DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Examiner Notes
Examiner cites particular columns and line numbers in the references as applied to the claims below for the convenience of the applicant. Although the specified citations are representative of the teachings in the art and are applied to the specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested that, in preparing responses, the applicant fully consider the references in entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner.
The examiner encourages Applicant to submit an authorization to communicate with the examiner via the Internet by making the following statement (from MPEP 502.03):
“Recognizing that Internet communications are not secure, I hereby authorize the USPTO to communicate with the undersigned and practitioners in accordance with 37 CFR 1.33 and 37 CFR 1.34 concerning any subject matter of this application by video conferencing, instant messaging, or electronic mail. I understand that a copy of these communications will be made of record in the application file.”
Please note that the above statement can only be submitted via Central Fax, Regular postal mail, or EFS Web (PTO/SB/439).
Examiner Notes
Applicant’s remarks filed on March 31, 2026 have been fully considered but are not persuasive. Applicant’s remarks are related to newly amended claim language and have been fully addressed in the rejections below. Examiner would like to note that the Gavali reference (US 20200364086 A1) has been added to address some of the amended claim language.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The following language is unclear and indefinite:
As per claim 1, it is unclear what is meant by “determine whether the application has a performance gain by the inference model not being executed in the information handling system” (i.e. The term “performance gain” in claim 1 is a relative term which renders the claim indefinite. The term “performance gain” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. For example, it is unclear that the first and second performance levels are able to be compared. The claims do not indicate that the first performance level and the second performance level are associated with the same set of metrics (i.e. the first performance level could be related to bandwidth and the second performance level could be related to CPU utilization or some other metric unrelated to bandwidth). If the performance levels consider metrics that cannot be compared, it was unclear how a performance gain could be ascertained based on these two performance levels. Independent claims 9 and 17 contain similar language and are rejected for the same reasons. Dependent claims 2-8, 10-16, and 18-20 are rejected due to their dependence on independent claims 1, 9, and 17.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-2, 6-7, 9-10, and 14-15, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Shukla et al. (US 2022/0318674) in view of Aygar et al. (US 2024/0039809) and Gavali et al. (US 20200364086).
As per claim 1, Shukla teaches the invention substantially as claimed including an information handling system ([0038], system 100 is configured specifically to enable the execution of AI workloads, such that the hardware, firmware, and/or software of the system 100 is configured to enable efficient execution of tasks associated with AI workloads... Alternatively, or additionally, the system 100 may include hardware, firmware, and/or software configured specifically to enable the execution of other types of workloads) comprising:
resource detection circuitry to collect data associated with resources being utilized in the information handling system ([0024], scheduler is further configured to monitor and/or track workloads that are currently running and hardware capacity that is currently available anywhere around the world in the cloud of the disclosed service); and
a processor to communicate with the resource detection circuitry ([0077], An example system for managing AI workloads in a cloud infrastructure platform comprises: at least one processor of the cloud infrastructure platform), the processor to:
determine resources for execution of an inference model ([0059], resource subsets of the distributed infrastructure resources are assigned to the received AI workloads...Assigning the resources may include determining resource requirements of an AI workload and then identifying a subset of infrastructure resources that satisfy those requirements; Examiner Note: Shukla’s ai workload includes inference workloads: [0017], The AI infrastructure service of the disclosure is operable with all AI workloads, including training (e.g., workloads for training new or updated AI models) and inferencing (e.g., workloads for using trained AI models to evaluate and make inferences from data; and [0025], The disclosed service is configured to manage AI workloads in a priority-driven and/or tier-driven manner. When the disclosed scheduler makes decisions regarding AI training or inference workloads, the scheduler may consider the designated tier of a given job (or an inferencing model));
receive the data associated with the resources from the resource detection circuitry ([0059], identifying a subset of infrastructure resources that satisfy those requirements);
based on the resources for the execution of the inference model, determine a first performance of an application if the inference model is executed in the information handling system ([0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform);
...migrate the inference model to an ...server for execution ([0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform and, based on that monitoring, the scheduling of the AI workloads is adjusted. The adjusting of the scheduling may include... migrating an AI workload).
