Performance Comparison between Multihomed Network Mobility Protocols Mohammed Atiquzzaman Telecommunications and Networking Research Lab The University of Oklahoma Md Shohrab Hossain University of Oklahoma William D. Ivancic NASA Glenn Research Center, Ohio IEEE GLOBECOM @ Anaheim, CA Dec 4, 2012
1
Outline of the Presentation Motivation and Problem Statement Multihomed NEMO and SINEMO Experimental Setup Results Conclusion
2
Why Mobility Protocols Satellites with IPenabled devices capture videos, images and send them to control centers on earth Need to maintain continuous connectivity with remote computer Mobility protocols are required to ensure session continuity
3
IETF Solution to IP Mobility: Mobile IP
Employs mechanism
similar to postal service mail forwarding Problems:
Correspondent Node (CN) Home Agent
Packets from CN to MH
Internet
Encapsulated Packet
Inefficient routing High handover latency Packet loss
Foreign Agent
Home Network
Decapsulated Packets
Visiting Network
4
Network Mobility (NEMO) A collection of nodes moving as a unit (Example: airplanes, trains, ships) Mobility can be managed in an aggregated way in NEMO Mobile Router acts as default gateway and manages mobility on behalf of mobile network nodes
HA
CN
5
NEMO Architecture
Inside NEMO MR: Mobile Router LFN: Local Fixed Node LMN: Local Mobile node VMN: Visiting Mobile Node Problems: Heavy load on HA Drop in throughput during handover
Data path
6
Multi-homed NEMO
Network layer solution Exploits Make-before-break Strategy and MCoA registration policy in HA We name it M-NEMO Problem: • Cannot sense network capacity and adjust accordingly
7
SIGMA Transport layer solution proposed by researchers at the TNRL lab Exploits IP-diversity of a mobile host Benefits: Establishes a new connection before disconnecting the old one Decouples location management from data transmission Less handover delay and packet loss, Optimal routing between MH-CN CN Location Manager
8
SINEMO SIGMA-based solution for mobile networks The MR maintains a translation table for all the mobile network nodes MNN’s private IPs do not change
Works in Transport layer Can sense network capacity in heterogeneous environment • Can adjust rate accordingly
Default gateway 9
Objective and Contributions Objective: To compare two multi-homed network mobility protocols working in two different layers: M-NEMO (Network layer) SINEMO (Transport layer)
Contributions: Build linux-based experimental testbeds for performance comparison of two multi-homed network mobility protocols that exploits makebefore-break strategy Illustrate and analyzing their handover performance in terms of throughput, queue occupancy and retransmissions
10
Experimental Evaluation Performed experimental evaluation of M-NEMO and SINEMO Built Linux-based experimental testbed To capture real network phenomena, the testbeds were connected to the University of Oklahoma operational network that carries production traffic
11
M-NEMO testbed
Mobile Network
12
Software and Hardware Configurations for M-NEMO testbed Device Type
Software Configuration
Hardware configuration
MR
Ubuntu 8.04 Kernel 2.6.23 + NEPL
CPU: Intel Pentium 4, 2.20 GHz, 512 MB RAM, NIC: 802.11 based Netgear MA111
LFN
Windows XP + FTP Client CPU: Intel Celeron, 2.19 GHz, 256 MB RAM
HA
Ubuntu 8.04 2.6.23 + NEPL
ARs
FC6 2.6.18-1 kernel + CPU: Intel P4, 1.50 GHz, 512 MB RAM radvd-1.0
APs
Channel 6 and Channel 11 DLink WBR-1310
CN
CN & Windows Vista + CPU: Intel Core 2 Duo, 2.2 GHz, 2 GB RAM FTP Server
Kernel CPU: Intel Core 2 Duo, 2.20 GHz, 2 GB RAM
13
SINEMO testbed
Mobile Network
14
Software and Hardware Configurations of SINEMO testbed Device Type
Software Configuration
Hardware configuration
MR
FC5 + iptables
CPU: Intel Pentium 4, 2.20 GHz, 512 MB RAM, NIC: 802.11 based Netgear MA111
LFN
FC5 + lksctp-tools 1.0.6
CPU: Intel Pentium 4, 1.73 GHz, 1 GB RAM
ARs
FC6 2.6.18-1 kernel + CPU: Intel P4, 1.50 GHz, 512 MB RAM radvd-1.0
APs
Channel 6 and Channel 11 DLink WBR-1319
CN
FC5 + lksctp-tools 1.0.6
CPU: Intel Celeron, 2.8 GHz, 512 MB RAM
15
Results To find out how the two protocols behave, we conducted two types of experiments: Handover between Homogeneous capacity networks Handover between Heterogeneous capacity networks
Measured following performance metrics to find out the impact of handover on the network components:
Throughput at LFN Handover delay Number of retransmission by CN Queue length at the AR
Captured packet flows withWireshark network protocol analyzer Mobile network moved from AR1 to AR2 16
Throughput at LFN during Homogeneous Handoff M-NEMO
SINEMO
M-NEMO handover
SINEMO handover
In both cases, throughput does not drop to zero •
Reason: Established connection with AR2 before disconnecting from AR1
LFN is able to receive data through MR’s other interface Hence, CN is NOT required to retransmit packets 17
Queue Length at AR2 during Homogeneous handoff M-NEMO
Subnet 1 7.80 Mbps
SINEMO
Not many packets queued
Subnet 2 7.62 Mbps
Subnet 1 7.75 Mbps
Subnet 2 7.50 Mbps
Not many packets queued at AR2 buffer for both protocols Reason: Not much difference between the capacity of the two access networks
18
Throughput at LFN during Heterogeneous handoff M-NEMO
M-NEMO handover
SINEMO
SINEMO handover
Both handoffs seem to similar, but there are issues that affect overall performance of the access network and other users As a network layer-based solution, M-NEMO cannot sense the change in capacity of the new access network (AR2), causing performance issues explained in the following slides 19
Queue Length at AR2 during Heterogeneous handoff M-NEMO
SINEMO
Larger queue length Subnet 1
7.80 Mbps
Subnet 2 720 Kbps
Smaller queue length Subnet 1 7.75 Mbps
Subnet 2 708 Kbps
M-NEMO: CN sending traffic at a previous rate causes queue buildup in the AR2 • Affects other users sharing AR2 in M-NEMO SINEMO: CN adjusts its rate and queue length is almost similar to the homogeneous handover scenario
21
Conclusion Performed experimental evaluation of two multi-homed-NEMO architectures operating in two different layers Presented results showing impact on the access networks for handovers between homogeneous and heterogeneous capacity networks Experimental results show that SINEMO (operating in Transport layer) can adjust quickly in response to the network capacity changes unlike M-NEMO (operating in Network layer)
22
Thank You http://cs.ou.edu/~atiq
[email protected]
The research work was funded by NASA
23
