A Collusion-Resistant, Fault-Tolerant Data Aggregation Method and System Based on Dual Servers
By using a dual-server architecture for independent tag generation and XOR reconstruction, the problem of tag leakage and result forgery caused by collusion between data providers and computing servers is solved. This achieves low-overhead, fault-tolerant data aggregation and verifiable results, ensuring data privacy and system security.
Patent Information
- Application Number
- CN202611470849.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-21
- Publication Date
- 2026-11-17
AI Technical Summary
In existing technologies, data providers and computing servers may collude to recover global secret differential values, leading to label leakage and result forgery. Furthermore, the overhead of unintentional bit-by-bit transmission is too large, making it difficult to achieve a reproducible and auditable fault-tolerant aggregation system.
The system employs a dual-server architecture, where the data provider generates independent source tags for two computing servers. The two servers respectively execute the check gate and the sharding filter gate, and the XOR reconstruction function circuit inputs the tags. The initiator decodes the two outputs to ensure consistency of the results.
It reduces the risk of label leakage and result forgery, reduces online interaction overhead, and achieves input privacy protection, output verification and result access control, while being compatible with existing fault-tolerant aggregation capabilities.
Smart Images

Figure CN123116336A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of electronic data processing technology, and more specifically to the fields of data security, privacy-preserving computing and multi-source data fusion technology, particularly to a collusion-resistant, secure, and fault-tolerant data aggregation method and system based on dual servers. Background Technology
[0002] In cross-personnel, cross-organizational, or large-scale IoT scenarios, multiple data providers typically need to submit their collected interval data to an aggregator to obtain a fusion result that represents the majority of honest participants. Traditional solutions often first aggregate plaintext data and then perform averaging, intersection, or fault-tolerant filtering. Therefore, the aggregator, business servers, or other participants may have access to sensitive data such as location, health indicators, and production status. At the same time, attacked participants may submit arbitrary interval data, corrupting the final result.
[0003] Existing privacy-preserving data fusion schemes can compile data aggregation tasks using the Marzullo interval fault-tolerant algorithm or its variants into obfuscated circuits, such as PG (short for "Byzantine Fault-Tolerant and Privacy-Preserving Sensor Fusion With Guaranteed Output Delivery," its technology originating from the eponymous research by ChengluJin, Chao Yin, Marten van Dijk, Sisi Duan, Fabio Massacci, Michael K. Reiter, and Haibin Zhang, published in ACM CCS 2024). The obfuscator generates two random labels representing logic 0 and logic 1 for each input line of the obfuscated circuit. The computer, unaware of the plaintext input, completes the circuit computation using only the labels corresponding to the actual input.
[0004] The existing PG solution system architecture is as follows: Figure 1 The diagram shows the system architecture before the improvement. The data provider connects to the computing server, meaning the data provider submits its selected input tags to the computing server; the obfuscator connects to the computing server, meaning the obfuscator sends obfuscated tables of check gates, filter gates, and functional circuits to the computing server, and receives obfuscated output tags from the computing server for decoding. In this architecture, all input tags from all data providers converge on the same computing server, which then performs the calculations for the check gates, filter gates, and subsequent functional circuits.
[0005] To identify illegal obfuscated inputs submitted by malicious data providers, the existing scheme PG sets up a check gate and a filter gate before the functional circuit. The check gate verifies whether the input label submitted by the data provider is a valid label generated according to predetermined obfuscation rules. The filter gate is set at the input of the functional circuit and determines whether to allow the corresponding input to proceed to subsequent calculations based on the check gate's result and the data provider's timeout status. For inputs submitted on time and passing the check gate, the filter gate passes its corresponding obfuscated label to the obfuscation circuit. For missing or failed-check inputs, it outputs a substitute label representing positive or negative infinity, allowing the Marzullo interval fault-tolerant algorithm to exclude the corresponding abnormal inputs and continue aggregation without interruption due to a single participant's missing or illegal label submission. Through this structure, the PG scheme can protect the input privacy of each data provider while resisting a certain number of abnormal inputs and ensuring the deliverability of the aggregation results. To improve the computational efficiency of the obfuscation circuit, engineering implementations typically employ further optimizations such as FreeXOR, allowing labels representing true and false values in the same circuit to be associated through a globally secret difference value, thereby reducing the computational and communication overhead of the XOR gate.
[0006] However, combined Figure 1 As can be seen, existing PG schemes employ a centralized computing architecture of "obfuscator—single computing server—multiple data providers," typically assuming that the data providers and the computing server executing the obfuscation circuit do not collude. If the data provider obtains a complete true / false tag pair for the same input bit during the initialization phase, or obtains sufficient remnant tag material to deduce the relationship between the two, it may collude with the computing server to recover the global secret differential value, further identify other input tags, decrypt intermediate states of the circuit, or even forge or replace the computation results. This problem makes the "non-collusion between the data provider and the computing server" a single point of security assumption for the system. Meanwhile, the single computing server structure has only one computing path; there is no second independent server to form complementary fragments for the same input bit, nor is there dual-output consistency verification. Although check gates and filter gates can identify illegal inputs, handle disconnections or timeouts, and output abnormal replacement tags, their function is concentrated on abnormal input filtering and fault-tolerant aggregation, and cannot solve the problem of recovering the global secret differential value after collusion between the data provider and the computing server.
