• [email protected]
  • +971 507 888 742
Submit Manuscript
SciAlert
  • Home
  • Journals
  • Information
    • For Authors
    • For Referees
    • For Librarian
    • For Societies
  • Contact
  1. Information Technology Journal
  2. Vol 12 (4), 2013
  3. 584-593
  • Issues
    Online First Current Issue All Issues
  • Information About
    Aims and Scope Editorial Board Guide to Authors Article Processing Charges
    Submit a Manuscript

Information Technology Journal

Year: 2013 | Volume: 12 | Issue: 4 | Page No.: 584-593
DOI: 10.3923/itj.2013.584.593
crossmark

Facebook Twitter Reddit Linkedin E-mail
Research Article

Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Chuanhuang Li
College of Computer Science and Technology, Zhejiang University, Hangzhou 310029, China

Sanyuan Zhang
College of Computer Science and Technology, Zhejiang University, Hangzhou 310029, China

Weiming Wang
College of Information and Electronic Engineering, Zhejiang Gongshang University, Hangzhou 310018, China

ABSTRACT


TML (Transport Mapping Layer) is the ForCES (Forwarding and control separation) protocol message transporting layer in the ForCES framework. Its communication performance will directly affect the efficiency and availability of ForCES NE (Network Element). In this study, the congestion control mechanism of SCTP (Stream Control Transmission Protocol) transport mapping layer is introduced and in a stochastic network calculus framework, the service model and service curve of SCTP which is in the congestion avoidance state by using the AIMD (Additive Increase and Multiplicative Decrease) algorithm is studied. At the same time, based on the EBB (Exponentially Bounded Burstiness) input model, the stochastic performance is also analyzed and the bounds on backlog and delay are derived. Numerical examples are presented to test the bounds. These performance analysis results can be used to optimize the SCTP TML to improve the efficiency of the ForCES communication channel.
PDF Abstract XML References Citation

Keywords


  • stochastic network calculus
  • AIMD service curve
  • SCTP congestion avoidance
  • SCTP performance analysis
  • ForCES TML
  • ForCES TML

Article History

Received: September 23, 2012;   Accepted: December 13, 2012;   Published: January 22, 2013

How to cite this article

Chuanhuang Li, Sanyuan Zhang and Weiming Wang, 2013. Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage. Information Technology Journal, 12: 584-593.

DOI: 10.3923/itj.2013.584.593

URL: https://scialert.net/abstract/?doi=itj.2013.584.593

INTRODUCTION


Transport Mapping Layer (TML) addresses the protocol message transportation issues between Control Element (CE) and Forwarding Element (FE) in the ForCES (Forwarding and control element separation) protocol architecture for IP Network Element (NE, Such as, router, firewall, or load balancer, etc.) (Yang et al., 2004). It isolates Protocol Layer (PL) from the underlying network to solve mapping problem in different link layer (such as TCP/IP layer, Ethernet layer, ATM, etc.) (Doria et al., 2010). That improves the flexibility of ForCES system.

SCTP (Stream Control Transmission Protocol) TML (Salim and Ogawa, 2010) is a TML standard which has been published by ForCES working group in IETF. Before it became the RFC, exactly, there were several TML drafts in the WG. Apart from the SCTP TML draft, Wang et al. (2007) also proposed a TML service primitives draft based on TCP/UDP transport protocol. The main idea of this draft is using TCP and UDP to transmit control protocol messages and redirection protocol messages separately. At the same time, some researches which are related with this TML were conducted. Such as, Li et al. (2012) proposed a scheme of reliable multicast between CE and FE in ForCES architecture NE based on TCP/UDP TML. Zhuge et al. (2011) proposed a scheduling scheme to schedule ForCES messages in this kind of TML. Khosravi et al. (2006) proposed a TCP/IP based TML which uses TCP transmit control protocol messages. However, it uses DCCP protocol to transmit redirection protocol messages.

