Low-cost deployment method of hint chain in zero-trust access network

Abstract

The invention discloses a low-cost deployment method for a prompt chain in a zero-trust access network. The method comprises the following steps: introducing a verification cost balance (VCB) factor to reduce the verification times of a zero-trust ZT; proposing a sub prompt chain branch limit SCB algorithm based on a verification cost balance VCB factor, gradually selecting an optimal PES node to deploy a prompt chain, and thus realizing the minimum service cost; the invention not only solves the problem of conflict between the deployment cost and verification cost of the prompt chain in the ZT network, but also provides an economical, efficient and high-adaptability deployment method, thereby enhancing the service quality and resource utilization efficiency of the generative artificial intelligence application in the ZT network.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of new generation information technology and artificial intelligence applications, and in particular to a low-cost deployment method of a prompt chain in a zero-trust access network. Background Art

[0002] In modern network and distributed computing environments, chained service deployment methods are widely used in multiple fields, including network function virtualization (NFV), software defined network (SDN) and cloud computing. Chained services usually refer to a data flow management method that is processed in an orderly manner through a series of service functions (SFs), namely service function chain (SFC). This chain structure enables data flows to pass through a series of virtual or physical functions such as firewalls, load balancing, encryption, etc. on demand, thereby improving the flexibility and controllability of the network. While ensuring that data flows pass through each service node in a specific order, the service function chain helps to achieve resource optimization, bandwidth management and latency control.

[0003] A large number of studies have proposed different methods for the optimization deployment of SFC. Some scholars have defined the minimum cost service function chain embedding problem (Minimum Cost Service Function Chaining and Embedding, MC-SFCE), and proposed a cost factor-based SFC optimization algorithm (COFO-SC) to minimize the cost of service function chain delivery. Some scholars have defined the hybrid service function chain embedding problem (Hybrid Service Function ChainEmbedding, HSFCE), and proposed an embedding algorithm based on hybrid service function chain optimization (HSFCE algorithm) to optimize resource utilization and load balancing in multi-access edge computing environments. However, it is difficult to ensure cost optimization by directly applying this chain service deployment method to implement the deployment of the prompt chain under the zero-trust network architecture.

[0004] Among the cost factor-based service function chaining optimization algorithm (COFO-SC) and the hybrid service function chaining optimization embedding algorithm (HSFCE algorithm), the COFO-SC (Cost Factor Optimization for Service Chaining) algorithm aims to solve the minimum cost service function chaining embedding problem (MC-SFCE) by introducing cost factors to optimize the embedding of service function chains. The COFO-SC algorithm considers network transmission and computing costs in resource allocation and path selection, thereby minimizing the overall cost in service function chain delivery. The algorithm not only has significant advantages in transmission efficiency, but also provides an embedding strategy with good resource utilization for different service function chain requirements. In addition, COFO-SC has a provable performance bound, ensuring the efficiency and effectiveness of the algorithm.

[0005] In addition, the embedding algorithm based on hybrid service function chain optimization for the hybrid service function chain embedding (HSFCE) problem allocates service function chains in a multi-access edge computing (MEC) environment, aiming to optimize resource utilization and load balancing. The algorithm can reduce latency, improve service quality, and achieve effective resource management by reasonably distributing the functional components of the service function chain between edge nodes and cloud nodes. The HSFCE algorithm takes into account the heterogeneity of edge and cloud resources and proposes a hybrid embedding strategy to make more efficient use of computing resources and network resources in a multi-access edge computing environment.

[0006] In summary, many studies have focused on how to efficiently deploy chained services (such as service function chains, SFCs) and proposed numerous deployment methods. However, it is still challenging to directly apply these chained service deployment methods to zero trust (ZT) networks to achieve economic deployment of tip chains. This is mainly because the deployment process of the tip chain is closely related to the ZT verification process, and optimizing the deployment cost of the tip chain does not necessarily optimize the ZT verification cost simultaneously, which may bring some additional problems.

[0007] Specifically, the deployment of the prompt chain usually prioritizes the deployment cost of the generative artificial intelligence content (AIGC) path, and then minimizes the ZT verification cost on this path as much as possible on this basis. However, this step-by-step optimization approach may result in higher overall service costs. First, simply optimizing the deployment cost of the AIGC path may affect the efficiency of ZT verification, especially when bypassing the policy enforcement point (PEP), which may increase unnecessary resource consumption. Second, the computing resources and costs required to implement the virtual policy enforcement point (vPEP) are often underestimated, and additional delays and resource overhead may be introduced in the actual verification process.

