Privacy-preserving distributed visual data processing
In one embodiment, an apparatus comprises a processor to: identify a workload comprising a plurality of tasks; generate a workload graph based on the workload, wherein the workload graph comprises information associated with the plurality of tasks; identify a device connectivity graph, wherein the device connectivity graph comprises device connectivity information associated with a plurality of processing devices; identify a privacy policy associated with the workload; identify privacy level information associated with the plurality of processing devices; identify a privacy constraint based on the privacy policy and the privacy level information; and determine a workload schedule, wherein the workload schedule comprises a mapping of the workload onto the plurality of processing devices, and wherein the workload schedule is determined based on the privacy constraint, the workload graph, and the device connectivity graph. The apparatus further comprises a communication interface to send the workload schedule to the plurality of processing devices.
1. A system, comprising:
communication circuitry to:
receive a first request to execute a first workload on a distributed computing infrastructure according to a workload schedule, wherein the distributed computing infrastructure comprises a plurality of processing devices;
receive available vision capability implementations from a vision capability repository, the available vision capability implementations indicating multiple implementations of a particular vision capability optimized for each of the processing devices;
receive resource telemetry information of the plurality of processing devices, the resource telemetry information indicating an availability of different resource types on each processing device;
receive a second request to execute a second workload on the distributed computing infrastructure, wherein the first workload and the second workload comprise a plurality of tasks for processing visual data captured by one or more cameras, and wherein each of the tasks is associated with a vision capability implementation and a resource type; and
transmit an updated workload schedule to the processing devices, wherein the updated workload schedule indicates assignments of the tasks of the first workload and the second workload to the processing devices; and
processing circuitry to execute the first workload and the second workload on the distributed computing infrastructure according to the updated workload schedule,
wherein the updated workload schedule further indicates a privacy constraint associated with the first workload and the second workload,
wherein respective tasks of the plurality of tasks of the first workload and the second workload are executed by corresponding processing devices of the distributed computing infrastructure according to the updated workload schedule,
wherein the privacy constraint indicates privacy policies associated with the respective tasks and security levels associated with the corresponding processing devices,
wherein the privacy constraint requires the respective tasks to be executed on processing devices whose associated security levels are sufficient for the privacy policies associated with the respective tasks, and
wherein the updated workload schedule is further based on an optimal solution to an integer linear programming model having constraints including the requested vision capability implementations, the resource telemetry information, the privacy policies, and the security levels.
2. The system of claim 1 , wherein:
the first workload and the second workload further comprise a plurality of task dependencies among the plurality of tasks; and
the distributed computing infrastructure further comprises a plurality of device connectivity links among the plurality of processing devices.
3. The system of claim 2 , wherein:
the privacy policies correspond to the task dependencies; and
the security levels correspond to the device connectivity links.
4. The system of claim 3 , wherein the updated workload schedule maps the plurality of task dependencies to corresponding device connectivity links.
5. The system of claim 1 , wherein:
the security levels include a first security level and a second security level, wherein the first security level is higher than the second security level; and
the privacy policies include a first privacy policy and a second privacy policy, wherein the first privacy policy requires the first security level or higher, and wherein the second privacy policy requires the second security level or higher.
6. The system of claim 5 , wherein:
the first security level comprises security above a threshold;
the second security level comprises security below the threshold;
the first privacy policy comprises unrestricted access to the visual data; and
the second privacy policy comprises restricted access to the visual data.
7. The system of claim 1 , further comprising the plurality of processing devices, wherein at least some of the processing circuitry is comprised in the processing devices.
