DETAILED ACTION
Claims 1-20 are pending.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefore, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (abstract idea) without significantly more.
As per claim 1, in step 1 of the 101 analysis, the examiner has determined that the claim is directed to an apparatus. Therefore, the claim is directed to one of the four statutory categories of invention.
In step 2A prong 1 of the 101 analysis, the examiner has determined that the claim recites a judicial exception. Specifically, the limitation “assigning…based on characteristics of the machine learning task and the characteristic of the compute system, the machine learning task to at least one of the one or more compute devices” recite mental processes. Determining how to assign tasks by comparing using a criteria/rule constitutes a mental process of observation, evaluation, and judgement because a human with the aid of pen and paper can analyze status information about the devices and task requirements to determine how to schedule the task.
In step 2A prong 2 of the 101 analysis, the examiner has determined that the additional elements, alone or in combination do not integrate the judicial exceptions into a practical application for the following rationale:
The limitations “at least one processor”, “compute system”, and “at least one memory” (claims 16 and 19), apply judicial exceptions on a generic computer. "Alappat 's rationale that an otherwise ineligible algorithm or software could be made patent-eligible by merely adding a generic computer to the claim was superseded by the Supreme Court's Bilski and Alice Corp. decisions" so therefore applying judicial exceptions on a generic computer do not integrate the judicial exceptions into a practical application (MPEP 2106.05(b)).
The limitation “determining…a characteristic of a machine learning task”, and “determining…a characteristic of a compute system” represent insignificant, extra-solution activities. The term "extra-solution activity" can be understood as "activities incidental to the primary process or product that are merely a nominal or tangential addition to the claim" (MPEP 2106.05(g)). The examiner has determined that the limitation is directed to mere data gathering activities which is a category of insignificant extra-solution activities (MPEP 2106.05(g)).
In step 2B of the 101 analysis, the examiner has determined that the additional elements, alone or in combination do not recite significantly more than the abstract ideas identified above for the following rationale:
The limitations “at least one processor”, “compute system”, and “at least one memory” apply judicial exceptions on a generic computer and therefore do not provide significantly more.
The limitation “determining…a characteristic of a machine learning task”, and “determining…a characteristic of a compute system” represent insignificant, extra-solution activities and are well-understood, routine, or conventional because they are directed to "receiving or transmitting data" (MPEP 2106.05(d)). These are additional elements that the courts have recognized as well understood, routine, or conventional (MPEP 2106.05(d)). The citation of court cases in the MPEP meets the Berkheimer evidentiary burden since citation of a court case in the MPEP is one of the 4 types of evidentiary support that can be used to prove that the additional elements are well-understood, routine, or conventional (see 125 USPQ2d 1649 Berkheimer v. HP, Inc.). Thus, the limitations do not amount to significantly more than the abstract idea.
Considering the additional elements individually and in combination and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. The claim is not patent eligible.
As per claims 16 and 19, they are system type claims of claim 1, so it is rejected for the same reasons as claim 1.
As per claims 4-5 (and similarly for claims 14-15 and 20), they recite limitations that further discuss the data gathered in claim 1. These limitations do not provide significantly more than the abstract idea and only describe the gathered data.
As per claims 6-13 (and similarly for claims 16 and 17), they recite limitations directed to perform the abstract idea as shown on claim 1. As such, these claims do not provide significantly more than the abstract idea. For example, the claims are directed to gathering characteristics, policies and compatibilities which are used to determine how to assign the workload among the computing devices. Accordingly, the analysis for claim 1 applies.
Claim Rejections - 35 USC § 103
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.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Wang et al. (WO 2021/101617 A1) in further view of Rajappa et al. (US 2016/0054783 A1).
Wang was cited in the IDS.
