New York Patent of the Month – January 2026

A Comprehensive Research Report on the “New York Patent of the Month” (January 2026)

Introduction: The Static Nature of the “Fluid” Cloud

The evolution of cloud computing has been defined by a relentless march toward abstraction. In the early 2000s, virtualization decoupled the operating system from bare metal, allowing for the consolidation of servers and the birth of the modern data center. In the 2010s, containerization—spearheaded by Docker and orchestrated by Kubernetes—further decoupled the application from the operating system, creating a lightweight, portable unit of software delivery. This “Cloud Native” revolution promised a world where software could run anywhere, moving seamlessly between development laptops, on-premise servers, and public cloud instances.

However, a critical limitation remained: while the deployment of containers became fluid, the execution of containers remained static. Once a containerized process starts running on a specific host, it is tethered to that host’s kernel and network identity until it terminates. If the host requires maintenance, or if a cheaper spot instance becomes available elsewhere, the standard orchestrator’s only recourse is to kill the process and restart it on a new node. This “kill-and-restart” paradigm, while acceptable for stateless web servers, is fundamentally incompatible with the growing demand for long-running, stateful workloads such as large language model (LLM) training, real-time gaming, and in-memory databases.

On January 6, 2026, the United States Patent and Trademark Office (USPTO) issued US Patent 12,517,715, titled “Methods and systems for instantiating and transparently migrating executing containerized processes”. Assigned to Exostellar, Inc. and invented by the renowned Cornell University research team of Zhiming Shen, Hakim Weatherspoon, and Robbert Van Renesse, this intellectual property represents the technological breakthrough required to solve the “statefulness” problem.

Recognized by Swanson Reed as the “New York Patent of the Month” for January 2026, this patent describes a system that enables the “live migration” of containerized processes. This capability allows a running application to be paused, serialized, transferred across a network, and resumed on a different physical machine—potentially in a different cloud provider’s data center—without dropping active network connections or losing memory state. This report provides an exhaustive analysis of Patent 12,517,715, exploring its technical architecture, its disruptive impact on the FinOps and AI sectors, and its qualification for Research & Experimentation (R&D) Tax Credits under Internal Revenue Code (IRC) Section 41.

---

The Innovation: Technical Anatomy of Patent 12,517,715

To appreciate the magnitude of the invention, one must first dismantle the prevailing architecture of container runtime environments and identify the specific friction points that Patent 12,517,715 eliminates. The patent does not merely propose an incremental improvement to container orchestration; it introduces a fundamental architectural “shim” that redefines the relationship between a process and the host operating system.

The “Shim” Architecture: Decoupling State from Host

In a standard Linux container environment (e.g., Docker runc), the containerized process interacts directly with the host kernel via system calls (syscalls). The kernel manages the process’s memory pages, file descriptors, and network sockets. Because these resources are identified by host-specific integers and memory addresses, they are non-transferable. Moving the process to another host would render these pointers invalid, causing the application to crash.

Patent 12,517,715 introduces a “Shim Process”—a virtualization layer that sits in userspace between the containerized application and the host kernel. This Shim intercepts all system calls made by the application. Instead of allowing the application to see the “real” host resources, the Shim provides virtualized handles.

• Virtualization of Identifiers: When the application requests to open a file or create a socket, the Shim records the request in a private translation table and returns a virtual file descriptor. The application operates in a “Matrix-like” simulation, unaware that its reality is being mediated.
• State Tracking: Because the Shim mediates every interaction, it maintains a perfect, implementation-independent record of the application’s state. This includes the instruction pointer (where the CPU is executing), the memory map (what data is in RAM), and the status of all open I/O channels.
• Userspace Execution: Crucially, this architecture operates entirely in userspace. Unlike legacy virtual machine migration (e.g., VMware vMotion), which requires deep integration with the hypervisor and moves the entire Guest OS, the Exostellar approach migrates only the application process and its immediate dependencies. This results in a migration footprint that is orders of magnitude smaller and faster.

The “Holy Grail”: Transparent Network Migration

The most technically challenging aspect of live migration—and the primary focus of the patent’s claims regarding “transparency”—is the preservation of established Transmission Control Protocol (TCP) connections. A TCP connection is uniquely identified by a 4-tuple: (Source IP, Source Port, Destination IP, Destination Port). In a standard network environment, moving a process to a new physical host invariably changes its Source IP address. This change invalidates the 4-tuple, causing the remote client (e.g., a user’s browser or an API consumer) to reset the connection.

