Transcript Designing Wireless IP systems for indoor environments

Study Goals & Talk summary
University of Padova
• Aims
 Evaluation of FTP performance over Bluetooth (BT) radio link in
different environmental conditions
 Influence of BT radio packet format on system performance
• Outline
 TCP over wireless link
 Bluetooth overview
 Methodology of analysis & Performance metrics
 Main results
 Conclusions & Future Work
1
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
TCP over Wireless Links
University of Padova
• A hard coexistence
 TCP is tuned to work well in wired networks
 Wireless Link can produce packet losses not related to congestion
 These events may trigger useless congestion reaction mechanisms,
resulting in sub-optimal performances
• The “Link Layer” solution
 Idea:
– providing radio link reliability by using local retransmissions
 Drawback:
– possibility of bad interaction between TCP and Link Layer retransmission
mechanisms
2
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: piconet
University of Padova
 Two up to eight Bluetooth units sharing the
same channel form a piconet.
slave2
slave3
master
 In each piconet, a unit acts as master.
 Channel access is organised on the bases of a
centralised polling scheme.
slave1
master
active slave
parked slave
standby
3
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: Frequency Hopping
f(2k)
f(2k+1)
University of Padova
f(2k+2)
master
t
slave
t
625 ms
 Each piconet is characterised by a pseudo-random frequency hopping
sequence, imposed by master.
 All the units in the same piconet hop synchronously.
 Time is divided into slots of 625 ms; each slot corresponds to a different hop
frequency.
 Consecutive packets are transmitted on different RF carriers.
4
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: ARQ scheme
MASTER
A
B
University of Padova
B
X
C
NAK
ACK
SLAVE 1
G
SLAVE 2
F
H
Z
Z
 An Automatic Retransmission Query (ARQ) mechanism grants the
reliability of asynchronous data traffic (ACL)
– 1-bit fast ACK/NAK
– 1-bit sequence number
– header piggy-backing
5
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: multi-slot packets
University of Padova
625 ms
f(k)
f(k+1)
f(k+2)
f(k)
f(k+3)
f(k+4)
f(k+5)
f(k+3)
f(k+4)
f(k+5)
f(k)
f(k+5)
 A baseband packet can extend over one, three or five consecutive slots.
 The carrier frequency remains unchanged for the whole packet duration.
 Multi-slot packets reduce bandwidth losses due to packet header and PLL
settling time (220m)
6
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: DH & DM packets
University of Padova
 Asynchronous data packet can be optionally protected
by a 2/3 Forward Error Correction (FEC).
Paylod Capacity
350
Bytes
 Protected packet formats realise medium
data rate and are noted with DM.
300
250
200
150
 Unprotected packet formats realise
100
higher payload capacity but are more
50
subject to errors. They are noted with DH.
0
1 slot
3 slots
Medium rate
5 slots
High rate
7
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Bluetooth overview: packet format
72
54
AC
HEC
access code packet header
University of Padova
0-2745
CRC
payload
• Access Code (AC)
 All packet exchanged in a piconet have the same AC.
 Packets that don’t satisfy AC test are immediately discarded.
•
Packet Header
 Contains, among other information, slave active member receiver address, ARQ
flags, payload format, header checksum field (HEC).
 If the HEC test fails, the packet is immediately discarded.
•
Payload
 If the CRC test fails, the packet is negative acknowledged.
8
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Methodology of Analysis
University of Padova
• Bluetooth radio link connects a nomadic
client to a FTP server.
Measurement Platform
FTP Server
• Snooping programs collect end-to-end and
point-to-point transmission statistics.
Client
Router
Bluetooth
Ethernet
Notebook
Notebook
• A series of large bulk data transfers have
been performed, with notebook in different
TCP
Snooper
positions.
BT
Statistics
• Data collected have been analysed to extract
system performances.
9
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Packet Dropping Probability
AC
University of Padova
HEC
CRC
PDP
PEP
• Packet Dropping Probability (PDP)
 Probability of packet drop due to Access Code or Header Checksum failures.
• Packet Error Probability (PEP)
 Probability of packet retransmission due to bad reception: Access Code or Header
Checksum or CRC failures
10
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Performance indexes
University of Padova
• Radio Link performance metrics
 Goodput:
– average number of bit transmitted successfully in master to slave
direction, in the unit of time.
• End-to-end performance metrics
 Segment Service Time (SST):
– time employed by the BT entities to transmit a TCP segment through
the radio link.
 TCP sender transmission window size.
11
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Goodput VS Packet Drop Probability
University of Padova
12
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Segment Service Time
University of Padova
13
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
End-to-end performance: transmission window
University of Padova
T itolo:
P:\WIRELE~1\doc s\UniPD\THYRRE~1\IMA_FL~1.PS
Autore:
GSview from P:\WIRELE~1\doc s\UniPD\T HYRRE~1\IMA_FL~1.PS
Anteprima:
L'immagine EPS non è s tata salvata
c on l'anteprima inclus a i n ess a.
Commento:
L'immagine EPS potrà es s ere s tampata c on una s tampante
Pos tScri pt e non con
altri tipi di stampante.
14
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
End-to-end performance: spurious retransmission
University of Padova
Title:
P:\WIRELE~1\docs\UniPD\THYRRE~1\IMA_ZO~1.PS
Creator:
GSview from P:\WIRELE~1\docs\UniPD\THYRRE~1\IMA_ZO~1.PS
Preview:
This EPS picture was not saved
with a preview included in it.
Comment:
This EPS picture will print to a
PostScript printer, but not to
other types of printers.
15
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Conclusions
University of Padova
• Longest unprotected packet format (DH5) realises the highest performance in
almost all the situations considered.
• FEC protected packets overrun unprotected ones only in particularly hostile
channel conditions.
• Mean and standard deviation of the Segment Service Time grow rapidly
when PDP moves near one.
• In general, TCP well follows RTT oscillations except when PDP changes
drastically during the transmission.
16
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Open Issues and Future Work
University of Padova
• Definition of a mathematical model for the Bluetooth radio connection.
• Study of methods to protect TCP sender against drastic variations of
environmental conditions.
• Performance analysis in systems with more than two units per piconet.
• Study of hand-off problem between piconets.
• Routing algorithms for scatternet.
17
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Segment Service Time: measurement problems
University of Padova
• At the moment, we cannot directly measure the SST, because
 probing programs suffer of some drawbacks:
– master & slave statistics are collected independently,
– probing time is not always constant;
 we are not able to distinguish radio packets belonging to the same TCP
segment.
• We can estimate the SST statistic by using traditional queue theory.
18
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
PST, SST and RTT
University of Padova
Source
Datagram
L2CAP Packet
DH5
DH5
DH5
RTT
Destination
Datagram
SST
L2CAP Packet
DH5
DH5
DH5
DH5
DH5
PST
BT link (ARQ S&W)
PEP
19
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Segment Service Time: statistic estimation
University of Padova
• Let PST be the number of transmissions attempts until positive
acknowledgement.
• PST is a modified geometric random variable, with mean 1/PEP:
PST  G '  PEP
• Let N be the number of radio packets needed to carry a whole TCP segment.
• Since each packet requires PST transmission, SST is given by the sum of N
i.i.d. random variables:
SST 
 PST
N
• Hence, SST results a random variable with modified Pascal distribution.
20
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Packet Error Probabilities
University of Padova
21
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach
Packet Error Probability
University of Padova
22
Tyrrhenina Workshop, Sept. 2000
Wireless Internet Access: the BT approach