Regarding claim 1, Wang teaches the invention substantially as claimed including a method comprising:
determining, by at least one [controller] processor, a characteristic of a machine learning task ([0053] For example, the controller 108 may scan the request 112 to identify parameters (e.g., protocol bits) in the request 112 that specify CPU, memory, and accelerator requirements of various tasks in the workload (304). Based on the parameters and values in the request 112, the controller 108 may determine that an example workload includes 16 tasks, where each of the tasks require a total resource allocation of 96 CPUs and 4 special- purpose circuits, e.g., hardware accelerators. For example, the request 112 can include a scalar resource parameter that specifies a quantity of hardware accelerators (4) that are to be used for executing each of the 16 tasks. In some implementations, the scalar resource parameter may include a sub-type that specifies a type of hardware accelerator to be used to process the workload. For example, the sub-type may specify that each of the 4 hardware accelerators be a neural net processor configured to accelerate running a model trained for feature recognition.; [0062] the workload is a training or inference workload related to a particular machine-learning operation);
determining, by the at least one [controller] processor, a characteristic of a compute system, wherein the compute system comprises one or more compute devices ([0034] For example, the status logic 110 is run by the controller 108 to send commands to a subset of hosts 104 to obtain information about a processing state of the host 104, and to receive responses from the hosts 104.; Fig. 1 Hosts 104-1-N; [0036] Resources 105 in a host 104 can be varied or heterogeneous in many respects. For example, each group of resources 105 managed by a host 104 can vary in terms of processing devices (e.g., CPU, RAM, disk, network), processor type, processing speed, performance, and capabilities such as an external IP address or flash storage.); and
assigning, by the at least one [controller] processor, based on the characteristic of the machine learning task and the characteristic of the compute system, the machine learning task to at least one of the one or more compute devices ([0052] In general, each host 104 is configured to run or execute one or more tasks of a workload, including tasks of multiple different workloads that may be assigned to the host 104; [0055] In response to identifying the parameters in the request 112, the controller 108 is operable to determine an assignment scheme for scheduling and assigning tasks to hosts 104 in the cluster 102 based on parameters of the request 112 and based on a hardware socket topology of resources 105 (or resource groups 200) in a host 104. The controller 108 generates a respective task specification based on the assignment scheme for scheduling and assigning tasks to the hosts 104.).
While Wang teaches the controller being configured to perform the steps above, Wang does not explicitly define the controller as a processor.
However, Rajappa teaches a processor ([0088] one or more processors to perform operations including receiving, by a calibration module executed by the one or more processors, a calibration request including (i) a workload type, (ii) a list of compute nodes belonging to a distributed computer system, and (iii) one or more frequencies, responsive to identifying the workload type as a clustered workload type, instructing a plurality of compute nodes on the list of compute nodes to begin processing a workload of the workload type, responsive to identifying the workload type as a clustered workload type, instructing a compute node on the list of compute nodes to begin processing the workload of the workload type and responsive to beginning processing of the workload of the workload type, sampling one or more measurements of one or more components of each compute node processing the workload to collect one or more measurements, wherein the sampling the one or more measurements includes continuously measuring of a compute node on the list of compute nodes, at a predetermined time interval, one or more of a power being currently consumed by a component of the compute node, a voltage applied to a component of the compute node, a current applied to a component of the compute node, a temperature of a component of the compute node, a configuration setting of the compute node, a latency of the compute node, or a bandwidth usage of the compute node.).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of Rajappa with Wang to use a controller/scheduler comprising a processor to perform task scheduling operations. The modification would have been motivated by the desire of combining known elements to yield predictable results.
Regarding claim 2, Wang teaches wherein the characteristic of the machine learning task comprises at least one of a compatibility, priority, order, size, or type ([0029] types of tasks; [0053] For example, the controller 108 may scan the request 112 to identify parameters (e.g., protocol bits) in the request 112 that specify CPU, memory, and accelerator requirements of various tasks in the workload (304). Based on the parameters and values in the request 112, the controller 108 may determine that an example workload includes 16 tasks, where each of the tasks require a total resource allocation of 96 CPUs and 4 special- purpose circuits, e.g., hardware accelerators. For example, the request 112 can include a scalar resource parameter that specifies a quantity of hardware accelerators (4) that are to be used for executing each of the 16 tasks. In some implementations, the scalar resource parameter may include a sub-type that specifies a type of hardware accelerator to be used to process the workload. For example, the sub-type may specify that each of the 4 hardware accelerators be a neural net processor configured to accelerate running a model trained for feature recognition.; [0055] In some implementations, the request 112 may assign a priority to each of the parameters to further constraint the scheduler 108 and the controller 108. For example, priorities assigned to the accelerator sub-types or CPU cores can constraint the controller 108 to certain hosts 104 that have particular types of hardware accelerators or a particular quantity of available CPUs).
