Comparing different handoff schemes in IP based Micro-Mobility Protocols Mona Ghassemian; A.Hamid.Aghvami Centre for Telecommunications Research, King’s College London, Strand, London, WC2R 2LS UK Tel: +44 2078482596 , email: Email:
[email protected]
ABSTRACT The semi-soft handoff scheme for Cellular IP is optimized for networks where the mobile host is able to listen/transmit to two or more base stations simultaneously for a short duration- the time that the mobile host spends in the overlap area- as in the case of a WaveLAN or Code Division Multiple Access (CDMA) network. This scheme improves the network efficiency off handoff in terms of packet loss but with more delay. However not all wireless technologies have simultaneous connection capability as does a Time Division Multiple Access (TDMA) network. A new scheme introduced in a cellular IP draft [2] has been implemented in a network simulator and is discussed in this paper. Indirect Handoff has been implemented for the networks where the mobile host is able to listen/transmit to only one base station. It will be shown in this paper that for the wireless technologies where a semi-soft scheme can-not be used, an indirect handoff scheme is a more efficient alternative to hard handoff in cellular IP micro mobility protocol. The behaviour of TCP and UDP at handoff will then be discussed. Finally, the handoff performance for Cellular IP, Hawaii [?] and HMIP[7] will be compared.
I. Introduction Typically, fixed hosts connected to the Internet remain online for extended periods of time even though most of the time they do not communicate. Mobile subscribers connected to the wireless Internet will expect similar services. If mobile hosts were to maintain their location information in support of being continuously reachable, it would require frequent location updates. This signaling overhead can be reduced through the introduction of paging. Mobile hosts are expected to typically operate on batteries with limited lifetime. This makes it important to save idle mobile hosts from having to transmit frequent location update messages. Idle mobile hosts do not have to register if they move within the same paging area. Rather, they only register if they change paging area. Paging has been implemented by a number of micro-mobility protocols including Cellular IP [2] and Hawaii [?], and recently
proposed as extensions to Hierarchical Mobile IP [5]. Hierarchical mobility management reduces the performance impact of mobility by handling local migrations locally and hiding them from home agents.
II. Cellular IP Protocol The Cellular IP (CIP) proposal [2] from Columbia University and Ericsson supports fast handoff and paging techniques. The protocol is intended to provide local mobility and handoff support. It can interwork with Mobile IP [1] to provide wide area mobility support. A mobile host may sometimes wish to maintain its routing cache mappings even though it is not regularly transmitting data packets. A typical example of this is when a mobile host receives a UDP stream of packets on the downlink but has no data to transmit on the uplink. To keep its routing cache mappings valid, the mobile host transmits route-update packets on the uplink at regular intervals called the route-update time. Route-update packets update routing cache mappings as for normal data packets. However, route-update messages do not leave the Cellular IP access network and will be filtered in the gateway. The loss of downlink packets when a mobile host moves between access points is reduced by customized handoff procedures. Cellular IP supports three types of handoff scheme. A. Cellular IP hard handoff: is based on a simple approach that trades off some packet loss in exchange for minimizing handoff signaling rather than trying to guarantee zero packet loss. Hard handoff causes packet losses proportional to the round-trip time and to the downlink packet rate. B. Cellular IP semi-soft handoff: exploits the notion that some mobile hosts can simultaneously receive packets from the new and old base stations during handoff. During semi-soft handoff a mobile host may be in contact with either the old or the new Base Station and receives packets from them. Packets intended for the mobile host are sent to both Base Stations, so when the mobile host eventually moves to the new location it can continue to receive packets without interruption. [2].
C. Cellular IP Indirect Handoff: As mentioned before, not all wireless technologies have simultaneous connection capability. e.g. TDMA. They cannot listen to the current BS while sending a route-update packet to the new Base Station. For this situation an alternative indirect technique is used for Cellular IP. It is assumed the network can obtain the IP address of the new BS. This is the case in many cellular networks.
A. UDP Performance During the simulation the mobile host receives 512 byte UDP packets at rates from 50Kbps to 1.2Mbps while making handoff from BS1 to BS4 and vice versa.
When the mobile host decides to make a handoff, instead of sending a route-update packet to the new BS directly (as it cannot), it sends the packet to the current BS. This packet will have as its destination IP address, the IP address of the new Base Station [2].