[0007] One direct approach is to use 1-out-of-2 oblivious transfer (OT) to distribute tags bit by bit, ensuring that each data provider receives only one tag. However, when the data interval width, the number of participants, and the task concurrency are large, bit-by-bit online interaction introduces handshake, serialization, and network round-trip overhead. In scenarios with small inputs, low latency requirements, or massive sensor counts, this overhead may exceed the computational cost of the obfuscation circuit itself. Therefore, to reduce communication overhead, a second approach has emerged: adding a second server. However, simply adding a second server without redesigning the tag generation, distribution, filtering gate output, fragment reconstruction, and output verification processes cannot automatically prevent tag leakage or result tampering. Existing PG solutions also do not provide tag fragment generation, session binding, task isolation, fragment exchange, XOR reconstruction, output consistency verification, and result disclosure permission control processes for dual computing servers. Therefore, dual-server deployments are unlikely to form a practically runnable, reproducible, and auditable fault-tolerant aggregation system. Furthermore, in the existing single-computation server architecture, the initiator can usually only receive and decode the obfuscated output returned by the server; if the server returns incorrect, missing, or forged results, the initiator lacks a second independent computation result as a basis for comparison, making it difficult to detect unilateral anomalies in a timely manner. Summary of the Invention
[0008] This invention provides a collusion-resistant, secure, and fault-tolerant data aggregation method and system based on dual servers. The technical problem it solves is: how to use a dual-computing-server fixed architecture to prevent the leakage of complete tag pairs and the forgery of results caused by collusion between data providers and a single computing server through trust dispersion, tag sharding, low-overhead XOR reconstruction, and dual-path redundant computing, thereby reducing the online interaction overhead of unintentional bit-by-bit transmission and achieving interval data fault tolerance, input privacy, output verifiability, and result access control.
[0009] To address the above technical issues, this invention provides a collusion-resistant, secure, and fault-tolerant data aggregation method based on dual servers, comprising: an initiator, multiple data providers, a business server, a first computing server, and a second computing server;
[0010] The initiator is used to create aggregation tasks, select participants, set result permissions, and generate obfuscation tables for inspection gates, fragmentation filtering gates and functional circuits as the obfuscator. The obfuscation tables are then sent to the first computing server and the second computing server to save the output tag decoding relationship.
[0011] The data provider is used to input range data locally, encode the range data into a binary string, generate a source label for the first computing server and a source label for the second computing server for the same bit, and send the source label for the first computing server that corresponds to the actual input bit to the first computing server, and send the source label for the second computing server that corresponds to the actual input bit to the second computing server.
[0012] The business server is used to manage accounts, tasks, participation relationships, task status, result permissions, and historical records;
[0013] The first computing server and the second computing server are each used to execute the inspection gate using the source label of their respective path, form a normal state or an abnormal state according to the inspection result and the state of the data provider, execute the fragmentation filtering gate to obtain fragments, swap the selected fragments and XOR to reconstruct the input label of the functional circuit, load the obfuscation table of the functional circuit and independently execute the fault-tolerant aggregation obfuscation circuit, and send the obfuscated output to the initiator.
[0014] The initiator is also used to decode the two obfuscated outputs, and accept the aggregated result only if both outputs are successfully decoded and the results are the same.
[0015] Preferably, when the data provider generates source tags for the first computing server and source tags for the second computing server for the same bit, the two sets of source tags are generated using different random materials and are independent of each other, and one set cannot be directly derived from the other set; the data provider only sends the source tags corresponding to the actual input bit and does not send the interval data.
[0016] Preferably, when the obfuscator generates the obfuscation table for the check gate, the segmented filter gate, and the functional circuit, it determines the structure of the check gate, the segmented filter gate, and the functional circuit based on the number of participants, the interval input bit width, and the fault-tolerant aggregation function; the obfuscator generates a candidate source tag set, generates the obfuscation table for the check gate, the segmented filter gate, and the functional circuit on both sides according to the general obfuscation circuit method, sends it to the first computing server and the second computing server, and saves the output tag decoding relationship.
[0017] Preferably, when the first computing server and the second computing server execute the check gate using the source tags of their respective paths, they input the selected source tags of the two input lines to be checked from the same data provider into the check gate in pairs. The combination of valid candidate tags outputs a predetermined all-zero value, and the combination of invalid tags outputs a non-zero or unusable value. The first computing server and the second computing server compare the output of the check gate with the predetermined all-zero value, and combine it with the disconnection or timeout status of the data provider to form a normal state or an abnormal state.
[0018] Preferably, when the first computing server and the second computing server each execute the fragmentation filtering gate to obtain fragments, the source tag selected in this path is used as the first input of the fragmentation filtering gate, and the control tag from the obfuscator is used as the second input of the fragmentation filtering gate; the fragmentation filtering gate selects the result corresponding to the actual input bit or the result corresponding to the existing abnormal substitution bit according to the normal state or abnormal state, and outputs its respective fragment; the first computing server and the second computing server only exchange the selected fragments obtained in this calculation, and each XORs them to obtain the complete input tag of the functional circuit.
[0019] Preferably, when the initiator decodes the two obfuscated outputs, it obtains two plaintext results. The aggregated result is accepted only when both outputs are successfully decoded and the plaintext results are identical. If either output is missing, cannot be decoded, or the plaintext results are inconsistent, the result is not adopted and the task is marked as a computation failure or recalculated. When the task is set to public, the business server displays the result accepted by the initiator to the participants. When the task is set to private, only the initiator can view the result, and the participants can only view the task completion status.