Regarding claim 3, Wang teaches wherein the characteristic of the machine learning task comprises at least one of a performance compatibility ([0053] task and performance compatibility such as a specific type of hardware accelerator for a task), efficiency compatibility, or a latency compatibility.
Regarding claim 4, Wang teaches wherein the characteristic of the compute system comprises at least one of a policy, topology, status ([0034] For example, the status logic 110 is run by the controller 108 to send commands to a subset of hosts 104 to obtain information about a processing state of the host 104, and to receive responses from the hosts 104.), operating parameter, or scheduling algorithm.
Regarding claim 5, Wang teaches wherein the characteristic of the compute system comprises at least one of a performance policy ([0028] The host is operable to bind or assign a set of machine-learning tasks of a task specification to a resource group (or control group) constructed at the host. For example, based on the information conveyed by the protocol bits, the host can bind a task to a given resource group to reduce or prevent the occurrence of non-local memory or data access operations that can be degrade performance or execution of computations for a given task.) or efficiency policy.
Regarding claim 6, Wang teaches wherein:
the characteristic of the compute system comprises a policy ([0028] The host is operable to bind or assign a set of machine-learning tasks of a task specification to a resource group (or control group) constructed at the host. For example, based on the information conveyed by the protocol bits, the host can bind a task to a given resource group to reduce or prevent the occurrence of non-local memory or data access operations that can be degrade performance or execution of computations for a given task.);
the characteristic of the machine learning task comprises: a first compatibility, based on the policy, with a first one of the one or more compute devices, and a second compatibility, based on the policy, with a second one of the one or more compute devices ([0053] For example, the controller 108 may scan the request 112 to identify parameters (e.g., protocol bits) in the request 112 that specify CPU, memory, and accelerator requirements of various tasks in the workload (304). Based on the parameters and values in the request 112, the controller 108 may determine that an example workload includes 16 tasks, where each of the tasks require a total resource allocation of 96 CPUs and 4 special- purpose circuits, e.g., hardware accelerators. For example, the request 112 can include a scalar resource parameter that specifies a quantity of hardware accelerators (4) that are to be used for executing each of the 16 tasks. In some implementations, the scalar resource parameter may include a sub-type that specifies a type of hardware accelerator to be used to process the workload. For example, the sub-type may specify that each of the 4 hardware accelerators be a neural net processor configured to accelerate running a model trained for feature recognition.; [0054] A package field of the computing logic specifies a task binary for executing each of the 16 tasks (306). For example, the task binary can be a particular type of neural network or inference model that is to be executed or run at the hardware accelerator to perform computations for executing a particular task of the 16 tasks. In some cases, the task binary is derived from the scalar resource sub-type that specifies the type of hardware accelerator to be used to process tasks of the workload.; [0063] The request 112 can be to perform a ML workload, such as an inference workload to detect an object in an image or to recognize terms in a speech utterance. In this context, one or more of the hardware accelerators may be configured to implement a neural network that includes multiple neural network layers, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). The received request 112 may include parameters that specify a particular type of neural network configuration (e.g., a CNN or RNN) that should be used to execute tasks of the workload.); and
the assigning comprises assigning, based on the policy and the first compatibility, the machine learning task to the first one of the one or more compute devices ([0009] In some implementations, performing the ML workload includes: processing instructions for the respective task specification using each resource of a control group of the host and based on data exchanged between the respective memory, the hardware accelerator, and a respective processor that is included among the resources of the host.; [0040] The determinations for assigning certain tasks to particular resource groups of a host 104 are formed with a particular focus on leveraging resource locality within a host 104 of a computing cluster 102.; [0055]; [0064]).
