Research activities

From Cng1wiki

(Difference between revisions)
Jump to: navigation, search
(Service Guarantees for Individual Flows with Aggregate Scheduling)
(Traffic Engineering)
Line 44: Line 44:
* the design and evaluation of an Inter-AS path computation protocol, which negotiates a path (at the AS-level) to a remote destination, subject to bandwidth availability and QoS guarantees. The AS path thus computed can then be used to setup an inter-AS MPLS-TE tunnel.
* the design and evaluation of an Inter-AS path computation protocol, which negotiates a path (at the AS-level) to a remote destination, subject to bandwidth availability and QoS guarantees. The AS path thus computed can then be used to setup an inter-AS MPLS-TE tunnel.
The defined algorithms, protocols and modules have been implemented in a real pan-European research testbed.
The defined algorithms, protocols and modules have been implemented in a real pan-European research testbed.
Current research directions on this topic are:
* devising heuristics for path computation which blend several criteria (e.g., path lenght, delay, path diversity)
== Simulation Techniques ==
== Simulation Techniques ==

Revision as of 06:05, 12 July 2009

<analytics uacct="UA-1580095-2" ></analytics>

The CNG group is active in the following research fields.


Scheduling Algorithms for QoS Provisioning

When traffic flows pertaining to multiple traffic classes, each with different QoS requirements, coexist in a network, resource management needs to be flexible enough to efficiently provide each flow with a different type of service according to its class, at a feasible computational complexity. The research of the CNG group in this area includes design and performance evaluation of scheduling algorithms, both in wired and in wireless networks. Some recent research achievements in this area are:

  • Several scheduling algorithms have been developed for wired networks, therein including: the Timed Token Service Discipline, which manages rate-guaranteed and best-effort flows simultaneously, appying rules similar to those used to control medium access by the Timed Token Protocol; the Active List Queue Method (ALIQUEM), an innovative implementation of the Deficit Round Robin scheduling algorithm which improves latency and fairness while keeping O(1) per-packet complexity; the Eligibility-Based Round Robin (EBRR), which is suitable for scheduling in interconnection networks of parallel and distributed systems.
  • The design of scheduling algorithms aimed at providing IEEE 802.16 users with strict QoS guarantees. This involves exploiting at the MAC layer information of the physical layer, due to the inherent properties of the IEEE 802.16, with a cross-layer approach. Our solutions have been evaluated by means of extensive simulation studies in the case of fixed IEEE 802.16 networks. Activities in this area are carried out in the framework of two contracts: one with the Nokia Research Centre (Helsinki) and the other with Telecom Italia LAB (Torino).
  • The definition and evaluation of the Real Time HCCA (RTH) and Wireless Timed Token Protocol scheduling algorithms for the HCCA access function in 802.11e networks. The former applies the well-known EDF scheduling to the HCCA environment, thus proving more effective than the sample scheduler in serving real-time traffic such as VoIP. The latter, instead, takes into account both CBR and VBR traffic simultaneously, and it provides CBR traffic with a fixed rate, while at the same time allowing VBR traffic to share any leftover bandwdith in order to reduce the delay of occasional bursts.

Current research activities on scheduling algorithms are:

  • The definition of scheduling algorithms for UMTS High Speed Downlink/Uplink Packet Acces (HSDPA/HSUPA) that take into account the constraints inherent to the specific MAC layer and are specifically tailored to accommodate both real-time and best-effort traffic simultaneously. This activity is carried out in the framework of a contract with Telecom Italia LAB (Torino).
  • Investigating mobility support, which has been specified in the IEEE 802.16e amendment to the standard in 2005.
  • The investigation of the distributed mesh mode of IEEE 802.16 tailored to support a wireless mesh architecture, which provides extended coverage through multi-hop communications.

