Network resource provisioning

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.


ith the desire of today's user to access more and more new Internet services and applications ubiquitously (at anytime, from any location and with any end device), uninterruptible communications, robustness and reliability are becoming key issues in future scenarios. Examples of such demanding applications can be real-time multimedia (VoD, VoIP, IPTV, e-learning, etc), where its time-sensitiveness require Quality of Service (QoS) guaranteed, beyond the traditional best-effort, over the time to propagate data with and low rates of propagation delay and packet loss. As current Internet was not designed for such environments, it cannot provide adequate user experience for resource-constrained sessions.

With the aim to overcome the limitations described hereinabove, several new applications and protocols, many of then operating on top of correct layers (overlay), have been added to provide functions not natively supported under Internet's conversational model, as QoS, multicast and mobility. However, it has contributed to damage Internet's performance, mainly by increasing complexity even more, as well as pushing excessive signaling load, energy consumption and CPU/memory overhead with functions to control congestion, resilience and so on [RFC4094].

In Future Internet scenarios, we expect the integration of many types of new sessions, such as health-care, location-based, environmental monitoring or energy management, which distinguish by their value added in terms of applicability and public benefits. Such sessions are currently sustained by complex infrastructures, whereas they can be much more efficient and sustainable as soon as the Internet is fully integrated into their basic functions and processes. However, new levels of efficiency and productivity gains will only be possible if technological roadblocks are overcome with multi-disciplinary and open innovation approaches.

Many efforts have been taken by the research community to specify Future Internet with a new architecture completely remodeled with major networking innovations for the efficient future session deployments. Future Internet is envisioned to allow the support of several types of sessions, which can be accessed ubiquitously and propagated with guaranteed resources over the time via multi-party connections (one-to-one, one-to-many or many-to-many). For instance sessions matched based on user profile (preferences, location, etc.) are intended to be simultaneously accessed by group of users with ensured quality experiences, independent of their underlying technologies.

In the scenario described above, IP multicast allows bandwidth-constrained data propagation to group of users, where packets are only replicated in branching points. However, the per-flow receiver-driven way of legacy IP multicast is well-known an excessive signaling generation and state storage approach, as well as QoS-unaware. In the latter, integration with a QoS approach is challenging due to contradicting principles, for instance QoS control is traditionally source-driven and intelligence pushed to edge nodes [RFC3754]. In what concerns QoS, admission control, fast resilience and efficient bandwidth provisioning control are required, so that allowing intermittent session propagation with required quality level guaranteed while keeping overall system performance. In this sense, aggregate bandwidth control is much better than per-flow, since the latter is well-known as inefficient due to the exceeding signaling overhead, thus placing a lot of CPU consumption [1]. Therefore, it is foreseen that Future Internet QoS and multicast control plane will require different realizations at different levels within the network for its success.

Independent of the way in which transport will be deployed in Future Internet, we consider that all mechanisms converge in the same requirement, efficient resource provisioning for QoS with scalable multicast support integration. Consequently, we propose the Multi-servIce Resource Allocation (MIRA) for the dynamic network resource provisioning through the efficient integration of QoS and IP multicast control strategies. MIRA allows the efficient support of many types of multi-party future applications by deploying: admission control, aggregate resource reservation, fast resilience and scalable QoS-aware IP multicast. MIRA supports two operation behaviors: (i) Base MIRA (B-MIRA), for session-driven (per-flow signaling) per-class network resource control; (ii) and Advanced MIRA (A-MIRA), with dynamic over-provisioning of bandwidth and distribution trees to drastically reduce signallings, thus optimizing overall system performance. The performance evaluation of MIRA suite is carried out in Network Simulator v2 (NS-2), which demonstrates B-MIRA benefits against related work, as well as A-MIRA performance improvements.

This paper is organized as in the following. Section 2 provides related work analysis in what concerns network resource provisioning. Section 3 describes MIRA proposal in details. Section 4 provides performance evaluation of MIRA proposal with the related work. Finely section 5 provides conclusion and future work.

