A Grid Proxy Architecture for Network Resources (GPAN) is proposed to allow Grid applications to access resources shared in communication network domains.
GPAN bridges Grid services serving user applications and network services controlling network devices through its proxy functions such as resource data and management proxies. Working with Grid resource index and broker services, GPAN employs distributed network service peers (NSP) in network domains to discover, negotiate and allocate network resources such as bandwidth for Grid applications. An elected master NSP is the unique Grid node that runs GPAN and represents the whole network to share network resources to Grids without Grid involvement of network devices. GPAN provides the Grid Proxy service (GPS) to interface with Grid services and applications, and the Grid Delegation service (GDS) to interface with network services to utilize network resources. Resource-based XML messaging is employed for the GPAN proxy communication.
Grid networks are emerging as the “killer application” of next-generation networks through the building of overlay networks upon existing network infrastructures for the purposes of network-wide computing and data collaboration.
In a Grid network, which is also known as a virtual organization, Grid nodes are distributed widely in the physical networks and share their available hardware and software resources such as CPU, storage, data mining, and visualization centers. The resource sharing is actually provided by the Grid services running on Grid nodes. Those Grid services form the Grid overlay over the Grid nodes as they function under the same Grid rule.
According to the available services and resources, a Grid user distributes portions of his/her application task in the Grid nodes and puts those services and resources together to complete the user task with very high performance. Benefits of this technology include preventing resource wasting in the network and saving the user capital expenditure on equipment. Typical use cases of Grids include data-intensive and computation-intensive business tasks such as enterprise concurrent product design, utility computing, large-scale data synchronization, and many high-performance computations including nuclear energy simulations and long-term global weather forecasting.
Distributed computing in general, and grid computing in particular, is desirable because complex and resource-intensive computing tasks can be accomplished without purchasing and maintaining costly super-computers. Rather, relatively small portions of the overall computing task can be distributed among multiple computers and devices of relatively modest capability. It will be appreciated that the grid network capability may also be more easily scalable than high performance devices and super-computers. Further, there is a potential business model for the rental of grid services.
The dispersed Grid nodes are connected by network infrastructure such as the Internet. Grid nodes include computers, data servers and network devices. Resource sharing of network devices in a Grid is critical because it provides the interconnectivity of Grid nodes to form a pipeline of resource supply in the Grid. In other words, interconnectivity enables reliable sharing of resources such as computing power and data services.
One problem hindering the advance of Grid technology is that the requisite resource sharing of network nodes is not entirely supported by traditional networks. First, Grid nodes generally require a piece of Grid software installed on each Grid node. While this installation is relatively easy on a computer, it can be impractical on network devices such as routers and switches. Without installation of the Grid software, a network device cannot purposefully make its resource available to the Grid, and a Grid node cannot access a network device through the same Grid rule.
Another problem hindering the advance of Grid technology is that network devices are typically located in autonomous network domains, and controlled by network management facilities such as network managers and services. The network management facilities form an actual overlay network which is the network service overlay and gives the API for network control functions. In other words, the network service overlay consists of network services that are running on network devices. Thus, in a Grid network, there are two overlay networks: the network service overlay and the Grid service overlay. But there is a gap between these two overlays. The gap is that they are not compatible because network services and Grid services use different service infrastructures. Moreover, due to the modest capability of CPU and memory on network devices, network services are generally implemented in lightweight network software while Grid services are generally implemented in heavyweight software and thus require much more CPU power and storage.
It would therefore be desirable to have a technique for causing the two overlays to work together. In other words, network nodes should be capable of acting as Grid nodes in order to provide its resources such as network links and bandwidth in a Grid.
The present invention overcomes the above-mentioned and other drawbacks by providing a Grid-based proxy mechanism to allow Grid services and applications to access and utilize resources available in the physical network domains.
