PTAB

IPR2025-00318

Microsoft Corp v. ParTec AG

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Application Runtime Determined Dynamical Allocation Of Heterogeneous Compute Resources
  • Brief Description: The ’442 patent relates to a method for dynamically allocating sub-tasks of a computation across a heterogeneous computing system containing both "computation nodes" and "booster nodes" (accelerators). The system adjusts the distribution of sub-tasks for subsequent computing iterations based on information generated during prior iterations to improve computational efficiency.

3. Grounds for Unpatentability

Ground 1: Claims 1-10 are obvious over Lippert in view of Budenske

  • Prior Art Relied Upon: Lippert (Application # 2013/0282787) and Budenske (The Journal of Supercomputing, Oct. 1998).
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued that Lippert disclosed a modern heterogeneous computing cluster with computation nodes and boosters for dynamic task assignment. However, Lippert lacked an explicit teaching of re-distributing tasks over multiple iterations based on runtime performance. Petitioner asserted that Budenske, a 1998 publication, supplied this missing element by teaching the core concept of processing an iterative application on a heterogeneous system and dynamically re-mapping sub-tasks between iterations. Budenske’s method used "dynamic parameters" from a preceding iteration to generate a new sub-task mapping for the next iteration specifically to reduce execution time.
    • Motivation to Combine: A POSITA would combine Budenske's decades-old iterative optimization method with Lippert's more modern cluster architecture to improve the performance of iterative applications. This combination would fully leverage the hardware flexibility described in Lippert to achieve the efficiency gains promised by Budenske, representing a predictable improvement of known elements.
    • Expectation of Success: Petitioner asserted a POSITA would have a high expectation of success because the combination only required implementing known software techniques. Furthermore, other publications available at the time confirmed that Budenske's approach was feasible and effective.

Ground 2: Claims 2-5 and 8-10 are obvious over Lippert in view of Budenske and Kambatla

  • Prior Art Relied Upon: Lippert (Application # 2013/0282787), Budenske (The Journal of Supercomputing, Oct. 1998), and Kambatla (Application # 2018/0074855).
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground reinforced the primary combination of Lippert and Budenske by adding Kambatla to provide express teachings for the specific software components recited in the dependent claims. Petitioner argued that Kambatla disclosed a resource management system for distributed clusters, analogous to Apache YARN, which explicitly detailed the use of daemons. Kambatla’s system used a central resource manager daemon (containing an application manager module) that communicated with per-node manager daemons to monitor and allocate resources. This structure directly taught the "application manager" (claim 2), the manager's role in determining distribution (claim 3), and the use of "daemons" to generate information (claims 5 and 8), which Ground 1 argued were merely within the ordinary skill of a POSITA to implement.
    • Motivation to Combine: A POSITA would combine Kambatla’s well-known, daemon-based framework with the Lippert-Budenske system as a straightforward way to implement the necessary inter-node communication, resource monitoring, and task management. This would enable the efficient execution of Budenske's dynamic re-mapping logic on Lippert's hardware, especially for managing multiple concurrent applications and optimizing resource utilization.
    • Expectation of Success: Kambatla's teachings were based on common industry frameworks (YARN), making the implementation details well-known and predictable for a POSITA.

4. Arguments Regarding Discretionary Denial

  • Petitioner argued that discretionary denial under Fintiv is improper. The petition was filed promptly (less than four months after receiving infringement contentions), and the co-pending district court case is in its earliest stages with no significant progress. Petitioner further contended that the scheduled trial date is likely to occur well after the Final Written Decision (FWD) deadline, based on the presiding judge's caseload and median time-to-trial statistics, minimizing concerns of duplicative efforts.

5. Relief Requested

  • Petitioner requests institution of an inter partes review (IPR) and cancellation of claims 1-10 of Patent 11,537,442 as unpatentable.