[0020] This invention also provides a collusion-resistant, secure, and fault-tolerant data aggregation method based on a dual-server architecture. The key aspect is that, applied to the aforementioned dual-server collusion-resistant, secure, and fault-tolerant data aggregation system, it includes the following steps:
[0021] The initiator, acting as the obfuscator, creates an aggregation task and sets result permissions. The business server manages accounts, tasks, and participation relationships. The obfuscator establishes a session with the first computing server and the second computing server.
[0022] The data provider inputs range data locally and encodes it into a binary string. For the same bit, it generates a source label for the first computing server and a source label for the second computing server. It only submits the source label corresponding to the actual input bit and does not submit the range data.
[0023] The obfuscation method determines the structure of the inspection gate, the segmented filtering gate, and the functional circuit, generates two obfuscation tables and sends them to the first computing server and the second computing server, and saves the output tag decoding relationship.
[0024] The first computing server and the second computing server each use the source tag of their respective path to execute the check gate. The combination of valid tags outputs a predetermined all-zero value, while the combination of invalid tags outputs a non-zero or unavailable value. Combined with the disconnection and timeout status, a normal or abnormal status is formed.
[0025] The first computing server and the second computing server each execute the segmentation filtering gate. In the normal state, the actual input bits are taken, while in the abnormal state, the existing replacement bits are taken. Only the currently selected segments are swapped and the complete input labels are reconstructed by XOR.
[0026] The first computing server and the second computing server each load the obfuscation table of the functional circuit, independently execute the existing fault-tolerant aggregation obfuscation circuit using the reconstructed input tags, and send the two obfuscated outputs back to the initiator;
[0027] The initiator decodes the two obfuscated outputs and accepts the aggregated result only if both outputs are successfully decoded and the results are the same; otherwise, it is marked as a computation failure. Public tasks are displayed by the business server, while non-public tasks can only be viewed by the initiator.
[0028] Furthermore, when the data provider inputs interval data locally and encodes it into a binary string, it encodes the lower limit and upper limit of the interval into the binary string according to the endpoint bit width and arrangement order specified by the functional circuit, and determines the actual input bit corresponding to each input bit.
[0029] Furthermore, when the first computing server and the second computing server each execute the segmented filtering gate, they use the source tag selected in their current path as the first input of the segmented filtering gate and the control tag from the obfuscator as the second input of the segmented filtering gate. In a normal state, the result corresponding to the actual input bit is selected; in an abnormal state, the result corresponding to the existing abnormal substitution bit is selected. The obfuscator selects a random quantity for each possible filtering result and constructs the output on the first computing server side and the output on the second computing server side. The first computing server and the second computing server only exchange the selected segments obtained in this calculation and XOR them to obtain the complete input tags of the functional circuit.
[0030] Furthermore, when the initiator decodes the two obfuscated outputs, it obtains two plaintext results. The aggregated result is accepted only when both outputs are successfully decoded and the plaintext results are identical. If either output is missing, cannot be decoded, or the plaintext results are inconsistent, the result is not adopted and the task is marked as a computation failure or recalculated. When the task is set to public result, the business server displays the result accepted by the initiator to the participants. When the task is set to private result, only the initiator can view the result.
[0031] This invention provides a dual-server-based anti-collusion security fault-tolerant data aggregation method and system. The initiator, acting as the obfuscator, generates and distributes check gates, fragmentation filtering gates, and functional circuit obfuscation tables. The data provider locally encodes the interval data and generates two independent source tags, one for the first computing server and the other for the second computing server, submitting only the source tags corresponding to the actual input bits. Each computing server executes the check gates and fragmentation filtering gates, combining disconnection and timeout states to determine normal or abnormal states. Only the input tags of the currently selected fragment and XOR reconstruction functional circuit are exchanged. Then, the existing fault-tolerant aggregation obfuscation circuit is executed independently. The initiator decodes both outputs and only accepts the aggregation result if both are successful and consistent. Simultaneously, the business server displays or restricts viewing based on result permissions. Therefore, neither a single computing server nor a data provider colluding with it can obtain a complete 0 / 1 tag pair for the same input line. This reduces the risk of tag leakage and result forgery, minimizes online interaction overhead caused by unintentional bit-by-bit transmission, maintains compatibility with existing fault-tolerant aggregation capabilities, and achieves overall effects of input privacy, anti-collusion, verifiable output, and result permission control. Attached Figure Description
[0032] Figure 1 This is a system architecture diagram of an existing PG solution provided by the background technology;
[0033] Figure 2 This is an architecture diagram of a dual-server-based anti-collusion security and fault-tolerant data aggregation system provided in an embodiment of the present invention.
[0034] Figure 3 This is a structural diagram of the inspection gate calculation, filter gate segmentation output, and XOR reconstruction provided in an embodiment of the present invention;
[0035] Figure 4 This is a flowchart of a dual-server-based anti-collusion security fault-tolerant data aggregation method provided in an embodiment of the present invention. Detailed Implementation
[0036] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. The embodiments are given for illustrative purposes only and should not be construed as limiting the present invention. The accompanying drawings are for reference and illustration only and do not constitute a limitation on the scope of patent protection of the present invention, because many changes can be made to the present invention without departing from the spirit and scope of the present invention.
[0037] The anti-collusion, secure, and fault-tolerant data aggregation system based on a dual-server architecture provided in this invention has the following structure: Figure 2 As shown, it includes: an initiator, multiple data providers, a business server (not shown), a first computing server, and a second computing server.