Shukla fails to specifically teach, determine a second performance level of the application when the inference model is not executed in the information handling system; based on the first and second performance levels, determine whether the application has a performance gain by the inference model not being executed in the information handling system as compared to the inference model being executed in the information handling system; and in response to the application having the performance gain, migrate the inference model to an edge server for execution.
However, Aygar teaches, determine a second performance level of the application when the inference model is not executed in the information handling system ([0100], the orchestrator shows to the user through the orchestrator UI the possible resource savings, operating expense reduction, and sustainability improvements on defragmenting the containerized workload... During the analysis stage, the orchestrator develops scenarios to gauge resource savings and in-turn expense reduction if the current workloads scheduled on a target node are migrated to a neighboring node and the target node is decommissioned. These scenarios are then presented to the user via the orchestrator UI. The defragmentation process can also be triggered when the system runs low on resources or when each node has some unutilized spare capacity which can't be repurposed for placing workloads or groups; and [0103], context triggered defragmentation includes three components: a) based on the telemetry data, the orchestrator identifies workloads with interdependencies including trigger dependency and data dependency, b) the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them);
based on the first and second performance levels, determine whether the application has a performance gain by the inference model not being executed in the information handling system as compared to the inference model being executed in the information handling system ([0103], the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them... These telemetry data collected over time and under different deployment and execution conditions can be used to model workload behavior. The model can then be used to create scenarios and predict any savings and reduction in resource costs, along with improvement in workload performance based on predicted resources saved, reduction in inter-workload trigger and data transfer latency, and context and runtime sharing); and
in response to the application having the performance gain, ... migrate the inference model to an edge server for execution ([0103], the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them; and [0104], once the defragmentation process is triggered by the user, the orchestrator tags the containerized workload for migration and marks the edge devices from which the containerized workload migrates as un-schedulable. The orchestrator then migrates the containerized workload to target edge devices of the edge based on the generated defragmentation plans).
Shukla and Aygar are analogous because they are each related to workload management. Shukla teaches a method of allocating AI workloads based on workload requirements and telemetry data. (Abstract, The disclosure herein describes managing artificial intelligence (AI) workloads in a cloud infrastructure platform... The described cloud infrastructure platform provides efficient, secure execution of AI workloads for many different tenants and enables the flexible use of a wide variety of both third-party and first-party infrastructure resources; and [0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform and, based on that monitoring, the scheduling of the AI workloads is adjusted. The adjusting of the scheduling may include preempting an AI workload, migrating an AI workload, scaling up an AI workload, scaling down an AI workload, and/or load-balancing between two or more AI workloads). Aygar teaches a method of allocating/migrating workloads to edge devices based on workload requirements and telemetry data. (Abstract, Computer-implemented methods, media, and systems for right-sizing containerized workloads running on edge devices are disclosed. One example method includes receiving a manifest file including one or more runtime service level agreement (SLA) requirements and one or more upper bounds on resource allocation for running the workload on a software defined wide area network (SD-WAN) edge device. The workload on the SD-WAN edge device is started based on the manifest file. Telemetry data from the SD-WAN edge device is monitored, where the telemetry data includes resource usage data for running the workload on the SD-WAN edge device. A model for resource usage behavior of the workload running on the SD-WAN edge device is established based on the monitored telemetry data). It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention that based on the combination, the global scheduler of Shukla would be modified to include Aygar’s migration mechanism resulting in a system that can allocate and migrate workloads based on telemetry. Therefore, it would have been obvious to combine the teachings of Shukla and Aygar.
The combination of Shukla and Aygar fails to specifically teach, determine whether a global flag is set to a first value or a second value, wherein the first value indicates that an edge server is not available and the second value indicates that the edge server is available; and in response to the application having ... the global flag being set to the second value migrate the inference model to an edge server for execution.
