A method, system, and program for the non-stop modernization of legacy web systems using human-machine collaboration and AI swarm collaborative control under authority constraints.

Abstract

The present invention provides an integrated migration technology that integrates (1) dynamic schema analysis and business logic estimation under authority constraints, (2) restoration of multimodal UI / UX, (3) collaborative implementation of AI groups and real-time conflict resolution, (4) transparent security migration, (5) quality assurance for non-disruptive switching, and (6) detection of bottlenecks during operation and continuous monitoring of business KPIs. [Solution] This invention runs the old and new systems in parallel using the same input obtained from mirroring or recording / playing back production traffic, acquires the output difference and execution metrics, calculates Q_mig from UIm / APIm / Lat / Lat_CVaR, and determines the stage switching based on the m-of-n criteria of T_window and T_hold. In case of non-delivery or critical errors, it automatically rolls back and audits the basis for the decision using a hash chain + time signature. The authentication method is also migrated through transparent hashing during the initial authentication.

Description

Technical Field

[0001] The present invention relates to a method, system, and program for developers and multiple large language models (LLMs) to cooperate in an environment where server access rights are limited (e.g., only database connection, SSH not available, lack of management screen and design documents) to migrate a legacy Web system to a modern architecture without service interruption.

Background Art

[0002] Large-scale systems built with end-of-life support frameworks (e.g., legacy PHP frameworks, Classic ASP, etc.) have problems such as plaintext authentication, design deficiencies, and lack of documentation, and conventional reverse engineering and incremental refactoring involve high man-hours and downtime risks. In particular, under permission restrictions that only allow database connections such as RDS, safe migration with zero downtime has been difficult.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Non-Patent Documents

[0004]

Non-Patent Document 1

Non-Patent Document 2

[0005] The present invention provides an integrated migration technology that integrates (1) dynamic schema analysis and business logic estimation under authority constraints, (2) restoration of multimodal UI / UX, (3) collaborative implementation of AI groups and real-time conflict resolution, (4) transparent security migration, (5) quality assurance for non-disruptive switching, and (6) detection of bottlenecks during operation and continuous monitoring of business KPIs. [Means for solving the problem]

[0006] Based on database metadata analysis and data distribution analysis, relationships are estimated even in environments lacking foreign keys, and business domains are automatically classified. The extraction results are presented to developers via a human-machine collaborative UI (120), allowing for verification and correction to improve accuracy.

[0007] Screen specifications are extracted from multimodal information such as screen captures and DOM structure, and mapped to modern UI components. Screen transitions and input / output specifications are verified and consistency is ensured through differential verification.

[0008] Multiple LLM agents are coordinately controlled according to their respective roles (analysis, design, implementation, and verification), and the following processes are controlled in a loop: (i) task decomposition, (ii) parallel implementation, (iii) automatic conflict resolution, (iv) difference verification between old and new systems, and (v) phased switching based on KPI judgment (non-stop release / continuous verification).

[0009] Assuming no impact on existing users, we will gradually transition from the old authentication method to a cryptographic hashing method. This will involve transparent hashing during initial authentication, parallel operation, and unification after the complete transition.

[0010] By integrating end-to-end (E2E) differential verification, automated regression testing, and performance / consistency monitoring, and optimizing based on objective functions and constraints, we aim to stabilize migration quality and automate switchover decisions. Furthermore, in learning consensus weights, exponential updates on a simple probability system (including Mirror Descent) can be applied to achieve updates that satisfy the normalization constraints (non-negative and sum to 1).

[0011] Continuously analyze the operation data, such as activity classification by stage, failure rate, and annual trend, identify bottlenecks, and present improvement priorities. Also, continuously monitor business KPIs such as retention and LTV, detect abnormalities and changes, and conduct warnings and root cause analysis.

