Rational Exchange – A Formal Model Based on Game Theory

Download Report

Transcript Rational Exchange – A Formal Model Based on Game Theory

UCAN: A Unified Cellular and
Ad-Hoc network Architecture
By
H. Luo, S. Lu
Computer Science Department
UCLA
1
What is new?
 The goal is to improve the throughput performance
of the wide-ares wireless network by taking
advantage of the adhoc network.
 Try to take the benifits of both networks: 3G cellular network
and adhoc network
 3G BTS forward packets to proxy client with better quality
 Proxy clients use adhoc network to forward packets to
destination client
 Maintaining fairness by refining the 3G BTS scheduling
algorithm
 Develop greedy and on-demand protocol: UCAN
 Develop a secure crediting mechanism to motivate users
2
Comparison of Wide-area wireless network and
Local-area wireless network
Wide-area wireless
network
Local-area wireless
network
Coverage
Up to tens of
kilometers, e.g.,
20kms.
Up to hundreds of
meters, e.g., 250m for
Wi-Fi
Throughput
Low.
Up to 2Mbps, mainly
for voice
High.
11Mbps for 802.11b,
54Mbps for 11a/11g
Organization
Infrastructure mode,
one hop.
Ad-hoc, multi-hops
3
Requirement for UCAN
 It requires that the mobile terminals are equipped with two
interfaces: 3G & IEEE802.11b
 A fundamental question: why should a mobile user relay
traffic for other users ?
 Contributions of this paper:
• A novel architecture of UCAN
• Develop protocols including new proxy discovery, ad-hoc
routing;
• Refining scheduling algo for 3G BTS to balance traffic
• Sercure crediting to encourage relaying
• Increase single / aggregate HDR downlink throughput
4
Organization of this paper
 Sec1: introduction
 Sec2: background fo HDR, 11b, and related work
 Sec3: motivation to develop UCAN with examples
 Sec4: architecture of UCAN
 Sec5: proxy discovery and adhoc routing
 Sec6: enhance of scheduling algorithm of 3G BTS
 Sec7: secure credit to motivate relay
 Sec8: simulation results
 Sec9: related discussions
 Sec10: conclusion
5
Background knowledge
 HDR—anywhere, always on
• Part of 3G CDMA2000 standard, for burst data.
• UL 154k, DL: 2.4M, shared in TDMA mode.
• Duration of each user in DL is defined by the
scheduling algo: Propotional scheduling algo.
 IEEE802.11b—named as Wi-Fi
• most popular among the 802.11 family
• 2 modes: infrastructured mode and ad hoc mode
• using routing protocol to relay data: multi-hop
6
Related work of combing two wireless networks
 Category:
• traffic model -- peer-peer, infrastructure mode
• relay model -- stationary, mobile
• # of interface – one or two
 UCAN: infrastructure mode traffic, two interface, and mobile
relays.
 Disavantages of other options:
• peer-peer traffic: inefficient to provide high availability
service
• one interface: throughput limited by the bandwidth limited
cellular system
• Stationary relays: iCar system, increased cost.
7
Motivated by a simple example
 Non-relay case:
FTP Server
HDR cellualr link
 Relay case:
FTP Server
HDR cellualr link
Laptop inside room
Relay client
in corridor,
better HDR DL
Laptop inside room
802.11 link
8
Throughput comparison of the simple example
9
Architecture of UCAN
 HDR DL quality is measured by clients within BTS’s coverage
 When the DL quality is below a certain level,HDR will forward the
data through possible Proxy client, Relay clients.
 Clients should be able to operate in dual mode: HDR and 11b. 10
Issues need to be addressed in UCAN system
How does the HDR BTS disceovery a proxy
server?
• Proxy discovery and routing
How does the HDR BTS maintain fairness
among mobile nodes?
• HDR Scheduling algorithm
How are the mobile nodes encouraged to
paticipate in the traffic relaying?
• Secure crediting algorithm
11
Proxy Discovery and Routing: Greedy algo
 Proactive; unicast from destination client;route recorded in the RTREQ
message until until it reaches proxy client, and then forwared to HDR
BTS via a proxy application.
12
Proxy Discovery and Routing: On-demand algo
 Reactive; broadcast from destination client;route recorded in the
RTREQ message until it reaches proxy client candidates, and then
forwarded to HDR BTS via independent proxy applications.
13
Route and Proxy Maintenance -1
 Route Failures and Recovery
• 11b Mac layer generate a callback function to inform the client
such failures
• Client reports the failure to HDR BTS
• HDR BTS use HDR DL in replacement of the relay path, eliminate
the Proxy table
• A new route request may be initiated: proxy re-discovery
 Proxy Maintenance
• If the HDR DL of the proxy degrades, this proxy should be
replaced
• Not a good idea to initiate proxy discovery periodically: too much
overhaed and difficult to determine the interval
• Solution is to pigggy back the channel rate of the proxy client to the
destination client, and let the destination client to decide whether
or not to initiate a new round of proxy discovery
14
Route and Proxy Maintenance -2
 Route consistency and loops
• Solution is to include the entire relay path in the RTREQ message
to exclude route loop
15
Refining scheduling algo in HDR
 Original shceduling of HDR
• Tradeoff between throughput and fairness: DL channel is shared
by users in TDMA mode
• BTS selects the minimum Tk(t)/Rk(t), where Tk(t) means the
avearage throughput of the kth user in time t (with an arbitary
window size of w), and Rk(t) is the current DL channel rate of user
k at time t.
 Refined scheduling algo in UCAN
• Tk(t) is represented by the number of bits received by destination
client within the arbitary window size
• How to represent Rk(t)?
 Choice one : use that of the proxy client, or
 Choice two: use that of the destination client
 The key idea is to maintain fairness between clients and encourage
mobile nodes to become proxy clients
16
Refining scheduling algo in HDR-example
 Suppose DL channate rate of destination and proxy client node is 1:2
 When no relay is used, slot scheduling between destination and proxy
client is 1:1
 When relay is used, use proxy’s rate as criterion, slot scheduling
between destination and proxy client is 1:1, resulting in unfairness
 When relay is used, use client’s rate as criterion, slot scheduling
between destination and proxy client is 1:2, no unfairness
17
Allow diversity in UCAN
 Due to fast fading, max average channel rate may not
stand for the max instant channel rate
 In UCAN, allow HDR BTS to forward data to the client
along the relay path with the highest data rate, instead of
the proxy server.
 The HDR BTS need to know the complete relay path
 Additional processing is needed to keep the packets in
order
18
Secure Crediting mechanism
 The goal is to enourage clients to act as proxy client for
other nodes
 Extra incentive is givne to the proxy client and all the
clinets along the relay path by accumulating credits for the
them, besides the refining scheduling algorithm
 Detailed crediting is discussed in other papers
 Focus on authentications of clients along the relay path
• The basic idea is to include an authentication key
between two neighbouting clients in the relay path, and
forward the keys to the HDr BTS
• HDR BTS can discriminate the cheating clients by
requesting the compuataion of the authentication keys
19
Experiments and performance evaluations-1
 HDR Channel model
20
Experiments and performance evaluations-2
 HDR Channel Rate: instant and average
21
Experiments and performance evaluations-3