Patent 12,517,715 solves this through a combination of Network Namespace Isolation and Software-Defined Networking (SDN) Tunneling.

1. Namespace Isolation: The container retains its original IP address inside its private network namespace, regardless of the physical host it resides on. The application continues to send packets with the old Source IP.
2. Packet Interception and Tunneling: The Shim on the new host captures these outgoing packets. It encapsulates them (tunneling) and routes them back to the original network gateway or uses a virtual switch to update the routing tables dynamically.
3. TCP Repair Mode: The system utilizes advanced Linux kernel features (likely TCP_REPAIR options) to freeze the socket state—including sequence numbers and window sizes—during the migration window. When the process resumes on the destination, the socket is “hydrated” with the exact sequence numbers, ensuring that the next packet sent is perfectly synchronized with what the client expects.

The Migration Workflow: Pre-Copy vs. Post-Copy

The patent describes methods for minimizing the “freeze time” (the duration the application is unresponsive during migration). The system employs an iterative Pre-Copy algorithm.

• Phase 1: Initialization: The migration target is selected.
• Phase 2: Iterative Memory Transfer: The system copies the memory pages from the source to the destination while the application continues to run. Since the application is running, it “dirties” (modifies) pages that have already been copied.
• Phase 3: Delta Tracking: The system tracks these dirty pages and performs subsequent copy rounds, transferring only the modified data. This continues until the rate of dirtying drops below a threshold or a deadline is reached.
• Phase 4: Stop-and-Copy: The application is briefly paused. The final set of dirty pages and the CPU register state are transferred. This phase is typically sub-second.
• Phase 5: Resumption: The application resumes on the destination.

This mechanism allows multi-gigabyte applications to be migrated with only milliseconds of perceptible downtime, a critical requirement for real-time services.

---

The Swanson Reed Award: “New York Patent of the Month” (January 2026)

Swanson Reed, a leading specialist R&D tax advisory firm, designates a “Patent of the Month” to highlight innovations that exemplify technical rigor, economic potential, and regional excellence. The selection of US Patent 12,517,715 for January 2026 underscores the resurgence of New York State—specifically the corridor connecting Cornell University in Ithaca to the tech ecosystem of New York City—as a hub for deep technology and systems engineering.

Selection Criteria and Significance

The award is not merely a badge of honor; it reflects a detailed assessment of the patent’s quality based on Swanson Reed’s proprietary inventionINDEX and evaluation metrics.

• Metric 1: High-Bar Innovation: As noted in Swanson Reed’s criteria, the bar for selection is “set extraordinarily high,” favoring inventions with “exceptional novelty”. In a crowded field of cloud patents, 12,517,715 stands out because it solves a problem (live process migration) that was widely considered solved only by heavy virtualization (VMs), proving that lightweight containers could achieve the same resilience.
• Metric 2: Economic Viability: The patent addresses the multi-billion dollar market of cloud optimization. By enabling the use of transient “Spot” instances for stateful workloads, the invention offers a direct path to reducing global cloud expenditures by 50-80%.
• Metric 3: Societal Impact (Sustainability): The ability to pack workloads tightly and move them dynamically allows for better data center utilization. This “defragmentation” of the cloud reduces the number of active servers required, directly lowering the energy, carbon, and water footprint of the digital economy.

The Cornell-New York Connection

The patent lists Zhiming Shen, Hakim Weatherspoon, and Robbert Van Renesse as inventors. All three are associated with Cornell University’s Department of Computer Science.

• Robbert Van Renesse is a distinguished researcher in distributed systems, known for his work on the Amoeba operating system and gossip protocols.
• Hakim Weatherspoon is a professor whose work focuses on the “SuperCloud” concept—a cloud-of-clouds that abstracts away the underlying provider.
• Zhiming Shen (the lead inventor) conducted this research as part of his doctoral work, which later spun out into the startup Exostellar (formerly Exotanium).

This trajectory—from academic hypothesis to federally funded research (NSF grants often underpin such work) to patented commercial application—is exactly the lifecycle that the R&D Tax Credit is designed to support. Swanson Reed’s recognition highlights this successful translation of academic theory into industrial practice.

