A Technique to Enable the Corruption-aware Transport Protocols … · 2017-02-11 · transport...
Transcript of A Technique to Enable the Corruption-aware Transport Protocols … · 2017-02-11 · transport...
I. INTRODUCTION
The traditional reliable transport protocols such as
transmission control protocol (TCP) [1] and Stream
Control Transmission Protocol (SCTP) [2] are originally
designed for the wired networks which regard any packet
losses as the indication of network congestion and halve
their congestion windows (cwnds) to alleviate the traffic
overload of network. Traditional transport protocols
perform well in wired networks since the packet losses are
mainly caused by network congestions rather than packet
corruptions.
In the wireless networks with high bit error rate
(BER), however, a high and variable packet corruption
rate (e.g., 10~50% erasure rate [3]) cannot be neglected
any more and those corrupted packets have to be
ultimately retransmitted in end-to-end manner [4]. This
will significantly degrade the end-to-end throughput of
traditional transport protocols due to incapability of
differentiating packet corruption from congestion losses.
In order to overcome the severe performance
degradation caused by corruption, some corruption-aware
transport protocols [5]~[8] have been presented among
which Datagram Congestion Control Protocol (DCCP) [7]
has been standardized as a congestion-controlled
unreliable transport protocol. The common ground of
corruption-aware transport protocols is that a sender can
avoid deflation of cwnd while only corrupted packets are
detected.
Unfortunately, the corruption-aware transport
protocols cannot work in realistic networks up to now
because the corrupted packets will be discarded by link
layer Cyclic Redundancy Check (CRC) checksum
mechanisms before they are delivered to transport layer.
To make the corruption-aware transport protocols
work in transport layer, there are two ways to be
129
The traditional reliable transport protocols are originally designed for the wired networks, which regard any packet
losses as the indication of network congestion and halve their congestion windows to alleviate the traffic overload of
network. However, unlike in wired networks, non-congestion losses will severely degrade the performance of
traditional transport protocols in wireless networks. Thus some corruption-aware transport protocols have been
proposed to overcome the performance degradation caused by corruption. Unfortunately, the corruption-aware transport
protocols cannot work in realistic networks up to now since the corrupted packets have been discarded by the link layer
checksum mechanisms before they are delivered to transport layer. This paper proposes a technique to overcome this
problem without disabling the link layer checksum mechanisms. Simulation results show that the performance of
corruption-aware transport protocols are still far better than that of traditional ones while the proposed scheme is
applied.
Keywords: TCP, SCTP, DCCP, Transport Protocol, FCS, checksum and Corruption
Lin Cui: Tianjin University of Technology and Education
Xin Cui: Shandong University
Seok J. Koh: Kyungpook National University
A Technique to Enable the Corruption-aware Transport
Protocols in Realistic Networks
Lin Cui ·Xin Cui ·Seok J. Koh
mentioned in previous studies. One is to disable the lower
layers' CRC checksums completely [5], the other is to set a
"skip flag" in every packet's header so that the lower
layer s CRC schemes can skip them [6]. Both suggestions
take the risks to overload the network traffic and/or to
deliver some garbage to upper layer. This is because some
packets with the corrupted MAC frame header may be
delivered to the wrong destinations or be acknowledged to
the wrong sources. More seriously, TCP may deliver some
garbage to application layer since both IP version 4 and
TCP itself employ the 16-bit 1's complement checksum
scheme which is relatively weaker than 32-bit CRC
checksum (IP version 6 has removed the checksum field in
order to reduce packet processing time in routers). On the
other hand, RFC 4340 [7] ignores the detailed suggestions
and only simply mentions that link layers may then reduce
their protection on unprotected parts of DCCP packets.
This paper introduces a detailed technique about how
to enable the corruption-aware transport protocols in
realistic networks without disabling the link layer
checksum mechanisms. The rest of this paper is organized
as follows. Section II introduces some related works.
Section III describes the proposed scheme in detail.
Section IV shows some simulation results and section V
concludes this paper.
