System And Method For Accessing And Using A Supercomputer

• US 20080082933A1
• Filed: 08/31/2007
• Published: 04/03/2008
• Est. Priority Date: 09/01/2006
• Status: Active Grant

• Current Assignee: Massively Parallel Technologies, Inc
• Original Assignee: Massively Parallel Technologies, Inc
• Inventors: Geske, Thomas, Howard, Kevin, Lupo, James, Robertson, Nick
• Granted Patent: US 8,108,512 B2
• Time in Patent Office: Days
• US Class Current: 715/771
• CPC Class Codes: G06F 9/5072 Grid computing

Abstract

Systems and methods form and control a supercomputer based upon a parallel processing architecture such as a Howard cascade. A graphical user interface allows a user to interact with one or more virtual power centers of the supercomputer facility. A plurality of processing nodes self-organize into one or more virtual power centers. The processing nodes utilize overlapped input and output for improved communication.

Claims

1. A graphical user interface for interacting with a supercomputer facility, comprising:

a main panel for providing login for a user to an existing user account;

a status panel for displaying status information of a job;

an input panel for specifying an input dataset for the job;

an output panel for displaying an output dataset for the job; and

a menu from which options may be selected by the user;

such that the graphical user interface interacts with at least one transaction server and at least one virtual power center to process the job.

2. The graphical user interface of claim 1, wherein the status panel displays a chart illustrating a relationship between job time, cost and number of processors allocated to the job, and a slider bar for requesting the number of processors to use for the job.

3. The graphical user interface of claim 2, wherein the chart indicates estimated performance based upon availability of processors for the job.

4. The graphical user interface of claim 1, wherein the main panel indicates current and projected costs for the user.

5. A method for determining the optimum node count for processing a job within a parallel computing system, comprising:

determining a best value cascade size based upon the maximum number of compute nodes where parallel processing efficiency remains above 90 percent;

determining a best performance cascade size based upon the maximum number of compute nodes where parallel processing efficiency remains above a predetermined limit, the predetermined limit being less than 90 percent;

determining a number of compute nodes for processing the job based upon the best value and best performance;

estimating a processing duration of the job based upon the number of compute nodes and a computational load of the job; and

estimating a cost for the job based upon the number of compute nodes and the estimated processing duration.

6. The method of claim 5, further comprising:

determining an actual cost of the job during processing based upon the processing time and the number of compute nodes used; and

stopping processing of the job if a user specified maximum cost is reached by the job.

7. The method of claim 5, further comprising modifying the number of compute nodes allocated to the job while the job is running based upon input from a user.

8. A self-organizing network of processing nodes for supercomputer utilization, comprising:

a plurality of processing nodes, each having one or more communication channels, wherein each processing node receives indifferent code to operate as an agent, a first of the processing nodes receiving a list of other agent addresses within the network;

at least one switch supporting multiple point-wise full-duplex communication between the agents;

wherein the first processing node operates as a list node and each other agent operates as a state machine that interacts with the list node and other agents to determine a role for the agent within the supercomputer.

9. The self-organizing network of claim 8, wherein the determined role within the super computer for each processing node is one of:

a home node, a compute node, a diagnostic node, and a spare node, and wherein the list node is determined to be a diagnostic node that also defines parameters for the supercomputer comprising one or more of a maximum association size and a system quiescence time.

10. The self-organizing network of claim 9, wherein the agents interact to form at least one association of agents with a size equal to the maximum association size, and wherein a first agent in each association having the maximum association size is determined a home node and other agents within the association are determined compute nodes, and wherein each agent in an association having less than the maximum association size is determined a spare node.

11. The self-organizing network of claim 10, wherein upon receipt of a job start request at one home node, the home node creates, from available home nodes and compute nodes within the self-organizing network, one of a Howard cascade, a Howard-Lupo manifold, and a Howard-Lupo Hyper-manifold, based upon the job start request that defines the number of compute nodes required to process the job.

12. A method for forming a parallel processing system based upon emergent behavior of a plurality of processing nodes, comprising:

initializing each of the processing nodes with identical functionality, each node operating as an agent based upon a state machine;

interacting amongst the agents to form one or more associations of agents based upon defined parameters of the parallel processing system, the defined parameters specifying a maximum association size; and

determining a role for each agent based upon the formed associations, the roles comprising one of a home node, a compute node, a spare node, and a diagnostic node.

13. The method of claim 12, further comprising:

storing a list of agents on a first of the processing nodes, the list defining addresses of agents of the other processing nodes; and

accessing the list from unassociated agents to identify other agents to contact to form the associations.

14. The method of claim 13, further comprising:

determining role of diagnostic node for the first processing node;

determining role of home node for each agent forming each of the associations having the maximum association size;

determining role of compute node for each of the agents not forming each of the associations having the maximum association size; and

determining role of spare node for agents of associations having less than the maximum association size.

15. A parallel processing system with overlapped communication, comprising:

a plurality of processing nodes, each processing node having an input communication channel and an output communication channel, each input communication channel having an associated input thread and each output communication channel having an associated output thread, the input and output threads operating to concurrently receive and transmit data on the input and output communication channels and cooperating, when the received data is to be sent to another processing node, to transmit data received via the input communication channel on the output communication channel with a delay T′

; and

at least one switch connected to the input and output communication channels of each of the processing nodes, the at least one switch being configurable to transfer data from the output channel of a first of the processing nodes to the input channel of a second of the processing nodes;

wherein the at least one switch is configured upon setup of the parallel processing system to provide communication between the processing nodes based upon a topology of the parallel processing system, the parallel processing system broadcasting data to each processing node with a minimum delay.

16. The method of claim 15, wherein the input channel and the output channel are part of a full duplex communication channel within the processing node.

17. A method for propagating data between interconnected nodes of a parallel processing system, comprising:

transmitting the data from a first node, via an output channel of the first node, to at least one second node, via an input channel of the second node, the transmission of the data having a duration of T;

receiving the data from the first node at the second node, the data being stored at least temporarily within the second node;

transmitting the data from the second node, via an output channel of the second node, to at least one third node, via an input channel of the third node, concurrently with receiving the data from the first node, the delay between receiving data at the second node and receiving data at the third node being T′

, where T′

is less than T and the time to receive the data at the third node is T′

+T.

18. The method of claim 17, further comprising transmitting the data from the third node, via an output channel of the third node, to at least one fourth node, via an input channel of the fourth node, concurrently with receiving the data from the second node, the delay between receiving data at the third node and receiving data at the fourth node being T″

, where T″

is less than T and the time to receive the data at the fourth node is T′

+T″

+T.

19. The method of claim 18, wherein T′

equals T″

.

20. The method of claim 19, wherein the maximum number of fourth nodes is determined by the total number of output channels on all of the third nodes, and the time to receive the data by all the fourth nodes is 2T′

+T, and wherein the maximum number of second nodes is determined by the number of output channels within the first node, and wherein the maximum number of third nodes is determined by the total number of output channels within all of the second nodes.

21. The method of claim 17, wherein the input channel and the output channel of the second node are each part of a full-duplex communication channel.

22. The method of claim 17, wherein connectivity between communicating input and output channels is managed by at least one switch.