There are a lot of different projects that attempt to do some form of
ResourceBrokering for networks and grids. This page tries to give an overview of the different projects, along with a short summary.
UCLP stands for User Controlled
LightPaths. Developed by CA*Net, has 4 different implementations created by different universities. Work is underway to develop version 2.0.
DRAC
DRAC stands for Dynamic Resource Allocation Controller, is developed by Nortel Networks.
In this demonstration, an application program is executed using computing resources and optical paths, both of which are scheduled and reserved in advance. The Grid Resource Scheduler (GRS) calls on a Network Resource Management System (NRM) to request optical paths on a GMPLS controlled network, specifying the end nodes' locations and other requirements, such as bandwidth. NRM, in turn, checks the availability of the requested paths, reserves the paths, and sets up the paths when the scheduled time arrives. The goal of this project is to define a standard web service interface (GNS-WSI) between GRS and NRM, which is acceptable for both application service providers and commercial network operators, and which can be used as a tool for realizing emerging new commercial services.
G-Lambda is a tool based on GMPLS created by a cooperation of several Japanese research institutes. It was demonstrated at iGrid2005, where they showed a live demo that appeared to be very scripted. Whether this was because of limited functionality or because they felt it showed the best example is unknown.
It provides a webservices interface to submit reservations for clusternodes and at the same time a connection between these cluster, with bandwidth requirements. In the demo at iGrid2005 they demonstrated that they had several lightpaths, which could be reserved.
Disadvantages:
- One network management node for the whole network.
- Lightpaths are reserved in combination with nodes.
- It was at the time of the demo not possible to reserve a lightpath for a part of the duration of the node reservation (e.g. reserve nodes for a full day, but only have a lightpath during the last hour to gather results).
- It appeared to be tied to the use of specific optical crossconnects, unknown if this is really the case.
The ability to selectively forward traffic on a per-flow basis requires us to develop the capability to dynamically reconfigure forwarding of specific flows within our local production-use routers. Investigating how to do this, we have determined that suitable infrastructure to accomplish this is missing. We will develop that infrastructure. We refer to it as
LambdaStation. If one envisions the optical network paths provided by
UltraNet, National Lambda Rail, and other advanced optical-based research networks as high bandwidth data railways, then
LambdaStation would be the railroad terminal that regulates which flows at the local site are directed onto the high bandwidth data railways.
They also have some nice pictures showing the use of different paths.
http://www.lambdastation.org/path-switching.html∞
DRAGON-based demonstrations were presented both at iGrid2005 and SC|05.
An interesting starting point for more on this project is the recent press release:
http://biz.yahoo.com/prnews/051116/dcw064.html?.v=27∞
The following are some of the notes I took during the meeting with the DRAGON team while they were visiting Amsterdam.
The DRAGON network in the Washington D.C. area is an all optical network with WSS switches in four locations. MOVEZ is the manifacturer of these switches.
At the base of the GMPLS operation there two components:
- the VLSR, Virtual Label Switching Router;
- and the NARB, Network Aware Resource Broker.
The VLSRs perform the signaling and routing tasks, and control the underlying network equipment via SNMP commands. These are simple Linux PCs located in the domain, with a 1:1 correspondence with the network equipment. The NARB listens to the OSPF advertizements, it peers to the NARB in other domains and decides what to advertize based on preconfigured policies. There is one NARB per domain.
Other important points regarding the DRAGON operational model:
- the control plane and data plane are separate. The control plane runs over GRE tunnels over the IP network between the locations.
- the OSPF database has been extended to include resource scheduling and temporal information needed for the booking of resources and setup of paths.
Category
CategoryTechnologies
There are no comments on this page. [Add comment]