II. RELATED WORKS
With exponential increment of mobile hosts during
recent two decades, wireless technologies have attracted
more and more attentions. So far a lot of related literatures
have been published and three kinds of approaches have
been mainly discussed in order to avoid the performance
degradation in wireless networks. They are split
connection schemes, link layer schemes and end-to-end
schemes.
Of the proposals, the most ones can be classified into
the end-to-end schemes. TCP Veno [8] estimates the
network congestion level by which the sender judges the
loss source between congestion and corruption.
Specifically 1) TCP Veno refines the multiplicative
decrease algorithm of TCP Reno [1], which is most widely
deployed in practice, by adjusting the slow-start threshold
according to the perceived network congestion level rather
than a fixed drop factor. 2) TCP Veno refines the linear
increase algorithm so that the connection can stay longer
in an operating region in order to fully utilize the network
bandwidth.
The explicit loss notification (ELN) scheme [13] uses
ELN signals to distinguish corruption from congestion. In
this scheme, a TCP-aware agent is requested to monitor
the passing packets at the base station. When the
corruption is detected over the wireless link, the agent sets
the ELN bit in the ACK s header to notify the sender not
to invoke the normal congestion control. In this way, this
scheme can avoid the degradation of TCP performance at
a certain extent.
TCP HACK [5] employs the two additional TCP
options for data segment at the sender side and for special
DUPACK at the receiver side. The former is the header
checksum option which covers the TCP header and the
pseudo IP header. The later is the ACK option which is
generated by the TCP receiver in response to a packet
corruption. In detail, on reception of a data segment, TCP
receiver firstly verifies the integrity of the whole segment
by checking the existing overall checksum. In case that the
segment is corrupted, the receiver then identifies the
integrity of the segment's header by verifying the
additional header checksum. Once corruption only occurs
in the portion of user data, the receiver can recover the
available sequence numbers from the integrated segment'sheader and timely report it to the sender. Therefore, in the
wireless environments with high BER, TCP HACK can be
used to improve the throughput performance of wireless
TCP by keeping it cwnd unchanged while only corruption
occurs.
TCP CAIAD [6] introduces a new error and
congestion control scheme using corruption-aware
adaptive increase and adaptive decrease algorithm. In [6],
the corrupted segments will be further processed by the
transport layer of the receiver, and a duplicate ACK is
generated to explicitly indicate both a real-time corruption
event and the congestion state of the link. Based on the
feedback information, the sender estimates the current
corruption strength and increases its cwnd by different
increments instead of entering fast recovery phase as long
as there is no concurrent loss. Meanwhile, the slow start
threshold will be estimated not only based on the received
integral packets but also based on the received corrupted
packets as per every round trip time and only updated
when the lost but not the corrupted segment is detected.
The authors argue that since the corrupted packets can still
arrive at the receiver side, they should reflect some
available bandwidth at a certain extent as well. Therefore,
TCP CAIAD [6] can estimate the network bandwidth
more precisely and can significantly improve TCP
performance over wireless networks.
The four schemes mentioned above focus on the
performance improvement by distinguishing random loss
130 Telecommunications Review·Vol. 19 No. 1·2009. 2
does not violate end-to-end semantics of TCP since the
packets cached in base station do not get through the base
station's transport layer.
As mentioned in introduction section, up to now there
are two approaches for the corruption-aware TCP to work
in transport layer. One is to disable the link layer CRC and
the other is to set a "skip flag" in every packet header for
the link layer CRC mechanism to skip the checking
procedure. The former may take the risk to deliver some
garbage to application layer and both schemes will
overload network traffic more or less since every
corrupted packet has to be delivered to a destination.
This paper introduces a new scheme that enables the
corruption-aware TCP variants to work in transport layer
properly and gets a tradeoff between performance
improvement and traffic overload, without the necessity of
disabling link layer CRC mechanism.
III. PROPOSED SCHEME
1. Technique to enable the corruption-aware protocols over CRC mechanism
We can recall the structures of IP header. IPv4 has a
''protocol'' field and IPv6 has a ''next header'' field. Both
fields indicate the type of transport layer protocol. Since
IPv4 is the de-facto IP standard currently, we present this
proposal in the environment of IPv4 (hereinafter, IP refers
to IPv4). The detailed base header structure of IPv4 is
shown in Figure 1.