SCTP is a reformative transport layer of next generation IP network, in order to overcome the shortcoming of traditional TCP/UDP protocol. The same as TCP, SCTP also uses AIMD (Additive Increase and Multiplicative Decrease) algorithm to avoid the congestion (Stewart, 2007). Many researchers paid attention to the AIMD congestion avoidance. Altman et al. (2005) analyzed the performance of AIMD-like flow control mechanism. They employ a fluid approach based on a multi-state to model the controlled flow. Jasem et al. (2010) introduced a new AIMD algorithm and analyzed its delay, efficiency and fairness. Kim and Hou (2009) proposed to speed up the simulation performance for TCP-operated networks by incorporating network calculus-based models in a simulation framework. They analyzed TCP AIMD behavior and the service process. We used the same method of this literature to obtain the general expression of AIMD service curve. Then with this service curve, we studied the performance of the SCTP TML under EBB (Exponentially Bounded Burstiness) input flow model.

PRELIMINARY WORKS

Network calculus is a tool which can be used to analyze the delay and other characteristics of QoS in computer network. It provides an effective approach to model the more and more complicated and diverse network. The network calculus’s theoretical basis is the min-plus algebra and max-plus algebra which were introduced by Cruz (1991a, b). Since then network calculus has been continuously developed and generates two branches: deterministic network calculus and stochastic network calculus. Deterministic network calculus is mainly used in the occasion which provides certain service guarantees. However, many actual applications can tolerate a small amount of packet loss and delay. Stochastic network calculus is used in this situation to get the statistic performance. The detailed principles of network calculus was studied by Le Boudec and Thiran (2001) and Chang (2000).

In traditional algebra, the two most basic operations are plus and multiply in the integer or real number field. We can express it as (R, +, x). In Min-Plus algebra, the plus in traditional algebra evolved into an operator to find a minimum, multiplication operation evolved into plus and +∞ is introduced into the computing field. Its algebraic structure can be expressed (R∪{+∞}, ∧, +), where, ∧ denotes the infimum or, when it exists, the minimum. For the convenience of description, we list some concepts and operations in min-plus algebra. The detailed principles of network calculus can be found by Jiang and Liu (2008).

• Wide-sense increasing functions set F:
 

F = [f (t)|f (0) = 0, ∀t<0; f (0)≥0, f (s)≤f (t),
∀s≤t, s, t∈ [0, +∞]

• Wide-sense decreasing functions set Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage:
 

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage = [f (t)|f (0) = 0, ∀t<0; f (0)≥0, f (s)≥f (t),
∀s≤t, s, t∈ [0, +∞]

• Pointwise infimum: f ∧ g (t) = inf [f (t), g(t)]
• Pointwise supremum: f ∨ g (t) = sup [f (t), g (t)]
• Min-plus convolution: (f⊗g)(t) = inf {f (t-s)+g (s)}, ∀f, g∈F
• Min-plus deconvolution: (føg)(t) = sup {f (t+s)-s (s)}, ∀f, g∈F
• Maximum horizontal distance h (f, g): h (f, g) = sups≥0 {inf {τ≥0: f (s)≤s (s+τ)}}, ∀f, g∈F
• Maximum vertical distance v (f, g): v (f, g) = sups≥0 {f (s)- g (s}, ∀f, g∈F

Arrival curve and service curve are two very important concepts in network calculus. Jiang and Liu (2008) introduced weak stochastic service curve, t.a.c (traffic-amount-centric) stochastic arrival curve.

Assume traffic arrival process is denoted by A(t), the output process, A*(t) and the accumulative arrivals in interval [0, t], A(s, t), respectively. A service system S is said to provide a weak stochastic service curve β∈F (non-negative wide-sense increasing functions) with bounding function g∈Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage (non-negative wide-sense decreasing functions), denoted by S~ws<g, β>, if for all t≥0 and x≥0, there holds:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(1)

A flow is said to have a t.a.c stochastic arrival curve α∈F with bounding function f∈Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage, denoted by A(t)~ta<f, α>, if for all 0≤s≤t, there holds:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(2)

When we analyze the performance of SCTP AIMD, we will use the EBB as the traffic arrival model (Yaron and Sidi, 1993). A flow is said to have EBB with parameters (ρ, a, b), if for all s, t≥0 and all x≥0, there holds:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(3)

Apparently, EBB is a special case of the definition of t.a.c stochastic arrival curve.

