End to End Service Provisioning
Hassan Hassan
UREC-CNRS
4 place de jussieu, 75252 Paris cedex 5 France
Catherine Grenet
UREC-CNRS
4 place de jussieu, 75252 Paris cedex 5 France
Bernard Rapacchi
UREC-CNRS
4 place de jussieu, 75252 Paris cedex 5 France
Abstract
Transparent access to network resources, high transfer data rates between distant points, real time sharing of scientific data are a new category of services needed by the research community. This new category of services involves different actors of network access at the local and the national level. Although technical solutions do exist to answer these new needs, the heterogeneous nature of access and backbone networks puts the administrative and organizational aspects at the top of implementation obstacles stack. In this paper, we give an overview of end to end service provisioning in order to initiate a discussion about the importance of this category of services. Our goal is to start a cooperation between different network actors and prepare the inclusion of end to end service provisioning in the network service portfolio offered to the research and education community.
Keywords
End to End, Provisioning, Service, Bandwidth, On demand.
E-science applications evolve rapidly and consume more network resources than ever. Today it is common that one application generates network flows measured in Gbps during fixed periods of time. Often these flows transit between a pre-defined set of nodes, and imply strict constraints on quality of service.
The needs of such applications can not be met with ‘best effort’ routed networks, and require ‘traffic engineered’ point-to-point circuits. Although point-to-point circuits are already included in the service portfolios of most national research and education networks (NRENs), the real bottleneck resides in technical and organizational factors at the level of metropolitan, campus and local infrastructures. Indeed, the cooperation between NRENs, metropolitan/regional, campus and local networks is a prior step to deliver successfully end-to-end services.
In this paper we present a state of art of end to end service provisioning in european and international NRENs. Our goal is to initiate a discussion about end to end service provisioning for the research and education community in France. This would help in preparing for the rapid evolution of network services with the end to end paradigm. The work initiated in this paper could be the starting point for a national workgroup handling end to end service provisioning issues for the french research and education community.
The paper is organized as follows. In the next section, the end to end service provisioning problem is stated with a brief introduction to some applications requiring end to end service provisioning. In section 3, we present a technical overview of solutions used to deliver end to end services. Next in section 4 we present an overview of end to end service provisioning in France. And in section 5 we present the major guidelines for work to be accomplished in order to make end to end service provisioning more accessible for the french research and education community.
The rapid development of e-science applications revealed a new branch of applications with specific needs. These applications can be classified in two categories:
Heavy computing applications: in these applications, very complex algorithms need to be executed repeatedly. Calculation times may become prohibitive even on powerful stations. The computation needs of such applications require the cooperation of a high number computing nodes.
Huge data transfer applications: in these applications, the algorithm may be complex or simple but multiple input data streams are needed for its execution. The data streams could be provided by very distant sites and require very high bit rates in order to be transfered in a reasonable time.
Grid computing came to answer the needs of heavy computing applications. International cooperation in grid computing like EGEE [1] resulted in powerful set of tools allowing researchers to execute their applications on a large network of computing nodes.
On the other hand, heavy data transfer applications require more than a distributed platform for the execution of algorithms. Indeed, they require a flexible network infrastructure allowing the transfer of huge data streams between well defined, distant set of nodes. Unfortunately, the routed model used by IP-shared networks is unable to meet these needs. In fact, the IP routing model can not offer the expected quality of service as there is no control on the routing path. As a consequence predefined paths, as end to end circuits, reappeared strongly as an alternative to achieve high bit rate data transfer between distant end points. However, there are multiple techniques to establish end to end circuits over the existing network infrastructure, and the main problem is how to make these technologies transparent to users (researchers). This problem was remarkably introduced by the Internet2 [2] dynamic circuit network project:
“A key concept in cyberinfrastructure vision is that network capacity should be available as a schedulable, on-demand resource. The idea is that researchers, scientists, artists, or faculty can tap into deep bandwidth resources whenever and wherever they need it”
The problem of end to end service provisioning can be then defined as ”providing simple, easy to use tools making network resources appear as a schedulable, on-demand service for users”.
In the following paragraph, we shall take a look about some application examples requiring end to end service provisioning capabilities.
The Large Hadron Collider (LHC) [3], the new particle accelerator at CERN in Switzerland, provides a massive facility for physicists around the world to investigate the origins of the universe. This important research project will produce huge data volumes measured in petabytes (million gigabytes) annually. These huge amounts of data should be studied and analyzed by researchers all around the world. Each researcher is expected to download or transmit terabytes of data. These repetitive data transfers will be done over a fixed time window several times per month.