To surely enable the corruption-aware transport
protocols to work in realistic networks, in this paper we
propose to assign a currently unused protocol type value to
the corruption-aware transport protocols. If so, when a
MAC frame is constructed at the source, the frame check
from congestion loss. On the other hand, Freeze TCP [14]
tries to boost the TCP performance in mobile scenarios. In
particular, with the indication of impending disconnection
from the network layer, the mobile host of Freeze TCP
[14] sends a zero window advertisement (ZWA) to the
fixed host so that the fixed host freezes all retransmission
timers and enters a persistent mode without transmission
as well as deflation of cwnd. In such a way, when the
mobile host uses triple DUPACKs to restart data transfer
after reconnection, the slow start phase can be avoided and
the sender can still transmit data at an unchanged sending
rate.
TCP Westwood [15] estimates the network bandwidth
based on the acked data amount by every ACK. In
addition, TCP Westwood [15] only revises the sender side
algorithms and does not need any feedback information
from the intermediated routers. Nevertheless, though TCP
Westwood [15] can better utilize the bandwidth of a single
wireless link, it tends to overestimate the available
bandwidth in the presence of ACK compression [16]. This
undesired feature may accelerate network collapse when
the network goes into incipient congestion. To avoid this
problem, TCP Westwood+ [16] estimates bandwidth per
RTT interval instead of every ACK. Therefore, TCP
Westwood+ can improve throughput performance even in
the presence of ACK compression.
The typical split connection scheme is I-TCP [11]
which splits a TCP connection between a fixed host and a
mobile host into the two separate connections. I-TCP
hides the TCP source from the wireless link by using a
protocol optimized for wireless link.
As for the link layer schemes, the Snoop TCP [12] is a
famous paradigm. Snoop installs a dedicated agent in the
base station and employs a local retransmission scheme
for wireless link errors so that only packet losses caused
by congestion can be detected by the source. Also, Snoop
A Technique to Enable the Corruption-aware Transport Protocols in Realistic Networks 131
32 BITS
Version
Flags
Protocol Header Checksum
Source Address
Destination Address
Figure 1. Structure of IP base header
Time to Live
Fragmented OffsetIdentifier
HeaderLength Type of Service Total Length
8 8 8 8
garbage to be delivered to application layer, TCP may
need an additional CRC checksum in its option field for
checking the checksum of entire segment. In contrary to
TCP, SCTP natively uses CRC scheme and DCCP also
has a CRC checksum option already. As a reference, the
detailed 802.11 MAC frame format is shown in Figure 2
[10].
Nevertheless, in whole travel of a MAC frame, not
only source and destination but also every intermediate
node has to correctly differentiate the partial FCS from the
normal FCS. For this purpose, either an additional type
value or a flag bit needs to be assigned in each frame
header. It is noted that different standards' MAC frames
may have different structures. Thus the sign of the partial
FCS can be expressed in different ways.
For instance, when a corruption-aware transport
protocol runs over a heterogeneous network which
consists of a WLAN and a wired network, the type
subfield of 802.11 MAC frame should be filled by ''10''and the subtype subfield should be filled by a currently
unused value. The former means this is a data frame and
the later means this frame has a partial FCS but not a
normal one. Notice that both type and subtype fields are
two subfields of 802.11 MAC frame's frame control field.
The detailed structure information is shown in Figure 2.
As for the Access Point, on reception of an 802.11
MAC frame, it needs to exchange the frame header
between 802.11 and 802.3 standards as shown in Figure 3.
At the same time, the protocol type field of 802.3
MAC frame should also be filled by a currently unused
value to inform the intermediate nodes that this frame will
be processed by a corruption-aware transport protocol
ultimately and a partial FCS is enclosed in the end of this
frame. The detailed 802.3 MAC frame format is shown in
Figure 4.
sequence (FCS) of this frame can be calculated based on
different checksum scopes, depending on the type value of
the protocol field contained in IP header.
