Kein Folientitel

Download Report

Transcript Kein Folientitel

MobiHealth
Communication Infrastructure
Performance Analysis of the MobiHealth
E2E communication link
MobiHealth WP3 meeting
January 30, 2003
Katarzyna Wac, University of Twente
email: [email protected]
Internet: www.mobihealth.org
Presentation Layout
1. Performance analysis of TCP and UDP transport
protocols over GPRS Network
•
•
•
Research issues
Measurement setup
Measurement results
2. State of the Measurement Strategy
•
•
•
Research issues
Proposed measurement strategy
Measurement assumptions
1. Performance analysis of TCP and UDP
transport protocols over GPRS Network
Research issues
Evaluation of TCP/UDP protocols performance in two GPRS networks:
•
•
O2 - The Netherlands
TME - Spain
Performance objectives are:
•
Available network bandwidth (& coding schemes, time slots)
•
Achieved Reliability (data loss)
•
Total delay (& Round Trip Delay sensitivity)
Segments of E2E communication link
Sensor
MBU
(Mobile Base Unit)
Enterprise
router
Internet Gate
Mobile Operator
Network
BAN (internal)
e.g. Bluetooth
Host
Enterprise
Network
(Intranet)
Internet
…and particularly when applying 2.5G technology - GPRS
MBU
Base Station
Enterprise
router
GGSN
Host
BAN
2.5 technology - GPRS
wireless link
Internet
Enterprise
Network
(Intranet)
Measurement setup
End-to-End communication model
Data
File
Sender
Viewer
(555 kB)
Server
(Win2000)
Nokia D211
card
GPRS
Link
Client
Internet (O2)
or TME LAB
(Linux OS)
Measurement procedures
TCP protocol
•
•
•
No control on packet rate (on the server side) applied
(e.g. control on: #packets/sec, #samples/packet).
Control on packet rate applied
Packet size derived from the available network
bandwidth and from the known data rate produced by
the sensor system front-end (TMSI): 64 samples/sec, 30B/sample
UDP protocol
•
•
Procedures as above
Additionally, every packet has 4B sequence number –
control on:
out-of-sequence packet delivery,
data loss rate
At the client side: function of data buffering, re-ordering
and insertion of lost packets.
Bandwidth = [ (30B / sample * #samples / packet) + TCP/UDP_header] * #packets / second * 8 bits [kbps]
Measurement results
GPRS “hick-up”
LAYERS
Sender application is
situated on this layer
Application
Transport
Network
Host-toNetwork
TCP / UDP
IP
(GPRS)
data buffer
Nokia Card
O2 Measurement results
TCP
Packet Size
(TCP header:
40bytes)
30+40
bytes
Packet rate
(samples/packet,
packets/second)
1/1 , no control
stream
Time used
Data lost rate
GPRS hick-up
Delay
Figure
283490 ms
N/A
YES.
Frequently
Large
delay
Perfect
Better than
↑
Perfect
150+40
bytes
5/1 , 10/1
383390 ms
N/A
YES.
Less than no
control
300+40
bytes
10/1 , 5/1
430679 ms
N/A
YES.
A few times
Better than
↑
Perfect
480+40
bytes
16/1 , 3/1
419022 ms
N/A
Almost NO.
Better than
↑
Perfect
750+40
bytes
25/1 , 2/1
400386 ms
N/A
NO.
No
Perfect
1500+40
bytes
50/1, 1/1
451419 ms
N/A
NO.
No
Perfect
O2 Measurement results (cont.)
UDP
Packet Size
(UDP header:
28bytes, seqheader: 4bytes)
Packet rate
(samples/packet,
packets/second)
Time used
Data lost rate
GPRS hick-up
Delay
Figure
YES.
Frequently
Too large
Very bad
1.36 %
YES.
Less than no
control
Larger delay,
lots of out of
sequence
delivery
Bad
Better than ↑
A few out of
sequence
Good
1/1, no
control
stream
Too long
because
of the
hick-up
5/1 , 10/1
634412
ms
300+28+4
bytes
10/1 , 5/1
420795
ms
0.21 %
YES.
Few times
480+28+4
bytes
16/1 , 3/1
385294
ms
0.34 %
NO.
No.
No out of
sequence
Perfect
750+28+4
bytes
25/1 , 2/1
371034
ms
0.67 %
NO.
No
No out of
sequence
Perfect
1500+28+4
bytes
50/1, 1/1
412503
ms
0.83 %
NO.
No
No out of
sequence
Good
30+28+4
bytes
150+28+4
bytes
Too much
lost
Viewer Application – TCP protocol
Measurement case with and without control on packet rate