[0038] The initiator is used to create aggregation tasks, select participants, set result permissions, and generate obfuscation tables for check gates, fragmentation filtering gates, and functional circuits as the obfuscator. The obfuscation tables are then distributed to the first and second computing servers, and the output tag decoding relationship is saved.
[0039] The data provider inputs range data locally, encodes the range data into a binary string, generates a source label for the first computing server and a source label for the second computing server for the same bit, and sends the source label for the first computing server that corresponds to the actual input bit to the first computing server, and sends the source label for the second computing server that corresponds to the actual input bit to the second computing server.
[0040] The business server is used to manage accounts, tasks, participation relationships, task status, result permissions, and historical records.
[0041] The first and second computing servers each use their own source tags to execute check gates, form normal or abnormal states based on the check results and the status of the data provider, execute fragmentation filtering gates to obtain fragments, exchange the selected fragments and XOR them to reconstruct the input tags of the functional circuits, load the obfuscation table of the functional circuits, independently execute the fault-tolerant aggregation obfuscation circuits, and send the obfuscated outputs to the initiator.
[0042] The initiator is also used to decode the two obfuscated outputs, and only accepts the aggregated result if both outputs are successfully decoded and the results are the same.
[0043] and Figure 1 Compared to a single-computing server architecture, Figure 2 The key change is that the computing side has been split into two servers. and These are referred to as the first computing server and the second computing server, respectively, and the data provider sends data to... , Submit two different source tags. and A fragmented switching channel is set up between the two servers. The obfuscator simultaneously sends check gates, fragmented filtering gates, and functional circuit obfuscation tables to both servers and receives two obfuscated outputs. The service server is not directly drawn into... Figure 2 It is part of the computing chain, but it is still responsible for managing accounts, tasks, participation relationships, task status, result permissions, and historical records as a platform component, and does not access the plaintext of the interval or the source label.
[0044] When generating the obfuscation table for the check gate, the segmented filter gate, and the functional circuit, the obfuscator determines the structure of the check gate, the segmented filter gate, and the functional circuit based on the number of participants, the range input bit width, and the fault-tolerant aggregation function. The obfuscator generates a set of candidate source tags, generates the obfuscation table for the check gate, the segmented filter gate, and the functional circuit on both sides according to the general obfuscation circuit method, sends it to the first computing server and the second computing server, and saves the output tag decoding relationship.
[0045] In this scheme, the initiator, acting as the obfuscator, creates the aggregation task, selects participants, and sets the result disclosure permissions. Simultaneously, it generates an obfuscation table for the check gate, fragmentation filter gate, and existing fault-tolerant aggregation circuitry, and distributes it to... and And save the output tag decoding relationship.
[0046] When the data provider generates source tags for the first computing server and the second computing server for the same bit, the two sets of source tags are generated using different random materials and are independent of each other. One set cannot be directly derived from the other. The data provider only sends the source tags corresponding to the actual input bit and does not send interval data.
[0047] Each data provider, after entering its local input range, encodes the data into a binary string according to the endpoint bit width and arrangement order specified by the functional circuit, and generates a target for the same bit. Source tags and target The source labels; the two sets of source labels are generated using different random materials and are independent of each other; one set cannot be directly derived from the other. The data provider only sends the source labels corresponding to the actual input bits to the respective source labels. and Plaintext within the range is not sent to the business server or computing server.
[0048] After entering the calculation stage, , Each source label is evaluated using a check gate to verify if the label is a valid candidate label. The gate combines the results of disconnection, timeout, and check to determine a normal or abnormal state, and obtains the corresponding control label. Then, each of the two servers executes a local fragmentation filter gate: in the normal state, it outputs the fragment corresponding to the actual input bit; in the abnormal state, it outputs the fragment corresponding to the PG's existing abnormal substitution bit. , pass Figure 2 The bidirectional channel exchange shows the selected fragment, and the functional circuit input tags are reconstructed using an XOR method. After repeating this process for all data providers and all input bits, both servers obtain complete obfuscated circuit input tags and load the functional circuit obfuscation table for the same aggregation task, independently executing the existing Marzullo fault-tolerant aggregation obfuscation circuit.
[0049] When the first computing server and the second computing server each execute the check gate using their respective source tags, they input the selected source tags of the two input lines to be checked from the same data provider into the check gate in pairs. The combination of valid candidate tags outputs a predetermined all-zero value, while invalid tags output a non-zero or unusable value. The first computing server and the second computing server compare the check gate output with the predetermined all-zero value and combine it with the data provider's offline or timeout status to form a normal or abnormal status.
[0050] After the calculation is completed, , The obfuscated outputs are sent separately to the initiator. The initiator decodes both outputs according to the saved decoding relationship, accepting the aggregated result only if both outputs are successfully decoded and the plaintext results are identical; if either output is missing, cannot be decoded, or the results are inconsistent, the result is not adopted, and the task is marked as computationally failed or recalculated. If the task is set to public result, the business server displays the result accepted by the initiator to the participants; if set to private, only the initiator can view the result. This framework is based on... , For security purposes, the two computing servers should not be compromised simultaneously and should not collude with each other. Therefore, they should be deployed in independent computing environments.
