Bluetooth UDP Performance over Bluetooth

Analysis of UDP Performance over Bluetooth – A White Paper by Vertoda

Copyright © Sykoinia Limited 2009 1


Please Read before reading this White Paper
This white paper is not distributed under a GPL license. Use of this white paper is subject to the
following terms:

 This white paper is copyrighted by Sykoinia Limited. Copyright © Sykoinia Limited
2009. All Rights Reserved.
 You may create a printed copy of this white paper solely for your own personal use.
 Conversion to other formats is allowed as long as the actual content is not altered or
edited in any way.
 You shall not publish or distribute this white paper in any form or on any media, except
if you distribute the documentation in a manner similar to how Sykoinia Limited
disseminates it (that is, electronically for download on a Web site with the software) or
on a CD-ROM or similar medium, provided however that the white paper is
disseminated together with the software on the same medium.
 Any other use, such as any dissemination of printed copies or use of this white paper, in
whole or in part, in another publication, requires the prior written consent from an
authorised representative of Sykoinia Limited.
 Sykoinia Limited reserves any and all rights to this white paper not expressly granted
above.

For more information on the terms of this license or if you are interested in doing a translation,
please contact us at info@vertoda.com.

If you find a typographical error in this white paper or if you have thought of a way to make this
white paper better please contact us at info@vertoda.com.

Please note that this white paper is for informational purposes. Sykoinia Limited accepts no
responsibility for any loss due to the use of this white paper.

If you have any comments please email us at info@vertoda.com with your feedback.



Abstract

The Bluetooth protocol is one of the better-known technologies in the field of pervasive
computing and plays a major role in the consumer electronics marketplace in devices such as
digital cameras, mobile phones and laptops. This paper examines the performance of the
Bluetooth protocol using a simulator developed as part of the University of Berkeley’s Network
Simulator (ns) [1]. In particular, the performance of the UDP transport protocol when Bluetooth
is used as the bearer carrier is examined.



Table of Contents

Table of Contents....................................................................................................................................5

Table of Figures....................................................................................................................................... 6

Glossary................................................................................................................................................... 7

1. Introduction ....................................................................................................................................8

2. Related Research ..............................................................................................................................10

3. Simulation of the Bluetooth Protocol ...............................................................................................11

4. UDP Performance..............................................................................................................................12

4.1 IP Packet Size...............................................................................................................................12

4.2 Piconet Inquiry Frequency ..........................................................................................................14

5. Conclusions and Future Research.....................................................................................................18

References ............................................................................................................................................19



Table of Figures

Figure 1: UDP over Bluetooth Performance - Throughput vs. Packet Size for a Single Slave Piconet..14

Figure 2: UDP over Bluetooth Performance - Throughput vs. Packet Size for a 7-Slave Piconet.........16

Figure 3: UDP over Bluetooth Performance - Impact of Periodic Inquiry on Piconet Throughput ......17



Glossary

IP Internet Protocol

ISM Industrial and Scientific Medical

L2CAP Logical Link Control And Adaptation Protocol

LAN Local Area Network

LMP Link Manager Protocol

ns Network Simulator

PDA Personal Digital Assistant

QoS Quality of Service

RF Radio Frequency

TCP Transmission Control Protocol

UDP User Datagram Protocol



1. Introduction

The main aim of the Bluetooth Protocol is to eliminate hard connections between devices.
Instead, devices are connected to each other via a universal radio link in the Industrial Scientific
and Medical (ISM) Band. Bluetooth is a single low-power wireless communication chip that,
when embedded within a device, allows that device to communicate with other Bluetooth-
enabled devices without the need for cables. The protocol is a Radio Frequency (RF)
specification that facilitates short-range point-to-multipoint voice and data transfer.

The Bluetooth protocol can be used to connect almost any device to another device. A typical
example is the wireless linking of a laptop to a printer. Once two devices have established a
connection with each other the resulting ad-hoc network is referred to as a piconet. There are two
distinct roles in a piconet - the Bluetooth device that initiates the connection between the two
devices is referred to as the master device while the device with which it establishes the
connection is referred to as the slave device. A slave in one piconet can act as the master or slave
in another piconet. This concept of overlapping piconets is termed a scatternet.

