this one would be more realistic. N1 through N5 are Cellular IP nodes and N0 is the gateway. Each transmission delay of links is 5 ms. CN (Correspondent Node) transmits UDP or TCP traffic to MH from time 3. An MH oscillates between BS1 and BS5 at the constant speed from time 5. The MH stays for about 10 seconds before moving to the next BS. Cellular IP semisoft delay is fixed to 50 ms, and crossover delay to 20 ms. CN 50ms N0 CN 50ms N0 N1 N2 N1 N2 10ms N3 N4 N5 N3 N4 N5 1 2 BS1 BS2 BS3 BS4 BS5 BS1 BS2 BS3 BS4 BS5 MH MH (a) Topology 1 (b) Topology 2 Fig. 1. Simulation Topologies 4.1 UDP Results Table 1 and Table 2 show the number of lost packets and duplicated packets for UDP traffic in the topology 1 and the topology 2 each. The CN transmits a UDP packet every 5 ms. We compared the numbers of lost packets and duplicated packets of LPM with those of hard handoff and semisoft handoff. Note that LPM has no packet loss or duplication in both topologies. In hard handoff, it always suffers from the packet loss. The number of lost packets is proportional to the sum of the transmission delay from the new BS to the crossover node and the transmission delay from the crossover node to the old BS. Semisoft handoff shows no packet loss in topology 1. However it still suffers from packet duplication in topology 1. These packet duplications are caused by the crossover delay in the crossover node. With the longer crossover delay, the number of packet duplication would increase. Though it is not shown at the tables, end-to-end delay of each packet during handoff is also increased by the crossover delay, which is not suitable for the real-time multimedia application. In topology 2, the performance of semisoft handoff get aggravated, it suffers either from the packet loss or from the packet duplication. As fixed semisoft and crossover delay do not adapt to the network dynamics, they are too small (packet loss) in some cases and too large (packet duplication) in other cases. Type BS1 BS2 BS2 BS3 BS3 BS4 BS4 BS5 BS5 BS4 BS4 BS3 BS3 BS2 BS2 BS1 Hard 4/0 6/0 8/0 5/0 4/0 9/0 6/0 4/0 Semisoft 0/2 0/2 0/2 0/2 0/2 0/2 0/1 0/2 LPM 0/0 0/0 0/0 0/0 0/0 0/0 0/0 0/0 Table 1 : UDP Number of Lost and Duplicated packets on Topology1 (Lost/Duplicated)

5 Type BS1 BS2 BS2 BS3 BS3 BS4 BS4 BS5 BS5 BS4 BS4 BS3 BS3 BS2 BS2 BS1 Hard 4/0 9/0 15/0 8/0 9/0 16/0 9/0 4/0 Semisoft 0/2 0/3 3/0 3/0 0/3 4/0 2/0 0/2 LPM 0/0 0/0 0/0 0/0 0/0 0/0 0/0 0/0 Table 2 : UDP Number of Lost and Duplicated packets on Topology2 (Lost/Duplicated) 4.2 TCP Results 1.4e+06 LPM Hard 1.4e+06 LPM Semisoft 1.2e e+06 Throughput (bps) 1e Throughput (bps) 1e (a) LPM vs. Hard handoff (b) LPM vs. Semisoft handoff Fig. 2. TCP throughput on Topology 1. Figure 2(a) shows TCP connection throughput of hard handoff and LPM as a function of time on topology 1. On every handoff, hard handoff shows abrupt drops caused by lost packets. It is well known that a packet loss decreases the TCP performance significantly due to the TCP congestion control. Figure 2(b) shows TCP connection throughput of semisoft handoff and LPM as a function of time on topology 1. Although semisoft handoff improves the number of abrupt drops, semisoft can t get rid of all drops. On the other hand, LPM shows no throughput drops on every handoff Sequence Number Sequence Number (a) Semisoft handoff (b) LPM Fig. 3. Sender and receiver traces of TCP connection on Topology 1. (BS4 to BS5) In figure 3(a), TCP packet traces of the sender and the receiver around 45.6 are shown. Abrupt drops of throughput on semisoft handoff are caused by duplicated packets, which had been buffered in crossover node. An MH receives packets which had been received from old BS. Although UDP shows duplicated packet at most 2, TCP traffic is relatively burst than

