Search All Applications in Edge Computing
Application US20190327506
Published 2019-10-24
Dynamic Rebalancing Of Edge Resources For Multi-camera Video Streaming
In one embodiment, an edge compute node comprises processing circuitry to: receive an incoming video stream captured by a camera, wherein the incoming video stream comprises a plurality of video segments; store the plurality of video segments in a receive buffer in a memory; perform a visual computing task on a first video segment in the receive buffer; detect a resource overload on the edge compute node; receive load information corresponding to a plurality of peer compute nodes; select a peer compute node to perform the visual computing task on a second video segment in the receive buffer; replicate the second video segment from the edge compute node to the peer compute node; and receive a compute result from the peer compute node, wherein the compute result is based on the peer compute node performing the visual computing task on the second video segment.
Much More than Average Length Specification
View the Patent Matrix® Diagram to Explore the Claim Relationships
USPTO Full Text Publication >
- 1. An edge compute node, comprising:
a network interface to communicate over a network; a memory; and processing circuitry to:
receive, via the network interface, an incoming video stream captured by a camera, wherein the incoming video stream comprises a plurality of video segments;
store the plurality of video segments in a receive buffer in the memory;
perform a visual computing task on a first video segment in the receive buffer;
detect a resource overload on the edge compute node, wherein the resource overload causes the edge compute node not to perform the visual computing task on a second video segment in the receive buffer;
receive, via the network interface, load information corresponding to a plurality of peer compute nodes;
select a peer compute node to perform the visual computing task on the second video segment, wherein the peer compute node is selected from the plurality of peer compute nodes based on the load information;
replicate, via the network interface, the second video segment from the edge compute node to the peer compute node; and
receive, via the network interface, a compute result from the peer compute node, wherein the compute result is based on the peer compute node performing the visual computing task on the second video segment.
- 11. At least one non-transitory machine accessible storage medium having instructions stored thereon, wherein the instructions, when executed on a machine, cause the machine to:
receive, via a network interface of an edge compute node, an incoming video stream captured by a camera, wherein the incoming video stream comprises a plurality of video segments; store the plurality of video segments in a receive buffer in a memory of the edge compute node; perform a visual computing task on a first video segment in the receive buffer; detect a resource overload on the edge compute node, wherein the resource overload causes the edge compute node not to perform the visual computing task on a second video segment in the receive buffer; receive, via the network interface, load information corresponding to a plurality of peer compute nodes; select a peer compute node to perform the visual computing task on the second video segment, wherein the peer compute node is selected from the plurality of peer compute nodes based on the load information; replicate, via the network interface, the second video segment from the edge compute node to the peer compute node; and receive, via the network interface, a compute result from the peer compute node, wherein the compute result is based on the peer compute node performing the visual computing task on the second video segment.
- 18. A method, comprising:
receiving, via a network interface of an edge compute node, an incoming video stream captured by a camera, wherein the incoming video stream comprises a plurality of video segments; storing the plurality of video segments in a receive buffer in a memory of the edge compute node; performing a visual computing task on a first video segment in the receive buffer; detecting a resource overload on the edge compute node, wherein the resource overload causes the edge compute node not to perform the visual computing task on a second video segment in the receive buffer; receiving, via the network interface, load information corresponding to a plurality of peer compute nodes; selecting a peer compute node to perform the visual computing task on the second video segment, wherein the peer compute node is selected from the plurality of peer compute nodes based on the load information; replicating, via the network interface, the second video segment from the edge compute node to the peer compute node, wherein a direct memory access transfer of the second video segment is performed between the edge compute node and the peer compute node; and receiving, via the network interface, a compute result from the peer compute node, wherein the compute result is based on the peer compute node performing the visual computing task on the second video segment.
- 22. A system, comprising:
a camera; and an edge compute node, comprising:
a network interface to communicate over a network;
a memory;
processing circuitry to:
receive, via the network interface, an incoming video stream captured by the camera, wherein the incoming video stream comprises a plurality of video segments;
store the plurality of video segments in a receive buffer in the memory;
perform a visual computing task on a first video segment in the receive buffer;
detect a resource overload on the edge compute node, wherein the resource overload causes the edge compute node not to perform the visual computing task on a second video segment in the receive buffer;
receive, via the network interface, load information corresponding to a plurality of peer compute nodes;
select a peer compute node to perform the visual computing task on the second video segment, wherein the peer compute node is selected from the plurality of peer compute nodes based on the load information;
replicate, via the network interface, the second video segment from the edge compute node to the peer compute node, wherein a remote direct memory access (RDMA) transfer of the second video segment is performed between the edge compute node and the peer compute node; and
receive, via the network interface, a compute result from the peer compute node, wherein the compute result is based on the peer compute node performing the visual computing task on the second video segment.