The lowest layer in the Bluetooth protocol stack is the radio system which is responsible for
moving information between two devices connected by a single physical link. The Baseband
layer defines the process by which devices search for each other and establish connections for
data transfer. The Baseband layer also defines the master and slave roles for devices, the device
initiating the connection being the master while the device responding to the connection request
becomes the slave. Communication security and bandwidth negotiation is a critical feature of any
data transport protocol. The Link Manager Protocol (LMP) manages this functionality. The LMP
also reserves a certain Quality of Service (QoS) for data traffic. This traffic is routed through the
Logical Link Control and Adaptation Protocol (L2CAP), which manages data traffic through
segmentation and reassembly functionality and protocol multiplexing to allow several different
protocols and applications to share the air-interface.

The above is a brief introduction to the components of the Bluetooth protocol stack. [2]
provides the details of the Bluetooth protocol specification in depth while [3] describes its
profiles and use cases.

One of the key areas of interest for a bearer protocol such as Bluetooth is the behaviour of
transport layer protocols that use it. In particular, the distinguishing features of the bearer
protocol may influence the behaviour of the transport layer. This paper investigates how the
performance of the UDP transport protocol is influenced by the use of Bluetooth as its bearer.
Any anomalies in performance found may be significant given the role UDP plays in several of
the Bluetooth use cases and profiles, for example the LAN Access Profile.



Section 2 of this paper will now discuss research that has been carried out for the Bluetooth
protocol. Section 3 will briefly describe the simulator and experiment methodology used. Section
4 will then discuss UDP performance, in particular the effect IP packet size and Baseband
inquiry interval has upon the behaviour of the transport protocol. Section 5 concludes the paper.



2. Related Research

The most basic area of interest for any protocol is relates to its performance. [4] shows that link
throughput is significantly degraded by interference if a very large number of interfering piconets
exist. While this interference varies depending on the packet type – DM1 and DH1 being the
most robust – degradation is present for all packet types. However, as shown in [5], if piconets
are spaced in the correct way up to 20 equally spaced piconets can be supported in a medium to
large size room.

Aside from interference the other major area of interest in assessing the performance of the
Bluetooth protocol is to examine the throughput of various transport and application layer
protocols using Bluetooth as a bearer protocol. Of course, this is one of the purposes of this
paper but for now it is worthwhile to refer to [6] which examines the performance of TCP/IP
over Bluetooth. The Bluetooth protocol utilises an ARQ system, which means that the
Bluetooth Link Layer automatically retransmits packets that have been lost as the result of a bad
channel. There is a substantial delay in the transmission of data packets as the error sequences
tend to come in large bursts as error correlation increases. As a result, the TCP/IP layer does not
receive an ACK before its transmission timeout, which of course results in the reduction of the
TCP window size and a corresponding drop in throughput.

The main perceived competitor with the Bluetooth protocol is undoubtedly 802.11b. Indeed, any
attempts to turn it into a LAN technology would face tough competition with 802.11b and
802.11a [7]. However, one must consider the rationale behind the Bluetooth protocol as that of
a short-distance wireless technology that is optimised for low cost and low power consumption.

If one examines the strengths of the Bluetooth protocol as outlined previously it becomes clear
that there are certain applications for which Bluetooth is the more suitable technology. For
example, given that it uses much less electrical power than 802.11b it can be incorporated into
pervasive devices such as mobile phones, digital cameras and PDAs. 802.11b, on the other hand,
will never be incorporated into such devices as it uses too much electrical power to be feasible.

It is clear then that Bluetooth, though superficially similar to 802.11b, is intended for very
different applications. It is ideally suited for pervasive computing as its low cost and low power
features make its integration possible for any type of device. 802.11b, however, is superior to
Bluetooth in the enabling of Wireless LANs and this will probably be the case for the foreseeable
future.



3. Simulation of the Bluetooth Protocol

The simulator used for carrying out the performance assessments presented in this paper use the
University of Berkeley’s Network Simulator (ns) as its base platform. The rationale behind this
decision is the amount of functionality provided by ns such as TCP agents, schedulers and
packet structures. The simulator was implemented using C++ and the TCL/TK programming
language and runs on any LINUX or UNIX operating system.

