Transcript Mobile IP: Introduction - National Chi Nan University

Mobile IP: Performance
Reference: “Performance evaluation of Mobile IP protocols in a
wireless environment”; Dell'Abate, M.; De Marco, M.;
Trecordi, V.; Proc. IEEE International Conference on
Communications (ICC), 1998; pp. 1810 -1816
(MobileIPUnicast-1.pdf)
Mobile IP (MIP)
2
Route Optimization Mobile IP (ROMIP)
3
MIP vs. ROMIP
• Inefficiencies of MIP
– Triangle routing
– Home Agent overloading
• Advantages of MIP
– Simple
– Exchange of control messages is limited
– Address bindings are highly consistent
4
MIP vs. ROMIP (cont)
• Advantages of ROMIP
– Direct routing
– Handover management
 A moving host informs its previous FA about the
new care-of-address, so that packets tunneled to
the old location can be forwarded to the current
location
 In MIP, those packets had to be discarded or sent
to the HA again
• Disadvantages of ROMIP
– Complex
 Control messages, processing overhead
– Cached bindings are possibly inconsistent
5
Hypothesis for Simulation
• Mobile hosts always obtain a dedicated
bandwidth wireless connection to the
currently visited subnet
• Update process model of ROMIP
– Binding acquisition
 HA, just after having tunneled the 1st packet,
sends a binding warning message (W) back to the
source
 The source, in response to this warning, sends a
binding request message (R) to the HA, keeping
on sending user packets in the meanwhile
 The HA replies with a binding update message (U),
containing the requested care-of-address
6
Hypothesis (cont)
– Direct routing
 The source caches the received binding and uses it
to tunnel its packets directly to the FA (FA1)
– Handover
 The destination suddenly moves under another FA
(FA2); just after its movement, it sends two binding
update messages (U), both to its HA and to its
previous FA (FA1)
 The source has no way to get aware of the
movement and keeps on emitting user packets to
FA1. These packets get lost until FA1 receives the
above update
 As soon as FA1 gets updated, it warns the source
and forwards incoming packets to the actual
location (FA2)
7
Hypothesis: Time Model for ROMIP
8
Hypothesis: Fixed Network Topology
9
Hypothesis: Mobile IP Router Model
home list
visitor list
binding cache
routing table
10
Hypothesis: Mobile host Model
11
Hypothesis: Traffic pattern
• Traffic pattern
– Packet group (geometric r.v.) at each arrival
of a Poisson process (Bulking Poisson
Process)
– All packets in a group share their destination
address, drawn uniformly among all mobile
hosts’ addresses
– Two traffic descriptors
 Average session length (S, in kbit)
 Average offered load (L, in kbit/s)
 L = S/T, where T is the mean group arrival time
12
Traffic Pattern
13
Hypothesis: Mobility pattern
• Mobility pattern
– Mobility events occur at the arrivals of a
Poisson process
 When a mobile host enters a new subnet, it stays
there for a negative exponential random time
 p.d.f. = le-lt , P.D.F. = 1 - e-lt
– Descriptor
 Average mobility rate (the inverse mean stay time)
14
Theoretical Analysis
• R (packets/s): Rate at which control packets are issued
by ROMIP protocol, normalized for a single user
• Tstay (sec): Mean stay time for a mobile host
• L (kbit/sec): Mean user offered load
• S (kbit): mean session duration
• Bradio (kbit/s) : available one-way bit rate on the
radio channel
# session/sec
Binding update to HA
Binding update to FA
W, R, U
# handover during a session
15
Theoretical Analysis (cont)
• Discussion
– The control load due to the birth of new
sessions decreases by increasing the session
length (at a parity of user load, L)
– The control load due to handover events could
be brought down by increasing the radio
channel capacity (at a parity of user load, L)
– As user load L increases, a proportional control
load increase is induced; this reaction does not
take place in MIP, for which is simply R = 1/Tstay
16
Theoretical Analysis (cont)
• Validation for simulation
– Little’s formula: Npkt =
lpkt * Tpkt
– Npkt is the average # of packets in the system
(user + control)
– lpkt (1/sec) is the overall offered load (user +