However, Gavali teaches, determine whether a global flag is set to a first value or a second value ([0067], The new node available flag is a configurable flag that indicates whether a new worker node is currently available to be added to the group when an unhealthy node is detected; and [0104], the computer makes a determination as to whether ...a new node available flag are set to true in the workload orchestration environment (step 908). It should be noted that the new node automatic flag and the new node available flag are global flags,), wherein the first value indicates that an edge server is not available ([0104], If the computer determines that both the new node automatic flag and the new node available flag are not set to true, no output of step 908, then the process proceeds to step 912; and [0105], . If the computer determines that another worker node does not exist in the worker node group, no output of step 914, then the process terminates thereafter) and the second value indicates that the edge server is available ([0105], If the computer determines that...the new node available flag are set to true, yes output of step 908, then the computer adds a new worker node to the worker node group (step 910) ); and
in response to the application having ... the global flag being set to the second value migrate the inference model to an edge server for execution ([0105], If the computer determines that both the new node automatic flag and the new node available flag are set to true, yes output of step 908, then the computer adds a new worker node to the worker node group (step 910; and [0107], If the computer determines that the average resource utilization of any worker node is greater than the upper threshold of resource utilization based on the collected resource utilization data, yes output of step 1008, then the computer triggers redistribution of the workload on the worker nodes (step 1010)).
The combination of Shukla-Aygar and Gavali are analogous because they are each related to workload management. Shukla teaches a method of allocating AI workloads based on workload requirements and telemetry data. Aygar teaches a method of allocating/migrating workloads to edge devices based on workload requirements and telemetry data. Gavali teaches a method of workload management, including workload distribution, in accordance with policies that optimize workload performance and node availability. ([0035], Workload orchestration manager 218 controls the process of rescheduling workloads across worker nodes of a workload orchestration environment to redistribute the workloads based on policy; and [0050], a need exists to optimize workload distribution using an objective function that takes into account a ratio of resource utilization to workload capacity and historical health characteristics of the worker nodes. The ratio of resource utilization to workload capacity ensures equitable distribution of consumption of compute, storage, and network resources, such as, for example, processor usage, memory usage, network usage, and the like). It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention that based on the combination, the global scheduler of the combination of Shukla-Aygar would be modified to include Gavali’s workload redistribution mechanism resulting in a system that can allocate workloads and migrate workloads based on workload optimization and node availablity. Therefore, it would have been obvious to combine the teachings of the combination of Shukla-Aygar and Gavali.
As per claim 2, Shukla teaches, further comprising workload detection circuitry to communicate with the processor, the workload detection circuitry to:
determine a current workload level within the information handling system ([0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform; and [0066], regional schedulers 504 monitor the current regional capacity data 516 of the infrastructure resources 508 associated with the respective regions); and
provide the workload level to the processor ([0042], The runtime plane 104 includes subsystems configured to enable the AI workloads to be distributed to and executed on the infrastructure plane 106 as described herein. Such subsystems may include a monitoring subsystem 114;[0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform and, based on that monitoring, the scheduling of the AI workloads is adjusted... Such schedule adjustment may be performed by a global scheduling subsystem 112 or other component of the system 100; and [0066], regional schedulers 504 monitor the current regional capacity data 516 of the infrastructure resources 508 associated with the respective regions and that regional capacity data 516 is provided to the global scheduler 502 ).
As per claim 6, Aygar teaches, wherein the processor further to in response to the application not having the performance gain, determine that the inference model is to be executed in the information handling system ([0107], If the migration fails, step 512 is performed to cancel the migration of the containerized workload and restore the initial state of the containerized workload at the edge devices from which the containerized workload is migrated).
As per claim 7, Aygar teaches, wherein a default state is for the inference model to be executed in the information handling system ([0100], once a containerized workload has been provisioned to run on an edge that has multiple edge devices, the compute service at the edge continually monitors and collects telemetry data associated with the containerized workload).
As per claim 9, this is the “method claim” corresponding to claim 1 and is rejected for the same reasons. The same motivation used in the rejection of claim 1 is applicable to the instant claim.
As per claim 10, this claim is similar to claim 2 and is rejected for the same reasons.
As per claim 14, this claim is similar to claim 6 and is rejected for the same reasons.
As per claim 15, this claim is similar to claim 7 and is rejected for the same reasons.
