Granted Patent B2

US 12,204,944 · App. 17/505,820 · Granted Jan 21, 2025

Inventors: Rômulo Teixeira de Abreu Pinho (Niterói, BR); Vinicius Michel Gottin (Rio de Janeiro, BR); Eduardo Vera Sousa (Niterói, BR)

Assignee: EMC IP HOLDING COMPANY LLC

G06F9/505G06F9/48G06F9/4806G06F9/4843G06F9/4881G06F9/50G06F9/5005G06F9/5027G06F9/5055G06F9/5083G06F11/3409G06F16/9024G06F18/22G06N3/02G06F2209/501G06F2209/5019

Abstract

Claims (53)

1. A method for workload placement based on subgraph similarity, the method comprising:

obtaining a workload graph corresponding to a workload to be executed;

encoding the workload graph to obtain an encoded workload graph;

performing, using the encoded workload graph, a lookup to obtain a set of entries corresponding to previous executions of the workload, wherein the set of entries are stored in a graph database management system (GDBMS);

selecting an encoded previous execution workload graph from the set of entries obtained by the lookup, wherein the encoded previous execution workload graph comprises a workload resource usage signature;

obtaining an encoded infrastructure subgraph from the GDBMS corresponding to the encoded previous execution workload graph, wherein the encoded infrastructure subgraph comprises an infrastructure resource usage signature and an infrastructure portion of a device ecosystem on which a previous workload corresponding to the encoded previous execution workload graph was executed;

obtaining an execution metric set associated with the encoded infrastructure subgraph from the GDBMS;

making a first determination, using the execution metric set and an execution requirement set associated with the workload, that the encoded infrastructure subgraph should remain in a set of filtered encoded infrastructure subgraphs;

performing, using the encoded infrastructure subgraph and the GDBMS, a subgraph similarity query to obtain a plurality of encoded infrastructure candidate subgraphs, each comprising respective current resource usage signatures, wherein each of the plurality of encoded infrastructure candidate subgraphs is obtained by computing a similarity computation relative to the workload that yields a result over a similarity threshold, wherein performing the subgraph similarity query comprises using at least one distance metric;

executing a machine learning (ML) prediction model using an encoded infrastructure candidate subgraph of the plurality of encoded infrastructure candidate subgraphs and an average resource usage associated with previous executions of the workload to obtain a predicted execution result, wherein the predicted execution result comprises a predicted execution time;

making a second determination, using the predicted execution result and the execution requirement set associated with the workload, that an execution of the workload is likely to be successful when executed on infrastructure associated with the encoded infrastructure candidate subgraph, wherein the predicted execution result associated with the execution of the workload that is likely to be successful has a lowest predicted execution time; and

deploying, based on the second determination, the workload on a device ecosystem portion represented by the encoded infrastructure candidate subgraph.

2. The method of claim 1 , wherein the execution requirement set is based on a service level agreement.

3. The method of claim 1 , wherein the workload resource usage signature comprises information related to processor, memory, storage, and network resource usage.

4. The method of claim 1 , wherein the encoded infrastructure subgraph is a vectorial representation of an infrastructure subgraph.

5. The method of claim 1 , wherein the set of filtered encoded infrastructure subgraphs includes only encoded infrastructure subgraphs having execution metrics within a tolerance of the execution requirement set.

6. The method of claim 1 , wherein the encoded workload graph, the encoded previous execution workload graph, the encoded infrastructure subgraph, and the encoded infrastructure candidate subgraph are encoded using a graph neural network.

7. A non-transitory computer readable medium comprising computer readable program code, which when executed by a computer processor enables the computer processor to perform a method for workload placement based on subgraph similarity, the method comprising:

obtaining a workload graph corresponding to a workload to be executed;

encoding the workload graph to obtain an encoded workload graph;

performing, using the encoded workload graph, a lookup to obtain a set of entries corresponding to previous executions of the workload, wherein the set of entries are stored in a graph database management system (GDBMS);

selecting an encoded previous execution workload graph the set of entries obtained by the lookup, wherein the encoded previous execution workload graph comprises a workload resource usage signature;

obtaining an encoded infrastructure subgraph from the GDBMS corresponding to the encoded previous execution workload graph, wherein the encoded infrastructure subgraph comprises an infrastructure resource usage signature and an infrastructure portion of a device ecosystem on which a previous workload corresponding to the encoded previous execution workload graph was executed;

obtaining an execution metric set associated with the encoded infrastructure subgraph from the GDBMS;

making a first determination, using the execution metric set and an execution requirement set associated with the workload, that the encoded infrastructure subgraph should remain in a set of filtered encoded infrastructure subgraphs;

