Collaborative Steering and Control: Agile transport protocols are needed to control remote computations. Typically a computation is monitored remotely, perhaps by visualizing a partial projection of the parameter space, and steered into regions of interest by interactively specifying the parameters. It is very important that the steering operations be supported on a robust channel to place the computation in an appropriate parameter region. Note that an inadequate control channel can result in undershoot or overshoot problems, wasting valuable computational resources, particularly on supercomputers. Also, undue delays in control messages to a waiting computation on a supercomputer could result in the "flops on the floor" phenomenon. The control problem is more acute in the remote control of experimental devices, where delays in control commands can result in severe damage. In an extreme case, high frequency components in jitter can result in resonance, which could lead to a complete loss of control. Furthermore, when the steering or control operations are to be performed by multiple users at geographically dispersed locations, the control channels must be suitably coordinated. Except for very simple steering and control operations, TCP on IP networks does not provide the desired stability levels. The approach based on dedicated channels together with associated transport methods must be investigated for this class of capabilities.
On-Demand Channel Allocation: Provisioning of on-demand dedicated channels or bandwidth pipes requires allocation policies and implementations that are absent in packet switched IP networks. Requests for dedicated channels will be sent by the applications to allocation servers, which maintain the "state" of the network. Once the request is grated and accepted, implementation servers will setup the channels, maintain them for the allocated duration, and then tear them down. Such a capability does not exist over IP networks and must be developed for this class of DOE large-scale applications. Note that the allocation servers must be capable of implementing higher level policies for granting the requests as well as scheduling the channels by maintaining the state of available bandwidth levels of various network links. In addition, suitable routing and switching hardware and software must be in place to enable on-demand setup, maintenance and tear down of the various channels. In particular, it is important to be able to allocate the channels in groups and at various bandwidth resolutions, for example, a high bandwidth data channel together with a low bandwidth control channel.
Previous Next Table
of Content for report: Network Provisioning Home