Effect of the Invention

[0012] According to the present invention, even under license restrictions, (i) estimation and visualization of structure and logic, (ii) restoration of multimodal UI, (iii) cooperative parallel implementation of AI groups, (iv) transparent security migration, (v) quality assurance for non-stop switching, and (vi) detection of bottlenecks during operation and KPI monitoring can be integrally realized. As a result, reduction of the stop risk, stabilization of migration quality, shortening of the migration period, and continuity of operation improvement can be expected. The present invention is not limited to merely reproducing the behavior of the old system, and allows redesign and optimization within a range that satisfies a predetermined compatibility threshold (such as Q_mig, etc.) based on observed information (database structure, operation log, UI behavior, etc.). Thereby, while reducing design constraints and technical debts derived from legacy systems, improvements in performance, maintainability, security, and user experience can be achieved, and its effectiveness is verified by differential verification of the same input and stage switching gates based on quality indicators.

Brief Description of the Drawings

[0013] [Figure 1] Overall system configuration (relationship between modules) [Figure 2] Schema analysis and human-machine cooperation UI flow [Figure 3] Non-stop orchestration and stage switching [Figure 4] Transparent security migration sequence [Figure 5] Optimization and evaluation flow of quality indicators [Figure 6] Bottleneck detection and KPI continuous monitoring flow

Mode for Carrying Out the Invention

[0014] Definition (Terms and Normalization Range) - Migration Quality Index Q_mig: A comprehensive quality index normalized to 0 - 1 (weighted synthesis of response time, data integrity, test pass rate, etc.). - UI Consistency UIm: An index of the consistency between the old and new UIs (0 - 1). - API Compatibility APIm: An index of the compatibility between the old and new APIs (0 - 1). - Delay Index Lat and Risk Index Lat_CVaR: Normalized indices of the level of delay and tail risk (0 - 1). - Threshold: A predetermined reference value, set and adjusted according to operation requirements. - Consensus Weight Vector w: A vector on a simplex where each element is non - negative and the sum is 1, representing the contribution degree of agents / indicators. - Mirror Descent: An extended method that performs the standard gradient descent method in a space based on more flexible projection / geometry (e.g., Bregman divergence), suitable for weight learning under simplex constraints (such as exponential updates). - Human - Machine Collaboration: A process in which, under rights restrictions where source code, specifications, etc. do not exist or are inaccessible, AI generates hypotheses (structures, specifications, behaviors), and humans repeatedly (and in parallel if necessary) verify and correct them in light of business context and actual operation, and gradually reconstruct the functions in actual operation. Note that general human - machine collaboration often refers to a structure where humans are in the lead and AI operates in an auxiliary manner. However, the "human - machine collaboration" in this specification focuses on (but is not limited to) a reverse collaboration structure where AI leads hypothesis generation and humans perform verification and correction based on business knowledge. - m - of - n Criterion: A criterion that requires at least N (N ≤ M) out of M observations within the decision window period T_window to meet a predetermined condition. (Supplementary Definition) - Decision Window Period T_window: The observation period used for stage - switching determination. - Hold Time T_hold: The minimum continuous time to maintain the switching permission after the permission condition is satisfied. - Critical Error: A failure event that meets specified criteria, such as data corruption, authentication failure chain, or SLA violation. - Rollback instruction: A control signal that automatically executes actions such as canceling a switchover, switching back to the old system, or returning to the previous stage. - In this specification, "simultaneous time" refers to recordings within the same transaction or event processing context, and may include a maximum allowable delay Δt (e.g., tens of ms) to absorb implementation-level clock differences. Furthermore, in this specification, "real-time" may refer to effective real-time (including the maximum allowable delay Δrt), including implementation-level processing delays. In addition, such recordings may be implemented by atomic commits within the same transaction. (Note) The above is an example for illustrative purposes only, and the present invention is not limited to specific calculation formulas or coefficient values.

[0015] (1) Schema analysis and business estimation UI: Visualizes estimated relationships, business classifications, and usage frequency estimations, and includes a workflow that developers can approve and modify.

