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 .
Response to Amendment
The amendment filed on 5/18/2026 has been entered. Claims 1-9, 11-15, and 17-22 remain pending in this application. Applicant’s amendments to claims 2-4, 8, 11-12, and 17-18 have overcome the 35 U.S.C. 112(b) rejections previously set forth in the Non-Final Office Action. Accordingly, Examiner withdraws the 35 U.S.C. 112(b) rejections of claims 2-4, 8, 11-12, and 17-18.
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.
Claim(s) 1-9, 11-15, and 17-22 are rejected under 35 U.S.C. 103 as being unpatentable over Goksoy et al. (NPL Document - "DAS: Dynamic Adaptive Scheduling for Energy-Efficient Heterogenous SOCs" hereinafter Goksoy) in view of Smith et al. (US Patent No. 8,893,130 B2 hereinafter Smith).
As per claim 1, Goksoy teaches a dynamic adaptive scheduling (DAS) computing system, comprising: a first operating system (OS) scheduler; a second OS scheduler that is slower than the first OS scheduler (Section III(A), “Overview and Preliminaries”, “Unlike the current practice, which is limited to a single scheduler, DAS allows the OS to choose one scheduling policy
π
∈
Π
S
=
F
,
S
, where F and S refer to the fast and slow schedulers, respectively. Once the predecessors of a task are completed, the OS can call either a fast
(
π
=
F
)
or a slow scheduler
(
π
=
S
)
as a function of the system state and workload.”); and a runtime preselection classifier that is operably connected to the first OS scheduler and the second OS scheduler, which runtime preselection classifier is configured to effect selective use of the first OS scheduler or the second OS scheduler to perform a given scheduling task (Section III(B), “Zero-Delay DAS Preselection Classifier”, “The OS periodically refreshes the performance counters to reflect the current system state. Each time the features are refreshed, DAS preselection classifier updates its scheduler selection which will be used for the next ready task. This decision will always be up to date since it uses the features that reflect the most recent system state. This way, DAS determines which scheduler should be called even before a task is ready for scheduling.”), wherein the runtime preselection classifier is configured to effect use of the first OS scheduler or the second OS scheduler to perform the given scheduling task (Section III(B), “Zero-Delay DAS Preselection Classifier”, “The first step of DAS is selecting the fast or slow scheduler. Since this decision is on the critical path of the fast scheduler, we must optimize it to approach our zero overhead goal. One of the novel contributions of DAS is recognizing this selection as a deterministic task that will eventually be executed with probability one.”).
Although Goksoy teaches selecting either a first OS scheduler or a second OS scheduler to perform scheduling of a given task, Goksoy fails to teach selecting a scheduler based on a number of application instances being processed.
However, Smith teaches effect use of a first scheduler or a second scheduler to perform the given scheduling task based on a number of application instances being processed (Col. 10, lines 6-35, “It is also understood and appreciated that each scheduler 112 has characteristics, such as for example, slow but very precise, or very fast and wasteful of resources. Each scheduler 112 may therefore be identified by one or more characteristics. More than once scheduler 112 may be appropriate for use in scheduling task from a given queue 106. The choice of which scheduler 112 to employ may be based on a preferential ranking of scheduler 112 characteristics or other factors, such as one scheduler being more or less busy then another scheduler…In at least one embodiment, the routing of tasks to a particular scheduler 112 is based on the characteristics of the queue 106, decision 414. For example, if the queue 106 is for near real time tasks, the scheduler 112 selected may be one that is extremely fast in how it adds tasks to the schedule, but not overly efficient in how the schedule is put together (there are gaps of unusable time), or wasteful in how the resources are used (e.g. path 504 in FIG. 5), block 418. At other times, the schedule manager 110 will route to a scheduler 112 based on the characteristics of the scheduler, block 416. Such a scheduler characteristic may include simply the current idle status of the scheduler, the efficiency of the scheduler, or other characteristics and combinations thereof.”).
Goksoy and Smith are considered to be analogous to the claimed invention because they are in the same field of task scheduling. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Goksoy with the scheduler selection process of Smith to arrive at the claimed invention. The motivation to modify Goksoy with the teachings of Smith is that scheduler selection process of Smith increases the efficiency of scheduling all tasks and is adaptable for a wide range of tasks which results in more reliable scheduling of future tasks (See Smith – Col 10, lines 45-53.).