In particular, once the protocol field indicates that a
corruption-aware transport protocol is employed in
transport layer, the FCS will be computed based on the
scope which covers the frame header (variable length for
different network standards), the possibly maximum
length of IP header (20-byte IP base header plus 40-byte
option field) and the maximum length among TCP base
header (20-byte), SCTP common header (12-byte) and
DCCP generic header (16-byte) (hereinafter, refers to
partial FCS). Otherwise, the FCS is calculated as normal
(hereinafter, refers to normal FCS).
The reason that we propose to calculate the partial
FCS based on above scope is that generally the
intermediate nodes have no the detailed knowledge of
every passing MAC frame and some IP headers may
contain option fields which extend their header scopes.
Thus, in order to ensure the integrity of transport
protocol's base header, the partial FCS has to be calculated
based on the possibly maximum header scope. However,
TCP option field can be omitted since the both useful
sequence number and checksum fields lie on TCP base
header.
For example, when a corruption-aware transport
protocol runs over a 802.11 wireless local area network
(WLAN), the FCS of a 802.11 MAC frame will be
calculated based on the first 110 bytes (the sum of 30
bytes' 802.11 MAC frame header, 60 bytes' possibly
maximum IP header and 20 bytes' maximum transport
segment s base header). The integrity of the rest part of
this frame can be responsible by the transport layer
checksum. Notice that internet checksum is relatively
weaker than CRC mechanism. In order to prevent some
132 Telecommunications Review·Vol. 19 No. 1·2009. 2
Framecontrol
Protocolversion
Type SubtypeToDS
FromDS
Morefrag
Retry PwrmgtMoredata
WEP Rsvd
2
2
Figure 2. Structure of 802.11 MAC frame
2 4 1 1 1 1 1 1 1 1
2 6 6
MAC header(bytes)
6 2 6 0-2312 4
Duration/ID
Address1
Address2
Address3
Sequencecontrol
Addres4
Framebody
FCS
It is noted that the structure of 802.3 MAC frame is
different from that of 802.11 MAC frame, hence the
coverage scopes of their partial FCS fields are also
different. In detail, each intermediate node along the
wired path will check the FCS value based on the first 94
bytes (the sum of the length of 802.3 MAC frame header
which is 14-byte, the possibly maximum length of IP
header which is 60-byte and the maximum length of
transport segments' base headers which is 20-byte) in
order to identify whether the header scope is corrupted or
not.
In this way, only those MAC frames with the
corrupted header will be dropped in link layer, and the
frames with the valid header will successfully arrive at the
A Technique to Enable the Corruption-aware Transport Protocols in Realistic Networks 133
H1
AP
R1router
R1 MAC addr
AP MAC addr H1 MAC addr R1 MAC addr
dest.address
address 1
Figure 3. Exchange of frame headers
address 2 address 3
source address
802.3 fame
802.11 frame
AP MAC addr
Internet
MACHeader
OSI Layer 3 and higherMinimum: 46bytesMaximum: 1500bytes
7 bytesPreamble
10101011
10101010
DestinationMAC Address
Source MACAddress
Start Delimiter
Protocol type
data field
Pad Field=0~46bytes
Frame checksequence
Minimum: 64bytesMaximm: 1518bytes
1 byte
2 bytes
4 bytes
Figure 4. Structure of 802.3 MAC frame
6 bytes
6 bytes
receiver side without traffic overload as well as wrong
delivery. Therefore, the corruption-aware transport
receiver can process these corrupted segments and feed the
detailed corruption information back to the sender so as to
avoid deflation of cwnd.
2. Tradeoff between overload and throughput
Generally, while a packet travels a wireless channel
with a high BER, not only the part of user data but also the
packet header suffers bit errors with various probabilities.
In current networks, those corrupted MAC frames will be
discarded directly by the intermediate routers because of
failed FCS checksum. Thus normally the corrupted MAC
frames cannot arrive at receiver side unless corruption
occurs in the last hop of their traveling path.