As per claim 17, Shukla teaches the invention substantially as claimed including a system comprising:
an information handling system ([0038], system 100 is configured specifically to enable the execution of AI workloads, such that the hardware, firmware, and/or software of the system 100 is configured to enable efficient execution of tasks associated with AI workloads... Alternatively, or additionally, the system 100 may include hardware, firmware, and/or software configured specifically to enable the execution of other types of workloads) including:
a memory to store data associated with resources being utilized in the information handling system ([0068], computing device 600 for implementing aspects disclosed herein, and is designated generally as computing device 600; and [0069], Computing device 600 includes a bus 610 that directly or indirectly couples the following devices: computer-storage memory 612); and
a processor to communicate with the resource detection circuitry ([0069], Computing device 600 includes a bus 610 that directly or indirectly couples the following devices: computer-storage memory 612, one or more processors 614), the processor to:
determine resources for execution of an inference model ([0059], resource subsets of the distributed infrastructure resources are assigned to the received AI workloads...Assigning the resources may include determining resource requirements of an AI workload and then identifying a subset of infrastructure resources that satisfy those requirements; Examiner Note: Shukla’s ai workload includes inference workloads: [0017], The AI infrastructure service of the disclosure is operable with all AI workloads, including training (e.g., workloads for training new or updated AI models) and inferencing (e.g., workloads for using trained AI models to evaluate and make inferences from data; and [0025], The disclosed service is configured to manage AI workloads in a priority-driven and/or tier-driven manner. When the disclosed scheduler makes decisions regarding AI training or inference workloads, the scheduler may consider the designated tier of a given job (or an inferencing model));
based on the resources for the execution of the inference model, determine a first performance of an application if the inference model is executed in the information handling system ([0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform); and
...migrate the inference model to an ... server for execution ([0064], executing AI workloads are monitored based on the performance of the cloud infrastructure platform and, based on that monitoring, the scheduling of the AI workloads is adjusted. The adjusting of the scheduling may include... migrating an AI workload).
Shukla fails to specifically teach, an edge server configured to execute an inference model; determine a second performance level of the application if the inference model is not executed in the information handling system; based on the first and second performance levels, determine whether the application has a performance gain by the inference model not being executed in the information handling system; and in response to the application having the performance gain, migrate the inference model to an edge server for execution.
However, Aygar teaches, an edge server configured to execute an inference model ([0067], the containerized workload can be run on any edge device registered to orchestrator 202);
determine a second performance level of the application if the inference model is not executed in the information handling system ([0100], the orchestrator shows to the user through the orchestrator UI the possible resource savings, operating expense reduction, and sustainability improvements on defragmenting the containerized workload... During the analysis stage, the orchestrator develops scenarios to gauge resource savings and in-turn expense reduction if the current workloads scheduled on a target node are migrated to a neighboring node and the target node is decommissioned. These scenarios are then presented to the user via the orchestrator UI. The defragmentation process can also be triggered when the system runs low on resources or when each node has some unutilized spare capacity which can't be repurposed for placing workloads or groups; and [0103], context triggered defragmentation includes three components: a) based on the telemetry data, the orchestrator identifies workloads with interdependencies including trigger dependency and data dependency, b) the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them);
based on the first and second performance levels, determine whether the application has a performance gain by the inference model not being executed in the information handling system as compared to the inference model being executed in the information handling system([0103], the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them... These telemetry data collected over time and under different deployment and execution conditions can be used to model workload behavior. The model can then be used to create scenarios and predict any savings and reduction in resource costs, along with improvement in workload performance based on predicted resources saved, reduction in inter-workload trigger and data transfer latency, and context and runtime sharing); and
in response to the application having the performance gain,... migrate the inference model to the edge server for execution ([0103], the orchestrator then calculates performance improvement in defragmenting these workloads and co-locating them; and [0104], once the defragmentation process is triggered by the user, the orchestrator tags the containerized workload for migration and marks the edge devices from which the containerized workload migrates as un-schedulable. The orchestrator then migrates the containerized workload to target edge devices of the edge based on the generated defragmentation plans).
The combination of Shukla and Aygar fails to specifically teach, determine whether a global flag is set to a first value or a second value, wherein the first value indicates that an edge server is not available and the second value indicates that the edge server is available; and in response to the application having ... the global flag being set to the second value migrate the inference model to an edge server for execution.