[0051] like Figure 3 As shown, the system in Figure 2 Based on the dual-computing server architecture, the process of checking, sharding filtering, and XOR reconstruction of a single input line is further presented. Figure 3 The upper and lower paths correspond to the first computing server respectively. Second computing server For the first The first data provider One input line, the data provider in the local input range ( , They represent the first After specifying the lower and upper limits of the data provider's interval, the data is encoded into a binary string according to the endpoint bit width and arrangement order specified by the functional circuit. And generate the same target source tag and orientation source tag ,in This represents the actual input bit. The two sets of source labels are generated using different random materials and are independent of each other; one set cannot be directly derived from the other. The data provider only sends the source labels corresponding to the actual input bits to the respective data providers. and Plaintext within the range is not sent to the business server or computing server.
[0052] During the door inspection process, Select the source label for the two input lines to be inspected from the same data provider, for example... and Enter the check gate in pairs; For the corresponding label of this route and Perform the same processing. The gate uses the existing PG structure for inspection; output valid candidate label combinations. Inputs with forged labels, misplaced labels, or those that cannot be decoded correctly will result in non-zero or unavailable values. The server will check the gate output against... By comparing and combining the data provider's offline and timeout status, a normal status is formed. or abnormal state Check that the gate does not output input bits. The plaintext is not output to the filter gate, nor is any new logic label output.
[0053] In the segmented filtration stage and Select the source tag for this path respectively As the first input to the filter gate, the control label from the confusion party is used. As the second input, where (0 corresponds to) , correspond ), Indicates the candidate bit value. The filter gate retains the original filter gate's selection function. :when hour, This indicates that the input from the data provider is permitted; when , This indicates that the existing exception in PG is used to replace the branch. This is the replacement bit corresponding to the input line. Unlike the original PG filter gate which directly outputs a complete function label, this invention modifies the filter gate output to dual-server fragmentation. The obfuscator considers each possible filtering result... Selecting random variables and construct Side output , Side output Two filter gates output. and XOR ( After that, I got That is, the complete confusion circuit input label corresponding to the input line of the functional circuit.
[0054] thus, , Swap only the selected fragments obtained in this calculation, and then XOR them to get... This eliminates the need for bit-by-bit accidental transmission of each input bit, and also avoids swapping complete 0 / 1 tag pairs on the same input line. After repeating the above checking, filtering, swapping, and XOR reconstruction process for all data providers and all input bits, the two servers each form a complete obfuscated circuit input tag vector. For normal data providers, and This vector contains the function labels corresponding to each bit in the actual interval; for data providers that are missing, timed out, or failed checks, and This vector contains the functional labels corresponding to each of the existing abnormal replacement values of PG.
[0055] When the initiator decodes the two obfuscated outputs, it obtains two plaintext results. The aggregated result is accepted only if both outputs are successfully decoded and the plaintext results are identical. If either output is missing, cannot be decoded, or the plaintext results are inconsistent, the result is not adopted and the task is marked as a computation failure or recalculated. When the task is set to public results, the business server displays the result accepted by the initiator to the participants. When the task is set to private results, only the initiator can view the result.
[0056] Subsequently, , Each component loads its own obfuscation table corresponding to the same aggregation task and independently executes the existing Marzullo fault-tolerant aggregation obfuscation circuit using the reconstructed input labels. Internally, the obfuscation circuit still performs interval endpoint processing, the Marzullo algorithm or its variants, and abnormal input substitution, outputting obfuscated output labels representing the aggregation interval or existing abnormal states. After computation, , Obfuscated output , The data is sent to the initiator, who decodes both outputs according to the saved output decoding relationship to obtain the desired results. , Only when both channels are successfully decoded and Accepts aggregated results; if any path is missing, cannot be decoded, or... and If there is an inconsistency, the result is not used, and the task is marked as a computation failure or to be recalculated.
[0057] Figure 3The security effect is that a single computing server can obtain at most the random shards from its own path and the selected complementary shards received from another server, which can be used to reconstruct the selected label. However, the missing fragment corresponding to another candidate result on the same input line prevents the acquisition of complete 0 / 1 label pairs. Even if the data provider colludes with one of the computing servers, the missing fragment cannot be filled in. Furthermore, unilateral server tampering with the result will manifest as a error when the initiator compares the two outputs. and Inconsistent. The above process is based on... , For security purposes, the two computing servers should not be compromised simultaneously and should not collude with each other. Therefore, they should be deployed in independent computing environments.
[0058] The system employs a multi-party security and fault-tolerant aggregation platform, consisting of: a Windows 10 / 11 64-bit client, a business server based on Java 8 and Spring Boot, MyBatis and MySQL 5.7 databases, and a primary computing server running in Linux x86-64 containers. Second computing server The underlying obfuscation circuitry is implemented in C++. , Run as two independent computing processes or containers, and load the check gate, the modified filter gate, and the confusion table of the existing fault-tolerant aggregation function circuit for this task respectively.
[0059] This invention also provides a collusion-resistant, fault-tolerant data aggregation method based on dual servers, the process of which is as follows: Figure 4 It can be summarized into seven steps:
[0060] The first step involves the initiator creating an aggregation task and setting result permissions as the obfuscator. The business server manages accounts, tasks, and participation relationships. The obfuscator establishes sessions with the first and second computing servers respectively. The two computing servers are deployed independently and do not collude with each other.
[0061] The second step is for the data provider to input range data locally and encode it into a binary string. For the same bit, two independent source labels are generated for the two computing servers. Only the source label corresponding to the actual input bit is submitted, and the range plaintext is not submitted.
[0062] The third step involves the obfuscating party determining the structure of the inspection gate, the segmented filtering gate, and the functional circuit, generating obfuscation tables on both sides and sending them to the two computing servers respectively, while saving the output tag decoding relationship.