[0008] In summary, the shortcoming of the existing solution is that it fails to achieve global optimization between deployment cost and verification cost. Relying on a step-by-step strategy of first optimizing path deployment and then minimizing verification cost, it is easy to ignore the mutual influence between the two, resulting in a single optimization strategy that is difficult to meet the actual needs of complex network scenarios. In addition, this approach is insufficient to cope with the dynamic verification requirements in ZT networks, which may lead to decreased service responsiveness and reduced resource utilization. Therefore, designing a joint optimization strategy that can comprehensively consider deployment and verification costs will be the key to improving the deployment efficiency of chained services in ZT networks. Summary of the invention

[0009] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a low-cost deployment method of the prompt chain in a zero-trust access network. The present invention not only solves the conflict between the deployment cost and the verification cost of the prompt chain in the ZT network, but also provides an economical, efficient and adaptable deployment method, thereby enhancing the service quality and resource utilization efficiency of generative artificial intelligence applications in the ZT network.

[0010] To achieve the above object, the technical solution adopted by the present invention is: a low-cost deployment method of a prompt chain in a zero-trust access network, comprising the following steps:

[0011] Step 1: Introduce the verification cost balance VCB factor to reduce the number of verifications and related expenses of zero trust ZT;

[0012] Step 2: Based on the verification cost balance VCB factor, a sub-prompt chain branch and bound SCB algorithm is proposed to gradually select the optimal PES node to deploy the prompt chain, thereby minimizing the service cost.

[0013] As a further improvement of the present invention, in step 1, the calculation method of the verification cost balance VCB factor specifically includes:

[0014] (1) In the no-verification scenario: when continuous prompts can be deployed on the same PES node, no additional verification is required due to the inherent security guarantees within the node, so the value of the verification cost balance VCB factor is equal to the average value of the computational cost of deploying prompts;

[0015] (2) In the PEP verification scenario, when continuous prompts are deployed on different PES nodes, the PEP is used for zero-trust ZT verification, and the routing cost is included in the VCB calculation to control the overall cost of the service.

[0016] (3) In the virtual policy enforcement point vPEP verification scenario: When using vPEP for verification, no additional routing is required. In the previous prompting instance, that is, the carrier of the prompting function installed on the PES is deployed, and then a vPEP instance is created. The VCB value includes the additional computational cost.

[0017] As a further improvement of the present invention, the step 2 specifically comprises the following steps:

[0018] Step 2.1, algorithm initialization: First, initialize the forwarding path, remove the zeroth prompt of the prompt chain from the source node, that is, the service starting point, and then gradually deploy from the first prompt;

[0019] Step 2.1, gradually deploy the prompt chain: In each iteration, the algorithm calculates the VCB value of each PES node to deploy the current possible prompt set, selects the node with the lowest VCB value as the optimal deployment node; updates the currently deployed sub-path and adds it to the total path;

[0020] Step 2.3, iterative update: After each set of prompts is deployed, update the index and continue to deploy the next set of prompts until the entire prompt chain is deployed;

[0021] Step 2.4, select GPT server: After all prompts are deployed, select the GPT server with the lowest cost to process the final query and update it into the forwarding path.

[0022] The present invention optimizes the deployment cost and verification cost of the prompt chain in a zero-trust (ZT) network, and solves the cost conflict problem of the traditional chain service deployment method in the ZT environment. Specifically, it includes:

[0023] (1) Reducing the overall service cost: By introducing the VCB factor and the SCBB algorithm, the invention aims to achieve efficient deployment of the prompt chain in the ZT network, minimize the additional computing and routing costs in the ZT verification process, and thus reduce the total service cost.

[0024] (2) Balancing deployment and verification costs: The present invention achieves a balance between deployment costs and verification costs through flexible selection of three verification methods and dynamic calculation of the VCB factor, thereby avoiding an overall cost increase caused by excessive focus on a single cost.

[0025] (3) Improve deployment efficiency: Through the sub-prompt chain branch and bound (SCBB) algorithm, the optimal node is gradually selected to allocate the prompt chain, which improves the efficiency of prompt chain deployment on the PES node, and selects the best GPT server to process requests, thereby improving the response speed and efficiency of the service.

[0026] (4) Adapting to dynamic requirements: The invention takes into account the dynamic verification requirements in the ZT network and provides a flexible prompt chain deployment solution that can adapt to the resource limitations and network status changes of different PES nodes, achieving a more robust deployment in a complex network environment.