ForCES SCTP TML

ForCES protocol interface include PL and TML. PL is mainly responsible for the work of encapsulating the data into the ForCES protocol message which can be transmitted by TML. It is also responsible for the management and control TML; TML is responsible for the transmission messages issued by the PL, shielding the underlying transport implementation details.

The message transmission mechanism between CE and FE restricts the performance of the entire NE. In order to guarantee the reliability and real-time, the choice of transport protocol is vital to the performance of the whole system.

The single channel communication model of SCTP TML: SCTP is an end-to-end transport protocol which can do most of what UDP, TCP, or DCCP can achieve. It also can do most of what a combination of them can achieve (e.g., TCP and DCCP or TCP and UDP). Comparing with TCP, SCTP can also provide reliable, ordered, connection-oriented, flow-controlled, congestion-controlled data exchange, but it cannot provide byte streaming and instead provides message boundaries. Comparing with UDP, SCTP can provide unreliable, unordered data exchange but it cannot provide multicast support. Comparing with DCCP, SCTP can provide unreliable, ordered, connection-oriented, congestion controlled data exchange (Salim and Ogawa, 2010; Stewart, 2007).

SCTP has all the features required to provide a robust TML. As a transport that is all-encompassing, it negates the need for having multiple transport protocols in order to satisfy the TML requirements that in Section 5 of [RFC5810] (Doria et al., 2010). As a result, it allows for simpler coding and therefore reduces a lot of the interoperability concerns.

Figure 1 shows the interfacing between PL and SCTP TML and the internals of the SCTP TML. TML core interfaces to the PL utilizing the TML API and interfaces to the SCTP layer utilizing the standard socket interface.

In TML core, there are three channels used to group and prioritize the work for different types of ForCES traffic. The Higher-priority (HP) channel is used for transporting configuration\query messages and their response messages. The Medium-priority (MP) channel is used for event notification messages. The Lower-priority (LP) channel is for redirect and heartbeat messages.

The SCTP TML processes the channel work in strict priority. For example, if there are messages incoming from an FE on the HP channel, they will be processed first until there are no more left before processing the next priority work.

In user’s implementation, if the all three channels using a single socket, analysis revealed that head-of-line blocking issues will be occurred. Packets in LP channel not needing reliable delivery could block packets in HP channel that needs reliable delivery under congestion situations for an indeterminate period of time. For this reason, we elected to go with mapping each of the three channels to a different SCTP socket (Salim and Ogawa, 2010).

SCTP slow-start and congestion avoidance: SCTP uses the same congestion control mechanism as TCP. The mechanism includes slow-start stage, congestion avoidance stage, fast retransmit and fast recovery stages. But the congestion control in SCTP is employed in regard to the association, not to an individual stream (Stewart, 2007).

Like TCP, an SCTP endpoint uses the following three control variables to regulate its transmission rate:

• Receiver advertised window size (rwnd, in bytes) which is set by the receiver based on its available buffer space for incoming packets
• Congestion control window (cwnd, in bytes) which is adjusted by the sender based on observed network conditions
• Slow-start threshold (ssthresh, in bytes) which is used by the sender to distinguish slow-start and congestion avoidance phases

Beginning data transmission into a network with unknown conditions or after a sufficiently long idle period requires SCTP to probe the network to determine the available capacity.

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 1: The communication architecture of SCTP TML

The slow-start algorithm is used for this purpose at the beginning of a transfer, or after repairing loss detected by the retransmission timer.

When cwnd is greater than ssthresh, the congestion avoidance mechanism will be started. SCTP use AIMD policy to avoid the congestion. The AIMD algorithm can be described as below: cwnd will be incremented by 1*MTU per RTT (Round Trip Time) if the sender has cwnd or more bytes of data outstanding for the corresponding transport address. This is the additive increase process. Same as in the slow start, when the sender does not transmit data on a given transport address, the cwnd of the transport address should be adjusted to max(cwnd/2, 4*MTU) per RTO (Retransmission Timeout). This is the multiplicative decrease process.