The main side-effect resulted from applying the
corruption-aware transport protocol in transport layer is
that a certain amount of garbage (that is, corrupted
packets) will overload the network traffic. In particular, it
is not realistic to completely disable link layer CRC
checksum due to two reasons. First, some applications
still run over CRC mechanism (e.g., FEC scheme and
ARQ scheme). Second, too much garbage will be
delivered to the non-corruption-aware destinations. On the
other hand, although setting a ''skip flag'' in every packet
header can significantly reduce the garbage amount of
network traffic (since only particular flows skip CRC
checksum), the packet with the corrupted MAC addresses
may be delivered to the wrong destination or be
acknowledged to the wrong source. Unfortunately, the
''skip flag'' method is helpless for this kind of problem.
In the ''partial FCS'' approach, however, the wrong
delivery mistakes will be completely forbidden and only a
little overload will be introduced in network traffic
compared to other schemes.
Moreover, the authors assume in [5] that corruptions
only occur at the data payload. In [6], the packet dropped
rate caused by corruption is set in simulations by the
proportion of the header size over the packet size where
the header scope covers IP header and TCP header.
Without the particular instructions for performance
estimation, RFC 4340 [7] only refers that ''checksum
coverage may eventually impact congestion control
mechanisms as well''. We argue in this paper that when
partial FCS is applied, the packet dropped rate incurred by
corruption should be set in simulations by the proportion
between the possibly maximum header scope and the
whole size of the MAC frame over a lossy link. This is
because MAC frame is the service data unit (SDU) in
134 Telecommunications Review·Vol. 19 No. 1·2009. 2
wireless channel. For example, if each IP packet has a
fixed size of 1040-byte and packet corruption rate is β, the
packet drop rate caused by corruption could be considered
as 110β/1074 (that is, 1074=30+1040+4) approximately in
802.11 WLAN.
Overhead is another drawback introduced by the
partial FCS scheme. It is noted that TCP employs internet
checksum in transport layer which is relatively weaker
than CRC mechanism. In order to prevent some garbage to
be delivered to application layer, TCP may need an
additional CRC checksum in its option field. This will
also result in some extra overhead. Nevertheless, the
overhead introduced by the partial FCS scheme could be
minor and can be ignored, compared to the improved
throughput.
IV. SIMULATION RESULTS
We use TCP CAIAD [6] as the corruption-aware
transport protocol in our experiments, and select TCP
Westwood+ [9] as the reference protocol from which TCP
CAIAD is evolved.
In the simulations, we only devote our mind to the
impacts of header corruption and neglect the checksum
procedure of partial FCS. For that, each IP packet uses a
fixed size of 1040-byte, and the packet drop rate incurred
by corruption is set to 110/1074 of packet corruption rate
for the proposed scheme. On the other hand, TCP
Westwood+ [9] regards packet corruption as packet loss.
Thus its packet drop rate is equal to the packet corruption
rate in simulations.
To minimize other impacts on performance
comparison, e.g., network layer's congestion, we use a
simple simulation topology as shown in Figure 5. In the
figure, the wired link has the link bandwidth of 10 Mbps
and the transmission delay of 35ms, whereas the wireless
link has the bandwidth of 2 Mbps and the transmission
delay of 1ms.
We compare the average end-to-end throughputs
during 100 seconds by performing the file transfer
application in ns2 simulator. In the comparison, each data
segment carries 1000 bytes' user data by TCP Westwood+,
whereas TCP CAIAD only sends 990 bytes' user data by
every data segment (we assume that each data segment
contains two additional checksums in option field. One is
CRC checksum option for entire segment and the other is
internet checksum option for header portion. Both options
need 10 bytes in all). Simulation results are shown through
Figure 6 to Figure 9.
A Technique to Enable the Corruption-aware Transport Protocols in Realistic Networks 135
A note on notation: in the Figures proposed scheme
refers to TCP CAIAD with partial FCS scheme, and TCP
CAIAD refers to TCP CAIAD with skip flag scheme.
From Figure 6, we can see that both the partial FCS
scheme and the skip flag scheme can provide the better
performances while the corruption rate is higher than
0.1%. Especially, when the wireless link experiences the
packet corruption rates ranged from 0.1% to 1%, it seems
that both corruption-aware schemes can almost fully
utilize the link bandwidth, while TCP Westwood+ [9]
already gets a drastic performance degradation then.
