Computer graphics processing and selective visual display system – Display driving control circuitry – Controlling the condition of display elements
Reexamination Certificate
1997-02-24
2001-12-11
Huynh, Ba (Department: 2173)
Computer graphics processing and selective visual display system
Display driving control circuitry
Controlling the condition of display elements
C345S215000, C345S215000, C709S230000, C709S231000, C709S232000, C709S235000, C710S029000
Reexamination Certificate
active
06330008
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to apparatuses and methods for parallel computing and particularly to methods and apparatuses which allow for monitoring the performance of such parallel computing.
BACKGROUND OF THE INVENTION
Despite the rapid increase in the power of individual computer processors, there are many present and potential applications which could benefit from much greater computing power than can be provided by any individual present or foreseeable processor. The major approach to such greater computing power is to use parallel computers, that is, computers having more than one processor. Many different types of parallel computers have been designed, ranging from Symmetric Multi-Processing systems in which the each of the multiple processors and some amount of cache memory share main memory and all of the computer's other resources, to so-called shared-nothing systems where each processor has its own separate, often relatively large, main memory and, often, its own mass storage device, and the processors are only connected by computer network. The number of processors in current parallel computers vary from two to tens of thousands.
Parallel computers can provide a huge amount of raw computational power, as measured by all of the instructions per second which their multiple processors can execute. One of the major problem restricting the use of parallel computing has been the difficulty in programming and debugging parallel computing programs because of the complexity of their computation. Also the execution of large computations on parallel computers can often fail or be slowed drastically because of resource limitations effecting all or a part of such computations. In addition, parallel computations can be lengthy, particularly if they are not properly designed. For all these reasons, it is important for those designing and running parallel programs to be able to better understand the computation processes with which they are dealing.
A form of computation which has been previously used is record-based data flow programming. This form of computation causes a flow of records to pass through a stream of operators which remove or add records to the stream, modify the values in records. or create new records. Such computation can be performed on one processor or in parallel on a plurality of processors. Parallel Relational Data Base Systems (parallel “RDBMSs”) run programs which respond to a user query written in a data base query language such as SQL, and then automatically create a corresponding parallel data flow graph. In such systems the user cannot explicitly create the graph, nor can he create, even indirectly, any graph other than one created in response to a query in a data base language.
A new approach to programming parallel record-based data flow programming is disclosed in U.S. patent application Ser. No. 08/627,801, filed by Michael Beckerle et al. on Mar. 25, 1996, entitled “Apparatuses And Methods For Programming Parallel Computers” (hereinafter the “Beckerle et al. Application”. The Beckerle et al. Application is hereby incorporated into this application in its entirety. The rights in the Beckerle et al. Application are owned by Torrent Systems, Inc., the assignee of the present application.
This prior application discloses a system in which a user can explicitly define a data flow graph by connecting together graph objects including data sets and operators with datalinks. The operators have input and output ports at which they can receive and output records, respectively, over a datalink. Each such port has a defined schema which defines the name and type of fields from which or to which the operator is to read or write data. The schema can define transfer operators which designate that all field of a record are to be supplied from one input port to one or more output ports. Field adapters objects can be placed between a datalink and an operator to change the name or type of fields in the records supplied to or output by such operators.
The user is given the capability to define new parallelizable operators, including new parallel operators containing programming written by the user, new parallel operators each instance of which executes a standard sequential program, or new parallel operators using subgraphs defined from combinations of other, previously defined operators. The system automatically parallelizes the execution of the user defined graph.
It would also be helpful for those programming and running data flow graph computations could better understand the performance of such computations.
SUMMARY OF THE INVENTION
It is an object of the present invention to provide apparatuses and methods for helping those who program and/or run parallel programs to better understand the parallel execution of those programs.
It is an object of the present invention to provide apparatuses and methods for helping those who program and/or run dataflow programs to better understand the execution of those programs.
One aspect of the invention provides a performance monitoring tool which provides simultaneous visual display of information on the performance of each of a plurality of processes on each of a plurality of processors.
Another aspect of the invention provides a computer system which computes a data flow graph, and which has a performance monitor which shows a graphical representation of the graph's execution. The graphical representation includes varying, performance information about various parts of the graph shown in locations corresponding to those graph parts.
In a preferred embodiment of the invention, the computer system is capable of providing parallel computation of such data flow graphs in which the parallel operators of the graph are executed by multiple instances of those operators and the data flow associated with the input or output of such parallelized operator is partitioned between its parallel instances. In such cases the performance monitor is capable of providing a visual representation of the parallelized graph. In this visualization, each parallel operator instance and its associated partitioned data flow, or datalink, can be shown, and performance information can be displayed at or near the visual representation of such parallel objects in the graph.
The preferred embodiment of such a parallel performance monitor is capable of providing both 2D and 3D visualization of parallel graphs in which the user can selectively change the focus, zoom and viewpoint. This allows the user to quickly zoom in on parts of the graph in which he or she is currently interested and to use the relative expansion and contraction and relative changes in position caused by changes in viewpoint, zoom, and perspective to allocate screen area to desired parts of the graph. In 3D views the preferred embodiment groups parallel instances of the same operator together in a 2D array, both to provide a compact representation and to provide an easy to perceive representation of both the unparallel and parallel structure of the graph.
In the preferred embodiment of the invention information about the flow of data between operator instances can be represented visual attributes of the lines representing datalinks. This further compacts the representation of complex parallel graphs, since it allows both datalinks and information about their performance to be represented in one line. The preferred embodiment can represent the data rate of an individual datalink by both the density and velocity of broken line segments along its line. This use of broken, moving line segments makes it easier to distinguish separate datalinks which are very close to each other or overlap in a visualization, since it allows the user to see through the broken parts of such lines. The ability to distinguish between such closely placed datalink lines is further increased by the fact that the velocity and segmentation pattern of such datalink lines varies.
The user can select to color the moving broken line segments representing data rate with other information
Beckerle Michael J.
Ives Jeffrey D.
Kohn Daniel W.
Razdow Allen M.
Huynh Ba
Torrent Systems, Inc.
Wolf Greenfield & Sacks P.C.
LandOfFree
Apparatuses and methods for monitoring performance of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Apparatuses and methods for monitoring performance of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Apparatuses and methods for monitoring performance of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2562400