6 UDP traffic, the number of packet duplication is more than UDP. As TCP uses the accumulated ACK mechanism, rather than relying only on the time-out mechanism, it uses sequential duplicate ACKs more than 3 as an indication of packet loss. If a TCP receiver receives packets which had been already received, it may generate duplicate ACKs that trigger the congestion control of the sender. The MH starts real handoff at 45.6 and starts receiving packets from the new BS at Through and 45.63, the MH receives packets that had already received from old BS and sends more than 3 duplicate ACKs to CN. CN interprets these duplicated ACKs as a signal of packet loss, it sends that packet again and starts congestion control at Whereas semisoft suffers packet duplication, LPM shows no duplication (figure 3(b)). 1.4e+06 LPM Hard 1.4e+06 LPM Semisoft 1.2e e+06 Throughput (bps) 1e Throughput (bps) 1e (a) LPM vs. Hard handoff (b) LPM vs. Semisoft handoff Fig. 4. TCP throughput on Topology 2 Figure 4 shows the TCP connection throughput on topology 2 as a function of time. The throughput of LPM drops from 1100 Kbps to 940 Kbps at time 25. This is because the RTT increases by 30ms as the MH moves from BS2 to BS3. TCP throughput decrease as the RTT increases. Hard and semisoft handoff shows more rapid drops of throughput than in topology 1. All hard handoff suffers from packet drops. Except handoff from BS1 to BS2, all semisoft handoff shows abrupt drops of throughput. Remember that semisoft handoff suffered from packet duplications as well as packet losses during handoffs. Duplicated packets may also decrease throughput by triggering the TCP fast recovery mechanism. Because LPM is free from packet loss and duplication, its throughput is only affected by the variance of RTT Packet Sequence Number Fig 5. Handoff from BS3 to BS4 : Packet Loss Semisoft delay is shorter than the sum of new and old path transmission delay

7 Packet Sequence Number Fig 6. Handoff from BS4 to BS5 : Packet Loss New path transmission delay is far shorter than that of old path Packet Sequence Number Fig. 7. Handoff from BS5 to BS4 : Packet Duplication New path transmission delay is longer than that of old path Throughput degradation of semisoft handoff can be classified as 3 categories. For each case, TCP sender and receiver traces are shown in figure 5, 6, 7. The first case is shown in figure 5. When MH handoffs from BS3 to BS4, semisoft delay (50ms) is shorter than the sum of new path transmission delay and old path transmission delay (6). When MH makes real handoff to BS4, although crossover node multicast data packets to both the new and old path, there still are unicast packets on old path which MH doesn t receive yet. These packets (packet 4075, 4076 and 4077) are destined to loss after handoff and incur TCP congestion control. But to get rid of this condition, setting up longer semisoft delay is not advisable. Link layer connection may be disconnected if real handoff is delayed too long. Handoff from BS4 to BS3 shows same result. In the second case shown in figure 6, new path transmission delay is far shorter than that of old path. New BS receives multicasted packets long before old BS receives the same packets. Crossover delay is not sufficient to delay these packets. Before the MH hands off to the new BS, packet 5057 through 5059 are already arrived to the new BS and these packets are lost. But increasing crossover delay is not advisable too. Because on other conditions like further new path, increasing crossover delay incurs more duplicated ACKs, that degrades TCP performance severely. Handoff from BS3 to BS2 shows similar results. In the third case shown in figure 7, new path transmission delay is longer than that of old path. At 53.55, real handoff is finished. Crossover node forwards packets, which had stayed more than crossover delay, to the new BS. The MH receives these packets from to

8 53.59, which the MH already received from old BS. So, the MH sends more than 3 duplicated ACKs to CN and this incurs packet retransmission and congestion control. Handoff from BS2 to BS3 shows similar result.on the contrary, for every case, LPM shows no duplication, no loss of packets and no throughput degradation. An MH hands off safely for TCP connection. Packet sequence number increase stably and the result pattern is similar to figure 2 (b). We omit the TCP sender and receiver traces of LPM because of space limitation. We studied the performance of LPM where delaying a handoff is restricted, where we complement the basic LPM with the packet forwarding and duplicate packet elimination mechanism described in Section 2. It also showed similar results. However due to the space limitation, we omit the result too. 5. Conclusion We have proposed a simple handoff scheme called LPM for Cellular IP that signals an MH the exact cue to handoff on optimal time point for micro-mobility. We also proposed efficient packet forwarding and duplicate packet elimination mechanism for the situation where delaying a handoff is restricted. We studied the performance of LPM using computer simulation. Our simulation study showed that LPM received all packets without duplication in case of UDP traffic. Also in case of TCP traffic, the throughput of LPM is significantly better than those of semisoft handoff and hard handoff. 6. References Andrew T. Campbell, Javier Gomez, Sanghyo Kim, Andras G. Valko, Chieh-Yih Wan, Design, Implementation, and Evaluation of Cellular IP, IEEE Personal communication, Vol. 7, No.4, Aug. 2000, pp D.B. Johnson, C. Perkins. Mobility Support in Ipv6, IETF Internet Draft, draft-ietf-mobileip-ipv6-12.txt, Apr Karim El Malki, Hesham Soliman, Fast Handoffs in Mobile IPv4, IETF Internet Draft, draft -elmalkimobileip-fast-handoffs-03.txt, Internet Draft, Sept 27, C. Perkins, editor, IP Mobility Support, IETF RFC 2002, Oct R. Ramjee, T. La Porta, S. Thuel, K.Varadhan and S. Wang, "HAWAII: A Domain-based Approach for Supporting Mobility in Wide-area Wireless networks", International Conference on Network Protocols, ICNP'99 S. Seshan, H. Balakrishnan, and R.H. Katz, "Handoffs cellular wireless networks: The Daedalus implementation and experience," Wireless Personal Communications, Vol. 4, No. 2, 1997, pp