Grid Computing Power Meter
Original Publication Date: 2003-May-12
Included in the Prior Art Database: 2003-May-12
Disclosed is a mechanism for measuring the available unused resources in highly dynamic networks with special focus on the available but unused computing power. The knowledge of the unused but available computing power allows for a planned usage of those resources for additional work. This work could be a task in a dynamic grid computing network, for example.
Grid Computing Power Meter
One area of grid computing deals with the usage of existing computer resources in a network. Many companies use large computer networks to run their daily business with those computers. Very often these computers are powerful machines and their computing capacity is only partly used. The companies usually don't know what an enormous amount of unused power is available in their network. So far it is only possible to make a rough estimate of that power by combining the power of each computer, its average workload and its average availability.
In highly dynamic networks, in which the number and types of the connected computers change permanently, the results of these rough estimates can be completely wrong. An example for a highly dynamic network is the internet or large intranets, because the computers can be connected to or disconnected from the network at any time. Even the hardware or operating system or the setup of a computer can change while it is not connected to the network.
A real measuring of the capacity of such highly dynamic networks would be a great improvement of the current situation and a necessary precondition for acommercial usage of such resources that can be planned.
Currently available technology
Today, grid systems are mainly built of dedicated resources which are administered and controlled by a central grid management component. In such systems the emphasis is put on the workload management, that is the determination of the current workload of each machine and the distribution of work to each machine based on its current workload. A typical example can be found in the product 'Platform Clusterware'. To decide which computer should work on a specific task one can usually define trigger rules. And usually the actual computing power delivered from a machine for the specific work is measured to be able to do some kind of accounting. See also the white paper 'Grid Service Accounting Extensions' for more details on that subject. It is important to notice that only the used resources for a specific work unit are measured and only after the work is done. An example of that kind of architecture is shown in figure 1.
In such a 'classical' grid system one knows in advance the participating resources and one can control them from a central point. It is therefore not really necessary to measure the available capacity of that system. It can be deduced from the capacity of each machine.
New mechanism necessary
In highly dynamic networks such as large intranets, which are usually not planned as part of a grid system, the situation is completely different. See figure 2. Beside the normal daily work for which the machines are used, additional tasks could be done using the idle time of the processor or other resources.
The principal steps to do some extra work in such a system would be: A client machine requests a work unit from a server (pull principle)
The server sends a work unit to the...