Professional Documents
Culture Documents
PWN v8
PWN v8
Abstract
1. Introduction
Bluetooth [1] is a representative technology for short range wireless communication. Bluetooth
enables cell phones, PDAs, and notebooks to be connected without wire and is used to form
Personal Area Network (PAN). Minimum communication unit of Bluetooth is piconet that
consists of one master and up to 7 slaves. To connect more than 8 Bluetooth device, scatternet is
proposed and researches are still in progress [3][4][5][6][8].
The goal of the most scatternet formation schemes proposed so far is either to minimize
scatternet formation time or to maximize the performance of formed scatternets. Traffic pattern
information is useful to construct an efficient scatternet [ref?], however it is very hard to reliably
guessing estimate traffic pattern at formation time is not obvious. In practice, a person will
communicate more frequently with other persons belonging to the same social group than with
strangers belonging to other social groups [2]. Thus, wWe can apply this observation to
scatternet formation scheme so that the resulting scatternet can be more efficient. We propose a
scheme that forms small sized scatternets of socially grouped devices and then interconnects
these small groups through tunnels. The proposed scheme enhances overall performance of
tunneled scatternet since devices that may communicate frequently will have low average path
length. Tunneled scatternets formed by the proposed scheme are able to share capacity more
fairly among communicating peers than an existing scheme.
A set of scatternets interconnected by tunnels is also a scatternet, so we can employ existing
metric when evaluating tunneled scatternets. We found that the existing metric is not useful
when evaluating interconnection of multiple scatternets. We propose a new metric that fits for is
more suitable for the evaluation of tunneled scatternets. We also propose a tunnel selection
scheme that maximizes the proposed metric. Tunneled scatternets formed by the proposed
scheme are able to share capacity more fairly among communicating peers than a naïve scheme.
(Following section~)
2. Related Work
3. Design Consideration
3.1 Scenario
Our work is based on the following scenario. In a conference room, most participants have
Bluetooth enabled devices and these devices form a scatternet to interact in an ad hoc manner.
A person may enter or leave the conference room in the middle of session, so member of the
network will may change frequently changed. OAnother characteristic of such the network is
that each person belongs to one of diverse social groups. (시나리오가 추상적이네요. 좀 더 구체
적으로 쓰면 어떨까요?)
3.2
Most of scatternet formation schemes form a scatternet by connecting all devices within an area.
Devices that frequently communicate may scatter in a throughout the scatternet so average path
length will be longer than they gather together. Since people belong to the same social group
would often interact each other, we should connect them first to enhance performance (먼저 연
결한다고 꼭 performance가 올라가는 것이 아니지요. 이유를 적어야 합니다.). Table 2 and
Figure 1 show performance comparison of group aware scatternets (group aware가 뭔지 정의도
안하고 여기서 바로 나오면 안되지요…) (e.g., best and worst cases) and group unaware
scatternet. The performance of group aware scatternet is usually higher than group unaware
scatternet. In the worst cast of group aware scatternet, even though it shows low average TCP
throughput, it shares network capacity more fairly than group unaware one. Figure 1 shows
TCP throughput of three cases. One point represents average TCP throughput per 10% of
throughput that are ordered descending order. In Figure 1, Group unaware scatternet shows
high throughput only for few (in this graph, 10%) traffic compared to group aware ones. (여기
서의 설명을 보면 뭔가 앞뒤가 바뀐 것 같습니다. 이미 group-aware formation scheme이 있고
그래서 그걸 보니까 좋더라. 이렇게 하지 말고 social group을 반영해서 만들려면 어떻게 하면 좋
을지 “논리”적인 설명을 해야지요.)
Table 2. Performance comparison of group aware scatternets and group unaware scatternet
250000
TCP throughput (bps)
200000
100000
50000
0
1 2 3 4 5 6 7 8 9 10
Figure 1. Performance comparison of group aware scatternets and group unaware scatternet
[8] identifies CGs by analyzing traffic flow and it provides more accurate group clustering than
social group method. Identifying CGs requires quite amount of time at first stage, it is hard to
apply to our scenario.
In conclusion, to form an efficient scatternet, small sized scatternets of socially grouped devices
are formed and then these small groups are interconnected through tunnels. Compared to
common scatternet, tunneled scatternet may experience congestion at tunnel. A tunnel selection
scheme should estimate load per link and distribute load per link by adjusting the number or
position of tunnel(s).
Problem definition이 여기에.
To share network capacity among several communication pairs, following things should be
considered: 1) average hop count, 2) the number of branch per node, and 3) load per link.