[0016] (2) Non-stop switching: The step switching conditions are defined as a combination of Q_mig threshold satisfaction, no errors detected, regression success, etc., and parallel execution is maintained until the conditions are satisfied. The same input includes mirroring or recording and replaying of production traffic. The determination of step switching may include hysteresis including m-of-n criteria and retention time (T_hold) within T_window. Specifically, within the judgment window period T_window, if Q_mig is above a predetermined threshold and Lat_CVaR is below a predetermined upper limit, and no critical errors are detected, a predetermined number of times (or consecutive periods) of conditions are met, then a stage switch is permitted. If these conditions are not met or a critical error is detected, the system will either continue parallel operation or issue an automatic rollback instruction. Constraint verification may be performed by detecting SSH handshake failures, verifying denial responses to OS privileged APIs, monitoring for unreachable management UIs, and checking the exclusivity of DB connection credentials.

[0017] (3) Proven scripts (examples): - bottleneck-analysis.ts: Classification by activity stage, failure rate calculation, annual trend analysis, priority presentation. - business-kpi-monitor.ts: Continuous monitoring, warnings, and factor analysis of retention, usage patterns, LTV, etc. (Note) The above script is an example of demonstration, and the names and implementations are illustrative and not limiting. [Examples]

[0018] (Example 1: Migration of a small subsystem under permission constraints) Prerequisites: No administrative access to the OS / application layer; only database connection is permitted. Target: View-only screens for member management. procedure: 1) The schema analysis module (110) estimates implicit relationships and presents a business classification (approved by the human-machine collaborative interface (120)). 2) Run the old system (200) and the new system (300) in parallel, and apply the same input (400) to verify the differences (calculate UIm, APIm, Lat, and Lat_CVaR). 3) Evaluate Q_mig(182) within the judgment window period T_window(181), and allow step switching if the threshold is satisfied. Result (example): UIm approximately 0.93, APIm approximately 0.97, Lat_CVaR approximately 0.12, Q_mig approximately 0.85, satisfying the switching gate (180) and switching without interruption. The audit log (190) records the judgment and execution at the same time.

[0019] (Example 2: Automatic rollback when the switching gate is not reached) Premise: Compatibility differences are detected at the boundary cases of API responses. procedure: 1) Real-time evaluation of the output difference for the same input (400) under parallel execution conditions. 2) An automatic rollback instruction (195) is issued when Q_mig(182) is below a predetermined threshold or a critical error is detected. Result (example): Due to a decrease in APIm at some endpoints and failure to reach Q_mig, a rollback instruction (195) was automatically triggered. The old system (200) was kept in operation, and re-evaluation was performed after the correction. The trigger and its rationale were recorded in the audit log (190) as an append-only entry.

[0020] (Example 3: Transparent Security Migration) Premise: A gradual transition from traditional plaintext authentication to hashing. procedure: 1) The security migration module (150) transparently hashes and stores the data during initial authentication, maintaining parallel operation. 2) After the conversion of the predetermined percentage is completed, the system switches to the unified authentication method. Result (example): Switchover completed without service interruption. Logs were anonymized / pseudonymized (510) and aggregated to visualize the trend of failure rate and delay associated with the migration. (remarks) - The specific examples, function forms, and script names in this specification are illustrative and do not preclude substitution with equivalents. - The claims do not use fixed numerical values, but instead describe them using terms such as "predetermined threshold" or "normalization index," and the embodiments can be made more specific according to operational requirements.

[0021] (1) Indicator calculation: - Q_mig = w_ui * UIm + w_api * APIm + w_lat * (1 - Lat) + w_tail * (1 - Lat_CVaR) - However, each element of w is non-negative and its sum = 1. Lat_CVaR is the conditional expectation of the top α% (e.g., 5%) of the delay distribution. T_window is, for example, a 10-minute moving window. (2) Renewal of consensus weight: - w_{t+1} ∝ w_t · exp(η * g_t) (exponential update based on Miller descent). g_t is the contribution gradient of each index / agent. (3) Irreversibility of audit logs: - For each sequential event e_i, H_i = SHA256(H_{i-1} || e_i || ts_i) is used, and a time signature from a time authentication authority is added at predetermined intervals to ensure the integrity of the chain. (Note) The formulas and values ​​in this section are examples only and do not limit the scope of the present invention. [Industrial applicability]