TCP parameters : 5 samples/packet, 10 packets/second
Viewer Application – UDP protocol
Measurement case without the control on packet rate
UDP parameters : 5 samples/packet, 10 packets/second
Viewer Application – UDP protocol (cont.)
Measurement case with the control on packet rate
UDP parameters : 16 samples/packet, 3 packets/second
TME
Measurement results
TCP
Packet Size
(TCP header:
40bytes)
Packet rate
(samples/packet,
packets/second)
Time used
Data lost rate
GPRS hick-up
Delay
Figure
30+40
bytes
1/1 , no control
stream
216503 ms
N/A
YES.
Frequently
Large
delay
Perfect
150+40
bytes
5/1 , 10/1
417520 ms
N/A
YES.
Less than no
control
Better than
↑
Perfect
300+40
bytes
10/1 , 5/1
383821 ms
N/A
Almost No.
Better than
↑
Perfect
480+40
bytes
16/1 , 3/1
488213 ms
N/A
NO.
No.
Perfect
750+40
bytes
25/1 , 2/1
490970 ms
N/A
NO.
No
Perfect
1500+40
bytes
50/1, 1/1
422667 ms
N/A
NO.
No
Perfect
TME
Measurement results (cont.)
UDP
Packet Size
(UDP header:
28bytes, seqheader: 4bytes)
30+28+4
bytes
Packet rate
(samples/packet,
packets/second)
1/1, no control
stream
Time used
Data lost
rate
GPRS hick-up
Delay
Figure
Too long
because of
the hick-up
Too much
lost
YES.
Frequently
Too large
Very bad
Larger delay,
lots of out of
sequence
delivery
Bad
150+28+4
bytes
5/1 , 10/1
564952 ms
0.5 1%
YES.
Less than no
control
300+28+4
bytes
10/1 , 5/1
426564 ms
0.10 %
YES.
Few times
Better than ↑
A few out of
sequence
Good
480+28+4
bytes
16/1 , 3/1
385312 ms
Around
0%
NO.
No.
No out of
sequence
Perfect
750+28+4
bytes
25/1 , 2/1
(100 s less
than TCP)
N/A
NO.
No.
No out of
sequence
Perfect
1500+28+4
bytes
50/1, 1/1
412663 ms
N/A
NO.
No.
No out of
sequence
Perfect
(100 s less
than TCP)
370133 ms
Conclusion on measurement results
TCP vs. UDP performance
•
•
•
•
TCP suitable when no control on packet rate applied
UDP performs better with the control on the packet rate
(low data loss, shorter time of transmission)
GPRS provides reliable data link service ->
UDP loss occurred on Internet or Enterprise Network
MobiHealth recommendation:
- UDP for the BAN monitoring data (if data loss allowed)
- TCP for the control data to BAN
O2 vs. TME GPRS Network
TME GPRS network has better performance
than O2 GPRS network.
Possible reasons:
Laboratory & Public network,
Voice priority in O2 network
2. State of measurement strategy
Layered structure of the m-health system
Application
Serv ice Aceess Point
(SAP)
HTTPS (WSB)
Transport
TCP / UDP / W-TCP *
Network
IP
Host-toNetwork
GPRS
UMTS
* with CMG consultancy
Proposed measurement strategy
High-level abstraction of the m-health system
MBU
Communication "black-box"
End-host
Traffic Sink
(TS)
Traffic Generator
(TG)
…and its decomposition
MBU
End-host
GGSN
Operator Network
Traffic Generator
(TG)
NET
Traffic Sink
(TS)
Intermediate
Measurement
Point
Proposed measurement strategy (cont.)
MBU
End-host
GGSN
Operator's Network
Internet
EN
Traffic Sink
(TS)
Traffic Generator
(TG)
MBU
Operator's Network
SGSN
GGSN
Radio Net
Traffic Generator
(TG)
Core Net
End-host
Internet
EN
Traffic Sink
(TS)
Proposed measurement strategy (cont.)
Characteristics of the link that will be measured
• “good put” experienced by end-user
& throughput (gross bandwidth – with overhead)
• Round Trip Delay
• Data Loss
• Packet Size
Measurement assumptions
•
Protocols:
Transport Layer: TCP/UDP/W-TCP
Network Layer:
IP
Data Link Layer: GPRS, UMTS
•
Measurements are done in the following directions
Terminal -> Network and Network -> Terminal
•
Measurements are done in the following positions:
Stable position & mobility (80-120 km/h)
•
Investigation of the real coverage of network will take place
•
Iterative process: Step-by-step derivation of system’s
performance characteristics
•
High number of measurements (samples) to guarantee
the statistical correctness
…any questions ?