II. Related Work

Future Internet control plane is envisioned to support efficient network resource control to allow intermittent session propagation with QoS guaranteed over heterogeneous environments. To that, resource reservation distinguishes itself by allowing data transport with bandwidth guaranteed, per individual flow or aggregation of flows, over the time. Bandwidth reservations can be updated as each flow is connected or disconnected (per-flow). Otherwise, a surplus amount of bandwidth (over-reservation) is initially assigned and is updated on-demand to accommodate more future flows without per-flow signallings.

In the scope of per-flow reservation control, existing standards and proposals are unfeasible for Future Internet scenarios due to the poor scaling properties, which are introduced by complexity, excessive state storage, signaling load and processing overhead. The Integrated Services model of current Internet relies on the complex Resource reservation Protocol (RSVP) [RFC2205] approach for signaling and bandwidth reservation. However, RSVP incurs additional processing and storage overheads on the routers [2]. For instance, the integration of RSVP and Multicast is inefficient in terms of signaling, since multicast tress are maintained by flooding operations [3]. Thus, the research community changed the focus to lightweight QoS control schemes. Motivated by that, several solutions have been proposed, such as Next Steps in Signalling (NSIS) [RFC4080], Resource Management in DiffServ (RMD) [4] and Boomerang [5], however the unicast-based scheme limits their scope in Future Internet. Moreover, in centralized approaches performance is shortcoming due to reduced flexibility and scalability properties.

In alternative to the mechanisms described above, over-provisioning has been proposed in the literature, which allows signaling reduction, being only expected re-provision congested over-reservations. Hence, the overall signaling load is decrease in comparison with the per-flow approach. In static over-provisioning approaches (e.g., Border Gateway Reservation Protocol - BGRP [6]), over-reservations are controlled based on fixed updating factors, such as by duplicating current configuration on-demand. Static over-provisioning inefficiently manages residual bandwidth, easily allowing potential session blocking even with available bandwidth (but assigned to other uncongested class). In contrast to BGRP, the Shared-segment Inter-domain Control Aggregation protocol (SICAP) [7] deploys a dynamic over-reservation control approach based on current network resource conditions. Besides more efficient in managing residual bandwidth than BGRP, SICAP overloads core network processing with additional functions (e.g., aggregations/de-aggregations), which should only be addressed to network border for scalability. Moreover, both SICAP and BGRP have no multicast support.

The literature also provides centralized over-provisioning proposals, such as the Dynamic Aggregation of Reservations for Internet Services (DARIS) [8] and Intra-Domain Resource Manager (IDRM) [9]. Both DARIS and IDRM specify a bandwidth broker responsible to control the dynamic re-sizing over-reservations when either their utilization target is exceeded (DARIS) or in periodic time-scales (IDRM). The main drawback of DARIS and IDRM resides in their total centralization, which results in scalability and reliability penalties. For instance, the BB can easily become the bottleneck of the system under heavily-loaded network conditions since re-sizing operations are high frequent invoked. Moreover, the timing capability of IDRM is inefficient, where a potential request can be denied under congestion conditions experienced in the interval between the last and the next re-sizing time.

The related work analysis pointed that current solution for the provisioning of network resources cannot fulfill the requirements of Future Internet. This way, MIRA is proposed to provide and efficient way for provisioning network resources.

III. MIRA Description

The Multi-service Resource Allocation (MIRA) proposal has as main goal allowing QoS-guaranteed session distribution over multi-party (unicast or multicast) connections. Therefore, MIRA supports mechanisms for the dynamic network resource provisioning in wired and wireless links. In MIRA suite, network resources are classified into QoS and connectivity to deal with the provisioning of bandwidth and communication path respectively. In order to provision QoS resources, MIRA implements a per-class bandwidth reservation mechanism, which is based on a session-driven approach (i.e., triggered whenever a session request is received). In what concerns connectivity resources, MIRA provides multi-party transport by controlling the creation of QoS-aware Single Source Multicast (SSM) [RFC3569] distribution trees in environments with routing asymmetries. Figure 1 depicts MIRA architecture, which is composed by one signaling protocol, two mechanisms and state tables. Moreover, internal and external interfaces are implemented to allow interactions between MIRA components, as well as expose them to existing solutions and standards respectively.

