Advanced Network Middleware
To develop a truly advanced Internet, some basic architectural issues must be resolved, especially with regard to matching precisely the requirements of advanced applications to the resources provided by the network. These interlinking tasks, processes and services must be accomplished through a mid-level set of technologies and capabilities. One common for this mid-level set of technologies is, appropriately, "middleware." Advanced network middleware provides capabilities have the potential to make networks significantly more reliable, adaptive, manageable, scalable, customizable, and intelligent than they are today.
A broad range of "middleware services" is required to provide for a persistent, high performance, reliable, high capacity network that can be rapidly scaled and readily managed. These services include many new types of network processes, systems, and technologies, including those that provide for access control, allow for the reservation of required network resources and ensure guarantees that network performance will match the resources requested by the application.
For example, such middleware provides sophisticated capabilities that allow networks to allocate, dynamically and more intelligently (precisely meeting requirement needs) various resources to the multiple applications utilizing them simultaneously. Advanced middleware can dynamically adjust to multiple changes in application resource requirements while network resources are also dynamically changing. These capabilities are especially important for large-scale high performance applications, such as those that are computationally and data intensive and that aggressively use network resources. When such capabilities are integrated into large scale distributed systems, they can be part of a metasystem that can control additional types of resources, including multi-terabyte storage systems, distributed data repositories, scientific instrumentation, and teraflop computational clusters. Much of the leading edge research in this area is being conducted by the community involved in developing terascale systems. Many of these concepts are being developed today through standardization forums such as the Global Grid Forum, in particular its Grid High Performance Networking Research Group. (www.ggf.org)
General Internet Architecture
Developing large-scale complex systems requires an architectural framework, especially to ensure optimal performance and interoperability with existing standards. iCAIR undertakes its research and development projects within the context of guidelines developed by a number of standards organizations, primarily the Internet Engineering Task Force (IETF, ietf.org). iCAIR also adheres to Internet related standards developed by other organizations including the GGF, W3C, IEEE, and ISO. iCAIR also participates in developing standards with these organizations.
Several key research themes in these areas are network service layer extension and integration, virtualization, and segmentation. Network services are usually described in terms of network layer, in accordance with the International Standard Organization's Open System Interconnect (ISO/OSI) model. To date, the Internet has been based on routed-packet, Layer 3 services. Increasingly, new methods are being created to complement those services with others at other levels. The majority of iCAIR research projects are focused on layers 1-4. Traditionally, these layers have been distinct separate services, which could not be addresses directly by processes external to the network. iCAIR is creating new methods to enable the integration and coordination of these layers, including through signaling from external processes.
Advanced Middleware Architecture
In cooperation with its research partners, iCAIR has established several projects focused on developing new types of architecture for network middleware. The delivery of optimal network capabilities that fully support advanced applications still remains a major challenge, in part because the full power of the network is underutilized because of traditional network architecture and technology, which was designed for analog not digital communications. The transition to digital communications is allowing for a parallel migration from a fairly primitive network middleware to extremely sophisticated network middleware. However, basic problems remain complex. A general lack of a common understanding of the requirements for middleware architecture has impeded the development of a consensus on the definition of middleware and its components. Detailed technical specifications are still in initial stages of formulation. The magnitude of this task is significant, and the technical challenges will be overcome only through cooperative efforts by the best networking experts within the research, education, and industry communities. However, progress is being made today in many of these areas.
Early Middleware Standards Activities
On December 3-4, 1998 An National Science Foundation-funded workshop on this important topic was held at iCAIR to begin the process of formulating a consensus, i.e. to begin the development of an architectural framework for advanced Internet middleware. This workshop also began an attempt to accelerate the process of enhancing middleware capabilities made available to users and applications through Grids. The workshop brought together middleware experts from throughout the US. (www.it.northwestern.edu/middleware/index.html) This workshop led to the development of an NSF report and to the first RFC on Middleware--2768 (www.ietf.org). The initiatives began at this workshop continue today through many forums, including through the GGF (www.ggf.org) and through the efforts of Globus (www.globus.org).
The Globus initiative began as part of the I-WAY project in 1995, led by the MCS Division of Argonne National Laboratory. The Globus Forum is developing an integrated set of basic Grid services, termed "Globus Services" - based on a Globus Toolkit. The toolkit is a collection of software components that provide basic services such as security, resource location, resource management, communication, interfacing and linking utilities, as well as related object libraries, development systems, and compilers. Globus components are available as an underlying component Grid resource that can be used by Grid applications.
iCAIR worked with the AT&T research labs for several years on various projects related to its experimental middleware system for networks, Geoplex, which was first implemented as a prototype in the labs at Northwestern in early 1998. This project was completed in the summer of 2000.