In accordance with the invention, a Grid Proxy Architecture for Network Resources (“GPAN) bridges a grid services overlay that includes of Grid services to serve user applications, with a network services overlay that includes network services to control the network devices. The GPAN proxy gives the Grid services in the Grid Service overlay the potential to access the network services in the Network Service overlay in order to control the networks for resource use. With GPAN, network nodes do not need to install and run a piece of Grid service software in order to provide their resources to the Grids. Network nodes are represented in a whole by the GPAN proxy and joined the Grids by a unique Grid node which runs the actual proxy software/system.
In one embodiment of the invention, GPAN works with the network service overlay to gather network resource information and provide that information to the Grid resource index service within the Grid network. GPAN may also facilitate brokerage and reservation of network resources upon the request of a Grid resource broker service within the Grid network by translating application requests from the Grid services to network operations of the network services.
In one embodiment the invention employs network service peers/agents (“NSPs”) distributed in the network domains in order to obtain network resource information and allocate network resources for the Grids. Each NSP may include an individual network domain, including various sub-nets, and may be operated by a different service provider. Of the NSPs, only the master NSP executes the GPAN proxy in a Grid network. The master NSP is usually selected from an NSP which is close to the location where run Grid resource services such as index, broker and scheduler. A master NSP may be elected based on how the Grid is organized and how easily the Grid accesses the network resources. Under the GPAN proxy, the master NSP sends network instructions to each NSP for resource operations such as information collection and allocation. Further, the master NSP provides Grid-based proxy APIs of resource operations to Grid services and applications. Consequently, the node running the master NSP is the unique Grid node representing the whole networks. The Grid services and applications talk to this master NSP node for network resource supply. They use the GPAN APIs which are provided in the form of Grid services to access the network resources. It will therefore be apparent that no actual network nodes need to execute native grid protocols.
In one embodiment of the invention GPAN provides at least two proxy functions, resource data proxy and resource management proxy. The resource data proxy function is to provide resource discovery, status (create/destroy/use), availability and service policy for the Grid network. The resource management proxy function provides resource scheduling, new/allocate/reallocate/release in the Grid network. GPAN sends resource information to the global Grid resource services such as index, through the resource data proxy. GPAN receives resource allocation requests of Grid apps from a Grid resource broker or scheduler service through the resource management proxy. Each proxy function may implement different proxy mechanisms. For example, the resource data proxy implements a network information provider (“NIP”) which is a Grid-based software to facilitate collection of network information and provide updates and reports to the Grid resource index service. The resource management proxy function implements a resource allocation and management provider (“RAMP”) which is also a Grid-based software to receive resource allocation requests from the Grid broker/meta-scheduler services and schedule resource operations in the communications network.
In one embodiment of the invention GPAN includes two main components: a Grid Proxy service (“GPS”) and a Grid Delegation service (“GDS”). GPS interfaces the GPAN proxy with the Grid overlay of Grid services to meet applications. GDS interfaces the proxy with the Network Service overlay of network services to utilize network resources. Resource-based XML messaging may be utilized for communication between GPS and GDS.
By implementing at least some of the features described above, GPAN advantageously causes the communications network to appear as grid resources from the perspective of the grid network. For example, the resource availability data is periodically gathered from each network service peer and provided to an index of the grid network in a format that is compliant with the protocols utilized by the index. Further, GPAN implements an API that is compliant with the signaling utilized by the grid manager in order to facilitate resource brokerage and reservation. Following receipt of a request from the Grid application, and identification of requisite resource from the index, a broker/meta-scheduler of the grid network signals to the grid resources and GPAN in order to reserve and utilize the resources. The broker/meta-scheduler signals for resource reservation to both the grid resources and GPAN in the same, grid-standard compliant manner. GPAN implements an API which is operative to translate commands from the broker/meta-scheduler into a format that may be utilized by the communications network nodes. Consequently, two-way communications between the network resources and grid manager are possible, thereby enabling operations such as bandwidth allocation, negotiation, agreement, information feedback, and status tracking Once the appropriate communications network and grid network resources are made available for the application, portions of the overall task are distributed to the grid resource services either directly by the broker/meta-scheduler or via GPAN.