[0063] The fourth step involves the two computing servers executing a check gate using their respective source tags. Valid tags output a predetermined all-zero value, while invalid tags output a non-zero or unavailable value. This, combined with disconnection and timeout statuses, forms a normal or abnormal status.
[0064] Fifth, the two computing servers execute the segmentation filtering gate respectively. In the normal state, the actual input bits are taken, while in the abnormal state, the existing replacement bits are taken. Only the selected segments are swapped and XORed to reconstruct the complete input labels.
[0065] Step 6: The two computing servers load the functional circuit obfuscation table respectively, use the reconstructed input tags to independently execute the existing fault-tolerant aggregation obfuscation circuit, and send the two obfuscated outputs back to the initiator respectively;
[0066] Step 7: The initiator decodes the two outputs separately. The aggregated result is accepted only if both outputs are successfully decoded and the results are the same; otherwise, it is marked as a calculation failure. For public tasks, the results are displayed by the business server. For non-public tasks, only the initiator can view them, and the participants can only view the task completion status.
[0067] The first step involves task creation and dual-server session initialization. Specifically, the initiator, acting as the obfuscator, creates the aggregation task, selects participants, and sets the aggregation results to be public or private. The business server sends invitations to the participants; participants either decline or accept and input a data range visible only locally. After all participants have completed processing, or after a preset deadline is reached, the business server changes the task status to "ready," with only the initiator having permission to initiate the task. The obfuscator interacts with the first computing server... Second computing server Establish a task session and send the obfuscation table corresponding to the same aggregated task to both sides. , It is advisable to deploy them in independent computing environments, with the premise that neither can be compromised at the same time and that they do not collude with each other for security.
[0068] The second step involves local range encoding and submission of two source labels. Specifically, the data provider inputs range data locally that is only visible to itself. The client, according to the endpoint bit width and arrangement order specified by the functional circuit, will... , Encoded as a binary string , for the The actual input bits are as follows: Data providers generate data targeting the same user. source tag and orientation source tag The two sets of source labels are generated using different random materials and are independent of each other; one set cannot be directly derived from the other. The data provider only sends the source labels corresponding to the actual input bits to the respective data set. , plaintext of the interval It is not sent to the business server or computing server.
[0069] The third step involves generating and distributing the obfuscation tables. Based on the number of participants, the range input bit width, and existing fault-tolerant aggregation functions, the obfuscator determines the check gates, fragmented filter gates, and functional circuit structures for this task. The obfuscator generates a set of candidate source labels needed to construct the check gates and fragmented filter gates, but does not obtain the data provider's actual selection for each bit. Subsequently, the obfuscator generates obfuscation tables for both check gates, fragmented filter gates, and functional circuits according to a general obfuscation circuit method, and sends them to the obfuscator. , And save the output tag decoding relationship. Figure 4 The confusion tables are sent to the left and right sides respectively in the direction of confusion, which means... , Each of them obtains its own inspection gate, segmented filter gate, and functional circuit confusion table.
[0070] The fourth step involves checking the gate values on both servers. Specifically, Select the source label for the two input lines to be inspected from the same data provider, for example... and Enter the check gate in pairs; For the corresponding label of this route and Perform the same processing. The gate uses the existing PG structure for inspection; output valid candidate label combinations. Inputs with forged labels, misplaced labels, or those that cannot be decoded correctly will result in non-zero or unavailable values. The server will check the gate output against... By comparing and combining the data provider's offline and timeout status, a normal status is formed. or abnormal state Check that the gate does not output input bits. The plaintext is not output to the filter gate, nor is any new logical label output. Its function is only to determine whether the input label of the data provider is valid and whether it should enter the normal aggregation branch. Figure 4 The "List of Abnormal Data Providers" in the table represents the abnormal statuses summarized by the server based on disconnections, timeouts, and inspection results.
[0071] The fifth step involves fragmented filtering and XOR reconstruction. Specifically, and Select the source tag for this path respectively As the first input to the filter gate, the control label from the confusion party is used. As the second input, the filter gate retains the selection function of the original PG filter gate. :when hour, This indicates that the input from the data provider is permitted; when , This indicates that the existing exception in PG is used to replace the branch. This is the replacement bit corresponding to the input line. Unlike the original PG filter gate which directly outputs a complete function label, this invention modifies the filter gate output to dual-server fragmentation. The obfuscator considers each possible filtering result... Selecting random variables and construct Side output , Side output . Figure 4 Two filter gates output and After exchanging the two-way arrows, XORing yields... This refers to the complete obfuscated circuit input tag corresponding to the input line of the functional circuit. After repeating the above checking, filtering, swapping, and XOR reconstruction process for all data providers and all input bits, the two servers each form a complete obfuscated circuit input tag vector. For normal data providers, and This vector contains the function labels corresponding to each bit in the actual interval; for data providers that are missing, timed out, or failed checks, and This vector contains the functional labels corresponding to each of the existing abnormal replacement values of PG.
[0072] The sixth step involves independent fault-tolerant aggregation on two servers. Specifically, , Each component loads its own obfuscation table corresponding to the same aggregation task and independently executes the existing Marzullo fault-tolerant aggregation obfuscation circuit using the reconstructed input labels. Internally, the obfuscation circuit still performs interval endpoint processing, the Marzullo algorithm or its variants, and abnormal input substitution, outputting obfuscated output labels representing the aggregation interval or existing abnormal states. After computation, , Obfuscated output , Send to the initiator.