Achieving such transfers over a fixed period of time requires dedicated circuits to guarantee that no other application traffic can interfere with the data transfers and analysis. Therefore, it is reasonable that these circuits should be provided to researchers participating in this project all over the world. Establishing such circuits over different networks is big challenge that was made possible within the LHC project.
Very-Long-Baseline Interferometry (VLBI) [4] involves high-resolution imaging of distant radio sources in the universe. Many radio telescopes located in different places around the world provide these images. In fact, signals received from telescopes should be carefully cross-correlated. The goal is to obtain accurate measurements of planet’s motion in space. This would allow detailed studies of distant objects in the universe.
The particularity of VLBI applications is that huge streams of data need to be transferred to achieve this cross-correlation. But moreover, this transfer should be allowed in “real time” among distant telescopes and laboratories. Something which is impossible to do without dedicated high bit rate circuits between involved locations.
The Laser Interferometer Gravitational Wave Observatory (LIGO) [5] project aims to detect cosmic gravitational waves and to develop gravitational-wave observations as an astronomical tool.
Research carried out by the LIGO Scientific Collaboration requires occasional access to several terabytes of data from other remotely-located LIGO participants. As a consequence, at all levels campus, regional and national backbone, network engineers are challenged by the presence of individual applications that have very high peak utilization.
In contrast with shared IP-based networks, a dynamic circuit reservation system offers greater control over dedicated network resources and allow setting up circuits in durations lasting from a few hours to a few weeks, to provide researchers with a flexible, cost-effective use of the resources.
The above mentioned applications are no exception today. Many applications in different fields: ecology, biology, robotics, medicine, … involve high quality sensors capable of producing several Gbps streams to be shared between applications in very distant locations and in real time. Those applications can not all benefit today from high networking capabilities such as dedicated circuits. And researchers in some fields have a biased vision of network as an obstacle rather than an efficient tool for their work. The future of e-science applications is tightly related to the development of easy to use tools making network resources available for research projects transparently.
In this section we present an overview of major projects dealing with end to end service provisioning. Our goal is to give an idea about different techniques and different visions concerning end to end service provisioning. But first we will review the different switching techniques used in networks today.
The following table resumes the two major kinds of switching techniques versus networking modes.
|
Switching/Networking |
Connectionless |
Connection oriented |
|
Packet switched |
e.g. Ethernet networks |
e.g. MPLS networks |
|
Circuit switched |
Not possible |
e.g. SONET networks |
Packet switching is the major switching technique used in the Internet today, with the connectionless networking mode. This mode was behind the great success of Internet. Connectionless packet switching mode allows flexible use of network based on a hope by hope behavior across the network.
However, the connectionless mode has too many limitations and many protocols like TCP/IP introduced a connection oriented mode at the application level to enhance flow transport across the Internet. Unfortunately, with these protocols there were no warranty that packets traveling from one source to one destination will follow the same path, making it very difficult to ensure any of quality of service for delivered packets.
In order to overcome these difficulties, a connection oriented mode using labels was introduced with Multi-protocol label switching protocol (MPLS). Packets in a MPLS domain are marked at the entry point of the domain with a predefined label, and then switched across the domain based on this label rather than being routed using the destination IP address. A label allows defining a virtual path between the ingress (input router) and egress (output router) for all packets marked with this specific label. Using different labels, different paths can be defined as virtual circuits. Controlling the path followed by packets made it possible to control the quality of service of delivered packets across the whole path. The connection oriented mode provides high quality virtual circuits from source to destination, with guaranteed service level agreement.
The introduction of virtual circuits with MPLS, was a turning point for the circuit switched mode to reappear on the Internet. In fact, new applications are very strict in their requirements and the circuit switched mode was an appropriate model to leverage quality of service offered by data networks. Besides the emergence of optical networks using wavelength multiplexing lead naturally to optical circuit switching. Optical networks technology is very promising for the new generation of e-science applications involving real time huge data streams transfer between distant points.
However, wether circuit switching is achieved by MPLS or by optical networks it requires a control over circuit establishment procedures across heterogenous domains. Particularly, the source and destination points may belong to different domains using two different techniques for establishing circuits (MPLS or optical). Switch controllers need to implement heterogeneous signaling protocols, to handle setup/release requests for circuits, increasing the complexity of circuit establishment process.
