Fast Selective ACK Scheme for Throughput Enhancement of Multi-Homed SCTP Hosts

Abstract

This Letter proposes a fast selective ACK scheme for Stream Control Transmission Protocol (SCTP) to enhance transmission throughput in multi-homing scenarios. In the proposed scheme, a multi-homed receiver sends SACK chunks to the sender over the fastest reverse path, which facilitates to inflate the congestion window and to retransmit the lost data packets as quickly as possible. Simulation results show that the proposed scheme could improve the throughput performance of multi-homed SCTP hosts under both normal SCTP and CMTSCTP cases if only the reverse paths have various one-way delays.

Full text

IEEE COMMUNICATIONS LETTERS, VOL. 14, NO. 6, JUNE 2010 587
Fast Selective ACK Scheme for Throughput Enhancement of
Multi-Homed SCTP Hosts
Lin Cui, Seok Joo Koh, and Woo Jin Lee
Abstract—This Letter proposes a fast selective ACK scheme
for Stream Control Transmission Protocol (SCTP) to enhance
transmission throughput in multi-homing scenarios. In the pro-
posed scheme, a multi-homed receiver sends SACK chunks to
the sender over the fastest reverse path, which facilitates to
inflate the congestion window and to retransmit the lost data
packets as quickly as possible. Simulation results show that the
proposed scheme could improve the throughput performance of
multi-homed SCTP hosts under both normal SCTP and CMT-
SCTP cases if only the reverse paths have various one-way delays.
Index Terms—SACK, Fast SACK, multi-homing, CMT, SCTP.
I. INTRODUCTION
FOR end-to-end transport protocols, one of the critical
issues is throughput degradation that may be induced
by asymmetric transmission delays. The works in [1], [2]
discussed TCP performances in the networks with asymmetric
delays for forward and reverse paths, but did not consider
the multi-homed hosts. Based on the multi-homing scenarios,
the authors in [3] tend to use a symmetric round-trip routing
with the minimum round-trip time (RTT) to improve the
throughput. But the authors in [4] argue that it is better for the
sender to use the asymmetric data and Selective ACK (SACK)
paths with the minimum RTT. Both works [3], [4] ignore the
Concurrent Multi-path Transport (CMT-SCTP) cases [5], [6].
In CMT cases, on one hand, the problem might get worse
since the increased out-of-order data and SACK segments
will result in the more unnecessary fast retransmissions, the
much reduction of congestion window and the higher traffic
of SACKs; on the other hand, data segments will be delivered
along all available forward paths, thus only the minimum RTT
does not matter for the CMT schemes [5], [6].
Our study is based on the notion that the throughput of
multi-homed SCTP hosts can be enhanced, if a receiver
transmits SACK chunks to a sender, as fast as possible,
which will be helpful to inflate the congestion window and
to retransmit the lost data segments quickly, especially in the
following cases: 1) static CMT case [5]; 2) case with frequent
Manuscript received November 30, 2009. The associate editor coordinating
the review of this letter and approving it for publication was F.-N. Pavlidou.
L. Cui is with the School of Information Technology Engineering,
Tianjin University of Technology and Education, 1310 South Dagu Road
(East of Liulin), Hexi District, Tianjin City, 300222, China (e-mail:
cuilin.academic@gmail.com).
S. J. Koh is with the School of Electronic Engineering and Computer Sci-
ence, Kyungpook National University, 1370 Sankyuk-dong, Buk-gu, Daegu,
702-701, Korea (e-mail: sjkoh@knu.ac.kr).
W. J. Lee is with the School of Electronic Engineering and Computer Sci-
ence, Kyungpook National University, 1370 Sankyuk-dong, Buk-gu, Daegu,
702-701, Korea (e-mail: woojin@knu.ac.kr).
Digital Object Identifier 10.1109/LCOMM.2010.06.092335
Fig. 1. Step 1 and 2 for configuration of the FSACK path.
slow start and/or frequent fast retransmission/recovery phases;
3) handover with CMT in overlap area between heterogeneous
networks, as described in [6].
In this Letter, we propose a Fast SACK (FSACK) scheme
of SCTP, in which a receiver sends SACK chunks, as many
as possible, to the sender over the fastest path among the
available return paths. The proposed scheme can be applied
to both SCTP [7] and CMT-SCTP cases [5], [6]. Different
from [3] and [4], we aim at selection of the optimal return
path only in this Letter because we argue that the optimal
forward path strongly depends on its bandwidth in addition to
its delay, and does not make sense for the CMT cases [5], [6]
yet.
II. PROPOSED FAST SACK SCHEME
In the proposed scheme, the sender determines the FSACK
path with the minimum one-way delay from the receiver, and
informs the receiver about its decision. Since then the receiver
will transmit all the subsequent SACK chunks to the sender
over the designated FSACK path. The FSACK path will be
dynamically updated, depending on the network conditions.
A. Configuration of FSACK Path
In data transmissions, a sender transmits the data chunks to
a receiver over the primary path (in SCTP) or multiple paths
(in CMT-SCTP). In the proposed scheme, FSACK path can
be determined in the course of data transmission as follows.
Step 1: Initially, the sender transmits a FSACK-INIT chunk
over the primary path, as shown in Fig. 1, which contains a
4-byte timestamp (current time).
Step 2: On reception of the FSACK-INIT chunk, the re-
ceiver responds with the FSACK-ACK chunks over all the
available return paths, as shown in Fig. 1. Once any FSACK-
ACK chunk reaches, the sender can immediately designate
the FSACK path as that one from which the fastest FSACK-
ACK chunk arrives since all FSACK-ACK chunks have the
same timestamp value of the FSACK-INIT chunk. The sender
also sets/resets “FSACK-RTT-threshold” to the arrival time
1089-7798/10$25.00 c
⃝2010 IEEE
Authorized licensed use limited to: Kyungpook National University. Downloaded on June 04,2010 at 01:16:17 UTC from IEEE Xplore. Restrictions apply.