III. HAWAII PROTOCOL
The Hawaii protocol defines two schemes for implementing confirmed handoff procedures within base stations: the forwarding and non-forwarding schemes. In the first type, packets are forwarded from the old base station to the new, whereas in the second, they are diverted at the Crossover router. The forwarding scheme is optimised for networks where the mobile host is able to listen/transmit to only one base station as in the case of a Time Division Multiple Access (TDMA) network. The non-forwarding scheme effectively implements route diversity and is optimised for networks where the mobile host is able to listen/transmit to two or more base stations simultaneously, as in the case of Code Division Multiple Access (CDMA) networks.
Figure 1. The Cellular IP platform As shown in Figure 2, for higher data rates a hard handoff scheme loses up to 30 packets. The experiment shows that, for wireless networks that cannot use a semi-soft scheme to decrease packet loss, indirect handoff could be used. The number of packets lost in Hard handoff is reduced by up to a half when using indirect handoff. This improvement is achieved because, before the mobile host moves to the new base station, the new base station has already been notified of the existence of the new mobile host in its coverage area through the packets forwarded from the old base station and has established the links so the data is on its way when handoff occurs. In this way those packets, which are lost in a hard handoff scheme during the establishment of the new link, are saved and arrive at the mobile host as soon as it switches to the new base station.
Pkloss (packets)
Handoff-Aware Wireless Access Internet Infrastructure (Hawaii) [4] is a domain-based approach for supporting local mobility. It uses path set-up schemes based on caching. Hawaii, like Cellular IP, aims at extending Mobile IP to allow efficient global multimedia mobility. It is reasonable to believe that Cellular IP may perform slightly better than the Hawaii protocol since a similar cache approach is used
35 30
Hard 25
Indirect
20
IV. HANDOFF EVALUATION
Semisoft
15 10 5
12 00
11 00
80 0 10 00
50 0
40 0
35 0
30 0
25 0
20 0
15 0
50
0 10 0
To compare different handoff schemes on Cellular IP, we used CIMS [3] codes implemented in ns-2. We also added indirect handoff scheme codes to CIMS and compare the three schemes for UDP and TCP applications. The platform used for the simulation is shown in Figure 1.
Data rate (kbps)
Figure 2. Number of packets lost when data rate is increasing for three handoff schemes in cellular IP.
Throuput (Mbps)
seconds due to the transmission delay variations encountered during handoff. The semi-soft handoff results look promising. Even at the highest handoff rate TCP throughput is almost identical to that of a stationary host as shown in Figure 4.
1.45 1.25 1.05
semi-soft indirect hard
0.85
140 120 100 80 60 40 20 0
0.65
SemiSoft
5
Indirect
10
Hard
15 20 Number of Handoff/min
Figure 4. Throughput in TCP Newreno with different speed of mobility for CIP handoff schemes 1.5
0.6 0.4 0.2 Time spent in overlap region(sec)
Figure 3.Number of packets lost when mobile host is moving with different speeds for three handoff schemes in cellular IP. B.
TCP Performance
In the next experiment, we study the impact of handoff performance on TCP Newreno throughput. The mobile host performs consecutive handoffs between B1 and B4 and vice versa. The average TCP throughput at the mobile host for all handoff schemes is shown in Figure 4. We observe that the performance of TCP degrades as the handoff frequency increases, due to packet loss. As the handoff rate increases TCP has less time to recover from loss. This forces TCP to operate below its optimal operational point resulting in a significant drop in transport performance as the handoff rate approaches one per 3 seconds. Note that a mobile handing off every 3 seconds is a high handoff rate. From Figure 4 we can observe that semi-soft handoff reduces packet loss and significantly improves the transport throughput in relation to the hard and indirect handoff schemes. In contrast to the semi-soft handoff experiment for UDP traffic, packet loss is not entirely eliminated with TCP. This can be observed in the decline in the measured throughput as the handoff frequency increases. We attribute the lack of synchronization and subsequent loss to the buffer delay device used. The buffering of packets is tied to the packet inter-arrival time, which is both shorter and more irregular in TCP streams than in the case of the UDP experiment [5]. A slight disturbance remains at handoff rates approaching one handoff per 3
As we observe from Figure 5, the packet loss is reduced in indirect handoff implemented compared to the hard handoff scheme, as, within the same simulation time, the maximum sequence number achieved in indirect handoff is higher. This shows that sender receives fewer duplicated acknowledgements and needs less retransmission, so the number of the sequence number of packets sent to the mobile host will increase more. As can be seen in Figure 5, greater packet loss is achieved in the implementation of indirect handoff in comparison to semi-soft handoff. This is due to the longer time required in indirect handoff to inform the new base station about the presence of the mobile host moving into its coverage area. This delay is approximately three times larger in indirect handoff than in semi-soft handoff where the mobile host can notify the new base station directly when the procedure for handoff is starting.