SERVICE CURVE OF SCTP AIMD

As we have showed, AIMD is the congestion avoidance algorithm that used in SCTP. When SCTP discovered there is no congestion, the sending rate will be increased linearity; otherwise, the sending rate will be decreased multiplicatively. SCTP will know the congestion state according whether the sender has received the ACK message from the receiver.

Suppose the number of AIMD period (One AIMD period includes an additive increase process and a multiplicative decrease process.) is (m-1) in the interval Γ, the initial window size is w0, the additive increase parameter AI (Additive Increase) is A and the multiplicative reduction parameters MD (Multiplicative Decrease) is B. As Fig. 2 showed, wi (t) is capacity of the ith SCTP congestion window. TRi (t) is the RTT experienced by the data sent at time t.

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 2: AIMD algorithm window size

From Kim and Hou (2009), we have:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(4)

Let Si be the cumulate data flow in the ith AIMD period, then we have the following total service process SΓ, m-1:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(5)

Assume that the link bandwidth is L, then the model can provide services β (t) = min (Lt, St,m-1), because the size of the window AIMD cannot be more than L, so service curve is St,m-1.

Remarks: We make the following assumptions on TML communication mode. (l) Network is relatively stable, free from other external factors, (2) RTT of SCTP packets are stable, (3) A = 20RTT2, B = 0.8 in AIMD (4). In case that the size of sending window can changes from 80 to 100 Mbps in 1 sec, In the network, the maximum data packet Lmax = 1 M. When there is enough data to send at all times, the AIMD window size will be Fig. 3.

When the constant rate arriving to a single SCTP channel is ρc<80 Mbps, because the data arrival rate is less than AIMD data service rate Vm = 90 Mbps, the buffer will be empty at some point and then the stagnation of SCTP send window will appear. The change of AIMD sending window depends on the network situation reflected by packet ACK. when there is no data to sent, AIMD sending window holds. The practical effect diagram is as Fig. 4.

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 3: The changes of AIMD window size in the given scenario

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 4: The stagnation phenomenon of the window size

When the constant rate arriving to a single SCTP channel is 90≤ρc≤100 Mbps, as the average data transmission rate of AIMD algorithm Vm = 90 Mbps, so SCTP channel buffer will not be empty. The change of sending window in this situation is shown in 0 and then we can get their arrival curve and service curve. Its data arrival curve is α (t) = ρct and service curve is β (t) = SΓ,m-1, where m is the complete number of cycles, Γ is the time experienced by the whole experiment.

STOCHASTIC PERFORMANCE ANALYSIS OF SCTP TML UNDER EBB INPUT MODEL

It is difficult to calculate St,m-1 accurately. Here, we study St,m-1 in an ideal AIMD situation which we called AMS (AIMD Metastable State) state.

Definition 1: Only when the AIMD window reaches a fixed threshold size, packet loss will appear on the link and this threshold may be total link capacity or a fixed bandwidth allocated to the link by the network node. AIMD window does not decrease consecutively while the packet loss occurred. To satisfy the above conditions, the system state is called AIMD Metastable State.

From the definition, we know the network will not be affected by other external conditions in the AMS state and the backlog will never be null.

We assume all packets’ RTTs are the same. The fixed threshold of the AIMD window is C, when SCTP is in the AMS state. The change of AIMD algorithm window size will be like Fig. 5.

We can get the service process in the time interval t:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(6)

where, Sm+1 is the flow of the (m+1)th AI stage which is incomplete and t1 = C-wo/k, t2 = t3 =þ= tm = T = C(1-B)/k, k = tanθ = A/t2Ri, m-1 = (t-ti)/T, tm+1 = t-(m-1)T-t2.

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 5: The changes of AIMD window size in AMS state

Ignoring the slow-start stage and after the system reached AMS state, from the time that the window size w0 = BC, in the time interval t , the total number of system services is:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(7)

where, T = C(1-B)/T, k = tanθ = A/t2Ri, m = t/T.