A. MIRA Components

The MIRA Protocol (MIRA-P) component provides integrated QoS and IP multicast signaling support compliant with the NSIS QoS-related specification [10]. MIRA-P follows a soft-state approach operating inside and between networks, where periodic signallings maintain activated resources, as well as collect current QoS capabilities (per-class bandwidth availability, propagation delay and packet loss) of bottleneck links to enforce admission control. MIRA-P minimizes the session setup cycle spent on processing and bandwidth consumption of per-flow related work by specifying only two signaling messages, as in the following. Whereas RESERVE messages indicates amount of QoS and connectivity resources for installation, releasing or maintenance, RESPONSE messages feedback requested operations. Moreover, RESPONSE messages are used to improve resilience, where re-routing is deployed upon detecting failures or handovers.

The Resource Provisioning (RP) component addresses to compute amount of QoS and connectivity resources for further enforcement in routers along communication paths. RP is triggered whenever a session request arrives at an ingress router, with information about: session/flow identifications; QoS requirements, in terms of bit-rate, Class of Service (CoS), and tolerance to delay, jitter and loss; and IP address of the destination (individual node or multicast address). In Base MIRA (B-MIRA) enabled networks, RP provisions each CoS with bandwidth reservations, being updated at each session establishment (adding) or finishing (releasing). Oppositely, RP over-provisions QoS and connectivity resources in the bootstrap of Advanced MIRA (A-MIRA) enabled systems, to allow session admissions without signaling exchanges.

Finally, the Resource Controller (RC) component implements local admission control (for bandwidth availability) and mechanisms to enforce both QoS and connectivity resources. Hence, RC supports interfaces to allow interactions with different network elements operating at layer 3 and/or layer 2 (wireless). For instance, RC deploys per-class resource reservation by correctly configuring packet scheduler behavior on queues of network interfaces. RC retrieves local network interfaces via unicast routing protocols, such as Open Shortest Path First (OSPF) [RFC2178], taking into account the destination IP address.

B. MIRA Agents

In MIRA suite, two major components running on edge routers (ingress-/egress-/access) and interior (core) routers are defined, namely MIRA Edge (MIRA-E) and MIRA-Core (MIRA-C) agents. The system scalability is increased by pushing the intelligence and complexity of resource control in the entire network to MIRA-E agents. Thus, MIRA-E agents implement full MIRA components and are statefull, keeping per-session and per-flow information, edge-to-edge per-class reservations, and the list of core-routers comprised in communication paths, as well as, the QoS capability of the bottleneck link. MIRA-C agents are with lightweight functions, implementing MIRA-P and RC components. Therefore, MIRA-C agents are reduced-state, keeping per-class reservations and the IP address of MIRA-E hosted at the ingress router that requested resource enforcement. In wireless networks, QoS Access Points (QAP) must host MIRA-C agent.

As MIRA mainly operates in the network layer, it is assumed the existence of mechanisms to control session announcement, subscribing and initiation. For instance, sessions can be announced by means of Session Description Protocol (SDP) [RFC4566], and accessed via hyperlinks published in SMS messages or web sites. The user interested in joining a multi-party session can deploy the subscription through Session Initiation Protocol (SIP) [RFC2543]. The SIP proxy at the MIRA-enabled network must be configured to re-direct SIP messages to the local ingress-router, both closest the refereed session source. Thus, MIRA-E at the ingress router closest to the session source derived in the SIP INVITE message can start the session setup towards the user which generated such message. Moreover, indications about session composition, identifications and QoS requirements must be provided.


The B-MIRA aims to control the provisioning of each CoS with an amount of bandwidth reservation so that aggregated flows can be delivered to experience their QoS requirements. The B-MIRA per-class reservation approach is driven by per-session request arrivals. It means that whenever required a session establishment, B-MIRA must be triggered at a MIRA-E agent hosted at an ingress router with information about each flow composing the session. Subsequently, B-MIRA inspects per-flow information, QoS requirements and destination IP to then start the session setup cycle. Whenever admission control fails (i.e., unavailable resources), the detecting node signals the requester MIRA-E agent, where all agents along the reverse path release the indicated amount of QoS and connectivity resources. Such approach envisions reducing session blocking by releasing unused resources as fast as possible. In comparison with per-flow related work, such approach improves bandwidth consumption and processing overhead of RSVP-A and RMD approaches. In RSVP-A, failure is detected by time out, and in RMD only egress routers can feedback, even when failure is detected by a core node.