Average hop count and the number of branch per node are already included in APC. In
addition to two factors, we should consider load per link. Load per link is defined as the
number of source-destination pair which passes that link when all possible source-destination
pairs in a scatternet communicate. A link that has the highest load per link becomes bottleneck
of the tunneled scatternet because many source-destination pairs compete to send data. Since a
quantity of traffic a source-destination pair can send is restricted by maximum value of load per
link on their routing path, we should minimize maximum value of load per link in a scatternet.
4. Proposed Scheme
To share network capacity among several communication pairs, following things should be
considered: 1) average hop count, 2) the number of branch per node, and 3) load per link.
Average hop count and the number of branch per node are already included in APC. In
addition to two factors, we should consider load per link. Load per link is defined as the
number of source-destination pairs which passes that link when all possible source-destination
pairs in a scatternet communicate. A link that has the highest load per link becomes bottleneck
of the tunneled scatternet because many source-destination pairs compete to send data. Since a
quantity of traffic a source-destination pair can send is restricted by maximum value of load per
link on their routing path, we should minimize maximum value of load per link in a scatternet
(전반적으로 공식을 사용해서 설명해야 할 것 같아요.)
In this section, we will explain how to calculate proposed metric. Figure 2(a) shows load per
link when two groups are interconnected by two tunnels. Since one node can utilize one link at
a time, actual load per link is load per link multiplied by the number of link a node has. (Here,
we assume that each node spends the same time to any link.) Final load per link is depicted at
figure 2(b).
19 19
51 50 49 50 51
19 19
50 50
46 49 50 50 46
19 19
19 19
LINK TUNNEL
(a)
57 57
153 150 98 150 153
57 57
150 150
57 57
57 57
LINK TUNNEL
(b)
Figure 2.Figure 1. Load per link
Based on load per link we can obtain total network flows. Total network flow is sum of
individual flow of each source-destination pair and individual flow means how much time a
pair can communicate per unit time. Individual flow of a pair is expressed as the reciprocal of
maximum load per link on their routing path. In some case, maximum total network flow leads
to low performance due to high average hop. Therefore we divide total network flow by
average hop count. It is final metric.
The relation between proposed metric and TCP throughput is depicted in figure 2. This
simulation also assumes that source-destination pair of intra traffic is twice compared to that of
inter traffic. A vertical axis is average value of TCP throughput of inter traffic from 20
experiments and a horizontal axis is metric value of given tunneled scatternet. We can insist that
proposed metric is suitable for evaluating tunneled scatternet since proposed metric and TCP
throughput of inter traffic has correlation.
5. Evaluation
6. Conclusion
7. Reference
(IEEE 방식에 따라 바꿔야죠~)
[1] Bluetooth Specification Version 1.1, Bluetooth Special Interest Group,
http://www.bluetooth.com, February 2001.
[2] B. Wang, J. Bodily, S. K. S. Gupta, “Supporting Persistent Social Groups in Ubiquitous
Computing Environments Using Context-Aware Ephemeral Group Service,” in Proceedings of
the Second IEEE Annual Conference on PERCOM 2004.
[3] C. Petrioli and S. Basagni, “Degree-constrained multihop scatternet formation for bluetooth
networks,” in Proceedings of the IEEE Globecom 2002, Taipei, Taiwan, November 2002.
[4] Z. Wang, R. J. Thomas, Z. Haas, “Bluenet - a new scatternet formation scheme,” in 35th
Hawaii International Conference on System Science (HICSS-35), Big Island, Hawaii, January
2002.
[5] F. Cuomo, G. Di Bacco, T. Melodia, “SHAPER: a self-healing algorithm producing multi-hop
Bluetooth scatternets,” in Proceedings of the IEEE Globecom 2003, San Francisco USA,
December 2003.
[6] G. Tan, A. Miu, J. Guttag, H. Balakrishnan, “An efficient scatternet formation algorithm for
dynamic environments,” in IASTED International Conference on Communications and
Computer Networks, Boston, MA, November 2002.
[7] T. Melodia, F. Cuomo, “Ad hoc networking with Bluetooth: key metrics and distributed
protocols for scatternet formation.” Ad Hoc Networks, vol. 2, no. 2, pp. 109–202, Apr. 2004.
[8] M. Kalia, S. Garg, R. Shorey, “Scatternet structure and inter-piconet communication in the
bluetooth system,” in IEEE National Conference on Communications New Dehli, India, 2000.
[9] G. Tan, “Blueware:Bluetooth Simulator for NS.” MIT Lab. Comput. Sci., Cambridge, MA,
October 2002.