[0027] The beneficial effects of the present invention are:

[0028] 1. Reduce total service cost: Compared with PCDF and LOFD schemes, the SCBB algorithm has a significant advantage in total service cost. Experimental results show that the average total service cost of SCBB is 4.67% and 13.89% lower than that of PCDF and LOFD, respectively. This is because SCBB reduces the verification requirements by deploying continuous prompts on the same PES node, thereby reducing the verification cost.

[0029] 2. Reduce verification costs: SCBB is also superior to other solutions in terms of verification costs. By reducing the number of PEP and vPEP verifications, service overhead is further saved. Experimental data shows that the verification cost of SCBB is 11.86% and 30.50% lower than that of PCDF and LOFD, respectively. This is due to the design of SCBB strategy that prioritizes the deployment of continuous prompts on the same PES node, reducing the number of ZT verifications and thus reducing verification-related resource consumption.

[0030] 3. Improve the acceptance rate: In terms of resource utilization efficiency, SCBB shows the highest acceptance rate. When deploying 4,000 requests, SCBB's acceptance rate reaches 56.24%, which is significantly higher than PCDF's 51.75% and LOFD's 28.18%. This is due to SCBB's optimization in resource allocation and prompt chain deployment path, which enables it to complete more requests more efficiently under limited resource conditions.

[0031] 4. Runtime: In terms of run time, SCBB shows high efficiency. The average time to process an AI-SR request is 133.14 milliseconds, which is close to LOFD's 123.73 milliseconds and significantly faster than PCDF's 191.95 milliseconds. This shows the competitiveness of SCBB in deployment efficiency and helps improve the system's real-time response capability.

[0032] In summary, SCBB has achieved lower total service costs and verification costs, higher acceptance rates, and better operating efficiency by optimizing the cost allocation of prompt chain deployment and ZT verification, proving its technical advantages in prompt chain deployment in the ZT network environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 An example diagram of the SCBB algorithm in an embodiment of the present invention;

[0034] Figure 2 A schematic diagram showing the effect of the length of the prompt chain in an embodiment of the present invention;

[0035] Figure 3 Schematic diagram of average total overhead vs. number of PEPs in an embodiment of the present invention;

[0036] Figure 4 Schematic diagram of reception rate vs. service demand in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0038] Example

[0039] A low-cost deployment method for the prompt chain in a zero-trust access network is proposed. A strategy for optimizing the deployment of the prompt chain is proposed. The service cost is minimized by introducing the verification cost balance (VCB) factor and the sub-prompt chain branch and bound (SCBB) algorithm. Among them, the VCB factor can help reduce the number of ZT verifications and thus achieve the purpose of reducing the overall service cost. The specific implementation steps are as follows:

[0040] 1. VCB factor calculation: Three VCB calculation methods are designed for different scenarios, which are applicable to the following situations:

[0041] (1) No Verification: When consecutive cues can be deployed on the same PES node, no additional ZT verification is required due to the inherent security, so the VCB value is equal to the average of the computational cost of deploying the set of cues.

[0042] (2) PEP Verification: When continuous prompting is deployed between different PES nodes, PEP (Policy Enforcement Point) is used for ZT verification and routing cost is included in VCB calculation to control the overall cost of the service.

[0043] (3) vPEP Verification: When using virtual PEP (vPEP) for verification, although no additional routing is required, a vPEP instance needs to be created after each new deployment prompt, and the VCB value includes additional computational costs.

[0044] 2. Sub-prompt chain branch and bound (SCBB) algorithm: The SCBB algorithm proposed based on the VCB factor gradually selects the optimal PES node to deploy the prompt chain, thereby minimizing the service cost. The core steps of the algorithm are as follows:

[0045] (1) Algorithm initialization: First, initialize the forwarding path (For_Path), starting from the source node (\mathbb{S}), and remove the first prompt (p0) of the prompt chain PrC, and then gradually deploy it starting from the first prompt.

[0046] (2) Gradually deploy the hint chain: In each iteration, the algorithm calculates the VCB value of each PES node to deploy the current hint set, selects the node with the lowest VCB value as the optimal deployment node, updates the currently deployed sub-path Sub_Path, and adds it to the total path (For_Path).

[0047] (3) Iterative update: After each set of prompts is deployed, the index is updated and the next set of prompts is deployed until the entire prompt chain is deployed.

[0048] (4) Select GPT server: After all prompts are deployed, the GPT server with the lowest cost is selected to process the final query and update it to the forwarding path.

[0049] This solution effectively solves the cost conflict between tip chain deployment and ZT verification, minimizes the global service cost through the calculation of the SCBB algorithm and VCB factor, and provides a cost-effective deployment solution for the tip chain in the ZT network.