In case of multicast sessions, the enforcement of connectivity resources in B-MIRA consists in allowing the creation of QoS-aware SSM trees. The default behavior of B-MIRA indicates the multicast routing protocol the QoS-capable path for further SSM tree state installation. To that, B-MIRA supplies the local Multicast Routing Information Base (MRIB) with the IP address of previous router network interface (since MRIB is receiver-driven). Thus, multicast routing messages are forced to follow the QoS-capable path, avoiding thus QoS violations commonly placed in environments with routing asymmetries [RFC3662]. Alternatively, B-MIRA can direct install SSM trees state during the session setup cycle, aiming at optimizing round trip times. Moreover, MIRA must allocate an IP multicast address for multi-party session distribution, which is done via standard solutions, such as Multicast Address Dynamic Client Allocation Protocol (MADCAP) [RFC2730].

A. B-MIRA Use-case

In order to clarify B-MIRA operations, the generic scenario provided in Figure 2 is considered, where a SIP-aware mobile device (SMD1) is interested in the multimedia session (S1) supplied by the source node linked to ingress router 1 (I1). User SMD1 is attached to the QoS Access Point 10 (QAP10), which is linked to the egress router 6 (E6). Assume that B-MIRA is responsible to install SSM tree state, and the SIP proxy is correctly configured to re-direct SIP messages to MIRA-E agents at local ingress routers.

Upon realizing the availability of session S1 by means of an SMS message, the user of laptop SMD1 clicks in the hyperlink accordingly. Therefore, SMD1 creates a SIP INVITE message that is propagated between SIP Proxies until reaching the MIRA-enabled network closest S1 source. I this use-case, we assume local SIP-proxy is implemented in a way that it is prepared to select the ingress router closest source node of S1 (I1 in the case), so that to re-direct the SIP INVITE accordingly. Hence, MIRA-E at I1 starts session setup by inspecting session-related information provided in the SIP INVITE message to map onto S1 QoS requirements (i.e., bit-rate, CoS, and IP destination). Firstly, RC uses the destination IP, derived from SIP INVITE message, to retrieve the local network interface that comprises the shortest path indicated in the unicast routing table. After that, RC composes a Traffic Specification (TSPEC) object filled with the QoS requirements of S1, and invokes RP with such information. Based on the TSPEC, RP compute the amount of QoS (bandwidth and CoS) and connectivity (IP multicast address allocation via MADCAP) resources for S1. After successful accomplishment, RP triggers back RC. Upon succeeding the admission control, RC installs both reservation and SSM tree state in the referred network interface, composes a RESERVE message correctly filled, and triggers MIRA-P to send it towards the destination IP. The RC component, at all agents visited by the RESERV message along the shortest path, enforces the QoS and connectivity resources in the same way as done in I1. The MIRA-E agent hosted at egress router E6 enforces QoS and connectivity at the network interface linked to the QAP10.

Before signaling MIRA-C at QAP10, MIRA-E at E6 must adapt current TSPEC information to an appropriated wireless class. This is allowed since MIRA-E at E6 interacts with MAC components in QAP10 at system bootstrap, and then periodically, to keep up-to-date wireless QoS capabilities. After mapping to the most suitable wireless class, MIRA-E signals MIRA-C at the QAP10 with a RESERVE message carrying the new TSPEC for QoS enforcement in the wireless channel associated to the destination SMD1. To that, MIRA-C must implement interfaces with components of wireless QoS framework. In Wi-Fi scenarios, RC interacts with the Hybrid Coordinator (HC) element to adjust the wireless scheduling accordingly [11]. A similar process is performed in WiMAX systems, where MIRA-C interacts with Connectivity Service Network (CSN) entity at the Access Service Network - Gateway (ASN-GW) to request the creation of QoS-aware service flows for the refereed session [12]. After accomplishing the session setup in the wireless network, MIRA-C at QAP10 feedbacks the MIRA-E at E6 with a RESPONSE (ok) message, which then forwards it to I1. After confirming the session setup, MIRA-E at I1 signals SMD1 with a SIP OK message to enable its interface for receiving S1 content. Other users interested in joining S1, same set of operations are deployed accordingly, where B-MIRA only installs new states in links without S1 activated.