The figures from Figure 7 to Figure 9 show the
comparisons of bandwidth utilization, evolutions of
congestion window and slow start threshold with 10%
Various Packet Corruption Rate
Souce Basestation
SinkWired Link
Figure 5. Simulation topology
Wireless Link2M 1ms10M 35ms
2000
1800
1600
1400
1200
1000
800
600
400
200
0
Throughput (Kbps)
Figure 6. Throughput comparison
Figure 7. Bandwidth utilization comparison for 10% packet corruption rate
Proposed scheme
TCP CAIAD
TCP Westwood +
0 0.1% 0.5% 1% 5% 10%
Packet corruption rate
136 Telecommunications Review·Vol. 19 No. 1·2009. 2
packet corruption rate, respectively. From the figures, we
can see that although all TCP variants get severe
performance degradation due to the heavy packet
corruption rate (e.g., 10%), the performance gains of the
corruption-aware schemes (including both the partial FCS
scheme and the skip flag scheme) are still prominent,
compared to that of TCP Westwood+ [9].
More seriously, once the initial control segments are
corrupted during connection establishment phase, the
traditional TCP will back-off the initial time exponentially
(See Figure 7). This will further degrade TCP
performance. On the contrary, both corruption-aware
schemes can retransmit the initial control segments
immediately so as to ensure the TCP connection to be
established in time (See Figure 7).
By comparisons, we can also see that partial FCS
scheme can still obviously improve the end-to-end
throughput compared to the traditional transport protocols,
though it cannot outperform TCP CAIAD. This is because
the proposed partial FCS scheme keeps the semantic of
differentiating packet corruption from packet loss. Hence
it inherits the native characteristics of TCP CAIAD. On
the other hand, although the TCP CAIAD with skip flag
scheme may lead to a higher throughput in simulation, the
reasons stated in introduction section make the proposed
partial FCS scheme more feasible for implementation in
realistic networks.
V. CONCLUSIONS
Although RFC 4340 points out that link layer should
reduce the protection on unprotected parts of DCCP
packets, it does not present any particular suggestion on
how to fulfill. In this paper, we present a detailed
technique to enable the corruption-aware transport
protocols to work in realistic networks without the
necessity to disable the link layer checksum mechanisms.
From simulation results, we can see that the proposed
scheme can still perform far better, compared to the
traditional transport protocols in wireless environment
with high BER.
On the other hand, even if disabling/skipping the link
layer CRC checksum can lead to the higher throughput in
Figure 8. Congestion window comparison for 10% packet corruption rate
Figure 9. Slow start threshold comparison for 10% packet corruption rate
simulations, the reasons mentioned before make the
proposed scheme more feasible to implement in realistic
networks.
AcknowledgementThis research was partially supported by the MKE
(Ministry of Knowledge Economy) of Korea, under the
ITRC support program supervised by the IITA (IITA-
2008-C1090-0804-0004).
[References][1] W. R. Stevens, ''TCP Slow Start, Congestion
Avoidance, Fast Retransmit, and Fast Recovery Algorithms,'' IETF, RFC 2001, Jan. 1997.
[2] R. Stewart et al., ''Stream control transmission protocol,'' IETF, RFC 2960, Oct. 2000.
[3] D. Aguayo, J. Bicket, S. Biswas, G. Judd and R. Morris, ''Link-level Measurements from an 802.11b Mesh Network,'' in Proceedings of the SIGCOMM 2004, Aug. 2004.
[4.] O. Tickoo, V. Subramanian, S. Kalyanaraman and K. K. Ramakrishnan, ''LT-TCP: End-to-End Framework to Improve TCP Performance over Networks with Lossy Channels,'' in Proceedings of the 13th IEEE International Workshop on Quality of service(IWQoS), Jun. 2005.
[5] R. K. Balan, B. P. Lee, K. R. Kumar, L. Jacob, et al, ''TCP HACK: TCP Header Checksum Option to Improve Performance over Lossy Links,'' in Proceedings of the IEEE INFOCOM 2001, Vol. 1, Apr. 2001, pp. 309?318.
[6] L. Cui, S. J. Koh, X. Cui, et al, ''Adaptive Increase and Adaptive Decrease Algorithm for Wireless TCP,'' in Proceedings of the ICNC2007, Vol. 2, Aug. 2007, pp. 392-398.