Apparently:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(8)

So, we can get the service envelop process of AIMD:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(9)

where, T = C(1-B)/k, k = tanθ = A/t2Ri, m = t/T

The backlog bound of SCTP AIMD: Considering a work conserving system, according the Theory 5.2 and 5.5 in Jiang and Liu (2008): if the input has a t.a.c stochastic arrival curve α∈F with bounding function f∈Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage and the system provides to the input a weak stochastic service curve β∈Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage with bounding function g∈Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage, then for all t≥0 and x≥0, the backlog Q(t) is bounded by:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(10)

and the delay D (t) is bounded by:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(11)

where, Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage, for any θ>0.

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 6: The maximum vertical distance υ (α, β)

For the EBB arrival process, fθ(x) = ae-bx (1+θb). If the SCTP is in AMS state, then g (x) = 0. So:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(12)

When, Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage as showed in Fig. 6, the maximum vertical distance υ(α, β) locates at t which is close to the T's jump point, then υ (α, β) = (ρ-BC)T.

So we can get:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(13)

Applying Eq. 13 to 12, we have:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(14)

where, T = C(1-B)/k, k = tanθ = A/t2Ri.

Remarks:

• If the SCTP is in AMS state, the backlog bound is related to the parameter ρ of EBB. When ρ is less than the smallest window size (BC), the backlog bound will be tighter. That is, the probability of existing backlog will be smaller. Moreover, The backlog bound is irrespective of the AIMD parameters and it’s only related with EBB parameters and x
• When ρ is greater than BC, the probability of existing backlog will be bigger. Furthermore, the bound is related with the AIMD parameters, x and the EBB parameters at the same time
• When, Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage ρ smaller or BC greater, the backlog bound will be tighter

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 7: The maximum horizontal distance h(α+x, β) (ρ≤BC)

The delay bound of SCTP AIMD: According Eq. 11 and the fact g (x) = 0 when the SCTP is in AMS state, we have:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(15)

We first compute the maximum horizontal distance h (α+x, β), then get the delay bound.

ρ≤BC: From Fig. 7, we can get:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(16)

where, n is any integer and n≥0.

• When Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(17)

Let Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage then Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage. So:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(18)

where, nT≤d<(n+1)T, n is any integer and n≥0.

• When:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(19)

Apparently, if x is greater, the probability will smaller and the bound is tighter. So:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(20)

Combined Eq. 18 and 20, we obtain:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(21)

where, nT≤d< (n+1)T, n is any integer and n≥0.

Though equivalent transformation to the condition, when ρ≤βC, for any d>0, we have the following final delay bound result:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(22)

where, n = [d/T].

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage: As showed in Fig. 8, with the changes of x, the h (α+x, β) is different. Assume n is any integer and n≥0.

• When Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(23)

Applying Eq. 22 to 15, so:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(24)

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 8: The maximum horizontal distance h (α+x, β) Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Let:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

then Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance StageWe obtain:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(25)

where, Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

• When:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(26)

Applying (26) to (15), so:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(27)

We obtain:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(28)

where, d = (n+1)T.

• When:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(29)

So:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(30)

Let:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

then x = (n+2) BCT+1/2 (n+1) TC (1-B)+ρ((n+2)T-d) . We obtain:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(31)

where, (n+1)T≤d< (n+2)T-BCT/ρ.

Based on the above results (25) (28) (31), when Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage, we have the following final delay bound result:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(32)

Where:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage : h (α+x, β) will be infinity.

Remarks:

• The delay bound is closely related with the delay value. If ρ≤BC and when d changed in a T-cycle, the change of the bound is continuous. But when d increases beyond T, a jump change of the bound will occur and the bound will be tighter
• When we fix the delay d, apparently, greater B is, then tighter the delay bound will be, that is, the probability of having that delay is smaller

Numerical example: This section shows numerically the performance of SCTP TML. The messages transmitted between CE and FEs can be considered to have the same flow model with different parameters. The MMOO (Markov-modulated On-Off) process can well describe these flows.