performing, using the encoded infrastructure subgraph and the GDBMS, a subgraph similarity query to obtain a plurality of encoded infrastructure candidate subgraphs, each comprising respective current resource usage signatures, wherein each of the plurality of encoded infrastructure candidate subgraphs is obtained by computing a similarity computation relative to the workload that yields a result over a similarity threshold, wherein performing the subgraph similarity query comprises using at least one distance metric;

executing a machine learning (ML) prediction model using an encoded infrastructure candidate subgraph of the plurality of encoded infrastructure candidate subgraphs and an average resource usage associated with previous executions of the workload to obtain a predicted execution result, wherein the predicted execution result comprises a predicted execution time;

making a second determination, using the predicted execution result and the execution requirement set associated with the workload, that an execution of the workload is likely to be successful when executed on infrastructure associated with the encoded infrastructure candidate subgraph, wherein the predicted execution result associated with the execution of the workload that is likely to be successful has a lowest predicted execution time; and

deploying, based on the second determination, the workload on a device ecosystem portion represented by the encoded infrastructure candidate subgraph.

8. The non-transitory computer readable medium of claim 7 , wherein the execution requirement set is based on a service level agreement.

9. The non-transitory computer readable medium of claim 7 , wherein the workload resource usage signature comprises information related to processor, memory, storage, and network resource usage.

10. The non-transitory computer readable medium of claim 7 , wherein the encoded infrastructure subgraph is a vectorial representation of an infrastructure subgraph.

11. The non-transitory computer readable medium of claim 7 , wherein the set of filtered encoded infrastructure subgraphs includes only encoded infrastructure subgraphs having execution metrics within a tolerance of the execution requirement set.

12. The non-transitory computer readable medium of claim 7 , wherein the encoded workload graph, the encoded previous execution workload graph, the encoded infrastructure subgraph, and the encoded infrastructure candidate subgraph are encoded using a graph neural network.

13. A system for workload placement based on subgraph similarity, the system comprising:

a processor comprising circuitry;

memory; and

a workload placement device operatively connected to a graph database management system (GDBMS), executing on the processor and using the memory, and configured to:

obtain a workload graph corresponding to a workload to be executed;

encode the workload graph to obtain an encoded workload graph;

perform, using the encoded workload graph, a lookup to obtain a set of entries corresponding to previous executions of the workload, wherein the set of entries are stored in a graph database management system (GDBMS);

select an encoded previous execution workload graph from the set of entries obtained by the lookup, wherein the encoded previous execution workload graph comprises a workload resource usage signature;

obtain an encoded infrastructure subgraph from the GDBMS corresponding to the encoded previous execution workload graph, wherein the encoded infrastructure subgraph comprises an infrastructure resource usage signature and an infrastructure portion of a device ecosystem on which a previous workload corresponding to the encoded previous execution workload graph was executed;

obtain an execution metric set associated with the encoded infrastructure subgraph from the GDBMS;

make a first determination, using the execution metric set and an execution requirement set associated with the workload, that the encoded infrastructure subgraph should remain in a set of filtered encoded infrastructure subgraphs;

perform, using the encoded infrastructure subgraph and the GDBMS, a subgraph similarity query to obtain a plurality of encoded infrastructure candidate subgraphs, each comprising respective current resource usage signatures, wherein each of the plurality of encoded infrastructure candidate subgraphs is obtained by computing a similarity computation relative to the workload that yields a result over a similarity threshold, wherein performing the subgraph similarity query comprises using at least one distance metric;

execute a machine learning (ML) prediction model using an encoded infrastructure candidate subgraph of the plurality of encoded infrastructure candidate subgraphs and an average resource usage associated with previous executions of the workload to obtain a predicted execution result, wherein the predicted execution result comprises a predicted execution time;

make a second determination, using the predicted execution result and the execution requirement set associated with the workload, that an execution of the workload is likely to be successful when executed on infrastructure associated with the encoded infrastructure candidate subgraph, wherein the predicted execution result associated with the execution of the workload that is likely to be successful has a lowest predicted execution time; and

deploy, based on the second determination, the workload on a device ecosystem portion represented by the encoded infrastructure candidate subgraph.

14. The system of claim 13 , wherein the workload resource usage signature comprises information related to processor, memory, storage, and network resource usage.

15. The system of claim 13 , wherein the encoded infrastructure subgraph is a vectorial representation of an infrastructure subgraph.

16. The system of claim 13 , wherein the set of filtered encoded infrastructure subgraphs includes only encoded infrastructure subgraphs having execution metrics within a tolerance of the execution requirement set.

17. The system of claim 13 , wherein the encoded workload graph, the encoded previous execution workload graph, the encoded infrastructure subgraph, and the encoded infrastructure candidate subgraph are encoded using a graph neural network.

Assignments (1)

Continuity (1)

References Cited (12)