Service Guarantees for Individual Flows with Aggregate Scheduling

The future Internet is foreseen to support the provision of reliable real-time services on a wide scale. A per aggregate resource management is nowadays regarded as the solution for achieving such scalable service differentiation. Two noticeable examples are the Differentiated Services (DiffServ) architecture, where flows traversing a domain are aggregated in a small number of classes or Behavior Aggregates (BA), and QoS is provisioned on a per-aggregate basis at each node, and MPLS, where flows are aggregated into Forwarding Equivalence Classes. However, many questions regarding these architectures, both theoretical and implementation-related, are still unanswered. First of all, it is not clear whether it is possible (and, if so, how) to derive QoS guarantees for single flows by only enforcing a per aggregate resource management. Another important problem – directly related to this – is setting up aggregation criteria for such architectures. In a DiffServ or MPLS domain, the coordination of the QoS policies can be performed by a centralized entity, e.g. a Bandwidth Broker (BB), whose task is to carry out a careful admission control, and to set up the appropriate configuration of the domain’s edge and core routers. While doing so, the BB knows about all requests for capacity of QoS classes. Therefore, decisions regarding the aggregation criteria would easily translate to algorithms implemented on a BB. Some recent research achievements in this area are:

  • The proof that Network Calculus allows one to compute good per-flow delay bounds from per-aggregate resource provisioning, although at the price of expensive computations
  • A closed-form end-to-end delay bound expression for the delay of single flows in sink-tree aggregate scheduling networks. The bound was proved to be tight, i.e. actually attainable.
  • The definition of an optimal admission control algorithm for real-time traffic in sink-tree networks which is based on the above mentioned bound
  • A formal proof that the delay bounds computed through Network Calculus in a FIFO feed-forward network can sometimes be overrated, i.e. larger than the worst-case delay, and an assessment of their tightness.

Current research directions on this topic are:

  • Devising traffic aggregation criteria so as to optimize the resource provisioning while still providing QoS guarantees
  • Devising per-aggregate resource provisioning algorithms based on expected per-flow delay requirements
  • Defining and implementing a Bandwidth Broker architecture, and interfacing it to the (core and edge) routers.

Traffic Engineering

The activity in this field, carried out in the framework of the EuQoS European project, involves the design of algorithms for interdomain traffic engineering and resource optimization in IP networks with QoS support. Specific well-known technologies such as MPLS-TE and DiffServ-TE have been used, as well as new technologies that are being standardized by the IETF (the Path Computation Element - PCE - architecture). More specifically, the current activities in this direction are:

  • the definition of a Traffic Engineering module, that manages inter-domain routing configuration and resource provisioning in a pure IP Autonomous System.
  • the definition of EQ-links, i.e. interdomain MPLS tunnels, with a guaranteed QoS, that are automatically negotiated between remote endpoints by means of inter-AS negotiation.
  • the design and evaluation of an Inter-AS path computation protocol, which negotiates a path (at the AS-level) to a remote destination, subject to bandwidth availability and QoS guarantees. The AS path thus computed can then be used to setup an inter-AS MPLS-TE tunnel.

The defined algorithms, protocols and modules have been implemented in a real pan-European research testbed.

Current research directions on this topic are:

  • devising heuristics for path computation which blend several criteria (e.g., path lenght, delay, path diversity)

Simulation Techniques

Simulation is one of the main techniques used in our performance evaluation activity. We have developed extensions to the Network Simulator 2 (ns2), a popular open-source simulator used in research environments. The following packages have been released as contributed code to ns2:

  • ns2mesh80216: an extension to ns2 to simulate IEEE 802.16 Wireless Mesh Networks
  • ns2voip: an extension to ns2 to carry out reliable performance evaluation studies with Voice over IP (VoIP) traffic
  • ns2measure: an integrated framework for enabling effective data collection and statistical analysis with ns2
  • ns2hcca: an extension to ns2 to support IEEE 802.11e HCCA
  • MDRR: modules to simulate CISCO and Juniper's Modified Deficit Round Robin schedulers

Furthermore, we have defined a common simulation framework for ns2 to carry over simulation experiments involving heterogeneous network domains (i.e., different access networks, core IP, etc.).

Performance Evaluation of MAC Protocols for Wireless LANs

Some recent achievement in this area have been:

  • The evaluation of the ETSI WLAN standard HIPERLAN Type 1. The analysis was carried out using realistic traffic models (such as those mirroring the behavior of WWW applications. In order to carry out a stability analysis of the protocol, its statistical properties (related to the contention resolution algorithm) were analytically derived. This also proved the optimality of the parameter selection done in the standard.
  • Evaluation of the ETSI standard HIPERLAN Type 2 for broadband WLAN access.
  • Participation (on behalf of Telecom Italia) in the standardization committee of ETSI BRAN (Broadband Radio Access Network).
  • The evaluation of the High Speed Downlink Packet Access (HSDPA) for the UMTS network.
  • The evaluation of the performance of IEEE 802.16 in Point-to-Multipoint (PMP) mode, with the OFDMA physical layer. The analysis is carried out by means of simulation with realistic traffic models, so as to assess the effectiveness of the IEEE 802.16 to support multimedia applications that have stringent QoS requirements. We are currently devising cross-layer MAC/physical scheduling algorithms to optimize the system performance in terms, e.g., of bandwidth utilization and admission control. Furthermore, we will investigate advanced features of the IEEE 802.16, such as automatic repeat request (ARQ) and adaptive modulation and coding (AMC).
  • The evaluation of downlink frame allocation strategies in the IEEE 802.16 with the OFDMA physical layer. This study is carried out by means of numerical analysis so as to devise algorithms and guidelines to maximize the transmission efficiency from the base station to the subscriber stations, in both fixed and mobile environments.

Energy Management in Mobile Computing Systems

Energy management is a key ingredient in mobile computing systems since portable computers are typically battery powered. We have proposed several proxy-based energy-saving schemes suitable for infra-structured wireless networks (e.g., the mobile Internet). In particular, we have considered both application-dependent and application-independent strategies operating at the middleware layer. We evaluated the performance of our strategies in a real test-bed and we have found that our solution allows a considerable saving in energy (up to 80%) without a significant impact on the QoS perceived by the user. Currently, we are also investigating schemes for ad hoc networks.

Reliable Multicast in Distributed Mobile Systems

We have designed a reliable multicast protocol for distributed mobile systems that supports dynamic membership and neatly accommodates three increasingly strong delivery ordering guarantees: FIFO, Causal and Total. The system model assumed by the protocol is quite general and includes incomplete spatial coverage of the wireless network while no limitations are posed on the mobility pattern of group members. The protocol is not based on hand-off and was designed bearing in mind that MHs may have scarce resources. Simulation results have shown that it has good scalability properties. Furthermore, simulation revealed that the absence of hand-off does not introduce significant costs in terms of memory requirements and/or power consumption.

Transport Protocols for Ad Hoc Networks

Recent simulation and experimental studies have highlighted that the behaviour of the TCP protocol in multi-hop ad hoc networks is far from ideal. In mobile ad hoc networks the performance of the TCP protocol may be very much influenced by route failures caused by the mobility of nodes. Even in static networks TCP performance may be affected by the interactions between the TCP mechanisms and the MAC protocol mechanisms. Currently, we are defining an innovative transport protocol to be used as an alternative to the TCP protocol in ad hoc networks.

Performance Measurements of Ad Hoc Networks

There is extensive literature on the performance of the TCP protocol in ad hoc networks. However, almost all these studies are based on simulation. We have set up an experimental test-bed based on laptops running the Linux Mandrake 8.2 operating system and equipped with D-LinkAir DWL 650 IEEE 802.11b cards operating at the nominal bit rate of 1, 2, 5.5 and 11 Mbps. The experimental analysis revealed several aspects that are usually neglected in simulation analysis. The transmission ranges are in practice much shorter than usually assumed in simulation analysis, and are not constant but highly variable (even in the same session). Furthermore, the througput achieved by a TCP/UDP application is much lower than the nominal bandwidth (i.e., 11 Mbps) even in the case of a single source-destination session. Currently we are performing additional measurements on our testbed.

Wireless Sensor Networks

The activity in this field includes:

  • Sleep/Wakeup Scheduling in Sensor Networks
  • Sensor Networks with Mobile Elements (Data Mules)
  • Performance Measurements of Sensor Networks

Wireless Mesh Networks

Wireless Mesh Networks (WMN) have recently been devised to allow mobile users a ubiquitous, QoS-based access to IP networks, such as the Internet, or metropolitan networks, or corporate networks of private and public companies. Our activity in this area focuses on the backbone part of the WMN, for which the following specific issues are researched

  • Channel assignment algorithms, which allow the various frequency channels to be assigned to the radio interfaces so as to guarantee network connectivity and optimize its performance
  • Routing protocols, which have to be scalable and QoS-oriented
  • Algorithms for traffic scheduling, provisioning, aggregation and admission control, which enable QoS in the presence of multimedia traffic
  • MAC Protocols, which by leveraging on multi-radio access, can guarantee QoS to several traffic classes
  • Transport protocols
  • Power management algorithms
  • Models and techniques for performance analysis and simulation of the above algorithms and protocols
Personal tools