Network Policy Services
A key component of network middleware services is one that enables access to resources through policy decisions. As capabilities are developed to match the requirements of advanced applications to the resources provided by the network, it is crucial to have a capability for determining the identification of the requesting entity (e.g., individual, application, group, organization, etc), for authenticating that identity and for determining whether that entity is authorized for the requested resources and services. Related to these capabilities are provisions to audit the network resources that are used. Links between specific types and priorities of applications and resource guarantees need to be defined and implemented through some set of policy servers, which, in turn, must be linked to flow control and other adjustment mechanisms. iCAIR has been researching and experimenting with many different types of policy mechanisms including those based on the IETF AAA standard, particularly the implementation developed by the University of Amsterdam.
Control Plane Architecture
Traditionally, network architecture includes consideration of categories of processes for interacting with network resources. Three key categories include the management plane, control plane, and data plane, and, generally some components of each of these three has been implemented at each network layer. A current architectural trend is the consideration and development of a "universal control plane," which can be used at all layers. iCAIR is involved in multiple research projects related to developing next generation control planes, especially for optical networks based on dynamic lightpath provisioning. For the last five years, many of these efforts have involved service overlays to IP-based control methods, such as using the IETF Generalized Multiprotocol Lambda Switching (GMPLS) standard. (Ref: Optical Networking section of this web site.)
With funding from DARPA, iCAIR and Nortel Advanced Technology division established a project that developed and demonstrated a novel architecture for data intensive services supported by distributed infrastructure based on optical networks with inherent dynamic lightpath provisioning. This type of architecture ("DWDM-RAM") can be used by multiple data intensive application communities. The architecture was designed for optimized, fault-resilient, dynamic management of services supporting large, n-way replicated immutable data objects over a large-scale MAN/LAN optical network testbed (OMNInet), interconnecting Grid computational clusters. The DWDM-RAM architecture is innovative in several respects. For example, it closely integrates application-level data resources with DWDM optical resources, resulting in high-performance and highly scalable data migration and management, for example, through optimal integrated data discovery and transfer operations. Like other OMNInet services innovations, this approach combines data services and dynamic wavelength-switched layer. Using this technique, high volumes of distributed data can be transferred in parallel using resources such as discovered light paths, data repository locations, and local and remote I/O capacity, replication sites, etc. Also, the DWDM-RAM architecture provides for a migration path, as a supplemental to services based on traditional performance-limited, limited layer 3 routing protocols.
A prototype implementation of the DWDM-RAM architecture integrated high volume high performance data services with dynamically switched wavelength optical networking, and demonstrated : 1) content-addressed data retrieval, 2) a meshed DWDM switched network capable to establish an end-to-end lightpath in seconds, 3) an signaling function between the application and the DWDM network, to allow the integration of application metadata and network metadata, 4) discovery functions operating on the combined application and network meta-data, 5) large scale data-transfer facilities exploiting circuit-switched networks, and 6) out-of-band functions for adaptive placement of data replicas. The architecture can be expandable to include additional functionality, for example, to include enhanced file systems semantics.
Layer 2 Services and Technologies
iCAIR has been researching and experimenting with issues relate to the virtualization, segmentation, and direct control of Layer 2 network services, particularly those based on IEEE Ethernet standards, such as vLANs. In part, these projects are extensions of iCAIR's research and development projects that are creating new optical control plane architecture. In addition, iCAIR has been exploring SONET related technologies, such as G.709 and the integration of Ethernet and SONET to provide for integrated services.
Layer 3 Services and Technologies
Today, the Internet is primarily a layer 3 packet routed network, which treats all packets the same, providing a "best effort" service. Recently, much research has been conducted to experiment with new architecture that can be used to differentiate among various classes of layer 3 services. When iCAIR was first established, its research projects in this area were guided by Brian Carpenter, who is now the Director of the Internet Engineering Task Force (IETF), the primary standards body for the Internet. Previously, he was a member of the IETF Internet Architecture Board (IAB), which he chaired from 1995 until March 2000 and he was the co-Chair of the IETF Differentiated Services (DiffSServ) working group. When serving as the Program Director for Internet Standards & Technology at IBM, he was on assignment at iCAIR and taught courses in the Computer Science Department at Northwestern. He was also active in IPv6 activities (isocbriefing01.pdf), and worked with the World Wide Web Consortium. iCAIR established multiple projects related to these technologies, especially DiffServ, including many that experimented with integrating them with specific applications.
DiffServ has been a particularly important architectural focus of a number of major iCAIR research projects. The current Internet provides "best-effort," undifferentiated services, which, essentially, provides the same common infrastructure to all applications and users. New techniquese being developed to provide different types of service guarantees depending on a variety of a variety of parameters. The IETF has established a differentiated services, DiffServ, architectural standards effort (previously, DSARCH, now RFC 2475). iCAIR has undertaken a wide range of DiffServ related research, experiments, and development of prototype technologies.
To provide appropriate support for applications over high performance networks, it is necessary to address many key QoS related issues, and to proving concepts through testbed networks. A number of iCAIR experiments involved creating DiffServ testbeds to experiment with these concepts. In 1999, the Center led an international consortium that established a trans-continental Diffserv testbed. Each participant implemented a QoS service regionally. However, to provide for end-to-end high quality service across national and international infrastructures, it was necessary to develop, test, and provide for early deployment of processes and functions for a range of integrated network services, including management policy options that would allow for differentiation of service categories and for distributed governance and resource allocation mechanisms across multiple domains.