V. Advanced MIRA

The per-flow signaling approach of B-MIRA fits well in small networks, due to the reduced session demands and frequency of times in which traffic requires more bandwidth than normally needed. However, large scale networks can experience an excessive signaling load, as well as state and processing overhead of per-flow B-MIRA approach. As a result, A-MIRA was specified to overcome the shortcomings described above by means of a dynamic QoS and connectivity resource over-provisioning strategy. Contrary to centralized approaches (e.g., DARIS and IDRM), A-MIRA envisions improve performance through operating in a distributed way. Moreover, scalability is achieved by taking all the complexity at ingress routers, leaving core routers simple, as in B-MIRA (as opposed in SICAP approach).

The main idea consists in bootstrapping both per-class over-reservations (QoS) and shortest communication paths (connectivity) to further allow their dynamic allocation without per-flow signallings. The bootstrapping mechanism requires each ingress router flooding the network so that each visiting router can initialize per-class over-reservations and collect all available communication paths (list of IP addresses of all nodes in each path associated with the QoS capabilities of their bottleneck links). As such scheme allows A-MIRA taking decisions without per-flow signallings, overall signaling load can be drastically reduced, as well as bandwidth, processing and memory overhead throughout a network. Next sessions describe how A-MIRA controls the over-provisioning and allocation of QoS and connectivity resources.

A. QoS Resource Over-provisioning

The QoS resource over-provisioning mechanism of A-MIRA aims at assigning surplus bandwidth (over-reservation) for each CoS at nodes throughout a network. However, in class-based networks with a variety of CoSs on links, the unpredictability of bandwidth utilization within different CoSs makes over-reservation approach very challenging, mainly in terms of waste of resources, session blocking probabilities and system performance. Such problem can be introduced by allocating too much resource to CoSs, which is generally done statically. A-MIRA distinguishes itself by supporting efficient techniques to control dynamically over-reservations, which include new functions for: (i) computing per-class over-reservations; (ii) efficient re-sizing of congested over-reservations; (iii) redistributing over-reserved, but unused, bandwidth (residual) to avoid wasting bandwidth.

Each CoS is assigned with both maximum and committed thresholds, so as to avoid class starvations and QoS violations respectively. In order to initialize per-class QoS resources, each MIRA agent visited during the flooding cycle (at system bootstrap) assigns local CoSs with an over-reservation, talking into account an factor (e.g., ½, ¼, etc) in function of the maximum reservation threshold, both assigned by network administrator. As the network is running, A-MIRA allows admission of multiple sessions without per-flow signaling, until detect that a session request cannot be admitted by a CoS. In such case, A-MIRA must compute new bandwidth over-reservation patterns to be installed for CoSs along the related communication path. In another words, A-MIRA attempts to increase the present over-reservation, of the congested CoS, taking into account: current utilization ratio; the bandwidth required by the session; and reservation thresholds. If succeeding, the congested CoS becomes over-provisioned again for future sessions, otherwise sessions are denied.

Approaches based on static reservation thresholds can place session blocking even when the required bandwidth is available in the network interface. A-MIRA efficiently beats such problem by implementing a dynamic resource management strategy to reduce the negative impact of residual bandwidth of remaining CoSs. Whenever an over-reservation re-sizing fails (maximum reservation threshold exceeding), A-MIRA re-adjusts the maximum reservation thresholds of all CoSs [15]. For this reason, A-MIRA attempts to increase the maximum reservation of a congested CoS based, reducing of others. Such computing takes into account residual bandwidth of remaining CoSs associated with their utilization ratio. Thus, A-MIRA improves session blocking under congestion conditions, and consequently, overall resource utilization.