---

Competitive Landscape: The Battle for Workload Mobility

The domain of workload orchestration is fiercely competitive, dominated by tech giants and open-source foundations. However, Patent 12,517,715 carves out a unique “Blue Ocean” by addressing the gap between the “heavy but mobile” Virtual Machines and the “light but static” Containers.

Kubernetes (The Incumbent Standard)

Kubernetes (K8s) is the de facto standard for container orchestration. Its philosophy, however, treats containers as “cattle, not pets”—ephemeral entities that can be destroyed and replaced at any time.

• Limitation: When a K8s node fails or needs draining, the scheduler terminates the pod on Node A and starts a new pod on Node B. For stateful applications (like a PostgreSQL database or a Minecraft server), this “restart” is traumatic. It involves a crash recovery process, cache warming, and a severed client connection.
• Patent Difference: Exostellar’s technology allows the “pet” to be moved without dying. It provides K8s with a capability it natively lacks: true live migration.

VMware vMotion (The Legacy Giant)

VMware’s vMotion technology has allowed for the live migration of Virtual Machines (VMs) for two decades.

• Limitation: vMotion migrates the entire Guest Operating System (OS), including the kernel and all background processes. Moving a 64GB VM over the network is bandwidth-intensive and slow (minutes). It suffers from “data gravity.”
• Patent Difference: Patent 12,517,715 migrates only the application process and its memory pages. This “process-level” migration is granular and lightweight, often requiring 10x to 100x less bandwidth than a full VM migration.

CRIU (The Open Source Component)

Checkpoint/Restore In Userspace (CRIU) is a Linux tool that can freeze a running application and save it to a disk file.

• Limitation: CRIU is a low-level utility, not a migration system. It freezes the process but does not natively handle the complex network handover (TCP repair) required to move that process to a different machine without breaking connections. It also struggles with external resources like GPU contexts.
• Patent Difference: The patent describes a complete system (the Shim architecture) that wraps CRIU-like checkpointing in a network virtualization layer. It transforms the raw capability of checkpointing into the seamless experience of transparent migration.

Spot.io (The Economic Rival)

Spot.io (acquired by NetApp) is a market leader in Spot Instance management.

• Limitation: Spot.io relies on predictive algorithms to guess when a Spot instance will be interrupted. When it predicts an interruption, it spins up a new node and “drains” the old one. It relies on the application handling a “graceful shutdown.” If the application cannot shut down and restart in under 2 minutes, data is lost.
• Patent Difference: Exostellar does not need to predict the future perfectly. Because it can migrate the running state, it can react to the actual interruption signal (the “2-minute warning” from AWS) and move the workload instantly, preserving the exact point of execution.

Feature	Kubernetes (Standard)	VMware vMotion	Spot.io	Exostellar (Patent 12,517,715)
Migration Unit	Pod (Restart Only)	Full VM (OS + App)	Node Replacement	Process/Container (Live)
State Preservation	No (Crash/Restart)	Yes	No (Graceful Shutdown)	Yes (Exact Execution State)
Network Continuity	Broken (New IP)	Preserved	Broken	Preserved (Transparent)
Bandwidth Cost	Low (Image Pull)	Very High	Low	Low/Medium (Memory Only)
Spot Compatibility	Stateless Only	High Cost	Batch Jobs Only	Stateful & AI Workloads

---

Real-World Impact: The “SuperCloud” and AI Economics

The commercial implications of Patent 12,517,715 extend far beyond technical novelty. The invention enables the “SuperCloud” vision—a computing environment where workloads flow like water to the lowest-cost or highest-performance hardware available, regardless of which vendor owns it.

The “Spot Arbitrage” Opportunity

Cloud providers like AWS, Google Cloud, and Azure sell excess capacity (“Spot Instances” or “Preemptible VMs”) at a 60% to 90% discount compared to on-demand pricing. The catch is volatility: the provider can reclaim these instances with very short notice (typically 30 seconds to 2 minutes).

• Before the Patent: Companies could only run interruption-tolerant, stateless batch jobs on Spot instances. Mission-critical databases, API servers, and long-running simulations had to run on expensive On-Demand instances to avoid the risk of termination.
• After the Patent: The Exostellar system allows these sensitive workloads to run on Spot instances safely. When the interruption signal is received, the system automatically “teleports” the running container to a new Spot instance (or a safe On-Demand fallback).
• Financial Impact: For a mid-sized SaaS company spending $10 million annually on cloud compute, shifting 50% of their stable workloads to Spot instances using this technology could yield $3-4 million in annual savings directly to the bottom line.