[0073] The seventh step involves dual-output decoding comparison and access control. Specifically, the initiator decodes both outputs according to the saved output decoding relationship to obtain... , Only when both channels are successfully decoded and Accepts aggregated results; if any path is missing, cannot be decoded, or... and If there is an inconsistency, the result will not be adopted, and the task will be marked as a computation failure or recalculated. If the task is set to public result, the business server will display the result accepted by the initiator to the participants; if it is set to private, only the initiator can view the result, and the participants can only view the task completion status.
[0074] Under normal conditions and Taking =1 as an example, At most, the local piecewise filtering gate can obtain random pieces of the possible outputs of the two candidate results. , And this time from Selected complementary fragments received It can reconstruct the currently selected label. But lacking Cannot obtain the same input line , Complete tag pair. Data provider and Even when conspiring, the missing fragment cannot be made up. Changing the output separately will also be compared with the initiator. , The results are inconsistent. The aforementioned safety effects are based on... , The two computing servers are deployed in independent computing environments, provided that they are not attacked at the same time and do not collude with each other.
[0075] In summary, the beneficial effects of the anti-collusion security and fault-tolerant data aggregation method and system based on dual servers provided by the embodiments of the present invention are as follows:
[0076] 1. Reduce the risk of tag leakage caused by collusion between data providers and a single computing server. The same actual input is processed into two independent source tags for two computing servers. The complete obfuscated circuit input tag is obtained by XORing the fragments generated by the two computing servers. A single computing server, or a data provider colluding with a single computing server, cannot obtain a complete 0 / 1 tag pair for the same input line, thus making it impossible to recover the global secret differential value, decrypt other inputs, or forge results.
[0077] 2. Reduce the burden of online interaction. Compared with the bitwise inadvertent transmission method, the data provider does not need to perform bitwise inadvertent transmission with a single server for each input bit. Instead, it generates two source tags locally and sends them to two computing servers respectively. The two computing servers calculate the fragments locally and exchange the fragments. The input reconstruction of the obfuscated circuit is completed by XORing, reducing the overhead of public network handshake, serialization and network round trip.
[0078] 3. Able to detect one-sided output anomalies. The two computing servers return obfuscated outputs respectively. The initiator decodes and compares the two plaintext results. If either output is missing, cannot be decoded, or the two results are inconsistent, the one-sided result is not accepted, and the task is marked as a computation failure or recalculated.
[0079] 4. Compatible with existing fault-tolerant aggregation capabilities. This invention retains the check gate and filter gate selection logic and Marzullo fault-tolerant aggregation function circuit of the existing PG scheme, without changing its internal abnormal input substitution and fault tolerance threshold processing. It only modifies the input label formation method and the output verification method, which can support continued aggregation in the case of data provider disconnection, timeout, or submission of illegal labels.
[0080] 5. Implement result access control and platform-based task collaboration. The business server manages accounts, tasks, participation relationships, task status, result permissions, and historical records, supporting the public or private nature of aggregated results; public tasks display the results accepted by the initiator to the participants, while private tasks allow only the initiator to view the results, facilitating actual deployment, reproduction, and auditing.
[0081] 6. Achieve distributed trust and independent deployment of dual servers. The first and second computing servers are deployed in independent computing environments, with the security premise that neither can be compromised at the same time and that they do not collude with each other, thus avoiding the concentration of system security dependence on a single computing server.
[0082] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.
Claims
1. A dual-server-based anti-collusion security and fault-tolerant data aggregation system, characterized in that, include: Initiator, multiple data providers, business server, first computing server and second computing server; The initiator is used to create aggregation tasks, select participants, set result permissions, and generate obfuscation tables for inspection gates, fragmentation filtering gates and functional circuits as the obfuscator. The obfuscation tables are then sent to the first computing server and the second computing server to save the output tag decoding relationship. The data provider is used to input range data locally, encode the range data into a binary string, generate a source label for the first computing server and a source label for the second computing server for the same bit, and send the source label for the first computing server that corresponds to the actual input bit to the first computing server, and send the source label for the second computing server that corresponds to the actual input bit to the second computing server. The business server is used to manage accounts, tasks, participation relationships, task status, result permissions, and historical records; The first computing server and the second computing server are each used to execute the inspection gate using the source label of their respective path, form a normal state or an abnormal state according to the inspection result and the state of the data provider, execute the fragmentation filtering gate to obtain fragments, swap the selected fragments and XOR to reconstruct the input label of the functional circuit, load the obfuscation table of the functional circuit and independently execute the fault-tolerant aggregation obfuscation circuit, and send the obfuscated output to the initiator. The initiator is also used to decode the two obfuscated outputs, and accept the aggregated result only if both outputs are successfully decoded and the results are the same.
2. The anti-collusion security and fault-tolerant data aggregation system based on dual servers according to claim 1, characterized in that, When the data provider generates source tags for the first computing server and source tags for the second computing server for the same bit, the two sets of source tags are generated using different random materials and are independent of each other, and one set cannot be directly derived from the other set; the data provider only sends the source tags corresponding to the actual input bit and does not send the interval data.