B. Connectivity Resource Over-provisioning

As previously described in section V, available communications paths are discovered along the flooding operations at the system bootstrap. In more details, each router visited by RESERVE messages during the flooding cycle adds the IP address of the incoming interface (among other operations), and floods the message next. Egress routers finish the flooding cycle by sending a RESPONSE message towards the refereed ingress router with the entire path derived from previous RESERVE message. In order to ensure keeping shortest paths, A-MIRA controls flooding to avoid infinite looping, such as by stopping the cycle whenever and ingress router is visited by such message, as well as when the incoming interface is still filled. Moreover, a maximum number of routers can also be defined, as done based on TTL information. After finishing the flooding cycle, A-MIRA has information of a large number of unicast paths within the network. Afterwards, a combinatorial algorithm is triggered to create multicast trees from the refereed ingress router towards all associated egress router. Since a very large number of combinations can be reached, A-MIRA implements filters to keep shortest multicast trees, such as by eliminating paths with redundancies, loops and etc.

After finished the connectivity resources initialization, A-MIRA is enforced with an efficient view of the underlying topology. This way, A-MIRA is allowed to select the most suitable shortest path (unicast or multicast) for a requesting session, which takes into account QoS requirements, destinations and current resource conditions (of path candidates). Moreover, resilience support is improved with the automatic switch of paths (or trees) by ingress routers without signallings, benefiting by: reducing session setup times while taking session continuity over the time; preventing waste of resources, such as by avoiding sending packets to leaf nodes without active destinations; session quality degradations, due to re-routing caused by unpredictable link failures, new join/leave events or handovers; etc. The session transport along the best communication path is deployed by means of tunneling. Thus, a communication path can aggregate sessions mapped in different CoSs, but they must share same ingress and egress routers.

VI. Performance Evaluation

The performance evaluation of MIRA suite was carried out in the Network Simulator v2.29 (NS-2) with extensions to MIRA functionalities. The simulation model uses random network topology composed by 14 routers interconnected with links 100Mb/s and varying propagation delay. For class-based QoS support, we used DiffServ with WFQ scheduling extension, and following CoS configurations: one EF-alike (Premium); two AF-alike (Gold and Silver); and one best-effort. The maximum reservation threshold (MRth) of EF and AF CoSs are 20% of the link, and the committed reservation threshold is set to 50% of the MRth defined for each CoS. The multicast support of NS-2.29 (still not deploying multicast signaling support for tree creation) was extended with a MRIB structure, to allow MIRA enforcing QoS-aware multicast.

The methodology applied for MIRA performance evaluation encompasses two set experiments, each one using different amount of session requests, which have been placed following a Poisson distribution. The B-MIRA evaluation experiments examined the impact that it takes in relation to the current Internet resource reservation approach, RSVP. Furthermore, the A-MIRA evaluation experiments addressed to deeper inspect the benefits of A-MIRA over-provisioning mechanisms over B-MIRA per-flow signaling approach.

A. B-MIRA Evaluation

The first set of experiment analyzes the scalability properties taken by B-MIRA and RSVP in the simulation model. To that, four set of experiments were carried out, each one varying from 100 to 400 the number of sessions admitted during the entire simulation. A basic measure for protocol efficiency deals with signaling load. In addition to consume precious bandwidth, exceeding signaling load seriously damages routers performance in terms of CPU, memory and energy overhead. Figure 3 illustrates the signaling load results of B-MIRA and RSVP in the simulation model.

The simulation results show that number of signalling messages generated in MIRA Basic experiments is reduced in 82% when it is compared with RSVP experiments. The numerical results reveal that RSVP introduces 5.64 times more signallings. The reasons reside mainly in the two-way signaling approach and the flooding-based multicast state discovering. Such operations are the well-known reasons for the high RSVP complexity, introduced by the high number of events spent in its operations. Thus, we can clearly realize the signaling load reduction capabilities of B-MIRA over RSVP at all experiments.

In addition to the signalling load, state storage is an important measure to examine scaling capabilities, for the reason that the processing burden increases with the number of control information stored, which exponentially overloads state look-up operations. For instance, packet forwarding latency increases with the number of entries in routing tables. In what concerns resource reservation, the maintenance of large state tables is costly to interior nodes, which should be mostly dedicated to data plane functions instead of resource control.