[7] E. Kohler, M. Handley and S. Floyd, ''Datagram Congestion Control Protocol,'' IETF, RFC 4340, Mar. 2006.
[8] C. P. Fu, S. C. Liew, ''TCP Veno: TCP Enhancement for Transmission Over Wireless Access Networks,'' IEEE Journal on Selected Areas in Communications, Vol. 21, No. 2, Feb. 2003, pp. 216-228.
[9] Westwood+ TCP - Modules for ns2 from http://193.204.59.68/mascolo/tcp%20westwood/modules.htm.
[10] A. Leon-Garcia and I. Widjaja, Communication Networks: Fundamental Concepts and Key Architectures, McGraw-Hill, 2000.
[11] Bakre and B. R. Badrinath ''I-TCP: Indirect TCP for Mobile Hosts,'' in Proceedings of the ICDCS, May 1995, pp. 136-143.
[12] E. Amir, H. Balakrishnan, S. Seshan, R. Katz. ''Efficient TCP over Networks with Wireless Links,'' in Proceedings of the 5th Workshop on Hot Topics in Operating Systems, May 1995, pp. 35-41.
[13] H. Balakrishnan, R. Katz, ''Explicit loss notification and wireless web performance,'' in Proceedings of the IEEE Globecom Internet Mini-Conference, Nov. 1998.
[14] T. Goff, J. Moronski, D. S. Phatak, V. Gupta, ''Freeze-TCP: A True End-to-end TCP Enhancement Mechanism for Mobile Environments,'' in Proceedings of the IEEE INFOCOM, Mar, 2000, pp. 1537-1545.
[15] S. Mascolo, C. Casetti, M. Gerla, M. Sanadidi, R. Wang, ''TCP Westwood: End-to-end bandwidth estimation for efficient transport over wired and wirelessnetworks,'' in Proceedings of the ACM Mobicom 2001, Jul. 2001, pp. 287-297.
[16] R. Ferorelli, L. A. Grieco, S. Mascolo, G. Piscitelli, P. Camarda, ''Live Internet Measure-ments Using Westwood+ TCP Congestion Control,'' in Proceedings of the IEEE GLOBECOM 2002. Vol. 3, Nov. 2002, pp. 2583-2587.
A Technique to Enable the Corruption-aware Transport Protocols in Realistic Networks 137
138 Telecommunications Review·Vol. 19 No. 1·2009. 2
Lin Cui
Lin Cui received B.S. degree in Electronic Engineering
from Tianjin University, China, in 1989. He also received
M.S. degree in Computer Engineering from Kyunghee
University, Korea, in 2005 and Ph.D. degree in Computer
Science from Kyungpook National University, Korea, in
2009, respectively. He is now with Tianjin University of
Technology and Education, China. His current research
interests include Transport Layer Protocols, Wireless
Communication and Internet Mobility.
E-mail: [email protected]
Xin Cui
Xin Cui received B.S. degree in Materials Science and
Engineering from Northwestern Polytechnical University,
China, and M.S. degree in Interdisciplinary Program of
Electronic Commerce from Chonnam National University,
Korea, in 1986 and 2002, respectively. He has been an
instructor in Shandong University at Weihai, China, since
January 2003. His current research interests include
development and application of Electronic Commerce,
Networks Security, Wireless Networks and Transport
Control Protocol.
E-mail: [email protected]
Seok Joo Koh
Seok Joo Koh received B.S. and M.S. degrees in
Management Science from KAIST in 1992 and 1994,
respectively. He also received Ph.D. degree in Industrial
Engineering from KAIST 1998. From August 1998 to
February 2004, he worked for Protocol Engineering
Center in ETRI. He is now an Associate Professor at
Electrical Engineering and Computer Science in the
Kyungpook National University since March 2004. His
current research interests include Mobility Management
for NGN, Internet Mobility and Transport Layer
Protocols. He has so far participated in the International
Standardization as an editor in ITU-T SG19, SG17, SG13,
ISO/IEC JTC1/SC6 and IETF.
E-mail: [email protected]