Regarding claim 7, Wang teaches ML workloads as cited above. In addition, Rajappa teaches wherein:
the characteristic of the compute system comprises a first policy and a second policy ([0026] In addition, the administrative policies 130 will guide the management of running the jobs 120 by providing an over-arching policy that defines the operation of the HPC system 100. Examples of policies that may be included in the administrative policies 130 include, but are not limited or restricted to, (1) maximize utilization of all hardware and software resources (e.g., instead of running fewer jobs at high power and leaving resources unused, run as many jobs as possible to use as much of the resources as possible); (2) a job with no power limit is given the highest priority among all running jobs; and/or (3) suspended jobs are at higher priority for resumption. Such administrative policies govern the way the HPC system 100 may schedule, launch, suspend and re-launch one or more jobs.)
the characteristic of the machine learning task comprises:
a first compatibility, based on the first policy, with a first one of the one or more compute devices, and a second compatibility, based on the second policy, with a second one of the one or more compute devices ([0018] Various embodiments of the disclosure relate to estimating the power performance of a job that is to be run on a distributed computer system. An estimation of the power performance of a job may be determined based on, at least in part, whether the owner of the job permits the job to be subject to a power limit, the job power policy limiting the power supplied to the job, whether the owner of the job permits the job to be suspended and/or terminated, and/or calibration data of the one or more nodes of the distributed computer system on which the job is to run. [0025] Each job includes a “power policy,” which will be discussed in-depth below. The power policy will assist the HPC system 100 in allocating power for the job and aid in the management of the one or more jobs 120 being run by the HPC system 100.; [0041] The each job requested by a user (e.g., the owner of the job) is accompanied by a user policy 205 (also illustrated in FIG. 1). The user policy includes at least a decision on whether the job 250 may be subjected to a power limit, the policy to limit the power when power limit is permitted (e.g., fixed frequency, minimum power required, or varying frequency and/or power determined by the resource manager 210), and whether the job 250 may be suspended. The user policy will be discussed in-depth below with FIG. 3.; [0042] In one embodiment, a power aware job scheduler 211 is configured to receive a selection of a mode for a job (e.g., included within the user policies 205), to determine an available power for the job based on the mode and to allocate a power for the job based on the available power); and
the assigning comprises assigning, based on the first policy and the first compatibility, the machine learning task to the first one of the one or more compute devices ([0043] The resource manager 210 uses power aware job scheduler 211 and power aware job launcher 212 to schedule and launch a job based on the received power inputs, e.g., the user policies 205 and the administrative policies 206.; [0045])
Regarding claim 8, Wang teaches wherein:
the machine learning task is a first machine learning task ([0063] The request 112 can be to perform a ML workload, such as an inference workload to detect an object in an image or to recognize terms in a speech utterance. In this context, one or more of the hardware accelerators may be configured to implement a neural network that includes multiple neural network layers, such as a convolutional neural network (CNN) or a recurrent neural network (RNN). The received request 112 may include parameters that specify a particular type of neural network configuration (e.g., a CNN or RNN) that should be used to execute tasks of the workload.);
the characteristic of the machine learning task is a first characteristic of the first machine learning task ([0053]; [0062]; [0063]); and
the assigning comprises: selecting, based on the first characteristic of the first machine learning task, a second characteristic of a second machine learning task, and a scheduling algorithm, the first machine learning task and assigning, based on the selecting, the first machine learning task to the at least one of the one or more compute devices ([0065] System 100 determines a resource requirement based on the request (404). The resource requirement can indicate certain details about resources of system 100 relative to the workload request 112, such as types and amounts of computational resources that are required to perform a suite of tasks representing the ML workload. For example, the resource requirement can specify a certain processor or processor type, processing power or speed, an amount of memory or memory size, a quantity of hardware accelerators, or a measure of resource locality for resources at the distributed system. [0066] The system 100 determines a quantity of hosts 104 that are assigned to execute a respective task of the ML workload based on the resource requirement and multiple hardware accelerators for each host (406). For each host 104 in the quantity of hosts 104, the system 100 generates a respective task specification based on a memory access topology of the host (408). The memory access topology of the host can be based on one of multiple respective NUMA topologies for each resource group 200 of the host 104.).
Regarding claim 9, Wang as cited above teaches ML workloads. In addition, Rajappa teaches wherein the machine learning task is a first machine learning task, the method further comprising:
modifying, based on a priority of the first machine learning task and a priority of a second machine learning task, an operation of the first machine learning task on the at least one of the one or more compute devices and assigning, based on the modifying, the second machine learning task to the at least one of the one or more compute devices ([0026] (2) a job with no power limit is given the highest priority among all running jobs; and/or (3) suspended jobs are at higher priority for resumption.).