Different projects lead by National Research and Education Networks (NRENs) tried to handle this problem by introducing intelligent tools based on middleware capable of driving circuit establishment across multi-domain networks. These tools range from simple manual establishment of a circuit between end points to completely automated establishment of circuits by applications using web services. The underlying network infrastructure may use heterogenous techniques where optical and MPLS circuits can be used between end points. In next sections we will give a brief overview of some of these projects.
The Automated Bandwidth Allocation across Heterogeneous Networks (AutoBAHN) [6] is a research project of GÉANT network. It aims at providing a dynamic circuit establishment facility over global research and education network infrastructures in Europe. The AutoBAHN system has been designed to allocate network bandwidth to users/applications in the form of dynamic circuits.
The fact that not all NRENs in Europe deploy the same type of technologies makes it very difficult to provide a unique method to establish dynamic end to end circuits. Instead there should be different ways of providing dynamic circuits. For this reason the AutoBAHN project has been designed to support:
Layer 2 switched circuits.
Layer 1 switched circuits.
Practically when an end-point in NREN 1 needs to establish a circuit with another end-point in NREN 2, via the GÉANT2 network, there is an important work of coordination of service requests across domains. In order to achieve this coordination, instances of the AutoBAHN system must be deployed in each involved domain in the circuit establishment process. It is the duty of each network domain to define its own policies and its own quality parameters that govern the use of its networking resources. These requirements are expressed through the AuotBAHN system. A primitive view of AutoBAHN system architecture is depicted on Figure 1
Figure 1: AutoBahn System Architecture (courtesy of GEANT network)
An AutoBahn prototype was developed and deployed over some european NRENs. However, many problems still need to be resolved in order to make the reservation system practical from a user point of view. The development of AutoBahn will be continued within GEANT3, making of AutoBahn a serious candidate for an end-to-end service provisioning tool to users connected through european NRENs
The Internet2 Dynamic Circuit Network (DCN) [7] is an optical circuit network that provides dedicated bandwidth for research applications. The dedicated bandwidth is provided by creating short-term circuits between end-points. The created circuits may have different life durations, and they can last from minutes to days. The goal is to offer to users ad-hoc point-to-point circuits across the Internet2 infrastructure. The process of setting up and tearing down the circuits is completely automated. This is performed by using a well developed control plane. The control plane consists of:
IDC (Inter Domain Controller) – the inter-domain controller accepts requests and coordinates circuit requests with other domains
DC (Domain Controller) – The domain controller controls circuits within its own domain.
The ambitious goal of DCN to make network resources available as schedulable circuits for researchers, requires the deployment of inter-domain controllers IDCs in all crossed domains including partner NRENs. In order to make this works, the DCN project collaborates with partners in Europe and other countries in order to provide a complete software suite based on standard protocols for circuits set up and tear down. GEANT is partner of the DCN project.
Figure 2: DCN Overview (courtesy of Internet2 network)
Many other projects were conducted in the united states to resolve the automated bandwidth reservation problem. However, the targeted applications are not always the same. While DCN gathered projects answering the needs of specialized e-science applications that generally has few number of users with important data transfer rates (> 10Gbps). Other projects tried to offer solutions to educational e-science applications with important number of end users and less important data transfer rates (around 1Gbps). For example medical applications involving high quality video imaging.
GLIF, the Global Lambda Integrated Facility [8], is an international organization promoting lambdas networking based on a completely optical infrastructure. Lambdas are dedicated optical wavelengths. The goal is to provide a high efficient infrastructure suitable for data-intensive scientific research. The GLIF initiative supports the development of middle wares facilitating the deployment of lambda networking across heterogeneous networks.
All lambdas used by the GLIF network are provided by GLIF participants. Those lambdas are interconnected through optical exchange points called GOLEs (GLIF Open Lightpath Exchanges). All used lambdas must terminate at an exchange points (GOLE).
The main task of optical exchange points GOLEs is to perform lightpath switching. A lightpath is simply a virtual circuit established over lambdas. Once established, the optical virtual circuit is used to connect end-points in the network.
Although the GLIF initiative is very promising, its applications are still very limited. This is because of its very expensive operation fees. Actually maintaining and operating an optical exchange point involves important investments, that most NRENs can not afford today.