Given that the principal reason for the implementation of the simulator is to assess the
performance of data traffic using Bluetooth as the bearer carrier only those components of the
protocol stack germane to data transmission were implemented i.e. the Baseband, the LMP and
the L2CAP. The design of the simulator module roughly correlated to a C++ class for each layer
of the protocol stack that is modelled. Of course, there are also a number of ancillary classes to
provide functionality such as transmission timers and clocks, as detailed in the specification.

A number of TCL/TK scripts were also implemented. One of the principal values of TCL/TK
for ns is its interface to C or C++ programs. Using this mechanism a TCL class corresponding
to a Bluetooth device consisting of a Baseband, LMP and L2CAP layer can be constructed.
Scripts are also used to construct ad-hoc networks consisting of 2 or more Bluetooth devices.
These scripts are configurable so that parameters such as the number of devices and the
transport and application layer protocols used can be specified.



4. UDP Performance

We will now examine the performance of UDP (User Datagram Protocol) over the Bluetooth
protocol. Specifically we will look at:

 The impact that packet size has on throughput performance.
 The impact that frequency of the periodic inquiry has on throughput performance.

UDP is a transport layer protocol and along with TCP is one of the main protocols used to
transfer information over the Internet using IP (Internet Protocol) as the network-layer protocol.
The key characteristics of UDP is that it is more lightweight than TCP but is unreliable as no
attempt is made to retransmit packets which have been dropped.

A number of simulations are run which typically vary between 60 and 200 seconds. The number
of slaves in the piconet and the periodic inquiry interval are also varied. In the section analysing
packet size, that particular parameter is varied between 50 and 1000 bytes. The data transmission
is always from a master to one of its slaves and is unicast rather than multicast. The duplex rate
and time is 1 Mb and 3 ms respectively and the exponential traffic model is used to generate
traffic.

4.1 IP Packet Size

The experiments used for analysing the effect of IP Packet size on the performance of UDP
over the Bluetooth protocol consist of a number of tests, which have a packet size varying
between 50 and 1000 bytes. In the case of a single slave piconet the simulation time is varied
between 60 and 140 seconds while the 7-slave piconet scenario has simulation times between 120
and 160 seconds so as to allow all slaves to join the piconet. The piconet suspension interval to
facilitate inquiry operations is scheduled to take place every 15 seconds until the maximum
number of slaves has joined the piconet.

The first scenario examined is that of a single slave piconet. It is clear from Figure 1 that
throughput does not vary greatly regardless of the size of the packets being sent, the mean
throughput being 24.091 kbs (Kilobits per second) with a standard deviation of 0.228 kbs. This is
a surprising result as one would expect that the throughput would be larger, the larger the packet
size. However two factors in the single slave piconet explain this behaviour. The first
consideration is the segmentation, which takes place in the L2CAP layer. For example, a 1000
byte packet is subdivided into 4 DM5 packets of 226 bytes each with the remainder being
transmitted in a single 123-byte DM3 packet. This means that the size of the packet being
transmitted through the Bluetooth protocol does not vary significantly. The other issue to bear
in mind is that there is little or no impact on throughput in a single slave piconet as though the


piconet is periodically suspended to facilitate inquiries no other slave device responds so the
impact of processing an additional inquiry does not occur.

Figure 2 illustrates the system throughput for a 7-slave piconet. Unlike the single-slave case the
system throughput varies widely from approximately 60 kbs to 140 kbs. The mean system
throughput is 100.577 kbs while the standard deviation of 37.646 kbs denotes a wide disparity in
the system throughput depending on the packet size. There thus appears to be a significant
disparity in the throughput vs. packet size case for single slave and multiple slave piconets.

Figure 2 also shows that this variation takes place for individual slaves in the piconet – in these
examples, the first, third and sixth slaves to join the piconet have mean throughputs of 14.292
kbs, 14.305 kbs and 14.999 kbs respectively. The system throughput is equitably divided among
the piconet slaves but the standard deviations of 4.819, 5.042 and 5.549 emphasise the point that
the packet size has an impact on the throughput for a multi-slave piconet. In addition to the
inconsistency with the single-slave piconet a further curious aspect of the impact that packet size
has on UDP throughput is the non-linear increase in throughput as packet size increases. In fact
in certain cases throughput declines – for example, though system throughput increases by 23%
if the packet size is increased from 500 to 750 bytes, throughput declines by 4.5% if the packet
size is increased from 750 to 1000 bytes.

