Fairness in Transfer Control Protocol for Congestion Control in Multiplicative Increase and Multiplicative Decrease

International Journal of Engineering Inventions
ISSN: 2278-7461, ISBN: 2319-6491
Volume 2, Issue 1 (January 2013) PP: 46-50
www.ijeijournal.com P a g e | 46
Fairness in Transfer Control Protocol for Congestion
Control in Multiplicative Increase and Multiplicative
Decrease
Sarika Gupta Agarwal1
, Parul Gupta2
, Mamta Narwaria3
1,3
Department of Computer Science, Greater Noida-201308 India
2
Department of Computer Science, Mumbai Maharashtra, India
Abstract:- High speed Network(Scalable TCP) uses Multiplicative Increase Multiplicative Decrease (MIMD)
algorithm . This paper investigates fairness among sessions sharing a common bottleneck link, where one or
more sessions use the MIMD algorithm. Congestion occur when Traffic reached Beyond the capacity of
window. We first study how two MIMD sessions share the capacity in the presence of generalcombinations of
synchronous and asynchronous losses. We show that, in the presence of rate dependent losses, the capacity is
fairly shared whereas rate independent losses provide high unfairness.
Keyword:- MIMD, Scalable TCP, AIMD,CUBIC,WESTWOOD,Fairness
I. INTRODUCTON
TCP provides the mechanisms that provide data to be transferred across networks that are dynamic and
have a large variety of resources. Until now the AIMD algorithm was found to provide satisfactory
performance.However, in high speed networks , the additive increasing the sender’s rate may lead to inefficient
link utilization[1].To overcome this drawback in high speed networks, the MIMD algorithm has been proposed
as an alternativeto the AIMD algorithm In the future, situations may arise when sessions using these two
algorithms would compete for the same network resource. The share of the capacity obtained by each of these
sessions will depend on the various parameters specific to the algorithms. The sharing of a resource gives rise to
the question of how fairly is this resource shared among the sessions. Fairness issues have been addressed in
several previous works. In [4], the authors considered a class of rate control algorithms in the presence of
synchronous control signals. They showed that the AIMD algorithm converged to fairness whereas the MIMD
algorithm did not converge. In [5], the author studied the MIMD algorithm under a more realistic assumption
of rate dependent losses and argued that MIMD algorithm also converges to fairness. In [6] the convergence to
fairness ofthe different flavors of TCP was studied both analyticallyand using simulations. The monotonic
convergence to fairness for algorithms in rate-based TCP-friendly applications was studied in [7].
II. BASIC CONCEPTS OFTCP CONGESTION CONTROL ALGORITHMS
2.1 AIMD Algorithim
The data transfer of TCP starts from a slow start, in which TCP tries to increase its sending rate
exponentially, until it encounters the first loss. It then switches to another stage, called congestion avoidance, in
which TCP employs the Additive Increase, Multiplicative decrease mechanism to slowly adapt to the available
bandwidth. Further congestion, the TCP goes into the Fast Recovery &Fast Retransmission stages. In this
scenario, when TCP do not receive an acknowledgment for a packet after some timeout period, it assumes that
this packet is lost. & then retransmits that packet and doubles its retransmission timeout value (RTO) detecting
packet loss. This process continues until the packet is successfully transmitted & acknowledged. TCP tries to
clear congestion by cutting its sending rate in half.
2.1.1 Additive Increment
After receiving an ACK for new data, congestion window is increment by (MSS)2
/Cwnd, where MSS
is maximum segment size, this formula is known as additive increment. The goal of additive increment is to
open congestion window by a maximum of one MSS per RTT. Additive increment can be described by using
the equation (1):
Cwnd = Cwnd + a*MSS2/Cwnd Eq (1)
Where the value of a is a constant, a = 1.
2.1.2 Multiplicative Decrement

