Microsoft Corp v. ThroughPuter Inc

1. Case Identification

• Case #: IPR2022-00527
• Patent #: 9,632,833
• Filed: February 4, 2022
• Petitioner(s): Microsoft Corp.
• Patent Owner(s): ThroughPuter, Inc.
• Challenged Claims: 10, 13-15, 17-22, 24-25, 27, 29, 32-34, and 37

2. Patent Overview

• Title: Scheduling Application Instances to Processor Cores Over Consecutive Allocation Periods Based on Application Requirements
• Brief Description: The ’833 patent describes techniques for managing the parallel execution of software program tasks on an array of processor cores. The system periodically allocates cores among programs for defined "Core Allocation Periods" (CAPs) based on the capacity demands of the programs.

3. Grounds for Unpatentability

Ground 1A: Claims 15, 17-18, 34, and 37 are obvious over Agrawal in view of Brent.

• Prior Art Relied Upon: Agrawal (a 2006 ACM paper, "Adaptive Scheduling with Parallelism Feedback") and Brent (Application # 2010/0131955).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued that Agrawal taught a "two-level scheduling" system where a high-level job scheduler allocates processors to different jobs (programs) for defined time periods ("scheduling quanta"), based on "parallelism feedback" from each job's task scheduler. This feedback indicates the number of processors the job desires. Brent was argued to disclose a multi-core system with an adaptive scheduler that operates on a heterogeneous array of processor cores (e.g., custom DSPs, general DSPs, eCPUs). Brent taught that tasks can have an "affinity" for a preferred core type, and its scheduler matches tasks to their preferred cores to optimize execution. The combination applied Agrawal’s high-level scheduling framework to Brent's specific implementation of a heterogeneous core array where tasks are matched to optimal core types.
  • Motivation to Combine: Petitioner contended a POSITA would combine these references because they addressed the highly similar problem of scheduling tasks in a parallel processing environment. It was argued that Brent provided a specific, advantageous implementation (using heterogeneous cores and affinity scheduling) for the more general framework described by Agrawal. The combination would achieve predictable benefits, such as improved efficiency and reduced task completion time, by executing tasks on cores with matching instruction sets.
  • Expectation of Success: Petitioner asserted a high expectation of success, as the two-level scheduling techniques in Agrawal were fully compatible with implementation on a heterogeneous core array like Brent’s. A POSITA would have found it straightforward to augment Agrawal’s numerical "parallelism feedback" with the core-type affinity information taught by Brent, as this information was already available to the task scheduler in Brent's system.

Ground 1B: Claims 10, 13-14, 19-22, 24-25, 27, 29, and 32-33 are obvious over Agrawal and Brent in view of Feitelson.

• Prior Art Relied Upon: Agrawal (a 2006 ACM paper), Brent (Application # 2010/0131955), and Feitelson (an IBM research report, "Job Scheduling in Multiprogrammed Parallel Systems," 1997).
• Core Argument for this Ground:
  • Prior Art Mapping: This ground built on the Agrawal/Brent combination but added Feitelson to address the specific problem of assigning tasks that continue execution across successive scheduling quanta. Feitelson taught "affinity scheduling," a policy of rescheduling threads (tasks) onto the same processing element (core) on which they last ran. This practice avoids the significant overhead and performance penalties associated with task migration, such as transferring the execution context and losing the benefit of data already in the core's local cache. Petitioner argued this teaching directly maps to claim limitations requiring that a program instance included in both a present and a new allocation is kept assigned to the same core. Feitelson also taught multi-round allocation, where unallocated cores from a first round (meeting entitled shares) are re-allocated in a second round to jobs with unmet demand, mapping to other claim limitations.
  • Motivation to Combine: Petitioner argued that while Agrawal and Brent addressed initial task placement, they did not explicitly detail how to handle tasks that persist across multiple scheduling periods. Feitelson addressed this known issue directly. A POSITA would combine Feitelson’s teachings to improve the efficiency of the Agrawal/Brent system, a well-known and common optimization practice. The motivation was to avoid unnecessary switching costs, reduce processor overhead, and leverage data locality in the cache, thereby increasing the "effective scheduling quantum" as explicitly stated by Feitelson.
  • Expectation of Success: Success was argued to be highly predictable. Affinity scheduling was a common and well-understood practice long before the patent's critical date. A POSITA would have readily understood the benefits of avoiding task migration and would have been capable of implementing Feitelson’s policy in the combined Agrawal/Brent system to achieve improved performance.

4. Key Claim Construction Positions

• "instance": Petitioner argued the term "instance," recited in all independent claims, should be construed to encompass a program "task." This construction was supported by dependent claim language explicitly stating an instance can correspond to a "task." Petitioner further argued that the claims do not require multiple "instances" of a program to be instances of the same task. This broader construction was asserted to be consistent with the specification, which discloses embodiments supporting multiple task types per application program, preventing the claims from being improperly limited to a narrower preferred embodiment.

5. Arguments Regarding Discretionary Denial

• Petitioner argued against discretionary denial under Fintiv, asserting that the relevant factors heavily favored institution. The primary arguments were:
  • The co-pending district court litigation was in its infancy, with no trial date set, minimal discovery conducted, and no claim construction performed.
  • Petitioner stipulated that, if the inter partes review (IPR) is instituted, it would not pursue in the district court any invalidity ground that was raised or reasonably could have been raised in the IPR. This stipulation was intended to prevent overlap and duplicative efforts between the two proceedings.

6. Relief Requested

• Petitioner requested institution of an IPR and cancellation of claims 10, 13-15, 17-22, 24-25, 27, 29, 32-34, and 37 of the ’833 patent as unpatentable.