Regarding claim 10, Wang teaches further comprising: determining an operating status of the at least one of the one or more compute devices and assigning, based on the operating status, the machine learning task to a data structure ([0032] In some implementations, each controller 108 includes status logic 110 that manages communications with subsets of hosts 104. For example, the status logic 110 is run by the controller 108 to send commands to a subset of hosts 104 to obtain information about a processing state of the host 104, and to receive responses from the hosts 104.; [0055]).
Regarding claim 11, Wang teaches wherein:
the at least one of the one or more compute devices comprises a first one of the one or more compute devices (Fig. 1, Hosts 104-1-N);
the characteristic of the machine learning task comprises:
a first compatibility with the first one of the one or more compute devices and a second compatibility with a second one of the one or more compute devices ([0053-54]; [0062-63]); and
the method further comprises:
determining an operating status of the first one of the one or more compute devices and assigning, based on the operating status and the second compatibility, the machine learning task to the second one of the one or more compute devices ([0032] In some implementations, each controller 108 includes status logic 110 that manages communications with subsets of hosts 104. For example, the status logic 110 is run by the controller 108 to send commands to a subset of hosts 104 to obtain information about a processing state of the host 104, and to receive responses from the hosts 104.; [0055]).
Regarding claim 12, Wang teaches wherein: the characteristic of the machine learning task comprises a size of the machine learning task ([0053-54]; [0065-66]) and the assigning comprises assigning, based on the size of the machine learning task, the machine learning task to the at least one of the one or more compute devices ([0055]).
Regarding claim 13, Wang teaches wherein the at least one of the one or more compute devices comprises a first one of the one or more compute devices (Fig. 1, Hosts 104-1-N), the method further comprising:
modifying the characteristic of the compute system, and assigning, based on the modifying, the machine learning task to a second one of the one or more compute devices ([0026]; [0034] The status logic 110 may determine that a task assigned to a host 104 has stalled if the host 104 fails to provide a status report after a threshold number of attempts to obtain information about the processing state of the host.).
Regarding claim 14, Wang teaches wherein the characteristic of the compute system comprises a policy ([0028] The host is operable to bind or assign a set of machine-learning tasks of a task specification to a resource group (or control group) constructed at the host. For example, based on the information conveyed by the protocol bits, the host can bind a task to a given resource group to reduce or prevent the occurrence of non-local memory or data access operations that can be degrade performance or execution of computations for a given task.).
Regarding claim 15, Wang teaches wherein the characteristic of the compute system comprises an operating parameter ([0034] For example, the status logic 110 is run by the controller 108 to send commands to a subset of hosts 104 to obtain information about a processing state of the host 104, and to receive responses from the hosts 104.; Fig. 1 Hosts 104-1-N; [0036] Resources 105 in a host 104 can be varied or heterogeneous in many respects. For example, each group of resources 105 managed by a host 104 can vary in terms of processing devices (e.g., CPU, RAM, disk, network), processor type, processing speed, performance, and capabilities such as an external IP address or flash storage).
Regarding claim 16, it is a system type claim having similar limitations as claim 1 above. Therefore, it is rejected under the same rationale above.
Regarding claim 17, it is a system type claim having similar limitations as claim 6 above. Therefore, it is rejected under the same rationale above.
Regarding claim 18, it is a system type claim having similar limitations as claim 7 above. Therefore, it is rejected under the same rationale above.
Regarding claim 19, it is a system type claim having similar limitations as claim 1 above. Therefore, it is rejected under the same rationale above.
Regarding claim 20, it is a system type claim having similar limitations as claim 4 above. Therefore, it is rejected under the same rationale above.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JORGE A CHU JOY-DAVILA whose telephone number is (571)270-0692. The examiner can normally be reached Monday-Friday, 6:00am-5:00pm.
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, Aimee J Li can be reached at (571)272-4169. 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.
/JORGE A CHU JOY-DAVILA/Primary Examiner, Art Unit 2195
Prosecution Insights