There are many other projects dealing with dynamic circuit reservation for research applications like the NetherLight project of SURFnet [9], Phosphorus [10], ... Most of these projects cooperate with GEANT, Internet2, or the GLIF initiative to make end to end service provisioning possible across heterogeneous networks. All these project prove the importance of end to end service provisioning in the NREN service protofolio.
The french research and education network RENATER participated in the GEANT2 AutoBAHN project. Besides RENATER provides static end to end circuits upon request [11]. However, few e-science applications of french laboratories expressed their needs for such service making it difficult to evaluate the need for advanced circuit reservation tools.
The need to implement end to end service provisioning solutions is naturally driven by e-science applications requirements. Unless there is a strong need for end to end circuits, making this service automatic is useless. The lack of strong demand can be explained by several factors:
The network infrastructure is heterogeneous involving different partners: campus, metropolitan/regional and national networks. The heterogeneous nature of these networks makes it difficult to set up end to end circuits without special arrangements. The absence of harmonization between different partners makes the deployment of end to end service provisioning solutions complicated.
There is a lack of communication about available services provided by national and metropolitan networks. And the final user (researcher) has no access to this information in most cases. An important effort of communication about end to end service provisioning is needed in order to make potential projects express their needs.
Big projects with sufficient financial support have already their own dedicated backbones as their needs are very important (e.g. IN2P3 with LHC). Accessing end to end service provisioning appears today like a reserved class of service for well funded projects. Fortunately, the evolution of this service in other countries proves the opposite. Ad hoc end to end circuits may be used for small and large applications, and it largely participates in accelerating research in many fields.
End to end service provisioning is not very well developed in the research and education community in France today, and many actions should be taken in order to make this service transparent for e-science applications.
Enhancing end to end service provisioning offer for the research community in France requires several actions on the local and the national level:
A wide communication and information campaign about end to end service provisioning possibilities should be lead in the french research community. The goal is to raise scientific research needs in laboratories. Analyzing the needs should allow a better understanding of the way users want to access the service. A user perspective is rarely the same as the provider's one.
Whatever the technical solution of choice to provide this service, it should result of a common agreement between different actors: local, campus, metropolitan/regional and national network actors. The involvement of the national network actor RENATER in european projects gives it a better understanding of the future trends in end to end service provisioning. However, regional/metropolitan and campus networks need to be informed about those needs in order to include them in future plans of development of their networks. As a consequence it is important to organize a national event gathering all network actors for the research and education community in order to exchange this kind of information.
A proof of concept prototype may give an insight of real administrative, technical and organizational aspects of the problem. The selection of pilot sites with facilities to access end to end services could help in implementing such prototype.
A national workgroup is needed for the coordination of these actions. End to end service provisioning concerns mostly distant laboratories connected via different network actors. As a consequence it can not be accomplished without the cooperation of all concerned actors.
End to end service provisioning is a promising brand new category of services. It should make it possible for researchers to transfer data in real time at high bit rates using dedicated circuits between distant end points. Dedicated circuits could be reserved in advance for defined periods of time. Many applications could take benefit of this new category of services, allowing new ways of collaboration between researchers.
However, end to end service provisioning raises complicated problems to resolve. The heterogeneous nature of access and backbone networks could only be overcome by an active collaboration between different network actors. Besides, the proposed tools should hide all the complexity of circuit establishment and tear down between distant end points from users, allowing them to access this service whatever the network technology used by their network infrastructure.
Different projects and initiatives handling end to end service provisioning in other countries should make it easier to promote this new category of services in France. But its success is tightly related to the cooperation between different actors providing network access to the french research and education community.
EGEE: Enabling Grid for E-sciencE http://www.eu-egee.org/
U.S. advanced networking consortium http://www.internet2.edu/
Large Hadron Collider http://public.web.cern.ch/Public/fr/LHC/LHC-fr.html.
Very-Long-Baseline Interferometry, European VLBI, JIVE project http://www.jive.nl/
Laser Interferometer Gravitational Wave Observatory http://www.ligo.caltech.edu/
Automated Bandwidth Allocation across Heterogeneous Networks http://www.geant2.net/server/show/ConWebDoc.2544
Internet2 Dynamic Circuit Network http://www.internet2.edu/network/dc/
Global Lambda Integrated Facility http://www.glif.is/
NetherLight http://www.surfnet.nl/nl/Thema/netherlight/Pages/Default.aspx
Phosphorus http://www.ist-phosphorus.eu/
Circuits Renater http://www.renater.fr/spip.php?article576#connex