Fairness in Transfer Control Protocol for Congestion Control in…
Multiplicative decrement occurs after a congestion event, such as a lost packet or a timeout. After a
congestion event occurs, the slow start threshold is set to half current congestion window. This update to slow
start threshold follows equation (2)
ssthresh = (1 – b)*CWND
Eq(2)
CWND is equal to amount of data that has been sent but not yet ACKed and b is a constant, b = 0.5. The
congestion window is adjusted accordingly. After a timeout occurs, congestion window is set to one MSS and
slow start algorithm is reuse.
2.2Multiplicative Increase and Multiplicative Decrease(MIMD)
Scalable TCP uses a different congestion avoidance algorithm than Standard TCP. Scalable TCP uses a
multiplicative increment multiplicative decrement (MIMD) rather than the AIMD of Standard TCP
2.2.1 Multiplicative Increment
The multiplicative increment occurs when standard additive increment would normally occurs. In
equation (8) shows the formula used to adjust congestion window after receiving a new ACK.
Cwnd = Cwnd + a*Cwnd Eq. (3 )
Where a is adjustable, the value of a used was 0.02
2.2.2 Multiplicative Decrement
The multiplicative decrement is same as Standard TCP except that the value of b in equation (2) is
adjustable, the value of b used 0.125. The connection starts in the slow start algorithm until channel is filled.
The connection uses the multiplicative increment portion of congestion avoidance to adjust congestion window.
After a single drop occur around 1.4 seconds, fast retransmit and recovery algorithms are used to cut congestion
window by 0.125, the value of b, and congestion avoidance is used again to reopen congestion window
III. COMPARISON BETWEEN STANDARD TCP AND SCALABLE TCP
Table 1:Comparison between Standard and Scalable TCP
Standard TCP(AIMD) Scalable TCP(MIMD)
no losses Wn+1=Wn+1=
linear increase
dw/dt=1/T
Wn+1=∝*Wn= multiplicative increase
dW/dt=log[∝]/T*W
=exponential growth
≥1 loss Wn+1=0.5*Wn
multiplicative decrease
Wn+1=𝜷*Wn
multiplicative decrease
IV. FAIRNESS IN MIMD SESSIONS (EQUAL RTTS)
We consider two sessions which share a link of capacity C. At time t, the rates obtained by the two
sessions are denoted by x(t) ≡ (x1(t), x2(t)). At each control instant, the controller sends a control signal to each
source. This control signal either informs on no congestion (a 0 signal) or of congestion (a 1signal). In the
absence of congestion, the sources increase theirrate exponentially xk(t + τ) = ατ/τ0 ・xk(t), k= 1, 2,
Table 2: (REACTION TO CONTROL SIGNALS)
control vector x1(tj+) x2(tj+)
(0,0) x1(tj) x2(tj)
(0,1) x1(tj) β ・x(tj)(A Synchronous Congestion)
(1,0) β ・x1(tj) x2(tj) (A Synchronous Congestion)
(1,1) β ・x1(tj) β ・x2(tj)(Synchronous Congestion)
Whereτ0 is the time constant (for example, the RTT) for the sessions, and α >1 is the increase factor
The above Formulation is a continuous time equivalent of a multiplicativealgorithm in which, for every RTT
without congestion signals, the sender multiplies the window by a factor of α. This can be seen by substituting t
= nτ0. We assume that the two sources receive the control signals at the same instant. However, unlike the
model in [4], the two sources can receive different control signals. That is, a congestion signal need not be sent
to both the sources at the same instant. Hence, the congestion signals could be asynchronous. Let β <1 be the