As per claim 2, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the performance of the DAS computing system is degraded if either the first OS scheduler or the second OS scheduler is inactive, the degradation being measured using one or more performance measures selected from the group consisting of: execution time, energy-delay product (EDP), and energy consumption (Section I, “Introduction”, “The following key observations enable us to design the DAS framework that outperforms both types of schedulers taken separately. First, the scheduling is not an ordinary process that may be called in the future with some probability. Instead, it will be called with 100% certainty and use a subset of available performance counters, i.e., features used for scheduling.” Section III(A), “Overview and Preliminaries”, “The goal of the fast scheduler F is to approach the theoretically minimum (i.e., zero) scheduling overhead by making decisions in a few cycles with a minimum number of operations. In contrast, the slow scheduler S aims to handle more complex scenarios when the task wait times dominate the execution times. The goal of DAS is to outperform both underlying schedulers in terms of execution time and EDP by dynamically switching between them as a function of system state and workload.”).
As per claim 3, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system achieves an average speedup of at least 1.2 times and at least 30% lower EDP relative to the first OS scheduler when a workload complexity increases (Section IV(C), “Performance and Scheduling Overhead Analysis”, “This section compares the DAS framework with LUT (fast), ETF (slow), and ETF-ideal schedulers. ETF-ideal is a version of the ETF scheduler which ignores the scheduling overhead…DAS achieves 1.28× speedup and 37% lower EDP than LUT (i.e., first OS scheduler), when the complexity increases. In summary, DAS consistently performs better than both of the underlying schedulers, successfully adapts to the workloads at runtime, and aptly chooses between LUT and ETF to achieve low execution time and EDP.”).
As per claim 4, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system achieves an average speedup of at least about 1.2 times and at least 40% lower EDP relative to the second OS scheduler at a low data rate (Section IV(C), “Performance and Scheduling Overhead Analysis”, “This section compares the DAS framework with LUT (fast), ETF (slow), and ETF-ideal schedulers. ETF-ideal is a version of the ETF scheduler which ignores the scheduling overhead…At low data rates, DAS achieves 1.29× speedup and 45% lower EDP than ETF (i.e., second OS scheduler)…”).
As per claim 5, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the runtime preselection classifier is configured to dynamically switch between use of the first OS scheduler and the second OS scheduler for the given scheduling task as a function of a state of system resources and/or workload characteristics (Section III(A), “Overview and Preliminaries”, “The OS collects a set of performance counters during the workload execution to enable two aspects for the DAS framework: 1) precise assessment of the system state and 2) desirable features for the classifier to dynamically switch between the fast (i.e., first) and slow (i.e., second) schedulers.”).
As per claim 6, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system comprises a heterogeneous computing system (Section V, “Conclusion”, “In this letter, we presented a DAS framework that combines the benefits of fast and sophisticated schedulers for heterogeneous SoCs.”).
As per claim 7, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system is implemented in a system that comprises scheduling algorithms comprising operating system kernels and a runtime software environment (Section III(B), “Zero-Delay DAS Preselection Classifier”, “The OS periodically refreshes the performance counters to reflect the current system state. Each time the features are refreshed, DAS preselection classifier updates its scheduler selection which will be used for the next ready task. This decision will always be up to date since it uses the features that reflect the most recent system state.” Section III(C), “Fast and Slow (Sophisticated) (F&S) Schedulers”, “The DAS framework can work with any choice of fast and slow scheduling algorithms. This work uses a LUT implementation as the fast scheduler since the goal of the fast scheduler is to achieve almost zero overhead. The LUT stores the most energy-efficient processor in the system for each known task in the target domain.”).
As per claim 8, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system achieves a scheduling overhead comprising less than 5 nJ energy and less than 10 ns runtime for a first workload and less than 30 nJ energy and less than 70 ns runtime for a second workload, the second workload being heavier than the first workload (Section I, “Introduction”, “The major contributions of this work are as follows. 1) The DAS framework that dynamically combines two schedulers and outperforms each of them. 2) Low Scheduling Overhead: 4.2 nJ energy and 6 ns runtime for low to medium loads; 27.2 nJ energy and 65 ns runtime for heavy workloads.” See also “Conclusion”.).
As per claim 9, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system comprises a processor and a memory communicatively coupled to the processor, the memory storing non-transitory computer executable instructions which, when executed by the processor, perform operations comprising: using the runtime preselection classifier to effect the selective use of the first OS scheduler or the second OS scheduler to perform the given scheduling task (Section III(B), “Zero-Delay DAS Preselection Classifier”, “At runtime, a background process periodically updates a preallocated local memory with a small subset of performance counters required by the classifier. After each update, the classifier determines whether the fast F or slow S scheduler should be used for the next available task.” Section III(C), “Fast and Slow (Sophisticated) (F&S) Schedulers”, “The DAS framework can work with any choice of fast and slow scheduling algorithms. This work uses a LUT implementation as the fast scheduler since the goal of the fast scheduler is to achieve almost zero overhead. The LUT stores the most energy-efficient processor in the system for each known task in the target domain. Unknown tasks are mapped to the next available CPU core.”).