control)
– Tpkt is the mean end-to-end packet delay
(obtained by weighting user and control delay)
– lpkt = L / Plength + R
– Plength is the IP data unit length
17
Simulation Result- Fig. 9
18
Simulation Result- Fig. 10
19
Discussion- Fig. 9 & 10
– 1. At null mobility rate, end-to-end delay always
increases as session duration increases
– 2. With S = 100 Kbit
 The minimum delay is obtained, i.e. a value slightly
higher than the time needed to transmit a packet over
the source and destination radio links (2*8)/19.2
 Any further remaining part of a delay rises up to in the
backbone
 For null mobility, the above gap ought to be ascribed
only to the increasing traffic burstiness
20
Discussion- Fig. 9 & 10 (cont)
– 3. Increasing the mobility rate, the MIP delay
also increases
 Owing to network load and tracking effort
– 4. A similar increase is observed for ROMIP too,
except for the 400 Kbit session, reason:
 Suppose that session end-points mobility results in
traffic scattering in the backbone, thus improving the
delay performance over that obtained with lower
mobility
 With ROMIP, source and destination mobility cuts up
the longest sessions into small pieces, thus canceling
burstiness effects in the backbone
21
Discussion- Fig. 9 & 10 (cont)
– 5. ROMIP may gain efficiency with longer
sessions, because of the source binding
acquisition process
– 6. Increasing session length, the MIP delay also
increases
 Since longer and longer traffic bursts make HA more
congested
 ROMIP seems to be much less sensible to session
duration
 However, it is evident that MIP delay performance
improves and gets closer to ROMIP’s for relatively
short sessions (100 Kbit)
22
Simulation Result- Fig. 11
23
Discussion- Fig. 11
– 1. For short sessions, MIP achieves much lower
delay than ROMIP
 In fact, short sessions hardly enter their direct routing
phase provided by ROMIP
 In these condition, ROMIP degenerates and delivers
packets by triangle routing;
 Moreover, it floods the network with useless control
messages, giving rise to a performance drawback
– 2. For longer sessions, ROMIP delay performance
improves
 Because of direct routing
 In MIP, the links surrounding the HA rapidly become
choked up by packet trains, giving rise to a huge delay
24
Simulation Result- Fig. 12
Exchange???
25
Discussion- Fig. 12
– Packet loss
 Due to transmissions to the wrong subnet
– Better performance for ROMIP
 Because of handover support
 But the performance is not substantial
– Loss probability could be reduced by increasing
the backbone bandwidth, to allow a more
effective tracking of mobile hosts
26
Simulation Result- Fig. 13
27
Discussion- Fig. 13
– Right side: cache agent overhead for tunneling
operations
 Linear relation between processing load and offered
traffic exists, but only for low traffic volumes
– For low mobility and low traffic, left-side diagram
 Redirected packets have been tunneled only once (ideal
operating region for ROMIP)
– For High traffic
 The location tracking algorithm lags behind
 On the average, more than one tunnel hop is needed for
a packet to catch the destination
28
Simulation Result- Fig. 14
For ROMIP
29
Discussion- Fig. 14
– Impact of cache size over quality of service
– A small cache capacity gives rise to a lower loss
and a higher delay
– A large capacity originates a higher loss and a
smaller delay
 A large amount of cached binding may be inconsistent,
but packets succeeding in reaching their destination
often travel along the shortest path
– Small lifetimes (timeout values)
 May keep the bindings up-to-date, but it is more likely
that a valid binding is removed and thus triangle
routing occurs
30
Conclusion
• MIP shows better performance, when
– The rate of birth and death of sessions is high
• Large session duration
– Exploit the optimization of routing by ROMIP
• As long as the traffic bursts last on average as
much as the average cell permanence time
– The direct routing of ROMIP allows to better
distribute the traffic offered to the fixed network
– Indirect routing (MIP) is subject to overload of
the HA
31