AI Training and GPU Scarcity

In the era of Generative AI, the training of Large Language Models (LLMs) represents a massive capital expenditure. These training runs often span weeks or months on clusters of thousands of GPUs (e.g., NVIDIA H100s).

• The Problem: Hardware failure is a statistical certainty at scale. If a node fails during a training run, the process usually crashes. While checkpoints are saved to disk, restoring from a disk checkpoint can take hours of “re-computation” and data loading. Furthermore, high-end GPUs are often unavailable in specific regions due to demand.
• The Solution: Patent 12,517,715 enables “Checkpoints on the Fly.” It allows the training job to be paused and migrated from a region with high energy costs or low availability to a region with surplus capacity. This supports a “Follow-the-Sun” model, where the AI training job physically moves around the globe to follow renewable energy availability or lower electricity rates.

Benchmarks and Performance

Research benchmarks conducted by the inventors at Cornell (published in ACM transactions) demonstrate the efficacy of the underlying technology:

• Migration Downtime: The “stop-the-world” phase is consistently measured between 50ms and 500ms, depending on the memory footprint and the “dirty rate” of the application. This is well within the TCP timeout window (typically 60 seconds), ensuring clients do not disconnect.
• Overhead: The “Shim” introduces a runtime CPU overhead of less than 5% for most workloads. This is a negligible price to pay for the insurance of mobility.
• Cross-Cloud Capability: The system has been successfully demonstrated migrating workloads between AWS and Google Cloud Platform, proving the viability of true multi-cloud redundancy.

---

R&D Tax Credit Eligibility: A Swanson Reed Analysis

One of the primary objectives of this report is to analyze the eligibility of the development work behind Patent 12,517,715 for the Research & Experimentation (R&D) Tax Credit (IRC Section 41). Swanson Reed, as a specialist advisory firm, utilizes a rigorous methodology to substantiate such claims. The development of the “Shim” architecture and the migration algorithms serves as a textbook example of a Qualified Research Activity (QRA).

To qualify for the credit, the research activities must satisfy all four prongs of the IRS Four-Part Test.

Test 1: Permitted Purpose

The Requirement: The activity must relate to a new or improved business component (product, process, computer software, technique, formula, or invention) held for sale, lease, or license, or used by the taxpayer in its trade or business. The intent must be to improve functionality, performance, reliability, or quality.

Application to Patent 12,517,715:

Exostellar’s development of the migration platform was clearly aimed at creating a new software product (the Exostellar Platform) for commercial sale. The specific improvements sought were:

• Reliability: Preventing data loss during instance termination.
• Functionality: Enabling live migration capabilities that did not previously exist for containerized workloads.
• Performance: Reducing the downtime of migration to sub-second intervals compared to the minutes required for cold restarts.

Note on Internal Use Software (IUS): While IUS regulations are strict, software developed to be sold as a service (SaaS) or licensed to third parties (as Exostellar intends) is generally exempt from the “High Threshold of Innovation” additional tests. However, even if deemed IUS, the patent’s groundbreaking nature would likely satisfy the higher threshold.

Test 2: Technological in Nature

The Requirement: The research must fundamentally rely on principles of the hard sciences—physical or biological sciences, engineering, or computer science. Activities based on economics, social sciences, or market research are excluded.

Application to Patent 12,517,715:

The development process was deeply rooted in Computer Science and Systems Engineering. The engineers had to leverage advanced principles of:

• Kernel Architecture: Understanding Linux namespaces, cgroups, and system call handling.
• Network Engineering: Manipulating TCP stacks, IP routing, and packet encapsulation.
• Memory Management: Implementing copy-on-write mechanisms and page tracking algorithms. The work required technical expertise in C++, Go, and Assembly language, satisfying this test unequivocally.

Test 3: Elimination of Uncertainty

The Requirement: At the outset of the project, there must be uncertainty regarding the capability to develop the component, the method of development, or the appropriate design of the component. Uncertainty exists if the information available to the taxpayer does not establish how to achieve the result or if the result is even possible.