[0050] Example 2

[0051] In this embodiment, the SCBB (sub-prompt chain branch and bound) algorithm is used to show the optimization deployment process of the prompt chain in detail (such as Figure 1 The following is a detailed description of each step:

[0052] 1. Iteration 1: First, the first prompt p1 needs to be deployed. According to the topology, both PES nodes A and B are capable of carrying p1. The VCB factors of PES A and B are calculated to be 5 and 6 respectively. p1 is chosen to be deployed on PES A, and a virtual PEP (vPEP) is installed on the source node S for zero trust (ZT) verification. This choice minimizes the deployment cost of the first step. At this point, the last deployed node in the forwarding path is updated to PES A. This operation not only meets the security requirements of the ZT network, but also reduces the number of verifications on the transmission path, saving additional verification costs.

[0053] 2. Step 2 (Iteration 2): Next, the second prompt p2 needs to be deployed. At this point, PES nodes C and D can host the deployment of p2. According to the optimization calculation of the SCBB algorithm, the second prompt p2 and the third prompt p3 are chosen to be jointly deployed on the same PES node D. This method further reduces the overhead of ZT verification by deploying continuous prompts on the same node. In order to ensure the ZT verification requirements, PEP verification is performed between PES A and PESD. The advantage of this choice is that PEP verification can provide security during transmission without having to add the cost of vPEP to each node (the cost of choosing PEP verification is lower). At the same time, since p2 and p3 are jointly deployed on PESD, the number of ZT verifications is reduced, thereby reducing the total verification cost.

[0054] 3. Iteration 3: When all prompts are deployed, the algorithm enters the final step, which is to select a GPT server to handle the final query. The SCBB algorithm selects GPT server G2 to minimize the final service cost. GPT server G2 not only meets the service requirements, but also provides a relatively low transmission cost, further optimizing the deployment cost of the entire prompt chain.

[0055] Detailed calculation of total service cost: The VCB factors in the first two steps are both 5, carrying 1 and 2 prompt functions respectively. In the third step, the server G2 with the lowest service cost is selected, and the cost is 3. Therefore, the total cost is: 5+(5*2)+3=18.

[0056] Through the deployment process of the SCBB algorithm, it is demonstrated how to minimize costs by reasonably selecting PES nodes and ZT verification strategies. This optimization scheme not only effectively reduces the total service cost by reducing the number and frequency of PEP and vPEP verification, but also fully considers the balance of computing resources and transmission resources of the path. The final results show that the SCBB algorithm has obvious advantages in resource efficiency and cost control, and provides a good solution for the economical and efficient deployment of the prompt chain in the zero-trust network.

[0057] Effect experiment:

[0058] 1. Experimental process:

[0059] In the experiment, this embodiment uses the NY-20 topology for network simulation. The network includes 20 nodes and provides 7 prompts, of which 4 nodes are randomly selected as GPT servers, 1 to 4 nodes are used as PEPs (policy enforcement points), and the remaining nodes are used as PESs (prompt execution servers). The computing power of each PES is 1000 to 2000 CPU cores and supports 4 prompt instances. The bandwidth capacity of the link is between 10 and 40Gbps, and the weights are continuously distributed between 5 and 20. Each AI-SR (artificial intelligence service request) contains 3 to 7 prompts, each prompt consumes 0.5 to 0.8 CPU cores, and each vPEP consumes 0.6 CPU cores. The bandwidth requirement is between 1 and 5Mbps, and the source nodes are randomly selected in the network.

[0060] The experiment was conducted in 80 different networks, with 4,000 AI-SR requests in each network. The following three solutions were used as benchmarks for comparison:

[0061] (1) Prompt Chain Deployment Only (PCDO): ​​Deploy AI-SR using only existing chain service deployment technology without zero trust (ZT) verification.

[0062] (2) Prompt Chain Deployment First Approach (PCDF): First, the prompt chain path is obtained through PCDO, and then the cheapest ZT verification method is selected for the adjacent prompt nodes in the path.

[0063] (3) Local Optimal First Deployment (LOFD): For each prompt node, the cheapest verification method is first selected, and then the nearest PES is found to deploy the next prompt instance.

[0064] 2. Results and data analysis:

[0065] Experimental results show that the SCBB (Sub-Cue Chain Branch and Bound) algorithm of this embodiment outperforms other benchmark solutions in various key performance indicators:

[0066] (1) The impact of chain length on cost:

[0067] ①Total Service Cost (TC):

[0068] Figure 2(a) in Fig. 3 shows the total service cost under different hint chain lengths. As the hint chain length increases, the service cost of all schemes increases, because more hints directly increase the computation cost and may extend the forwarding path.