Sequence numbers
PkLoss(packets)
As the mobile host moves faster, it spends less time in the overlap region so the packet loss increases, as shown in Figure 3. This result can be explained if we compare the handoff delay to the time the mobile host spent in the overlap region. In the semi-soft handoff the increase in packet loss is less because the mobile host can hear from both base stations once the routing update packet has arrived at the crossover router. However unlike the semi-soft scheme, a hard or indirect handoff mobile host can only send and receive packets to one base station at a time so it can-not receive all the packets sent to it and loses more packets.
7000
Indirect-Handoff
6000
Semi-soft-Handoff
5000
Hard-Handoff
4000 3000 2000 1000 0 0
20
40
60
Simulation time(sec)
Figure 5. Packet sequence number sent in TCP Newreno during 6 handoffs
The major difference between the two approaches is related to handoff procedures and to path refresh (i.e. cache management). Hawaii involves dynamic routing whilst only selected nodes implement the mechanisms for path Caching [5]. In comparison with Cellular IP, Hawaii only uses one cache refresh frequency, and the user data is not involved in cache management.
number received at the end of the simulation. The degradation caused by packet loss increases with the increasing handoff frequency. This phenomenon is illustrated in Figure 7 where the mobile host moves at the speed of 20 meters/second and performs 3 handoffs during a 30 seconds simulation time.
Max Seq no.
V. COMPARING CELLULAR IP , HAWAII AND HMIP
Handoff latency is proportional to the round-trip time between mobile host and crossover node. When mobile is doing a handoff with smaller Handoff distance the time that packets could be lost is shorter which results less packet loss in total comparing to bigger handoff distances.
In contrast to Cellular IP, Hawaii and HMIP[7] updates routing only when registration messages reach the GFA[6]. Therefore, Hierarchical Mobile IP cannot benefit from the fact that a crossover node is topologically close to the BSs. This phenomenon is illustrated in Figure 6 where the handoff delay for HMIP is shown to be independent of the crossover distance, and is equal to the handoff performance in the case of the maximum crossover distance for Cellular IP and Hawaii.
semisoft 1600
indirect hard
1100
Hawaii-F 600
Hawaii-NF
100 0
9
10
12
15
20
21
Handoff/min
Figure 7.Maximum sequence number for TCP Newreno flow control As mentioned before, when the receiver receives nonsequential packets, it sends duplicate acknowledgements to notify the sender that it should retransmit the lost packets. Figure 8 shows the number of duplicated acknowledgements for the same scenario as Figure 7. The number of non-sequential packets arriving at the mobile host is higher in Hard handoff than in the other two schemes in Cellular IP. In Hawaii, on the other hand, we can see that the number of duplicate acknowledgements is greater in the forwarding scheme. 140 duplicate acks
Comparing packet loss numbers for UDP applications in Cellular IP and Hawaii as shown in Figure 6, Cellular IP hard handoff and Hawaii MSF are very similar. When the mobile host moves between BS1 and BS2 the delay is small. If the crossover distance is larger, the handoff delay increases with an extra packet delay for each additional hop. The results are a direct consequence of the similarity between these two protocols, particularly in the way in which the protocols build up the route between a crossover node and new BS.
2100
120 100 80 60
Number of packet loss
40 2 1.5
Hawaii_MSF
20
CIP-Hard
Semi-soft Hawaii-NF
HMIP
indirect Hawaii-F
hard
Figure 8. Number of duplicated acknowledgements the mobile host sends when it receives disordered packets
1 0.5 0 HD1
HD2
HD3
crossover router distance
Figure 6. Comparing Hawaii, CIP and HMIP for UDP applications For TCP applications, the more the sender needs to retransmit packets, the smaller is the last sequence
Therefore evaluation and comparison of the performance of the three CIP schemes for both UDP and TCP was performed. Indirect handoff is a better alternative to hard handoff in cellular IP for wireless networks that do not support simultaneously sending/receiving.
