Parallel computer
First Claim
1. In a parallel processor including a control station coupled to a plurality of processing nodes, the control station serving to partition a processing task and distribute partitioned subtasks thereof to the processing nodes for processing, the control station further serving to collect processed results from said nodes and use said results to produce a final output to said processing task, an improvement comprising an asymmetrical network coupling each of the processing nodes to the control station, said network providing a first bandwidth link for coupling data from the control station to the processing nodes, and a second bandwidth link for coupling data from the processing nodes back to the control station, said first bandwidth being at least twice the second bandwidth, wherein the network resources are efficiently allocated to handle processing tasks that can be partitioned into subtasks which have more input data than output data.
1 Assignment
0 Petitions
Accused Products
Abstract
A massively parallel computer system employs hundreds or thousands of otherwise-idle microprocessors in set-top boxes distributed throughout a cable television system. Coupling between these processing nodes is effected by the cable television distribution network, which provides an asymmetrical transmission speed characteristic that is well suited for many parallel applications. The system is tolerant of interruptions due to preemption of set-top boxes for television viewing.
-
Citations
37 Claims
- 1. In a parallel processor including a control station coupled to a plurality of processing nodes, the control station serving to partition a processing task and distribute partitioned subtasks thereof to the processing nodes for processing, the control station further serving to collect processed results from said nodes and use said results to produce a final output to said processing task, an improvement comprising an asymmetrical network coupling each of the processing nodes to the control station, said network providing a first bandwidth link for coupling data from the control station to the processing nodes, and a second bandwidth link for coupling data from the processing nodes back to the control station, said first bandwidth being at least twice the second bandwidth, wherein the network resources are efficiently allocated to handle processing tasks that can be partitioned into subtasks which have more input data than output data.
- 10. In a parallel processor including a control station coupled to a plurality of remote processing nodes, the control station serving to partition a processing task and distribute partitioned subtasks thereof to the processing nodes for processing, the control station further serving to collect processed results from said nodes and use said results to produce a final output to said processing task, an improvement wherein each of the processing nodes is switchable between primary and secondary modes of operation, in the secondary mode the node being available to process a subtask distributed thereto by the control station, in the primary mode the node being preempted from subtask processing by local usage of the node for a task unrelated to said processing task, said local usage not being invoked by the control station but being scheduled by a human user, and wherein each node includes means for automatically indicating to the control station if it is in said secondary mode of operation.
- 16. In a parallel processor including a control station coupled to a plurality of processing nodes, the control station serving to partition a processing task and distribute partitioned subtasks thereof to the processing nodes for processing, the control station further serving to collect processed results from said nodes and use said results to produce a final output to said processing task, an improvement comprising means responsive to human intervention at each of said nodes for preempting a processing subtask at said node that is in-process so as to perform a different processing job unrelated to said processing task, and wherein the control system includes means for redundantly distributing a plurality of said subtasks to at least two processing nodes so as to reduce delays in completing the processing task due to such preemptions.
-
20. In a multi-processor system employing a central control processor linked to a plurality of distributed processors, the central processor serving to partition a processing task and distribute partitioned subtasks thereof to the distributed processors for processing, the control station further serving to collect processed results from said nodes and to use said results to produce a final output to said processing task, an improvement wherein:
-
each of said distributed processors is linked to the central control processor, but not with others of said distributed processors; the system includes means for providing subtask data as well as control signals from the central control processor to each distributed processor; the central control processor comprises circuitry different than each of the distributed processors; and each link provides a first bandwidth path for coupling data from the central control processor to a distributed processor corresponding thereto, and provides a second bandwidth path for coupling data from said distributed processor back to the central control processor, the first bandwidth being at least twice the second. - View Dependent Claims (21, 22, 23)
-
-
24. A remotely distributed processing system including:
-
a central facility; a plurality of remotely distributed processing units, each unit being switchable between at least first and second modes, in the first mode the unit being unavailable to process a subtask assigned thereto by the central facility, in the second mode the unit being preempted from subtask processing by local usage of said unit, said local usage not being invoked by the central facility, each unit indicating its mode to the central facility; a network coupling each processing unit to the central facility; the central facility including means for partitioning a processing task into a plurality of subtasks, distributing said subtasks to remote processing units indicated as being in the first mode, and receiving subtask results back therefrom; and completion means for completing processing of said processing task notwithstanding interruption of at least one subtask due to switching of a processing unit executing said subtask from the first mode to the second mode before execution of said subtask is completed. - View Dependent Claims (25, 26, 27, 28, 29, 30, 31, 32, 33)
-
- 34. A remotely distributed processing system including a central facility and a plurality of remotely distributed processing units, each of the processing units being switchable between first and second modes, in the first mode the processing unit being available to process a subtask assigned thereto by the central facility, in the second mode the processing unit being preempted from subtask processing by local usage of the unit, said local usage not being invoked by the central facility, the central facility including means for partitioning a processing task into a plurality of subtasks, distributing said subtasks to processing units in the first mode, and receiving subtask results back therefrom, the central facility including a database containing data relating to historical patterns of local usage for said processing units, wherein said subtasks can be allocated among said processing units in accordance with said historical usage patterns.
Specification