In order to expose state storage, we measured all information kept by B-MIRA, operating to interact with multicast routing protocols (B-MIRA-d) or direct creating SSM trees (B-MIRA-o), and RSVP agents at ingress and core routers. In the case of B-MIRA, we noticed a marginal variation of reservation state in agents hosted at edge and core routers. The difference lies in multicast information, which is only required by B-MIRA-o. Therefore, the output information of the simulation results is simplified by presenting the state storage measurements of RSVP and B-MIRA at the ingress router (since MIRA-E is statefull), and its neighbour core router, because it is not detected variation in the amount of state kept by the remaining MIRA-C agents. The state storage load at the ingress is introduced in Figure 4 .

The graphical analysis of Figure 4 exposes the capabilities of MIRA in introducing a less amount of state at the ingress router in comparison with RSVP. The deeply view of the numerical results divulges that the state storage minimization of MIRA at the ingress router is of 67% (B-MIRA-d), and 64% (B-MIRA-o). At the core router, B-MIRA kept 0.3KB (B-MIRA-d) and 1.1KB (B-MIRA-o), while RSVP introduced 24.7KB (100), 49KB (200), 67.9KB (300) and 76KB (400). Hence, the state reduction capability of MIRA is of 99.3% (B-MIRA-d) and 97.6% (B-MIRA-o). The numerical results reveal that B-MIRA is the cost-effective solution in terms of state storage in comparison with RSVP. The reason for RSVP inefficiency deals with requiring a pair of state for reservation (per-class) and path (increasing with the number of visited routers) at all routers along the SSM tree. The difference of MIRA-E and MIRA-C resides in path state, which is stored only at the ingress routers.

B. A-MIRA Evaluation

The A-MIRA set of experiments has as main goal accurately attest the performance benefits of its over-provisioning mechanism against per-flow signaling basis of B-MIRA. To that, the simulation model is scaled up to 1,000 multicast sessions, with 333, 333 and 334 session requests for Premium Gold and Silver CoSs respectively. The sessions have a lifetime that varies from 20s (short-live) to 120s (long-live), and have a constant bit rate of 224Kb/s, to emulate a three flows scalable rate composition (32Kb/s, 64Kb/s and 128Kb/s). Thus, each session request means the setup of three flows with different requirements and characteristics. A-MIRA initializes QoS resources taking the init factor of ¼ (25%), and same maximum and committed thresholds of previous experiment. MARA is configured to filter shortest trees taking a maximum of 6 hops.

The first performance analysis consists in realizing the scalability properties of A-MIRA and B-MIRA in the simulation model. The overall amount of signallings generated by A-MIRA and B-MIRA along the simulation time (120s) is shown in Figure 5. We can clearly observe the network performance optimization taken by A-MIRA experiment. The integration of admission control and over-provisioning of QoS resources allows A-MIRA introducing signallings only to re-size over-reservations of CoSs noticed with not enough bandwidth to accommodate a session. In contrast, B-MIRA signals the network according session demands.

The numerical results of signaling load reveal that the bandwidth consumption of B-MIRA and A-MIRA signallings are of 0.0048% and 0.0013% respectively. Hence, A-MIRA allowed reducing in 88.1% the signaling load of B-MIRA in the simulation model. As B-MIRA introduces ~3.5 times more signaling load to establish sessions in the signaling model, the QoS over-provisioning mechanism demonstrates its efficiency by its cost-effectiveness signaling approach.

In addition to the bandwidth consumption improvements over B-MIRA, A-MIRA implements a new resource management strategy to optimize waste of resources by allocation residual bandwidth among available CoSs based on session demands and current resource conditions. The efficiency of A-MIRA resource management strategy must be studied, thus we analyze the session blocking behavior over the simulation time. The unpredictability of session demands is contrary to static thresholds approaches (i.e., B-MIRA), where potential sessions can be blocked even with presence of residual bandwidth in other classes. The overall session blocking is depicted in Figure 7.