A special consideration was ensuring that DiffServ can be implemented across multiple domains and systems. As part of this process, it was necessary to arrive at a common understanding of optimal DiffServ design and implementation among many possible technologies and options related to selecting components and parameters. This process starts with selecting and provisioning for specific service categories and determining service category parameters and attributes.
iCAIR was a research participant in the EMERGE experimental testbed research project, which was one of the first in the world to attempt to closely integrate edge process control of core network resources, using innovative network middleware. This Department of Energy (DoE) funded initiative, which was led by UIC (www.evl.uic.edu), designed and developed a Science Grid testbed - the ESnet/MREN Regional Grid Experimental NGI Testbed. The project experimented with new concepts of middleware on a large scale, DiffServ (Differentiated Services) enabled network. Testbed extensions included an experimental DiffServ IPv6 testbed, international experimentation, and QoS-enabled host systems with special TCP stacks.
A key goal of EMERGE was established to design, deploy and test DiffServ on an IP/ATM Regional GigaPoP network (MREN, ref: www.mren.org) inter-operating with ESnet for applications in combustion, climate and high-energy physics.The initiative also worked with the DoE, NSF and NASA supported groups to deploy "Grid Services", and document and evaluate the performance of emerging NGI technologies, such as multi-domain authentication and resource brokering services, adaptive network APIs, high performance transport protocols and IETF architectures such as such as Differentiated Network Services (DiffServ) on real large-scale scientific applications.
In part, this project was undertaken to achieve and demonstrate DiffServ over MREN as a representative model for DoE/University connectivity, to support DoE-specific Next-Generation Internet (NGI) applications and attempt to motivate inter-operability across other GigaPoPs as well as the Abilene network and ESnet. This project provided DiffServ capabilities as a part of advanced Grid Services, and implemented capabilities for: access control (identification, authorization, authentication, and resource utilization); directory services via the Lightweight Directory Access Protocol (LDAP) ; delivery of multimedia data through sequence numbering, time stamping, and contents identification using Real-Time Transport Protocol (RTP); and Real-Time Control Protocol (RTCP) to control RTP data transfers; and network management including instrumentation. This project has concentrated on facilitating advanced data flows, extremely large computed datasets, ultra-high resolution rendered imagery, and real-time unicast/multicast digital video (including implementations of the 1394 (Firewire) protocol encapsulated within IP). iCAIR testbed extensions included an experimental DiffServ IPv6 testbed, international experimentation, and QoS-enabled host systems with special TCP stacks.
Middleware Services and Applications
iCAIR has participated in a number of projects that have linked middleware services techniques with specific applications, including many large scale e-science applications, datamining, visualization, digital media - distribution, digital video transmission, and video conferencing - and large scale interactive gaming. One project was the design and implementation of a prototype national digital video network based, in part, on advanced middleware technologies to provide high-quality digital video services through a national digital video network, enabled by technologies. iCAIR has also established a number of middleware-related projects that will provide for network performance and measurements, including those related to specific applications.
IPv6 and 6Bone
iCAIR has participated in multiple IPv6 research projects, including many using new middleware techniques. IPv6 is a particularly important successor to IPv4 because it extends address space, which is a requirement to deploy billions of IP deveices. The MREN community was one of the first to implement IPv6 through 6Bone, a research implementation of IPv6 based on tunnels. iCAIR participated in a project to allow for provisioning native IPv6, instead of tunneled IPv6. Another project created the world's first IETF DIffServ QoS IPv6 testbed. iCAIR also investigated MBGP peering between a specified IPv6 routers and peer IPv6 routers. See isocbriefing01.pdf.
Multicast and SGM
Many iCAIR digital media research projects have integrated network middleware and new types of multicast techniques. Multicast is an important technique for optimizing bandwidth for a wide range of applications, especially digital multimedia, by allowing for the total number of data streams to be managed more efficiently. For example, by using multicast techniques, it is possible to direct streams only to nodes where they are required and can reduce the total number of streams required in network segments by avoiding duplication. While today's multicast schemes are scaleable in the sense that they can support very large multicast groups, these schemes have problems when a network needs to support a very large number of distinct multicast groups. The Small Group Multicast (SGM) project centers on a new approach to multicast that complements today's multicast schemes. The URL for the internet-draft on Small Group Multicast (SGM) can be found at http://www.ietf.org/internet-drafts/draft-boivie-sgm-00.txt. A slightly expanded version of the draft has also been published in the May/June 2000 issue of Internet Computing. Following is a URL for the online version: http://computer.org/internet/v4n3/w3onwire-a.htm. For further information, see Small Group Multicast.
About iCAIR Mission & Projects Announcements
Corporate & Research Partners Advanced Internet Initiatives Other URLs Home