Again, segmentation, which occurs at the L2CAP layer, plays a role in the throughput
performance. The reason why throughput does not differ significantly as packet size increases in
certain circumstances is due to the wrapping of a UDP packet at the L2CAP layer in the smallest
Bluetooth Data Packet available in terms of size. For example, a 50-byte and a 100-byte UDP
packet are both wrapped in a 123-byte DM3 packet – the smallest packet size available to both.
Segmentation also explains why system throughput is greater for a 750-byte packet compared to
a 1000-byte packet. A 750-byte packet consists of 4 Bluetooth Data packets. The payload in the
first 3 226 DM5 packets is completely occupied while the payload of the last packet consists of
only 72 bytes and is thus transported in a 123-byte DM3 packet. Similar segmentation takes place
for a 1000-byte packet. However in this case 4 DM5 packets and 1 DM3 packet are required to
send 1 UDP packet from the Master to one of its Slaves. This increase in the volume of packet
traffic sent over the Bluetooth network to transmit one UDP packet and indeed the cost of
segmenting and reassembling the packet offsets the increase in the packet size and therefore
leads to an overall decline in system throughput. The corollary of this situation occurs in the case
of a 500-byte packet where despite requiring only 2 DM5 and 1 DM3 packets to transport a
single UDP packet the decrease in size compared to a 750-byte packet results in less throughput
than in the case of the larger packet size.



UDP over Bluetooth Performance - Throughput vs. Packet Size for a
Single Slave Piconet

30

25
Throughput (kbs)

20

15

10

5

0
0 100 200 300 400 500 600 700 800 900 1000
Packet Size (bytes)

Figure 1: UDP over Bluetooth Performance - Throughput vs. Packet Size for a Single Slave Piconet

The other question that needs to be addressed is the large deviation between a single-slave and a
multiple slave piconet. Looking at the physical number of packets passed through a Bluetooth
Piconet over 140 seconds given a periodic inquiry operation of 15 seconds gives a mean
percentage decrease in packet traffic of 45.618% with a very large standard deviation of
50.041%. If the figures in Table 1 are examined more closely we see that the percentage decrease
in packet traffic increases as the packets become smaller. It is contended that this occurs because
of the sheer volume of data traffic that can be sent as the packet size becomes smaller. The
disparity in traffic volume becomes more acute as fair bandwidth allocation becomes a
significant factor to ensure that each slave in a 7-member piconet receives a reasonable amount
of data traffic.

4.2 Piconet Inquiry Frequency

A key metric that may affect the performance of UDP over the Bluetooth protocol is how often
piconet operations may be suspended. The Bluetooth specification does not denote a range for
this interval possibly because of the diverse range of Bluetooth applications. For example, while
an interval of 20 seconds may be appropriate for a personal computer this interval may be much
too large for another application such as a Bluetooth enabled PDA. A number of simulations of
between 80 and 160 seconds are also run here. Both the number of slaves and the frequency of
the inquiry interval are varied here. The packet size is fixed at 210 bytes.



Figure 3 displays the throughput performance of a single slave piconet when the periodic inquiry
period is varied from 20 seconds down to 0.01 seconds. As one would expect, the less frequent
the periodic inquiry operation the higher the overall throughput. This lack of consistency is
emphasised by a large standard deviation of 28.815 kbs given a mean of 15.268 kbs. What is
surprising though is that there is some throughput for periodic inquiry intervals of less than one
second. This can reasonably be explained for a single slave piconet as though the piconet inquiry
interval may be quite small there is no other slave responding to the inquiry, which lessens the
impact of the operation.

Figure 3 also illustrates the 3 and 6 slave piconet scenarios respectively and shows that the
system throughput for a multi-slave piconet behaves in a similar fashion to that of the single-
slave piconet. Again there are large standard deviations of 18.75 kbs and 22.165 kbs for the 3-
slave and 6-slave scenarios respectively given means of 27.286 kbs and 32.086 kbs. Again there is
some throughput for piconet inquiry intervals of less than one second. While this may be
explained by the fact that though a number of operations whereby a slave will join the piconet
will takes place in both piconet scenarios a number of inquiries that will not be replied to will
take place, it must also be noted that when the piconet inquiry interval is set at a very small
frequency the difference in system throughput ostensibly appears to decline dramatically.
However while the difference in system throughput for a 3-Slave and 6-Slave piconet where the
piconet inquiry interval is 0.01 appears to be small (3.260 kbs vs. 2.783 kbs respectively) the
percentage decline is in fact 17.139% which does not differ significantly from the percentage
difference in system throughput for a 20 second piconet inquiry interval.