Figure 7 exposes that B-MIRA denied 118 Silver-alike sessions, 111 Gold-alike sessions and 112 Premium-alike sessions. Moreover, in A-MIRA experiments the simulation model experienced 144 Silver-alike, 103 Gold-alike and 96 Premium-alike denials. The numerical results tell that the B-MIRA static threshold approach introduced homogeneous session denials, while optimizing in 0.005% the overall number of session denials. Besides introducing “more” session denials (2 sessions in the case), A-MIRA allowed establishing much more Premium-alike sessions than the remaining CoSs. Such behaviour is justified by the highest demand of Premium-alike sessions, thus placing firstly congestion experience and its granting with part of remaining classes' residual bandwidth. This means that A-MIRA resource management mechanism is able to favour potential CoSs, where sessions can be admitted even under hostile network conditions. Therefore, A-MIRA allows a more efficient allocation of bandwidth, where waste of resources is optimized by dynamically managing residual network resources.

Finally, we study the impact that A-MIRA and B-MIRA takes in the system performance by examining their connectivity state storage. In this sense, we measured the amount of multicast state introduced by both MIRA approaches at the ingress router, the intelligence point of A-MIRA-enabled networks. Figure 7 gives a picture of the multicast state storage behavior over the simulation time. The numerical results of the connectivity resource control analysis show that A-MIRA multicast state minimization rate, in comparison with the results collected in MIRA Basic experiments, is of 67.5% in the ingress router.

Consequently, we can evidently confirm that A-MIRA installs significantly less multicast state, as a result of filtering all the available paths to keep the shortest SSM trees. Contrary, B-MIRA approach faces legacy IP multicast behavior, where per-flow state is manipulated.

VII. Conclusion and Future Work



[1] D. Partain, A. Bergsten, M. Greis, G. Karagiannis, J. Manner, J. Murphy, P. Pan, V. Rexhepi, L. Westberg and H. Zheng, “NSIS QoS Signalling Requirements”, IETF Internet Draft, February 2002.

[2] A. Gani, L. Yang, N. Anuar, O. Zakaria, R. Taher, “Implementing lightweight reservation protocol for mobile network using hybrid schema”, In WSEAS Transactions on Computers, vol. 8 (3), pp. 417-428, 2009.

[3] X. Fu, and C. Kappler, “Towards RSVP Lite: Light-Weight RSVP for Generic Signaling”, In 17th Int. Conf. on Advanced Information Networks and Applications, Xi'an, China, March 2003.

[4] L. Westberg et al, “Resource MNanagement in DiffServ (RMD): A Functionality and Performance Behavior Overview”, Lecture Notes in Computer Science 2334, Springer Verlag, pp. 17-34, April 2002.

[5] G. Fehér, K. Németh, M. Maliosz, I. Cselényi, J. Bergkvist, D. Ahlard, T. Engborg, Boomerang--a simple protocol for resource reservation in IP networks, In Proc. of the IEEE Workshop on QoS Support for Real-time Internet Applications, Vancouver, Canada, June 1999

[6] P. Pan, E. Hahne, and H. Schulzrinne, "BGRP: A Tree-Based Aggregation Protocol for Inter-domain Reservations", In Communications and Networks Journal, Vol. 2, June 2000.

[7] R. Sofia,r R. Guerin, and P. Veiga, “SICAP, a Shared-segment Inter-domain Control Aggregation Protocol”, In Proc. of the Int. Conf. in High Performance Switching and Routing (HPSR), Turin, Italy, June 2003.

[8] R. Bless, “Dynamic Aggregation of Reservations for Internet Services”, in Proc. of 10th Conference on Telecommunication. Systems -Modelling and Analysis, vol. 1, pp. 26-38, Monterey - California, USA, Oct. 2002.

[9] H. Mantar, T. Okumus, J. Hwang, and S. Chapin, “A scalable intra-domain resource management architecture for DiffServ networks”, In Journal of High Speed Networks, vol.15 (2), pp. 185-205, January 2006.

[10] J. Manner, G. Karagiannis, and A. McDonald, "NSLP for Quality-of-Service Signaling", Work in Progress, October 2009, Available:

[11] X. Chen et al, “Supporting QoS in IEEE 802.11e Wireless LANS”, IEEE Transactions on Wireless Communications, Vol. 5, pp. 2217- 2227, August 2006.

[12] E. Cerqueira, A. Neto, M. Curado, P. Mendes, and E. Monteiro, “WIRA: An Approach to Resource Control in WiMAX Systems”, in Next Generation Internet Networks (NGI), Kraków, Poland, April 2008.