However, Gavali teaches, determine whether a global flag is set to a first value or a second value ([0067], The new node available flag is a configurable flag that indicates whether a new worker node is currently available to be added to the group when an unhealthy node is detected; and [0104], the computer makes a determination as to whether ...a new node available flag are set to true in the workload orchestration environment (step 908). It should be noted that the new node automatic flag and the new node available flag are global flags,), wherein the first value indicates that an edge server is not available ([0104], If the computer determines that both the new node automatic flag and the new node available flag are not set to true, no output of step 908, then the process proceeds to step 912; and [0105], . If the computer determines that another worker node does not exist in the worker node group, no output of step 914, then the process terminates thereafter) and the second value indicates that the edge server is available ([0105], If the computer determines that...the new node available flag are set to true, yes output of step 908, then the computer adds a new worker node to the worker node group (step 910) ); and
in response to the application having ... the global flag being set to the second value migrate the inference model to the edge server for execution ([0105], If the computer determines that both the new node automatic flag and the new node available flag are set to true, yes output of step 908, then the computer adds a new worker node to the worker node group (step 910; and [0107], If the computer determines that the average resource utilization of any worker node is greater than the upper threshold of resource utilization based on the collected resource utilization data, yes output of step 1008, then the computer triggers redistribution of the workload on the worker nodes (step 1010)).
The same motivation used in the rejection of claim 1 is applicable to the instant claim.
Claims 3, 11, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Shukla-Aygar-Gavali as applied to independent claims 1, 9, and 17 and in further view of Cortez et al. (US 20240420019 A1).
As per claim 3, Aygar teaches, wherein prior to the migration of the inference model , the processor further to:
determine a latency associated with the migration of the inference model to the edge server ([0162], t is determined that a violation to at least one of the one or more runtime SLA requirements occurs. In response to the determination that the violation to at least one of the one or more runtime SLA requirements occurs, the workload is migrated to another SD-WAN edge device with resources satisfying the one or more updated upper bounds on resource allocation; and [0163] The one or more runtime SLA requirements for running the workload includes one or more configurable requirements for at least one of ...network latency).
The combination of Shukla-Aygar-Gavali fails to specifically teach, compare the latency to a threshold quality of service latency; and in response to the latency being less than the threshold quality of service latency, determine that the inference model is to be migrated to the edge server.
However, Cortez teaches, wherein prior to the migration of the inference model ([0003], validating trained models in stage environments prior to deployment to production environments is described), the processor further to:
compare the latency to a threshold quality of service latency ([0025], the inference service 134 collects timing data and/or other performance data while the registered model 122 is performing inference operations and that timing data is compared a defined latency threshold and/or other performance metric thresholds); and
in response to the latency being less than the threshold quality of service latency, determine that the inference model is to be migrated to the edge server ([0046], a validity test applied to the deployed model is a latency validity test, wherein collected latency data of the performed inference operations is compared to a defined latency range and/or threshold. If the collected latency data fits into the defined latency range and/or is under the defined latency threshold, the deployed model is valid with respect to that particular validity test).
The combination of Shukla-Aygar-Gavali and Cortez are analogous because they are each related to workload management. Shukla teaches a method of allocating AI workloads based on workload requirements and telemetry data. Aygar teaches a method of allocating/migrating workloads to edge devices based on workload requirements and telemetry data. Gavali teaches a method of workload management, including workload distribution, in accordance with policies that optimize workload performance and node availability. Cortez teaches a method of deploying AI models based on requirements including latency constraints. ([0025], the inference service 134 collects timing data and/or other performance data while the registered model 122 is performing inference operations and that timing data is compared a defined latency threshold and/or other performance metric thresholds. If the timing data collected by the inference service 134 meets or is lower than the defined latency threshold, the registered model 122 is validated for latency). It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention that based on the combination, the global scheduler of the combination of Shukla-Aygar-Gavali would be modified to include Cortez’s latency validation test resulting in a system that can allocate workloads and migrate workloads based on latency constraints. Therefore, it would have been obvious to combine the teachings of the combination of Shukla-Aygar-Gavali and Cortez.