decrease factor. Let thejthcontrol signal be received at time tj. Then, the four possibilities for the rate vector,
x(tj+), just after tj, are givenin Table 1 The source continues with the increase algorithm on the reception of 0
signal. On the other hand, when a source receives a 1 signal, it instantaneously reduces its rate.
V. INTER PROTOCOL FAIRNESS (SAME RTT)
In this we study the fairness issue when sessions using two different congestion control algorithms
share a common link, and the losses are synchronous. Recently Scalable TCP, which uses the MIMD algorithm,
has been proposed as an enhancement for TCP in high-speed networks. Situations may, therefore, arise in which
a user with Scalable TCP shares a link with a user with standard TCP. Specifically, we study the equilibrium
behaviour of the window size, and the throughput obtained by a session of each algorithm at equilibrium in the
presence of synchronous losses
only. We also look at conditions under which a user of one algorithm can obtain a better throughput than a user
of the other algorithm. Previous work (e.g., [6], [8]) mainly studied the behaviour of sessions using the same
algorithm.
In this section, we assume that each session has the same RTT, τ . As mentioned in Section II, window-based
notation is equivalent to rate-based notation. In the rest of this paper, we use the window-based notation since
we are interested in obtaining the equilibrium window sizes for the sessions
5.1 System Model
Consider l sessions which share a link of capacity C bits/s.Each session transmits data using packets of
size M bits. Let Λ be the bandwidth-delay product (BDP) of the network. We assume that each session has the
same RTT, τ , and that the RTT is mainly determined by the propagation delay and, hence, can be considered to
be a constant. Let x(t) = (x1(t) x2(t) . . . xl(t)) denote the vector of window sizes of the k sessions at time t. A
synchronous loss (i.e., a loss for each session) is assumed to occur at time t if
x𝑖(𝑡) > Λ𝑙
𝑖=1 Eq.4
The above condition is equivalent to saying that a synchronous loss occurs when the total number of
outstanding packets in the network exceeds the total number of packets that the network can handle. Without
loss of generality, let sessions 1, 2, k use the MIMD congestion control algorithm and the rest of the l –ksessions
use the AIMD congestion control algorithm. In the absence of losses, the two algorithms increase the window in
the following way
xi(t+Δ) ={ x(t)αΔ/τ, 1 ≤ i ≤ k Eq.5
xi(t+Δ) ={ x(t) +, αa Δ/ τ, k+ 1 ≤ i ≤ l, Eq.6
Whereαm and αa are the increase parameters of the MIMD and the AIMD algorithm, respectively. For
example, αm =1.01 for Scalable TCP, and αa= 1 for standard TCP. Let tndenote the time instant when the nth
congestion signal is received. We note that a congestion signal is generated when a synchronous loss occurs.
In response to a congestion signal the two algorithms decrease the window in the following way.
xi(t+n) = { βmx(tn), 1 ≤ i ≤ k, Eq.7
xi(t+n) = { βax(tn), k+ 1 ≤ i ≤ l, Eq.8
_
Whereβm and βa are the decrease parameters of the MIMD and the AIMD algorithm, respectively. For example,
βm =0.875 for Scalable TCP, and βa = 0.5 for standard TCP.
VI. SIMULATIONS
We now compare these observations with simulations performed using ns-2 (version 2.26) and
wireshark. Unless stated otherwise, the simulation had the same set of parameters. The MIMD sessions used
Scalable TCP, and the AIMD sessions used TCP New Reno. The packet size, M, for each session was set to
1040 bytes (1000 bytes of data + 40 bytes of header). The propagation delay, σ, was taken to be 100ms. The
increase and decrease parameters for the two algorithms were set to αm = 1.01, αa= 1.0, βm = 0.75, and βa =
0.5. Since the ψifor AIMD increases with decrease in βm, we set βm to a value smaller than its recommended
value so that the AIMD sessions also obtain a certain throughputwe note that the AIMD sessions indeed
converge to the same equilibrium window size whereas the equilibrium window size of an MIMD session
depends on its window just before the first synchronous loss. The ψifor AIMD sessions remains the same even
though the link capacity is increased from 200Mbps to 300Mbps and the total number of sessions is increased

from six to twelve. Let ηaand ψadenote the throughput and the equilibrium window size, respectively, of any
one of the AIMD sessions.
Fig:1 Window Evolution of Sessions[source: Ref.11]