[0022] This invention provides a technology for gradually migrating legacy web systems operating in environments with restricted administrative access to servers to a modern architecture without interruption. It is a software solution that can be implemented on general-purpose computers (on-premise, cloud, or hybrid). It works in conjunction with common components such as load balancers, reverse proxies, application servers, and RDBMS, and does not require special dedicated hardware. Each function (parallel execution, differential verification, quality gate, automatic rollback, audit logging, etc.) can be implemented as a module and provided as a standalone product, library / SDK, CI / CD plugin, or managed service. The program can be distributed on a storage medium or provided over a network. The method of the present invention is particularly useful in industrial sectors with low downtime tolerance and strict change management and audit requirements. Specifically, it can be applied to legacy web systems in diverse fields such as finance (accounting systems, channel systems), public / local government (resident information, tax, core business operations), healthcare (reservation, electronic medical record viewing systems), telecommunications / media (billing, membership management), distribution / e-commerce (catalogs, payment-related systems), manufacturing (bills of materials, quality records), education (academic affairs, LMS), and SaaS / platform providers (multi-tenant environments). Even in outsourced operations or external data centers where only DB connections are permitted, parallel verification based on the same input and switching control using quality indicators (Q_mig, T_window, Lat_CVaR) contribute to reducing migration risks and maintaining SLAs. Furthermore, by providing transparent authentication method migration and tamper-proof audit trails, it is also useful for addressing compliance requirements (e.g., internal controls, ISO / IEC 27001, SOC 2, etc.) required in financial institutions, medical institutions, and the public sector. Furthermore, this invention can be applied to gradual function replacement (screen-level / API-level), advanced A / B / canary distribution, bottleneck detection during operation, and continuous KPI monitoring, contributing to extending the lifespan, improving quality, and reducing operating costs of existing systems. System integrators, cloud providers, software vendors, and operational outsourcing companies can also integrate it into their own tools and services, giving it broad industrial applicability across the entire legacy system modernization market. [Explanation of symbols]

[0023] 100 Modern Orchestration (Entire System) 110 Schema Analysis Module 120-person machine collaborative interface 130 Cooperative Control Modules (LLM Group) 140 UI Recovery Module 150 Security Migration Modules 160 Quality Assurance Modules 170 Concurrent Execution Differential Verification Module 180-step adjustable gate 181 Judgment window period (T_window) 182. Transition Quality Metric (Q_mig) 183. Lat-tail risk (Lat_CVaR) 190 Audit Log (Append Only + Hash Chain + Time Signature) 191 hash chain 192 Time signature (PKI) 195 Rollback Instruction 200 Old environment 210 Managed Database (RDBMS; DB connection only) 220 Older versions of the app (SSH unavailable / no management UI) 300 New Service 400 Identical input (e.g., mirroring production traffic) 410 Output 1 (Old System) 420 Output 2 (New System) 500 logs / events / metrics 510 Anonymization / pseudonymization (PII removal) 520 Activity Classification (by Stage) 530 Failure rate calculation 540 Time Series Analysis 550 Improvement priority presentation 560 Business KPI Monitoring (LTV / Retention / Frequency) 570 Anomaly detection 580 Warning / Factor Analysis / Corrective Proposal

Claims

1. A non-disruptive migration method executed by an information processing device to migrate a legacy web system, which is running under privilege restrictions where database connections are permitted but server access privileges are limited (for example, administrative access to the OS or application layer is not possible, and design documents and management screens do not exist), to a modern architecture without interruption. The aforementioned information processing device provides identical inputs acquired by mirroring or recording and replaying production traffic to the old and new systems in a synchronized manner, running them in parallel, and collects the output of the old system, the output of the new system, and execution metrics in a parallel running process. The information processing device performs a calculation process in which it normalizes the following values ​​to 0 to 1, based on the difference between the old system output and the new system output, and the execution metrics: UI match (UIm) based on the difference between the old system and the new system's UI output such as screen captures and DOM structure; API compatibility (APIm) based on the difference between the old system's API response and the new system's API response for the same input; delay index (Lat) for the old system and the new system; and delay tail risk (Lat_CVaR) based on the conditional expected value in the top α% of the response delay distribution. The information processing device performs a synthesis step to generate a transition quality index (Q_mig) by weighting the UI match (UIm), API compatibility (APIm), delay index (Lat), and delay tail risk (Lat_CVaR) with a weight vector w in which each element is non-negative and the sum is 1. The information processing device includes a determination step in which, within a determination window period (T_window), if the transition quality index (Q_mig) is above a predetermined threshold, the delay tail risk (Lat_CVaR) is below a predetermined upper limit, and no critical errors are detected, it permits a stepwise switch to the new system. A non-stop transition method characterized by including the following.