As per claim 11, this claim is similar to claim 3 and is rejected for the same reasons.
As per claim 18, this claim is similar to claim 3 and is rejected for the same reasons.
Claims 4-5, 8, 12-13, 16, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of Shukla-Aygar-Gavali as applied to independent claims 1, 9, and 17 and in further view of Issac et al. (US 2024/0232039).
As per claim 4, the combination of Shukla-Aygar-Gavali fails to specifically teach, wherein prior to the migration of the inference model, the processor further to: determine a power associated with the migration of the inference model to the edge server; compare the power to a threshold power; and in response to the power being less than the threshold power, determine that the inference model is to be migrated to the edge server.
However, Issac teaches, wherein prior to the migration of the inference model ([0058], processing logic identifies usage data associated with an application that is to be executed using a computing system), the processor further to:
determine a power associated with the migration of the inference model to the edge server ([0041], user performance metrics can include an amount of power consumed by user device 106 during execution of the application using an iGPU and/or a dGPU of the user device 106; Examiner Notes: 1. Under the broadest reasonable interpretation, “power associated with migration of the inference model” has been interpreted to include power consumption resulting from migration of the inference model; and 2. Issac teaches migration to an edge device: [0032],disclosed embodiments may be comprised in a variety of different systems such as ... systems for performing deep learning operations, systems for performing simulation operations, systems implemented using an edge device; and [0046], functions described in implementations as being performed by computing device 102 and/or or server machines 130, 140, 150 may also be performed on one or more edge devices);
compare the power to a threshold power ([0030], The computing system can further assign the application to be run using the iGPU responsive to determining that an application power consumption metric value for the iGPU meets or falls below an optimal power consumption threshold value of the one or more application processing criteria and/or a user experience metric value for the dGPU exceeds an optimal power consumption threshold value); and
in response to the power being less than the threshold power, determine that the inference model is to be migrated to the edge server ([0030], The computing system can further assign the application to be run using the iGPU responsive to determining that an application power consumption metric value for the iGPU meets or falls below an optimal power consumption threshold value of the one or more application processing criteria and/or a user experience metric value for the dGPU exceeds an optimal power consumption threshold value).
The combination of Shukla-Aygar-Gavali and Issac are analogous because they are each related to workload management. Shukla teaches a method of allocating AI workloads based on workload requirements and telemetry data. Aygar teaches a method of allocating/migrating workloads to edge devices based on workload requirements and telemetry data. Gavali teaches a method of workload management, including workload distribution, in accordance with policies that optimize workload performance and node availability. Issac teaches a method of deploying AI models based on requirements including power constraints. (Abstract, techniques for assigning execution of applications to various processing units using machine learning are disclosed herein. ... At least a portion of operations of the application to be executed using the integrated processing unit or the discrete processing unit based on the usage data and in view of at least one of one or more system performance metrics or one or more user experience metrics associated with executing the application using the integrated processing unit and the discrete processing unit; [0019], Integrated GPUs can share memory resources and power resources with other processing units integrated with the circuit board (e.g., central processing units (CPUs), etc.); and [0025], The processor can provide the usage data as input to a machine learning model. The machine learning model can be trained to predict values for one or more system performance metrics (e.g., clock data associated with the iGPUs and/or the dGPUs, a measured efficiency associated with executing the application, etc.) and/or one or more user experience metrics (e.g., an amount of power consumed from a battery of the computing system during execution of the application). It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention that based on the combination, the global scheduler of the combination of Shukla-Aygar-Gavali would be modified to include Issac’s power metrics resulting in a system that can allocate workloads and migrate workloads based on power constraints. Therefore, it would have been obvious to combine the teachings of the combination of Shukla-Aygar-Gavali and Issac.
As per claim 5, the combination of Shukla-Aygar-Gavali fails to specifically teach, wherein prior to the migration of the inference model, the processor further to: determine a power associated with the migration of the inference model to the edge server; compare the power to a threshold power; and in response to the power being less than the threshold power, determine that the inference model is to be migrated to the edge server.