[0069] SCBB consistently outperforms PCDF and LOFD in total service cost. SCBB exhibits lower overhead in both CPU computation cost and routing cost. Specifically, the average total service cost of SCBB is 4.67% lower than PCDF and 13.89% lower than LOFD.

[0070] ②Verification Cost:

[0071] SCBB deploys continuous prompts on the same PES node, thus reducing the number of PEP or vPEP verifications and significantly reducing the verification cost. In terms of verification cost, SCBB is 11.86% and 30.50% lower than PCDF and LOFD, respectively. Figure 2 (b) in Figure 2 further verifies this point. The SCBB scheme always has the highest "No Verification" ratio under different prompt chain lengths, showing the advantage of SCBB in reducing verification costs. For example, when the prompt chain length is 3, the No Verification ratio of SCBB is 43.38%, while PCDF and LOFD are 34.4% and 30.01% respectively; when the prompt chain length is 7, the No Verification ratio of SCBB increases to 54.50%, while PCDF and LOFD are 45.2% and 33.78% respectively.

[0072] (2) Impact of PEP quantity on cost:

[0073] Figure 3 The effect of different numbers of PEPs on the total service cost is shown. As the number of PEPs increases, the cost of the PCDO scheme increases because more PESs are selected as PEPs and prompt instances cannot be deployed. The performance of the SCBB, PCDF, and LOFD schemes changes slightly because the convenience of PEP verification and the deployment cost caused by the reduction of PESs offset each other. In this scenario, SCBB can still maintain low computing and routing costs, which is advantageous over other schemes.

[0074] (3) Acceptance Ratio (AcP):

[0075] Figure 4The average acceptance rate when 4,000 requests were deployed in 80 networks is shown. Since the SCBB scheme has the highest "no verification" ratio, its resource utilization efficiency is the best, which is reflected in the lower total service cost. The numerical value shows that the acceptance rate of SCBB is 56.24%, while the acceptance rates of PCDF and LOFD are 51.75% and 28.18% respectively. SCBB is significantly better than other schemes.

[0076] (4) Runtime:

[0077] In terms of running time, SCBB shows high efficiency. The average time required to deploy an AI-SR request is 133.14 milliseconds, which is faster than PCDF's 191.95 milliseconds and slightly higher than LOFD's 123.73 milliseconds. This shows that SCBB not only has advantages in cost control, but also maintains a relatively fast deployment efficiency.

[0078] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A low-cost deployment method for prompt chaining in a zero-trust access network, characterized in that: The following steps are involved: Step 1: Introduce the verification cost balance VCB factor to reduce the number of verifications and related expenses of zero trust ZT; Step 2: Based on the verification cost balance VCB factor, a sub-prompt chain branch and bound SCB algorithm is proposed to gradually select the optimal PES node to deploy the prompt chain, thereby minimizing the service cost.

2. The low-cost deployment method of the prompt chain in a zero-trust access network according to claim 1, characterized in that: In step 1, the calculation method of the verification cost balance VCB factor specifically includes: (1) In the no-verification scenario: when continuous prompts can be deployed on the same PES node, no additional verification is required due to the inherent security guarantees within the node, so the value of the verification cost balance VCB factor is equal to the average value of the computational cost of deploying prompts; (2) In the PEP verification scenario, when continuous prompts are deployed on different PES nodes, the PEP is used for zero-trust ZT verification, and the routing cost is included in the VCB calculation to control the overall cost of the service. (3) In the virtual policy enforcement point vPEP verification scenario: When using vPEP for verification, no additional routing is required. In the previous prompting instance, that is, the carrier of the prompting function installed on the PES is deployed, and then a vPEP instance is created. The VCB value includes the additional computational cost.

3. The low-cost deployment method of prompt chain in zero-trust access network according to claim 2, characterized in that: The step 2 specifically includes the following steps: Step 2.1, algorithm initialization: First, initialize the forwarding path, remove the zeroth prompt of the prompt chain from the source node, that is, the service starting point, and then gradually deploy from the first prompt; Step 2.1, gradually deploy the prompt chain: In each iteration, the algorithm calculates the VCB value of each PES node to deploy the current possible prompt set, selects the node with the lowest VCB value as the optimal deployment node; updates the currently deployed sub-path and adds it to the total path; Step 2.3, iterative update: After each set of prompts is deployed, update the index and continue to deploy the next set of prompts until the entire prompt chain is deployed; Step 2.4, select GPT server: After all prompts are deployed, select the GPT server with the lowest cost to process the final query and update it into the forwarding path.

Images