Application to Patent 12,517,715:

The team faced significant technical uncertainty:

• Capability Uncertainty: “Is it possible to decouple a running process from the host kernel entirely in userspace without severe performance degradation?”
• Methodological Uncertainty: “How do we handle the restoration of a TCP socket that is in a CLOSE_WAIT or FIN_WAIT2 state? Can standard Linux APIs support this, or do we need to patch the kernel?”
• Design Uncertainty: “Should the Shim be implemented as a library loaded via LD_PRELOAD (high compatibility, low security) or as a ptrace-based monitor (high security, high overhead)?”

The mere existence of competing solutions (like vMotion) does not negate uncertainty if the taxpayer is attempting a fundamentally different approach (container-level vs. OS-level).

Test 4: Process of Experimentation

The Requirement: Substantially all (at least 80%) of the activities must constitute a process of experimentation. This involves the identification of uncertainty, the formulation of hypotheses, the testing of alternatives, and the refinement of the design.

Application to Patent 12,517,715: The research papers published by the inventors document a classic scientific method approach:

• Hypothesis: “A userspace shim can virtualization system calls with less than 10% overhead.”
• Experiment 1 (Prototype): Building “ElasticDocker.”
• Analysis: Initial benchmarks showed high overhead on I/O-intensive workloads.
• Refinement: The team optimized the interception logic, perhaps switching from ptrace to a seccomp-bpf filter to reduce context switches.
• Re-testing: Running standard benchmarks (SPECjbb, TPC-W) to validate the improvement.
• Alternatives Investigated: The team likely evaluated and discarded alternative approaches, such as modifying the Docker daemon directly or using nested virtualization, before settling on the Shim architecture.

Documentation and Audit Defense (Swanson Reed Advisory)

To substantiate these claims during an IRS audit, Swanson Reed advises that companies like Exostellar maintain robust contemporaneous documentation.

• Technical Nexus: Linking financial records (wages, cloud costs) to specific technical projects (e.g., “Project Shim Optimization”).
• Evidence of Experimentation: Git commit logs, JIRA tickets labeled “Research” or “Spike,” architecture design documents (versions 1, 2, and 3), and test failure logs.
• The Patent as Evidence: While a patent is not definitive proof of R&D eligibility, it strongly supports the arguments for “Technological in Nature” and “Elimination of Uncertainty.” It serves as a public declaration of the novelty and technical complexity of the work.

Qualified Research Expenses (QREs)

Under Section 41, Exostellar can claim:

1. Wages: The taxable W-2 wages of the engineers (Shen, Weatherspoon, et al.) and product managers directly involved in the research.
2. Supplies: The cost of cloud compute resources (AWS EC2 instances, Google Cloud VMs) used specifically for development and testing environments. (Production hosting is excluded).
3. Contract Research: 65% of the fees paid to external US-based contractors or research labs (potentially including payments to Cornell University for sponsored research).

---

Future Outlook: The Commoditization of Compute

US Patent 12,517,715 is a foundational building block for the next era of cloud computing. By solving the technical problem of state mobility, it solves the economic problem of vendor lock-in.

• The “Priceline” of Cloud: In the future, companies may not buy “AWS instances” or “Azure VMs.” They will buy “Compute Capacity” from a broker. The broker (using technology like Exostellar’s) will dynamically place the workload on whichever provider is offering the cheapest spot price at that exact second.
• The End of “Regions”: Concepts like “us-east-1” will become irrelevant abstractions. Workloads will simply exist in the “SuperCloud,” flowing globally to follow capacity, energy, and cost efficiencies.
• Regulatory Impact: As data sovereignty laws (GDPR) tighten, this technology allows for precise, policy-based migration. A system could automatically migrate a container out of a US data center and into a German data center the moment it detects it is processing data belonging to a German citizen.

Final Thoughts

The recognition of US Patent 12,517,715 as the “New York Patent of the Month” is a testament to the thriving intersection of academic research and commercial innovation in New York State. This patent does not merely describe a clever software trick; it describes a fundamental re-architecture of the cloud operating model.

For the engineering community, it offers a solution to the decade-old trade-off between stability and cost. For the finance office, it unlocks the massive savings of the Spot market. And for the tax professional, it represents the ideal case study of a Qualified Research Activity—a high-risk, high-reward technical endeavor that pushes the boundaries of computer science to drive economic growth. As the cloud market matures from a collection of walled gardens into a fluid, interoperable utility, the technology described in Patent 12,517,715 will likely be viewed as a critical turning point in that evolution.