2. A non-disruptive migration method according to claim 1, further comprising a constraint verification step for verifying that the information processing device is under the authority constraint, wherein the constraint verification step is performed by at least one of detecting an SSH connection failure, confirming a rejection response to an OS privileged API, monitoring for unreachability of the management UI, and checking the restrictiveness of database connection credentials.

3. A method for a non-stop transition according to claim 1 or 2, further comprising the step of issuing an automatic rollback instruction to cancel the switch to the new system and continue operation on the old system, or to return to the previous stage, if at least one of the predetermined threshold or the predetermined upper limit is not met or a serious error is detected within the determination window period (T_window).

4. A method for a non-stop transition according to claim 1, further comprising the steps of: the information processing device recording the determination result and its basis data in an audit log in an append-only format at the same time as the determination of whether or not to allow or deny the stage switching in the determination step; concatenating the audit log with a hash chain; and further adding a time signature to ensure difficulty in tampering.

5. A non-stop transition method according to claim 1, characterized in that a determination of stage switching is made based on an m-of-n criterion or a predetermined number of consecutive achievements, which requires that at least N (N ≤ M) of the M observations within the determination window period (T_window) satisfy the threshold condition of the transition quality index (Q_mig) and the upper limit condition of the delay tail risk (Lat_CVaR).

6. A non-stop migration method according to claim 1, further comprising the steps of: the information processing device estimating inter-table relationships and business domains even in the absence of foreign keys based on metadata or samples obtained solely from a read connection to a managed database; presenting the estimation results to a developer via a human-machine collaborative interface; and accepting approval or modification by the developer.

7. The method for non-stop migration according to claim 1, further comprising an authentication method migration step in which the information processing device hashs the new method at the time of the user's initial authentication and stores it, allows verification using either the old method or the new method during the parallel operation period, and unifies to the new method after the conversion of all accounts is complete.

8. A non-stop modernization system for legacy web systems, A schema analysis module (110) that estimates inter-table relationships and business domains based on metadata of a managed database, A human-machine collaborative interface (120) presents the aforementioned estimation results to the developer and accepts their approval or modification, A cooperative control module (130) that coordinately controls multiple LLM agents according to their respective roles, A UI restoration module (140) extracts screen specifications from multimodal information such as screen captures and DOM structure and maps them to modern UI components, A security migration module (150) that transparently migrates authentication methods, A quality assurance module (160) that evaluates migration quality through differential verification and regression testing, A concurrent execution difference verification module (170) calculates and outputs, normalized to 0 to 1, the UI match score (UIm) based on the difference in UI outputs such as screen captures and DOM structure between the old and new systems for the same input, API compatibility (APIm) based on the difference in API responses between the old and new systems for the same input, the delay index (Lat) between the old and new systems, and the delay tail risk (Lat_CVaR) based on the conditional expected value in the top α% of the response delay distribution, based on the difference between the old and new system outputs for the same input and execution metrics. A step-by-step switching gate (180) outputs permission to switch to the new system or a rollback instruction within a judgment window period (T_window) based on a transition quality index (Q_mig) obtained by weighting and combining the respective indicators output from the concurrent execution differential verification module (170), An audit log module (190) records the determination result and basis data of the switching permission or rollback instruction output from the stage switching gate (180) in an append-only format, concatenates them with a hash chain, and assigns a time signature. A system characterized by comprising the following features.

9. A method for non-stop migration according to claim 1, further comprising the steps of: collecting logs, events, and metrics anonymized or pseudonymized; calculating failure rates based on staged activity classification, performing time-series trend analysis and anomaly detection on the logs, events, and metrics; and continuously monitoring business KPIs during operation.

Images