As per claim 11, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the first OS scheduler comprises a scheduling overhead having less than 10 nJ energy and less than 10 nanoseconds of runtime (Section III(C), “Fast and Slow (Sophisticated) (F&S) Scheduler”, “To profile the scheduling overhead, we developed an implementation using C with inline assembly code. Experiments show that our fast scheduler takes ~7.2 cycles (6 ns on Arm Cortex-A53 at 1.2 GHz) on average and incurs negligible (2.3 nJ) energy overhead.”).
As per claim 12, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the second OS scheduler comprises a scheduling overhead having more than 10 nJ energy and more than 10 nanoseconds of runtime (See Table 1, Fig. 2, and Fig. 3. Section I, “Introduction”, “The major contributions of this work are as follows. 1) The DAS framework that dynamically combines two schedulers and outperforms each of them. 2) Low Scheduling Overhead: 4.2 nJ energy and 6 ns runtime for low to medium loads; 27.2 nJ energy and 65 ns runtime for heavy workloads.”).
As per claim 13, Goksoy and Smith teach the system of claim 1. Goksoy also teaches wherein the DAS computing system comprises a heterogeneous systems-on-chip (SoCs), a high-performance computing system, and/or an embedded device (Section V, “Conclusion”, “In this letter, we presented a DAS framework that combines the benefits of fast and sophisticated schedulers for heterogeneous SoCs.”).
As per claim 14, Goksoy and Smith teach the system of claim 13. Goksoy also teaches wherein the heterogeneous SoC comprises a domain-specific SoC (DSSoCs) (Section IV(A), “Simulation Environment: We use DS3 [8], an open-source DSSoC simulation framework, for the detailed evaluation of DAS. DS3 includes built-in scheduling algorithms, models for PEs, interconnect, and memory systems. The framework has been validated with Xilinx Zynq ZCU102 and Odroid-XU3. DSSoC Configuration: We construct a DSSoC configuration that comprises clusters of general-purpose cores and hardware accelerators. The application domains used in this study are wireless communications and radar systems. The DSSoC used in our experiments uses the Arm big.LITTLE architecture with four cores each.”).
As per claim 15, Goksoy teaches a method of scheduling a runtime task in a heterogeneous multi-core computing system, the method comprising using a runtime preselection classifier of the heterogeneous multi-core computing system to effect selective use of a first operating system (OS) scheduler or a second OS scheduler that is slower than the first scheduler to perform a given scheduling task, thereby scheduling the runtime task in the heterogeneous multi-core computing system (Section III(A), “Overview and Preliminaries”, “Unlike the current practice, which is limited to a single scheduler, DAS allows the OS to choose one scheduling policy
π
∈
Π
S
=
F
,
S
, where F and S refer to the fast and slow schedulers, respectively. Once the predecessors of a task are completed, the OS can call either a fast
(
π
=
F
)
or a slow scheduler
(
π
=
S
)
as a function of the system state and workload.” Section III(B), “Zero-Delay DAS Preselection Classifier”, “The OS periodically refreshes the performance counters to reflect the current system state. Each time the features are refreshed, DAS preselection classifier updates its scheduler selection which will be used for the next ready task. This decision will always be up to date since it uses the features that reflect the most recent system state. This way, DAS determines which scheduler should be called even before a task is ready for scheduling.”), wherein the runtime preselection classifier is configured to effect use of the first OS scheduler or the second OS scheduler to perform the given scheduling task (Section III(B), “Zero-Delay DAS Preselection Classifier”, “The first step of DAS is selecting the fast or slow scheduler. Since this decision is on the critical path of the fast scheduler, we must optimize it to approach our zero overhead goal. One of the novel contributions of DAS is recognizing this selection as a deterministic task that will eventually be executed with probability one.”).
Although Goksoy teaches selecting either a first OS scheduler or a second OS scheduler to perform scheduling of a given task, Goksoy fails to teach selecting a scheduler based on a number of application instances being processed.
However, Smith teaches effect use of a first scheduler or a second scheduler to perform the given scheduling task based on a number of application instances being processed (Col. 10, lines 6-35, “It is also understood and appreciated that each scheduler 112 has characteristics, such as for example, slow but very precise, or very fast and wasteful of resources. Each scheduler 112 may therefore be identified by one or more characteristics. More than once scheduler 112 may be appropriate for use in scheduling task from a given queue 106. The choice of which scheduler 112 to employ may be based on a preferential ranking of scheduler 112 characteristics or other factors, such as one scheduler being more or less busy then another scheduler…In at least one embodiment, the routing of tasks to a particular scheduler 112 is based on the characteristics of the queue 106, decision 414. For example, if the queue 106 is for near real time tasks, the scheduler 112 selected may be one that is extremely fast in how it adds tasks to the schedule, but not overly efficient in how the schedule is put together (there are gaps of unusable time), or wasteful in how the resources are used (e.g. path 504 in FIG. 5), block 418. At other times, the schedule manager 110 will route to a scheduler 112 based on the characteristics of the scheduler, block 416. Such a scheduler characteristic may include simply the current idle status of the scheduler, the efficiency of the scheduler, or other characteristics and combinations thereof.”).