8. At least one hardware machine-readable storage medium having instructions stored thereon, wherein the instructions, when executed on processing circuitry, cause the processing circuitry to:
receive a first request to execute a first workload on a distributed computing infrastructure according to a workload schedule, wherein the distributed computing infrastructure comprises a plurality of processing devices;
receive available vision capability implementations from a vision capability repository, the available vision capability implementations indicating multiple implementations of a particular vision capability optimized for each of the processing devices;
receive resource telemetry information of the plurality of processing devices, the resource telemetry information indicating an availability of different resource types on each processing device;
receive a second request to execute a second workload on the distributed computing infrastructure, wherein the first workload and the second workload comprise a plurality of tasks for processing visual data captured by one or more cameras, and wherein each of the tasks is associated with a vision capability implementation and a resource type;
transmit an updated workload schedule to the processing devices, wherein the updated workload schedule indicates assignments of the tasks of the first workload and the second workload to the processing devices; and
execute the first workload and the second workload on the distributed computing infrastructure according to the updated workload schedule,
wherein the updated workload schedule further indicates a privacy constraint associated with the first workload and the second workload,
wherein respective tasks of the plurality of tasks of the first workload and the second workload are executed by corresponding processing devices of the distributed computing infrastructure according to the updated workload schedule,
wherein the privacy constraint indicates privacy policies associated with the respective tasks and security levels associated with the corresponding processing devices,
wherein the privacy constraint requires the respective tasks to be executed on processing devices whose associated security levels are sufficient for the privacy policies associated with the respective tasks, and
wherein the updated workload schedule is further based on an optimal solution to an integer linear programming model having constraints including the requested vision capability implementations, the resource telemetry information, the privacy policies, and the security levels.
9. The storage medium of claim 8 , wherein:
the first workload and the second workload further comprise a plurality of task dependencies among the plurality of tasks; and
the distributed computing infrastructure further comprises a plurality of device connectivity links among the plurality of processing devices.
10. The storage medium of claim 9 , wherein:
the privacy policies correspond to the task dependencies; and
the security levels correspond to the device connectivity links.
11. The storage medium of claim 10 , wherein the updated workload schedule maps the plurality of task dependencies to corresponding device connectivity links.
12. The storage medium of claim 8 , wherein:
the security levels include a first security level and a second security level, wherein the first security level is higher than the second security level; and
the privacy policies include a first privacy policy and a second privacy policy, wherein the first privacy policy requires the first security level or higher, and wherein the second privacy policy requires the second security level or higher.
13. The storage medium of claim 12 , wherein:
the first security level comprises security above a threshold;
the second security level comprises security below the threshold;
the first privacy policy comprises unrestricted access to the visual data; and
the second privacy policy comprises restricted access to the visual data.
14. A method, comprising:
receiving a first request to execute a first workload on a distributed computing infrastructure according to a workload schedule, wherein the distributed computing infrastructure comprises a plurality of processing devices;
receiving available vision capability implementations from a vision capability repository, the available vision capability implementations indicating multiple implementations of a particular vision capability optimized for each of the processing devices;
receiving resource telemetry information of the plurality of processing devices, the resource telemetry information indicating an availability of different resource types on each processing device;
receiving a second request to execute a second workload on the distributed computing infrastructure, wherein the first workload and the second workload comprise a plurality of tasks for processing visual data captured by one or more cameras, and wherein each of the tasks is associated with a vision capability implementation and a resource type;
transmitting an updated workload schedule to the processing devices, wherein the updated workload schedule indicates assignments of the tasks of the first workload and the second workload to the processing devices; and
executing the first workload and the second workload on the distributed computing infrastructure according to the updated workload schedule,
wherein the updated workload schedule further indicates a privacy constraint associated with the first workload and the second workload,
wherein respective tasks of the plurality of tasks of the first workload and the second workload are executed by corresponding processing devices of the distributed computing infrastructure according to the updated workload schedule,
wherein the privacy constraint indicates privacy policies associated with the respective tasks and security levels associated with the corresponding processing devices,
wherein the privacy constraint requires the respective tasks to be executed on processing devices whose associated security levels are sufficient for the privacy policies associated with the respective tasks, and
wherein the updated workload schedule is further based on an optimal solution to an integer linear programming model having constraints including the requested vision capability implementations, the resource telemetry information, the privacy policies, and the security levels.
15. The method of claim 14 , wherein:
the security levels include a first security level and a second security level, wherein the first security level is higher than the second security level; and
the privacy policies include a first privacy policy and a second privacy policy, wherein the first privacy policy requires the first security level or higher, and wherein the second privacy policy requires the second security level or higher.