UDP over Bluetooth Performance - Throughput vs. Packet Size for a 7-Slave
Piconet

150

140

130

120

110

100

1st Slave to join the Piconet
Throughput (kbs)

90

80 3rd Slave to join the piconet

70 6th Slave to join the piconet

60 System Throughput for the
piconet
50

40

30

20

10

0

0 200 400 600 800 1000
Packet Size (Bytes)

Figure 2: UDP over Bluetooth Performance - Throughput vs. Packet Size for a 7-Slave Piconet

Table 1: Comparison of Packet Size vs. Packet Volume for Single-Slave and 7-Slave Piconets

Packet Size No. Packets – Slave 1 No. Packets – Slave 2 % Decrease in Packet
(Single Slave Piconet) (7-Slave Piconet) Volume
(Bytes)

1000 424 332 21.698

750 564 469 16.844

500 848 545 35.731

250 1671 696 58.348

210 2020 827 59.059

100 4163 1552 62.719

50 8471 2971 64.927



UDP over Bluetooth Performance - Impact of Periodic Inquiry on
Piconet Throughput

60

50
Piconet Throughput (kbs)

40

6-Slave Piconet

30 3-Slave Piconet

Single-Slave Piconet

20

10

0
0 5 10 15 20
Periodic Inquiry Period (Seconds)

Figure 3: UDP over Bluetooth Performance - Impact of Periodic Inquiry on Piconet Throughput



5. Conclusions and Future Research

The Bluetooth technology is a key player in the fields of ad-hoc networking and pervasive
computing. Though quite similar to 802.11 on a superficial level the technology has quite
distinctive applications for devices such as PDAs, mobile phones and digital cameras. To date,
the majority of research relating to the protocol has addressed its impact in terms of interference
with 802.11. This paper attempts to partially redress the lack of performance-based analyses of
the protocol.

This paper presents an assessment of the performance of UDP over the Bluetooth protocol. The
most significant findings here perhaps are that UDP packet size can affect system performance
in certain circumstances while in general the higher the frequency of the inquiry period for a
UDP piconet the worse the per-second throughput performance. Further research in the area
would include TCP performance over Bluetooth. Given the complexity of TCP compared to its
UDP counterpart one would expect more unexpected results. In particular, one would expect
that the requirement for a TCP acknowledgement and the resultants decrease in window size
when TCP-senders do not send ACKs would lead to erratic results in Bluetooth piconets given
the presence of a periodic inquiry interval which results in the suspension of piconet traffic.



References

[1] The Network Simulator, University of Berkeley, California,
http://www.isi.edu/nsnam/ns

[2] Sundry Contributors, Specification of the Bluetooth System Version 1.0B, Volume 1
Core, December 1st 1999, http://www.bluetooth.com

[3] Sundry Contributors, Specification of the Bluetooth System Version 1.0B, Volume 2
Profiles, December 1st 1999, http://www.bluetooth.com

[4] Zurbes, Stefan. “Considerations on Link and System Throughput of Bluetooth
Networks”, The 11th IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications, 2000. Volume: 2, Page(s): 1315 –1319

[5] Barton, M, Galli, S, Wong, K.D. and Koshy, B.J., “Bluetooth Technology: Link
Performance and Networking Issues”, European Wireless Conference 2000 EW ’00

[6] Primak, Serguei and Snow, Chris, “Performance Evaluation of TCP/IP in Bluetooth
Based Systems”, IEEE Vehicular Technology Conference (VTC), Spring 2002

[7] Bhagwat, Pravin, “Bluetooth Vs. 802.11”, Hot Interconnects, Aug 24, 2001


Bluetooth UDP Performance over Bluetooth

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Bluetooth UDP Performance over Bluetooth (20)

Recently uploaded (20)

Bluetooth UDP Performance over Bluetooth