Goksoy and Smith are considered to be analogous to the claimed invention because they are in the same field of task scheduling. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Goksoy with the scheduler selection process of Smith to arrive at the claimed invention. The motivation to modify Goksoy with the teachings of Smith is that scheduler selection process of Smith increases the efficiency of scheduling all tasks and is adaptable for a wide range of tasks which results in more reliable scheduling of future tasks (See Smith – Col 10, lines 45-53.).
As per claim 17, it is a method claim comprising similar limitations to claim 11, so it is rejected for similar reasons.
As per claim 18, it is a method claim comprising similar limitations to claim 12, so it is rejected for similar reasons.
As per claim 19, Goksoy and Smith teach the method of claim 15. Goksoy also teaches wherein the method comprises: generating an oracle; selecting one or more features; and, training a model for the runtime preselection classifier (Section III(B), “Zero-Delay DAS Preselection Classifier”, “Offline Classifier Design: The first step to designing the preselection classifier is generating the training data based on the domain applications known at design time. Each scenario in the training data consists of concurrent applications and their respective data rates (e.g., a combination of WiFi transmitter and receiver chains, at a specific upload and download speed).” See also Fig. 1 which is a flowchart describing the flow of the Das framework: oracle generation, feature selection, and training a model for the classifier.).
As per claim 20, it is a computer readable media claim comprising similar limitations to claim 15, so it is rejected for similar reasons.
As per claim 21, Goksoy and Smith teach the system of claim 1. Goksoy teaches wherein the DAS computing system achieves at least 39% lower energy-delay product (EDP) relative to the first OS scheduler, and wherein the DAS computing system achieves at least 46% lower EDP relative to the second OS scheduler (Section IV(C), “Performance and Scheduling Overhead”, “For workloads 1 and 2, the SoC is not congested at low data rates. Hence, DAS performs similar to LUT. As data rates increase, DAS aptly chooses between LUT and ETF at runtime. Its execution time and EDP are 14% and 15% lower than LUT, and 15% and 42% lower than ETF. For workload-3, the execution time and EDP of ETF are significantly higher than LUT. DAS chooses LUT for >99% of the decisions and closely follows its trend. This study is extended to all 40 workloads. At low data rates, DAS achieves 1.29× speedup and 45% lower EDP than ETF, and 1.28× speedup and 37% lower EDP than LUT, when the complexity increases. In summary, DAS consistently performs better than both of the underlying schedulers, successfully adapts to the workloads at runtime, and aptly chooses between LUT and ETF to achieve low execution time and EDP.” Examiner Note: Although the above citation indicates 37% and 45% lower EDP with respect to each OS scheduler, para. 0009 of the specification of the instant application lists a range of percentages that the phrase “at least” is inclusive of and both 37% and 45% are included in the respective lists of lower EDP. Since the percentages indicated in Goksoy are within the margins of the specification, Goksoy teaches the above claim limitations.).
As per claim 22, Goksoy and Smith teach the system of claim 1. Goksoy teaches wherein the Das computing system achieves a scheduling overhead comprising: less than 4.1 nJ energy; and a runtime less than 5 ns (See Table 1, Fig. 2, and Fig. 3. Section I, “Introduction”, “The major contributions of this work are as follows. 1) The DAS framework that dynamically combines two schedulers and outperforms each of them. 2) Low Scheduling Overhead: 4.2 nJ energy and 6 ns runtime for low to medium loads;” Examiner Note: Although the above citation indicates 4.2 nJ energy and a runtime of 6 ns, para. 0010 of the specification of the instant application lists a range of percentages that the phrase “at least” is inclusive of and both 4.2 nJ and 6 ns runtime are included in the respective lists. Since the measurements indicated in Goksoy are within the margins of the specification, Goksoy teaches the above claim limitations.).
Response to Arguments
Applicant’s arguments with respect to claim(s) 1-9, 11-15, and 17-20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Applicant has amended the claims with new limitations that change the scope of the claimed invention. Therefore, the amended claims necessitate new rejections, as addressed above. The amended claims are not allowable over prior art cited previously along with an additional reference, necessitated by amendment, for reasons indicated above.
Conclusion
Applicant's amendment necessitated the new ground(s) 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 JOHN ROBERT DAKITA EWALD whose telephone number is (703)756-1845. The examiner can normally be reached Monday-Friday: 9:00-5:30 ET.
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, Lewis Bullock can be reached at (571)272-3759. 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.
/J.D.E./Examiner, Art Unit 2199
/LEWIS A BULLOCK JR/Supervisory Patent Examiner, Art Unit 2199
Prosecution Insights