Download Vertical optimization of data transmission for mobile wireless terminals

Vertical Optimization Of Data
Transmission For Mobile Wireless
Terminals
MICHAEL METHFESSEL, KAI F. DOMBROWSKI,
PETER LANGENDORFER, HORST FRANKENFELDT,
IRINA BABANSKAJA, IRINA MATTHAEI, AND ROLF
KRAEMER, IHP
IEEE Wireless Communications • December 2002
邱家偉
Outline









Introduction
Protocol Optimization On A Mobile End Device
Optimization Strategies For The Mobile Device
Interference Between TCP And The Wireless Link
Simulation Setup: TCP Running Over The IEEE
802.11 MAC
Optimizing The Uplink
Retransmission On The TCP And MAC Layers
Packet Fragmentation
Summary And Conclusions
Introduction




The integration of wireless and mobile systems.
A major bottleneck is enough processing power and
battery life.
This article discusses a more conservative approach
in which only the mobile device is available to be
modified.
Optimization of the wireless communication in the
uplink and in the downlink.
Protocol Optimization On A Mobile End
Device
The end-to-end TCP connection between a mobile device and a remote
server runs through a wireless link in a local WLAN and the Internet.
Protocol Optimization On A Mobile End
Device


TCP runs as an end-to-end protocol on the
mobile device and server, while a wireless
protocol such as IEEE 802.11 controls
transmission over the local wireless link.
Data transmission through the Internet can
involve longer and unpredictable delays, and
can collapse when congestion occurs.
Optimization Strategies For The Mobile
Device




A main part of the incentive for mobile devices is the
ability to roam through different networks.
The mobile device is completely available for
modification as long as it obeys the MAC and TCP/IP
standards itself.
If control is only over the mobile device, it cannot be
assumed that the base station will do the same in
the downlink.
The base station will choose its own transmission
characteristics by an algorithm.
Interference Between TCP And The
Wireless Link



The throughput can be decreased drastically even if
only a few packets are lost in the channel.
Consequently, losses in the wireless transmission
are incorrectly interpreted as a sign of congestion,
leading to a reduction of the data flow by the TCP
congestion control mechanism.
The fast retransmit mechanism is not able to cope,
and data transfer recommences only after a
retransmission timeout (RTO).
Interference Between TCP And The
Wireless Link
Each value shown is the average of 10 simulations for
the transfer of 2 Mbytes. The TCP packet length was 500 bytes.
Simulation Setup: TCP Running Over The
IEEE 802.11 MAC




Use the BONeS tool
A C implementation of TCP in the New Reno variant
was used.
It has been checked by extensive interoperability
tests against Linux, FreeBSD, and Microsoft
versions of TCP under a lossy channel.
The MAC layer was implemented in the BONeS
language, based on a model of the PHY layer that
takes account the correct timings and delays.
Simulation Setup: TCP Running Over The
IEEE 802.11 MAC




The network delay was kept at 1000 ms, and the
network data rate was set to 0.08 Mb/s, yielding a
pipe capacity of 20 kbytes.
The data rate of the wireless transmission was set to
2Mb/s.
This parameter does not play a very large role since
the overall rate is restricted by the much lower
network data rate.
The main difference is that data is sent in separated
bursts instead of a continuous stream.
Optimizing The Uplink



Considerable earlier work has been done on adaptive link layer
strategies as a means to improve transmission over a timevarying faulty wireless channel.
By sensing the quality of the channel and adjusting the
transmission characteristics accordingly, power can be saved
and data transfer improved.
Specific mechanisms include packet fragmentation, switching
to a slower but more robust data rate, modifying the error
correction mechanism, and channel probing to detect the end of
a channel downtime efficiently.
Optimizing The Uplink


One can increase the error correction only at
the price of a lower data rate, and fragment
the packets only at the price of increased
overhead.
A decision must be made based on at least
two thresholds: how bad the channel must be
before switching to a different method and
the maximum permitted transmission delay.
Retransmission On The TCP And MAC
Layers



The MAC layer discards a packet, the consequence
is that TCP must retransmit it, albeit at a later time.
It is the delays associated with these events
(retransmission& fast) that cause the reduced data
rate when LRL is smaller.
Timeouts at the TCP level are generally much larger
than the delays in the wireless channel, so there is
enough time available for numerous attempts.
Retransmission On The TCP And MAC
Layers
TCP throughput as a function of the BER in the wireless channel for various
values of the MAC long retry limit.
Retransmission On The TCP And MAC
Layers
Total number of bytes transmitted into the wireless channel as a measure of the
consumed energy
Packet Fragmentation



In the case of good channel conditions, the only
effect of reducing the packet length is to increase the
overhead needed to transmit a certain chunk of data.
The packet length can be changed by either
reducing the size of a TCP packet or introducing
fragmentation at the MAC level.
That basically the same conclusions apply when the
focus is on energy consumption instead of
throughput.
Summary And Conclusions


Specific targets are to reduce power
consumption, increase throughput and
availability, and reduce delay and jitter.
As a major benefit, it is then possible to
optimize the system performance without any
modifications to the base stations or to the
TCP protocol on the remote hosts.