3. The anti-collusion security and fault-tolerant data aggregation system based on dual servers according to claim 1, characterized in that, When the obfuscator generates the obfuscation table for the check gate, the segmented filter gate, and the functional circuit, it determines the structure of the check gate, the segmented filter gate, and the functional circuit based on the number of participants, the interval input bit width, and the fault-tolerant aggregation function. The obfuscator generates a set of candidate source tags, generates the obfuscation tables for the check gate, the segmented filter gate, and the functional circuit on both sides according to the general obfuscation circuit method, sends them to the first computing server and the second computing server, and saves the output tag decoding relationship.
4. The anti-collusion security and fault-tolerant data aggregation system based on dual servers according to claim 1, characterized in that, When the first computing server and the second computing server execute the check gate using the source tags of their respective paths, they input the selected source tags of the two input lines to be checked from the same data provider into the check gate in pairs. The combination of valid candidate tags outputs a predetermined all-zero value, and the combination of invalid tags outputs a non-zero or unusable value. The first computing server and the second computing server compare the output of the check gate with the predetermined all-zero value, and combine it with the disconnection or timeout status of the data provider to form a normal state or an abnormal state.
5. The anti-collusion security and fault-tolerant data aggregation system based on dual servers according to claim 1, characterized in that, When the first computing server and the second computing server each execute the fragmentation filtering gate to obtain fragments, they use the source tag selected in this path as the first input of the fragmentation filtering gate and the control tag from the obfuscator as the second input of the fragmentation filtering gate. The fragmentation filtering gate selects the result corresponding to the actual input bit or the result corresponding to the existing abnormal substitution bit according to the normal state or abnormal state, and outputs its respective fragment. The first computing server and the second computing server only exchange the selected fragments obtained in this calculation, and each XORs them to obtain the complete input tag of the functional circuit.
6. The anti-collusion security and fault-tolerant data aggregation system based on dual servers according to claim 1, characterized in that, When the initiator decodes the two obfuscated outputs, it obtains two plaintext results. It only accepts the aggregated result when both outputs are successfully decoded and the two plaintext results are the same. If any path is missing, cannot be decoded, or the plaintext results of the two paths are inconsistent, the result will not be adopted and the task will be marked as a calculation failure or recalculated. When the task is set to public result, the business server will show the result accepted by the initiator to the participants. When the task is set to private, only the initiator can view the result, and the participants can only view the task completion status.
7. A dual-server-based anti-collusion security and fault-tolerant data aggregation method, characterized in that, The anti-collusion security fault-tolerant data aggregation system based on any one of claims 1 to 6 includes the following steps: The initiator, acting as the obfuscator, creates an aggregation task and sets result permissions. The business server manages accounts, tasks, and participation relationships. The obfuscator establishes a session with the first computing server and the second computing server. The data provider inputs range data locally and encodes it into a binary string. For the same bit, it generates a source label for the first computing server and a source label for the second computing server. It only submits the source label corresponding to the actual input bit and does not submit the range data. The obfuscation method determines the structure of the inspection gate, the segmented filtering gate, and the functional circuit, generates two obfuscation tables and sends them to the first computing server and the second computing server, and saves the output tag decoding relationship. The first computing server and the second computing server each use the source tag of their respective path to execute the check gate. The combination of valid tags outputs a predetermined all-zero value, while the combination of invalid tags outputs a non-zero or unavailable value. Combined with the disconnection and timeout status, a normal or abnormal status is formed. The first computing server and the second computing server each execute the segmentation filtering gate. In the normal state, the actual input bits are taken, while in the abnormal state, the existing replacement bits are taken. Only the currently selected segments are swapped and the complete input labels are reconstructed by XOR. The first computing server and the second computing server each load the obfuscation table of the functional circuit, independently execute the existing fault-tolerant aggregation obfuscation circuit using the reconstructed input tags, and send the two obfuscated outputs back to the initiator; The initiator decodes the two obfuscated outputs and accepts the aggregated result only if both outputs are successfully decoded and the results are the same; otherwise, it is marked as a computation failure. Public tasks are displayed by the business server, while non-public tasks can only be viewed by the initiator.
8. The anti-collusion security and fault-tolerant data aggregation method based on dual servers according to claim 7, characterized in that, When the data provider inputs interval data locally and encodes it into a binary string, it encodes the lower limit and upper limit of the interval into the binary string according to the endpoint bit width and arrangement order specified by the functional circuit, and determines the actual input bit corresponding to each input bit.
9. The anti-collusion security and fault-tolerant data aggregation method based on dual servers according to claim 7, characterized in that, When the first computing server and the second computing server execute the fragmentation filtering gate, the source label selected in this path is used as the first input of the fragmentation filtering gate, and the control label from the obfuscator is used as the second input of the fragmentation filtering gate; in the normal state, the result corresponding to the actual input bit is selected, and in the abnormal state, the result corresponding to the existing abnormal substitution bit is selected; the obfuscator selects a random quantity for each possible filtering result and constructs the output of the first computing server and the output of the second computing server. The first computing server and the second computing server only exchange the selected fragments obtained in this calculation, and each XORs them to obtain the complete input tags of the functional circuit.
10. The anti-collusion security and fault-tolerant data aggregation method based on dual servers according to claim 7, characterized in that, When the initiator decodes the two obfuscated outputs, it obtains two plaintext results. It only accepts the aggregated result when both outputs are successfully decoded and the two plaintext results are the same. If any path is missing, cannot be decoded, or the plaintext results of the two paths are inconsistent, the result will not be adopted and the task will be marked as a calculation failure or recalculated. When the task is set to public result, the business server will show the result accepted by the initiator to the participants. When the task is set to private result, only the initiator can view the result.