However, Issac teaches, wherein prior to the migration of the inference model, the processor further to:
determine a power associated with an execution of the inference model in the information handling system ([0041], user performance metrics can include an amount of power consumed by user device 106 during execution of the application using an iGPU and/or a dGPU of the user device 106);
compare the power to a threshold power ([0030], The computing system can further assign the application to be run using the iGPU responsive to determining that an application power consumption metric value for the iGPU meets or falls below an optimal power consumption threshold value of the one or more application processing criteria and/or a user experience metric value for the dGPU exceeds an optimal power consumption threshold value); and
in response to the power being greater than the threshold power, determine that the inference model is to be migrated to the edge server ([0030], The computing system can further assign the application to be run using the iGPU responsive to determining that an application power consumption metric value for the iGPU meets or falls below an optimal power consumption threshold value of the one or more application processing criteria and/or a user experience metric value for the dGPU exceeds an optimal power consumption threshold value).
The same motivation used in the rejection of claim 4 is applicable to the instant claim.
As per claim 8, Shukla teaches, wherein the resources include a graphics processor ([0055], infrastructure resources 328, such as compute resources 380, may be linked to GPUs 388, for instance, such that a compute resource 380 provides instructions to the GPU 388 for how to execute steps of the AI workload. Such execution then takes advantage of specialized architecture of the GPU 388).
Shukla fails to specifically teach, wherein the resources include ...a memory, and a power capability.
However, Aygar teaches, wherein the resources include ... a memory ([0070], edge devices are sorted based on the continuous/quantitative values of context elements such as CPU and memory, and the best fit edge device to run the containerized workload is identified).
The combination of Shukla-Aygar-Gavali fails to specifically teach, wherein the resources include ... a power capability.
However, Issac teaches, wherein the resources include ... a power capability ([0049], CPUs 212 and/or GPUs 214 can share computing resources (e.g., memory 216, power source 218) of circuit board 210).
The same motivation used in the rejection of claim 4 is applicable to the instant claim.
As per claim 12, this claim is similar to claim 4 and is rejected for the same reasons. The same motivation used in the rejection of claim 4 is applicable to the instant claim.
As per claim 13, this claim is similar to claim 5 and is rejected for the same reasons. The same motivation used in the rejection of claim 5 is applicable to the instant claim.
As per claim 16, this claim is similar to claim 8 and is rejected for the same reasons. The same motivation used in the rejection of claim 8 is applicable to the instant claim.
As per claim 19, this claim is similar to claim 4 and is rejected for the same reasons. The same motivation used in the rejection of claim 4 is applicable to the instant claim.
As per claim 20, this claim is similar to claim 5 and is rejected for the same reasons. The same motivation used in the rejection of claim 5 is applicable to the instant claim.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure and is as follows:
Duggan et al. (EP 3182283 A1)-Teaches deploying AI models:
[0015], The resource allocation circuitry 116 determines resource loads for analytical models and determines resource load capabilities of a compute engine118 or other resources; and [0044], The resource allocation circuitry 116 may be configured to determine a resource load capability of the compute engine 118 (504). For example, the resource allocation circuitry 116 may have access to data describing the processing and memory resources available for use by a particular user or client. In other embodiments, the resource allocation circuitry 116 may have testing models with known resource loads which it can deploy to test the available resources of the compute engine 118.
Das et al. (US 20230418663 A1)- Teaches dynamic workload migration based on a variety of factors:
Abstract, dynamic migration of jobs (e.g., workloads, containers, service requests, etc.) between execution environments. The disclosed systems and methods may utilize monitoring techniques to determine when a migration should occur and/or forecasting techniques to predict optimal times when a migration should occur. Upon determining a migration should occur, a target execution environment for a job may be identified and a migration process may be initiated
Applicant's amendment necessitated the new grounds of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MELISSA A HEADLY whose telephone number is (571)272-1972. The examiner can normally be reached Monday- Friday 9-5:30pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Bradley Teets can be reached at 571-272-3338. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MELISSA A. HEADLY/
Examiner Art Unit 2197
/GREGORY A KESSLER/Primary Examiner, Art Unit 2197
Prosecution Insights