Simulator: ns-2
Application: FTP/TCP and CBR/UDP
Speed of mobile clients: 0, 2, 5, 10, 15m/s
Radius of HDR Cell: 500m, only one cell
Number destination clients : one or more
Number of other relay/proxy clients: 30~100
Two srecial techniques:
• Aggregate of data frames at the proxy client: because
each HDR DL frame is 128 bytes in average at speed
of 600kbps and 1.67 ms per slot. Inefficiency in IP.
• Scoped Neighborhood advertisement: In greedy proxy
discovery algorithm, use TTL in HDR DL channel rate
advertisement to reduce the overhead
22
Experiments and performance evaluations-4
 The relationship of the packet size vs relay hops
23
Experiments and performance evaluations-5
 Single destination client : throughput gain
• client placed at d = 400m to the HDR BTS, channel rate=340kbps
• radius of 802.11b is 115m at 11Mbps, so 3 hops are expected,
with channel rate = 1.25Mbps
24
Experiments and performance evaluations-6
 Single destination client : HDR uplink overhead
• The ratio of on-demand / greedy proxy discovery
algorithms
25
Experiments and performance evaluations-7
 Single destination client : Energy consumption
• The energy consume ratio of on-demand / greedy proxy
discovery algorithms
26
Experiments and performance evaluations-8
 Multiple destination clients : greedy algorithm
• Use variable TTL (from 1 to 4) in RTREQ thus result in different
length of relay path
• All the max/min throughput gains are greater than 1
27
Experiments and performance evaluations-9
 Multiple destination clients : On-demand algorithm
• Use variable TTL (from 1 to 4) in RTREQ thus result in different
length of relay path
• All the max/min throughput gain ratios are greater than 1
28
Discussions
 Strategies used in UCAN and some open issues
• Frugal Usage of HDR Links
• Base Station Pull v.s. Client Push
• Variable Data Rate and Transmit Range in 802.11
• HDR Uplink Proxy
• Co-located HDR BS and IEEE 802.11 AP
• Interaction with Peer-to-Peer Traffic
• HDR Scheduling and End-to-end Delay
• Multiple Cell Relay
• Application Scenarios In Section
29
Summary of UCAN
 Unified Cellular and Adhoc Network to improve the
throughput of cellular system
 Two approaches: greedy / on-demand proxy discovery
 Slightly better performance of on-demand with substantial
higher signaling overhead , compared with greedy algo.
 The deficiency of greedy algo is consuming more power
than on-demand algo.
 Refining scheduling algo for HDR
 Secure crediting system
30
THANK YOU !
31