Fig:2RTT Graph for TCP CUBIC+Vegas+Reno Fig:3 Throughput Graph for TCP Cubic+Vegas+Reno
VII. CONCLUSIONS
Scalable TCP implements simple changes to the currently used congestion control algorithm. These
changes have both a positive and negative effects on the existing network traffic. Each algorithm provides
higher channel utilization for high speed and long delay environment. However, the alternative algorithms do
not shares chainnnel equally, when mixed with Standard TCP traffic. In a homogenous environment, the overall
channel utilization and sharing between streams increments as compared to a mixed environment. Future work
is needed to study the effects of more than two competing streams. In This Paper , we studied the fairness in
sessions using MIMD congestion control algorithm. For sessions with the same RTT, it was observed that there
was extreme unfairness when the asynchronous losses were rate independent. It was shown that fair sharing
could be achieved by introducing a stream of rate dependent losses. For sessions with different RTTs, it was
observed that the arrival rate of these rate dependent losses had to be greater than a certain minimum rate in
order to achieve fairness. Therefore, in networks with sessions using MIMD algorithms, a stream of rate
dependent losses, using, for example, some
buffer management scheme, would be necessary to ensure fair sharing.
REFERENCES
1) KulvinderSingh”Experimental Study of TCP Congestion Control Algorithms”,IJCEM International
Journal of Computational Engineering & Management, Vol. 14, October 2011 ISSN (Online): 2230-
7893
2) S. Floyd. HighSpeed TCP for Large Congestion Windows. RFC 3649, Experimental, December 2003.
3) T. Kelly. Scalable TCP: Improving Performance in Highspeed Wide Area Networks. Computer Comm.
Review, 33(2):83–91, April 2003.
4) G. Vinnicombe. On the Stability of Networks Operating TCP-like Congestion Control. In Proceedings
of the IFAC World Congress, 2002.
5) D. Chiu and R. Jain. Analysis of the Increase/Decrease Algorithms for Congestion Avoidance in
Computer Networks. Journal of ComputerNetworks and ISDN, 17(1):1–14, 1989.
6) S. Gorinsky. Feedback Modeling in Internet Congestion Control. In Proceedings of the NEW2AN,
February 2004. Also see a techincal report at http://www.arl.wustl.edu/∼gorinsky/ TR2002-39.ps.
7) G. Hasegawa, M. Murata, and H. Miyahara. Fairness and Stability of congestion control mechanisms
of TCP. Telecommunication Systems,November 2000.
8) ]D. Loguinov and H. Radha. End-to-End Rate-Based Congestion Control:Convergence Properties and
Scalability Analysis. IEEE/ACM Transactionson Networking, 11(4), August 2003
9) L. Xu, K. Harfoush, and I. Rhee. Binary Increase Congestion Control(BIC) for Fast Long-Distance
Network. In Proceedings of the IEEEINFOCOM, March 2004.
10) A. Budhiraja, F. Hernndez-Campos, V.G. Kulkarni, and F. D. Smith. Stochastic Differential Equation
for TCP Window Size: Analysis and Experimental Validation. Prob. in the Engg. and Informational
Sciences,18:111–140, 2004.
11) E. Altman K., E. Avrachenkov, B. J. Prabhu. Fairness in MIMD Congestion Control Algorithim 2004.

Fairness in Transfer Control Protocol for Congestion Control in Multiplicative Increase and Multiplicative Decrease

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to Fairness in Transfer Control Protocol for Congestion Control in Multiplicative Increase and Multiplicative Decrease (20)

More from International Journal of Engineering Inventions www.ijeijournal.com (20)

Recently uploaded (20)

Fairness in Transfer Control Protocol for Congestion Control in Multiplicative Increase and Multiplicative Decrease