VI. An overview on Micro-Mobility Protocols
spent in the overlap region then the Semi-soft handoff scheme has the best performance for Cellular IP protocol.
We assume here that the delay between a MN and the gateway is T(gate), between a MN and its previous base station is T(prev) and the delay between the MN and the crossover base station is T(cross). we can say T(gate)>T(prev)>T(cross). Gateway Crossover router
Global Internet
Tgate
Tcross
OAR
NAR
In a Cellular IP network the number of hops from source to destination is constant since all routing-update packets must reach the gateway. However in Hawaii, nodes upper than Crossover router are not involved when handoff is happening. Therefore Hawaii is more reliable and, compared with Cellular IP, has less control signalling in the nodes upper than the Crossover router compared to Cellular IP. The longer handoff delay in the Hawaii handoff scheme causes greater packet loss than in Cellular IP handoffs. Comparing packet loss numbers for UDP applications in Cellular IP and Hawaii, at higher handoff rates the performance of both the forwarding and the nonforwarding schemes in Hawaii is worse than that of all handoff schemes in Cellular IP.
Tprev
Figure 10: A simple model for handoff parameter comparison
REFERENCES
Table 1 summarises how some of the proposed micromobility protocols tackle each Protocol Design Issue. A detailed description of how each of the selected protocols works can be found in the appropriate reference; in this paper we assume that the reader is reasonably familiar with them.
? icroMobility BRAIN [9] CIP HAWAII
HMIP EMA[8]
[1] C. Perkins, editor, “IP Mobility Support”, Internet RFC 2002, Work in Progress, October 1996. [2] Z. D. Shelby, D. Gatzounas, A. T. Campbell, C-Y. Wan, “Cellular IPv6”, Internet Draft, draft-shelbyseamoby-cellularipv6-00.txt, November 2000. [3] CIMS: The Columbia IP Micro-mobility Suite NS Source Code Distribution for Cellular IP, Hawaii and Hierarchical Mobile IP, April 2001
HO type
HO Latency
HO Losses
Soft/ Hard
2* T(anchor)
R* T(anchor)
L3
semi-soft/ hard/ Indirect
2*T(gate)
0/ R*T(cross)
[4] R. Ramjee, T. La Porta, S. Thuel, K. Varadhan ”IP micro mobility support using HAWAII”, Internet Engineering Task Force, Internet-draft, 19 February 1999, Work in Progress.
L3
[5] Grønbæk, “Cellular IP: overview and enhancements
Forward/ 2*T(prev) NonForward
R*T(prev)/ R*T(cross)
L3
HO Layer
“, March 1999.
Hard
2*T(gate)
R*T(gate)
‘L3.5’
MBB/BBM
3*T(prev) +T(tora)
0
L3
[6] A.T.Campbell,JGomez,S. Kim, Turanyi, Z., Wan, CY. and A, Valko "Comparison of IP MicroMobility Protocols", IEEE Wireless Communications Magazine, Vol. 9, No. 1, February 2002.
Table 1: comparative chart for handoff parameters
For the case of EMA and BRAIN micro-mobility protocols a higher scalability and resilience are expected since they are not only based on tree_like topologies and could support mesh architecture. Therefore if a link or a node fails there are other pathes that a mobile node could send or receive data.
VI. Conclusions The greater the handoff distance, the more packet loss results in all handoff schemes in Cellular IP. If the handoff delay is less than the time that the mobile host
[7] E. Gustafsson, A. Jonsson, and C. Perkins, “Mobile IP Regional Registration,” Internet draft, draft-ietfmobileip-reg-tunnel-03, work in progress, July 2000. [8] A. O'Neill, G. Tsirtsis, and S. Corson, "Edge Mobility Architecture", Internet Draft (work in progress), draft-oneill-ema-01.txt,March ‘00. [9] R. Hancock, Hamid Aghvami, M. Kojo, M. Liljeberg,"The Architecture of the BRAIN Network Layer", Proc. IST Mobile Communications Summit 2000.