MMOO is defined by considering a homogenous and continuous-time Markov chain x(t) with two states denoted by ‘On’ and ‘Off’ and with the transition matrix:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

Here, μ is the transition rates from the ‘On’ state to the ‘Off’ state and vice-versa for λ. So, in the steady state, the average dwell time of the process in ‘On’ state is 1/μ and the average dwell time in ‘Off’ state is 1/λ. Let the instantaneous arrival rate is either P (in ‘On’ state) or zero (in ‘Off’ state). According Courcoubetis and Weber (1996), Cuicu (2007), this MMOO process is an EBB process with:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
(33)

and a = 1, b = Δ, where Δ is subject to numerical optimizations.

Assume the allocated upper limit bandwidth to SCTP channel C is 20 M and AIMD parameter A is 20RTT2, B = 0.6. We can obtain:

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage

We select four sessions with different MMOO parameters showed in Table 1. Among them, ρ1<BC, ρ2~ρ4 are all in (BC, 1/2C (1+B)). Through (14), we can get the backlog bounds showed in Fig. 9. Through using (22) and (32), we can get the delay bounds showed in Fig. 10.

From the simulation results, we can see the backlog and delay bounds for session 1 is tighter than the others. Moreover, the bounds with smaller average arrival rate decay much slower.

Table 1: Four sets of MMOO model parameters
Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
1/μ: The average dwell time of the process in ‘On’ state, 1/λ: The average dwell time in ‘Off’ state; P: The instantaneous arrival rate in ‘On’ state; ρ: the parameter in EBB process

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 9: Backlog bounds computed with Eq. 14

Image for - Network Calculus Based Performance Analysis of ForCES SCTP TML in Congestion Avoidance Stage
Fig. 10: Delay bounds computed with Eq. 22 and 32

These indicate that if the average arrival rate is smaller, the probability of backlog and delay will be also smaller. We also can see the bound of backlog will be near to zero, when the backlog is greater than 7 Mb and the bound of delay will be near to zero too, when the delay is greater than 0.5 sec. That is to say the backlog of the SCTP in AMS state with the given parameters will not exceed 7 Mb basically and the delay produced in SCTP congestion avoidance will not exceed 0.5 sec basically.

CONCLUSIONS


The performance of congestion control mechanism of SCTP TML is analyzed. In a stochastic network calculus framework, the service model and service curve of AIMD algorithm was studied. At the same time, in the assumption of the system being in the AMS state and based on the EBB input model, the bounds on backlog and delay were derived. Numerical examples were presented to test the bounds.

We had added many conditions when analyzing the service process of AIMD. Those would affect the accuracy and make the results not having universality. But if we didn’t base on those conditions, the service curve would be very complex. It would bring more difficulty to derive the bounds of backlog and delay. So how to coordinate these problems and let the results having more universality is a work that needs to be studied.

Moreover, we only studied the performance of congestion avoidance stage of SCTP TML. The end to end performance between CE and FE is also one of our research directions.

ACKNOWLEDGMENTS


The authors would like to thank the anonymous reviewers for helpful comments. This work is supported by the National 973 Program (No. 2012CB315902, No.2009CB320804) of China, National Natural Science Foundation of China (No. 61272304, No.61170215).

REFERENCES


  1. Yang, L., R. Dantu, T. Anderson and R. Gopal, 2004. Forwarding and control element separation (ForCES) framework. RFC 3746, IETF, April 2004. https://tools.ietf.org/html/rfc3746.

  2. Doria, A., J.H. Salim, R. Haas, H. Khosravi and W.M. Wang et al., 2010. Forwarding and control element separation (ForCES) protocol specification. RFC 5810. http://tools.ietf.org/pdf/rfc5810.pdf.

  3. Salim, J.H. and K. Ogawa, 2010. SCTP-based Transport Mapping Layer (TML) for the forwarding and control element. RFC 5811. http://www.faqs.org/rfcs/rfc5811.html.

