MPLS Based Adaptive Concurrent Multipath Packet Dispersion Architecture for VoIP Networks

INTRODUCTION

Voice over IP is a popular cost effective service that supports the transmission of voice packets over Internet protocol to replace the traditional PSTN telephone services with less cost. Currently the IP network provides best effort service and does not support QoS. Thus QoS powered IP networks are becoming essential to support many real time applications such as voice.

In general, VoIP applications get shammed because of codec type, packet loss, delay and jitter (Singh et al., 2012). In Fixed load traffic networks, QOS is decided based on the available resources (Al-Irhayim et al., 2000). However, to provide strict QoS guarantees, adaptive resource allocation becomes important criteria. MPLS network offers an efficient transport and traffic engineering capabilities for real time applications, but it does not distribute load evenly.

MPLS is developed by IETF in 2001. IP networks in general uses destination based hop to hop forwarding. A router that supports the MPLS mechanism is called Label Switching router (LSR) (Mellah and Abbou, 2006) and terminates at egress LSR so that it traverses through the MPLS cloud. Moreover, MPLS-TE method (Portoles-Comeras et al., 2007) permits explicit path setup with the given QoS constraint including bandwidth, throughput etc.

This study proposes a MPLS based dispersion architecture for network resource utilization and to perform efficient load balancing for the VoIP and Non VoIP flows with assured QoS guarantees. During the overload condition traffic flows are splitted into multiple paths disjoint paths. The selection of multiple paths depends on bandwidth and delay prerequisites.

Fig. 1:	SCTP’s Multistreaming and Multihoming Transmission

More specifically voice transmission bounds on delay and delay variance, so that it can be transmitted using User Datagram Protocol (UDP/IP) (Lizambri et al., 2001). Whereas non voice data traffic is transmitted using SCTP/IP protocols, since it needs high reliability and less susceptible to delay and jitter. SCTP protocol has several advantages when compared with Transmission Control Protocol. It is message oriented, multi stream, multihoming protocol as show in Fig. 1. SCTP transmits messages as chunk (collection of several bytes).

Multistreaming allows stream of data chunks to be sent in parallel onto one channel (called association in SCTP). The receiver stores all the received data chunks in a stream buffer. Sequencing is done within the buffer. When a data chunk is lost from a particular stream, the following data chunk are received and it is stored and request for retransmission of the lost data chunk. So TCP blocking is avoided, in which a particular stream hold all upper layer information (Fu and Atiquzzaman, 2004).

There are several studies are performed to perform load balancing, resource utilization and to improve QoS for VoIP applications. SCIP and MPLS-TE in multi homed environment are used in VoIP architecture for fault recovery system and resource utilization (Chang et al., 2009).

In this proposed architecture, SCTP was employed to transmit SIP signaling messages with SIP proxy server interfaces while MPLS-TE mechanism was applied to transmit the voice data. One SIP interface connects to the primary path and the other connects backup paths. This fast reroute mechanism can reduce the loss rate of voice data transmission upon network congestion or network link/node failures but delay increases.

Efficient bandwidth utilization is associated with complexity, algorithmic and processing delay. An algorithm for efficient bandwidth utilization for VoIP over a WAN is proposed (Bhanu et al., 2010). Bandwidth utilization is calculated using Erlang Bandwidth Data rate Moderation (BDM) technique and then packet is dispersed on the desired path. Packet size, CODEC and compression decides the bandwidth for a packet on a particular route. Only the packet header (RTP+UDP+IP) fields are compressed. When several users are working simultaneously, the real bandwidth is divided among the whole users. This improves the performance and voice quality of VoIP. Therefore, efficient bandwidth allocation in LAN network is becoming more essential.

Packet loss is one of the most important QoS parameter. It is caused by propagation delay, congestion and channel impairments like noise and interferences. These impairments provide audible gaps in the speech sequence. To improve the packet loss rate, scheme based on forward error correction technique for VoIP applications is proposed (Faritha Banu and Ramachandran, 2011a). Vandermonde encoding is used to recover packet losses due to failures in network.

Then the VOIP packets approaches the Rendezvous Point without considering the destination by dividing the traffic at that point. Then each RP node routes the encoded packets to their respective destinations. Erasure recovery and error correction techniques are used to recover from packet losses at the destination. VoIP packets can be routed with Quality-of-Service (QoS) guarantees while balancing the load without reconfiguration of the network in response to traffic changes.

Congestion control is a major issue to reduce packet loss ratio. To reduce congestion several studies are performed. To evaluate and minimize congestion in TCP/IP Networks, virtual laboratory scheme is proposed (Babainejad et al., 2010). A virtual machine is placed between real router and corresponding interfaces are created, traffic is routed via virtual router and then to the real router. Original traffic is received by the virtual router and congestion is analyzed. With this approach a real network can be emulated easily.