588 IEEE COMMUNICATIONS LETTERS, VOL. 14, NO. 6, JUNE 2010
Fig. 2. Step 3 for configuration of the FSACK path.
Fig. 3. FSACK-INIT and FSACK-ACK chunks.
minus the timestamp value for asymmetric FASCK-INIT and
FSACK-ACK paths (sometimes they might be symmetric too).
Step 3: After that, the sender informs the receiver about the
FSACK path by sending the FSACK-COMPLETE chunk, as
shown in Fig. 2 and 4. Once the FSACK path is determined,
the receiver transmits all the subsequent SACK chunks to the
sender over the FSACK path until the FSACK path is changed.
During data transmission, these three-step procedures can be
performed repeatedly, as long as the sender realizes that the
current asymmetric FSACK RTT is greater than the ‘FSACK-
RTT-threshold’ value. This ensures that the dynamic network
conditions are reflected on the FSACK path configuration. In
addition, we will apply the maximum time interval for FSACK
path update (e.g., 1 second), so as to guarantee that the FSACK
path can be confirmed or updated at least once during a certain
time period.
B. Extensions of SCTP
To apply the proposed scheme, we define the following
new three types of SCTP chunks: FSACK-INIT, FSACK-ACK
and FSACK-COMPLETE. The formats of FSACK-INIT and
FSACK-ACK chunks are similar, as shown in Fig. 3, in which
“X” represents the new chunk type number. The FSACK-
COMPLETE chunk is used to inform the IP address of the
determined FSACK path, as shown in Fig. 4.
III. NUMERICAL RESULTS
We implemented the proposed FSACK scheme on top of
both SCTP and CMT-SCTP for performance evaluation over
heterogeneous and homogeneous networks, using the ns-2 net-
work simulator [8]. The simulation topology for heterogeneous
networks, where all available paths may experience different
propagation delays, is given in Fig. 5.
In figure 5, all links have the bandwidth of 2Mbps and
the upper link has the end-to-end propagation delay of 45ms,
while the end-to-end propagation delay of the lower link is
variable. The other simulation parameters are configured based
on the sctp-cmt-2paths-64K function for CMT-SCTP and the
sctp-multihome2-2Rtx1 function for normal SCTP. Both are
already given in the test-suite-sctp.tcl of ns-2.30 [8] where
the initial slow start threshold is set to 16000 bytes and each
data chunk has a fixed size of 1468 bytes, without background
traffic.
Fig. 4. FSACK-COMPLETE chunk.
Fig. 5. Simulation topology for heterogeneous networks.
In the simulation suit, we lasted a file transfer application
with 1% loss rate over the forward path for 100 seconds, and
set the maximum time interval to 1 second for FSACK path
update. The simulation result is shown in Fig. 6.
From the results, fist of all, we can see in the comparison
of SCTP and SCTP with FSACK that when the end-to-
end propagation delay of the alternative network (i.e., the
lower path of Fig. 5) is smaller, the proposed scheme (SCTP
with FSACK) outperforms the normal SCTP. However, with
increase of the propagation delay, the proposed scheme tends
to give nearly similar performance with the normal SCTP.
This is resulted in due to dynamic SACK routing capacity
of the proposed scheme. In particular, when the delay of the
alternative network gets much smaller (e.g., less than 45 ms
in Fig. 5), the proposed scheme will switch (i.e., reconfigure)
the FSACK path to the alternative network in order to inflate
its congestion window and/or retransmitting the lost segments
more quickly than the existing scheme, while the normal SCTP
uses the original primary path all along (i.e., the upper path
of Fig. 5).
On the other hand, in the comparison of the CMT-SCTP
and CMT-SCTP with FSACK, as the propagation delay of
the alternative network is larger, the proposed scheme (CMT-
SCTP with FSACK) outperforms the CMT-SCTP scheme
Fig. 6. Comparison of throughputs over heterogeneous networks.
Authorized licensed use limited to: Kyungpook National University. Downloaded on June 04,2010 at 01:16:17 UTC from IEEE Xplore. Restrictions apply.