  4. Wang, W.M., J.H. Salim and A. Audu, 2007. ForCES Transport Mapping Layer (TML) service primitives. http://tools.ietf.org/html/draft-ietf-forces-tmlsp-01.

  5. Li, C., K. Jin, W. Wang and D. Hu, 2012. A protocol of reliable multicast between CE and FE in ForCES architecture network element. Inform. Technol. J., 11: 959-970.
    CrossRefDirect Link

  6. Jasem, H.N., Z.A. Zukarnain, M. Othman and S. Subramaniam, 2010. The delay with new-additive increase multiplicative decrease congestion avoidance and control algorithm. Inform. Technol. J., 4: 1327-1335.
    CrossRefDirect Link

  7. Zhuge, B., C. Yu, K.P. Liu and W.M. Wang, 2011. Research on internal flow control mechanism of for CES routers. Inform. Technol. J., 10: 626-638.
    CrossRefDirect Link

  8. Khosravi, H., S. Chawla, F. Ansari and J. Maloy, 2006. TCP/IP based TML (transport mapping layer) for ForCES protocol. Working Group: ForCES. http://tools.ietf.org/html/draft-ietf-forces-tcptml-04.

  9. Stewart, R., 2007. Stream control transmission protocol. RFC 4960. http://tools.ietf.org/pdf/rfc4960.pdf.

  10. Altman, E., K. Avrachenkov, C. Barakat and P. Dube, 2005. Performance analysis of AIMD mechanisms over a multi-state Markovian path. Comput. Networks, 47: 307-326.
    CrossRef

  11. Kim, H. and J.C. Hou, 2009. Enabling network calculus-based simulation for TCP congestion control. Comput. Networks, 53: 1-24.
    CrossRef

  12. Cruz, R.L., 1991. A calculus for network delay. I. Network elements in isolation. IEEE Trans. Inform. Theory, 37: 114-131.
    CrossRefDirect Link

  13. Cruz, R.L., 1991. A calculus for network delay. II. Network analysis. IEEE Trans. Inform. Theory, 37: 132-141.
    CrossRef

  14. Le Boudec, J.Y. and P. Thiran, 2001. Network Calculus: A Theory of Deterministic Queuing Systems for the Internet. Springer, New York, USA., ISBN-13: 9783540421849, Pages: 274.

  15. Chang, C.S., 2000. Performance Guarantees in Communication Networks. Springer, New York, USA., ISBN-13: 9781852332266, Pages: 392.

  16. Jiang, Y.M. and Y. Liu, 2008. Stochastic Network Calculus. Springer, New York, USA., ISBN-13: 9781848001268, Pages: 232.

  17. Yaron, O. and M. Sidi, 1993. Performance and stability of communication networks via robust exponential bounds. IEEE/ACM Trans. Networking, 1: 372-385.
    CrossRef

  18. Courcoubetis, C. and R. Weber, 1996. Buffer overflow asymptotics for a buffer handling many traffic sources. J. Applied Probab., 33: 886-903.
    Direct Link

  19. Cuicu, F., 2007. Scaling properties in the stochastic network calculus. Ph.D. Thesis, University of Virginia, USA.

Related Articles

The Delay with New-Additive Increase Multiplicative Decrease Congestion Avoidance and Control Algorithm
A Protocol of Reliable Multicast between CE and FE in ForCES Architecture Network Element
Research on Internal Flow Control Mechanism of For CES Routers

Leave a Comment


Your email address will not be published. Required fields are marked *

Useful Links

  • Journals
  • For Authors
  • For Referees
  • For Librarian
  • For Socities

Contact Us

Office Number 1128,
Tamani Arts Building,
Business Bay,
Deira, Dubai, UAE

Phone: +971 507 888 742
Email: [email protected]

About Science Alert

Science Alert is a technology platform and service provider for scholarly publishers, helping them to publish and distribute their content online. We provide a range of services, including hosting, design, and digital marketing, as well as analytics and other tools to help publishers understand their audience and optimize their content. Science Alert works with a wide variety of publishers, including academic societies, universities, and commercial publishers.

Follow Us
© Copyright Science Alert. All Rights Reserved