To perform load balancing and reduce packet delay in MPLS network superior erratic splitting ratio routine is advised (Murugesan et al., 2008). This study divides the flow into multiple streams. During link failure, flows are rerouted amid the backup path. Only one alternate path is considered. An alternate approach for performing load harmonizing is discussed (Levy and Zlatokrilov, 2006). This study also throws light on performance of the network through loss rate as well as loss burstiness. NLR metrics have been analyzed to several dispersion line of attacks to reduce rate of packet loss.

Another strategy for minimizing the packet loss rate uses a special loss recovery buffer at each Label Switch Router (LSR). Packet Losses are measured at the egress router for a specific time interval. If the packet loss rate is increased beyond the threshold level, then the high priority VoIP flows are rerouted to the alternate way (Faritha Banu and Ramachanran, 2011b). There are different strategies to achieve QoS models. Best effort model provides equal priority and bandwidth for all traffic but no guarantees of bandwidth or priority.

Integrated Service Model (IntServ) is a guaranteed service for some specific level of traffic in a specific period of time. It also needs the resource reservation before it starts transmission of data. If the network cannot provide the required bandwidth the session is not allowed (Shi et al., 2005). Differentiated service model (DiffServ) includes classification tools, queuing mechanisms, policy and traffic shapping. It offers different QoS for different traffic but still with no guarantee (Altalhi et al., 2012). Therefore customized QoS approach is needed as proposed in this study.

PROPOSED METHODOLOGY

This system analyses VoIP and Non VoIP flows in MPLS network. Deterministic Periodic round robin packet diffusion scheduler courses packets to multiple QoS parallel conduits. Non-VoIP flow is transmitted using SCTP Concurrent Multipath Transfer (CMT). Figure 2 gives the clear view about the proposal.

Flow classification: Edge-to-edge snooping subdivides the flow of traffic. All LSP’s router sends packets from way in to way out which gets received by Gap. Former sends feedback to way out router.

Fig. 2:	Load balancing in proposed system

Load balancing for Non VoIP flow: The non VoIP flows are routed using SCTP’s concurrent multipath packet dispersion with the specified QoS. Before data can be exchanged, the two SCTP hosts must establish association by exchanging the communications state, IP Addresses and port number using a four way Handshaking signal. SCTP’s association shutdown is a three-way handshake.

SCTP multihoming assigns primary address whereas remaining IP addresses are back up. When primary link (association) fails the data chunks are transmitted over the backup. Control chunk can be transmitted using backup paths without disturbing normal sequence of data transmission. This improves upturn due to any link failures devoid of violating transmission.

Generally, SCTP has several paths for transmission. SCTP sender is well informed about one primary path and has minimal information about other paths to a receiver. But Constraint based Concurrent multipath SCTP sender transfers data concurrently between multiple end to end paths. That is, it distributes data over multiple independent paths. All the available paths are utilized to send data chunk. The multipath selection process is an important criterion. This approach uses the shortest compound paths satisfying required bandwidth, min impediment to route the packets.

Load balancing for VoIP flow: System load balancing can be measured via BOT strategy. Queue tenure threshold is positioned to ¾ of whole buffer space (Banu and Ramachandran, 2012). Bandwidth plus delay are considered here.

QoS limit based Multipath routing gives straight paths satisfying QoS conditions. Net bandwidth ought to be>or = to that for end-to-end runs to meet specified QoS prerequisites.

Triggering invokes load adapter to supersede defaulting routing rule. The Ns-2 Script of attaching different traffic to MPLS network for Periodic round robin dispersion With 2 State Gilbert Error Model is given in Fig. 3.

Fig. 3:	Sample Ns-2 Script for Round robin dispersion of different traffic flows

EXPERIMENTAL RESULTS

Simulation results: This modus operandi is simulated in ns-2. This topology has ingress, egress nodes connected with 20 MPLS facilitated LSRs. Unlike delay and link bandwidth are allotted to 6 paths. Results are in terms of QoS traits. This proposal is equated in the company of no dispersion best effort routing stratagem.

Packet loss rate, throughput, delay are calculated for diverse intervals having constant size of load are given in Fig. 3-5, respectively. The delay calculations in-terms of Time and delay is depicted in Fig. 6. Throughput of 1.5% is maximized and 3% of minimal packet loss rate and 6% of observed minimum delay are achieved.

Fig. 4:	Packet loss rate interms of time vs. throughput

Fig. 5:	Throughput interms of time vs. packet loss

Fig. 6:	Delay calculations interms of time Vs delay