CUI et al.: FAST SELECTIVE ACK SCHEME FOR THROUGHPUT ENHANCEMENT OF MULTI-HOMED SCTP HOSTS 589
Fig. 7. Simulation topology for homogeneous networks.
Fig. 8. Comparison of throughputs for homogeneous networks.
much more. This also benefits from the dynamic SACK
routing capacity. That is, in the proposed scheme all SACK
chunks are delivered to the FSACK path with a shorter delay,
while the existing CMT-SCTP scheme sends SACK chunks
through both heterogeneous networks without consideration
on their delay difference.
We then compare the schemes’ throughput performances
over homogeneous networks. The related simulation topology
is given in Fig. 7. In the figure, we employ three independent
links between sender and receiver. Each link has the capacity
of 2 Mbps, 1% packet loss rate and 45ms propagation delay.
Each node has a queue buffer of 256 packets. To simulate
dynamic network conditions, a various number of TCP traffics
(ranged from 0 to 8) are generated over the upper reverse
path. For each of the other forward and reverse paths, 4 TCP
background traffics are randomly generated.
Notice that all background traffics are randomly generated
by using FTP applications with 1500 bytes per packet, and
thus the transmission delays and error rates of the forward and
reverse paths will vary dynamically. The other parameters are
set based on the sctp-cmt-3paths-64K function for CMT-SCTP
and the sctp-multihome3-3timeout function for SCTP, which
are also given in the test-suite-sctp.tcl of ns-2.30 [8].
For SCTP, we perform an FTP application for 100 seconds,
and the maximum time interval is also set to 1 second for
FSACK path update. Fig. 8 shows the simulation results with
different number of TCP background traffics over the upper
reverse path. From Fig. 8, specifically, we saw that for a
single data flow the normal SCTP tends to experience severe
performance degradation as background traffics increase along
the primary reverse path (i.e., the upper link of Fig. 7),
compared to the SCTP with FSACK scheme. This is because
when the background traffics are lighter, both SCTP and SCTP
with FSACK schemes will deliver most of SACK chunks to
the same primary reverse path. However, once the background
traffics get heavier, the proposed scheme changes its FSACK
path to the alternative path with a shorter transmission delay,
whereas SCTP uses the existing primary path all the time.
On the contrary, in case that the TCP background traffics
are relatively low along the upper reverse path, performance
of the existing CMT-SCTP scheme will get worse than that
of the proposed CMT-SCTP with FSACK scheme because its
receiver always sends SACK chunks to all of reverse paths in
response to the arrival of data chunks, whereas the proposed
scheme only delivers SACK chunks to the FSACK path with
the shortest delay.
IV. CONCLUSION
This Letter proposed a fast SACK scheme for SCTP to
enhance the throughputs of SCTP and CMT-SCTP schemes.
From the simulation results, we saw that the proposed scheme
can improve the transmission throughputs of SCTP schemes
over both heterogeneous and homogeneous networks by ef-
fectively exploiting the fast SACK path. This implies that
the proposed scheme has deep potential to enhance the trans-
mission throughputs of SCTP schemes and warrants a further
exhaustive study.
ACKNOWLEDGMENT
This research was partially supported by the MKE, Korea,
under the ITRC support program supervised by the NIPA:
(NIPA-2009-C1090-0902-0009).
REFERENCES
[1] C. Barakat, et al., “On TCP performance in a heterogeneous network: a
survey,” IEEE Commun. Mag., vol. 38, pp. 40-46, Jan. 2000
[2] C. P. Fu and S. C. Liew, “A remedy for performance degradation of TCP
Vegas in asymmetric networks,” IEEE Commun. Lett., vol. 7, pp. 42-44,
Jan. 2003.
[3] A. Kelly, G. Muntean, P. Perry, and J. Murphy, “Delay-centric handover
in SCTP over WLAN,” Trans. Automat. Contr. and Computer Sci.,vol.
49, pp. 211-216, 2004.
[4] E. P. Ribeiro and V. C. M. Leung, “Minimum delay path selection in
multi-homed systems with path asymmetry,” IEEE Commun. Lett.,vol.
10, no. 3, pp. 135-137, Mar. 2006.
[5] J. Iyengar, et al., “Concurrent multipath transfer using SCTP multi-
homing over independent end-to-end paths,” IEEE/ACM Trans. Network-
ing, vol. 14, no. 5, pp. 951-964, Oct. 2006.
[6] C. M. Huang and C. H. Tsai, “The handover control mechanism for multi-
path transmission using Stream Control Transmission Protocol (SCTP),”
Computer Commun., vol. 30, no. 17, pp. 3239-3256, 2007.
[7] R. Stewart, Stream Control Transmission Protocol, IETF RFC 4960, Sep.
2007.
[8] Network Simulator (NS-2), http://www.isi.edu/nsnam.ns
Authorized licensed use limited to: Kyungpook National University. Downloaded on June 04,2010 at 01:16:17 UTC from IEEE Xplore. Restrictions apply.