A game intelligent distribution method and system
By using content-based variable-length chunking and Merkle hash tree-based data block management, combined with multi-source scheduling and P2P node reputation feedback, the game distribution process is optimized, solving the problems of long initial playtime and repeated transmission, and achieving efficient game resource distribution.
Patent Information
- Application Number
- CN202611224065.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-13
- Publication Date
- 2026-11-17
AI Technical Summary
The existing game distribution process suffers from issues such as long initial playtime, duplicate data transmissions due to data block anomalies, and a high proportion of network traffic for high-priced content distribution.
We employ content-based variable-length block segmentation technology, combined with Merkle hash trees and block hashes, to perform data block integrity verification and prediction. Through multi-source scheduling and P2P node reputation feedback, we optimize download source selection and edge preheating, achieving block-by-block verification and near-end caching.
Shorten the initial play time, reduce the repeated transmission of abnormal data blocks, reduce the proportion of traffic on high-priced content distribution networks, and improve download efficiency and resource utilization.
Smart Images

Figure CN123123331A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer networks and content distribution technology, and in particular to a game intelligent distribution method and system. Background Technology
[0002] As game graphics quality, resource quantity, and interactive content increase, the data volume of game installation packages and version update packages distributed by game platforms is gradually increasing. Existing game platforms typically store game installation packages and patch files at a central origin server and provide download services to clients through edge nodes of a Content Delivery Network (CDN). After receiving a version update notification, the client downloads the corresponding files according to the file list provided by the platform's downloader, and performs verification and installation upon completion.
[0003] To alleviate bandwidth pressure on central origin servers and content delivery networks, some game platforms employ peer-to-peer (P2P) transmission, allowing clients that have acquired some game resources to share data with other clients. However, existing P2P-assisted distribution typically uses methods such as random selection, geographical proximity selection, or fixed-weight selection to determine peer nodes. When the available bandwidth, network latency, or online status of peer nodes changes, download speed fluctuations can easily occur. Furthermore, without a node evaluation mechanism that combines data block integrity verification results, low-quality nodes or abnormal data blocks can increase duplicate transmissions.
[0004] During game version updates, some existing solutions still use files or complete patch packages as the basic update unit. When the content of a resource file changes, the client may still need to re-fetch the larger file. Even with chunked downloads, the same content between different versions may not be directly reused, resulting in duplicate transmissions. On the other hand, when a new game or version is launched, a large number of clients initiate download requests at close intervals. Fixed caching strategies struggle to deploy high-demand content to edge nodes of the corresponding regions and carrier networks in a timely manner, easily creating bandwidth spikes.
[0005] Furthermore, game resources typically have startup dependencies. Executable files, login scenes, initial character resources, or the first level's resources need to be downloaded before other resources can be downloaded for the client to enter the playable state for the first time. Existing methods of sequential downloading or scheduling based solely on file size do not place resource dependencies, the quality of candidate download sources, data block integrity, and transmission costs in the same control process, thus making it difficult to balance the initial playability wait time, data integrity, and distribution costs.
[0006] Therefore, there is a need to provide a game intelligent distribution solution that is based on data block content identification and interconnects demand forecasting, edge warm-up, multi-source scheduling, block-by-block verification, and node reputation feedback. Summary of the Invention
[0007] In view of the shortcomings of the prior art described above, the purpose of this invention is to provide a game intelligent distribution method and system to solve the problems of long waiting time for first playability, repeated transmission due to data block anomalies, and high network traffic proportion for high-priced content distribution in the existing game distribution process.
[0008] To achieve the above and other related objectives, the present invention provides a game intelligent distribution method, comprising the following steps:
[0009] S1. Perform content-based variable-length chunking on the game version file to obtain multiple data blocks. Calculate the block hash for each data block. Construct content-addressable storage based on the block hash. Construct a Merkle hash tree based on the block hash of each data block. Generate a version list including data block logical offset, block hash, playable priority, and Merkle root.
[0010] S2. Based on the historical download data and operation data of multiple regional network nodes, predict the demand popularity of each game version within a future preset time period, determine the edge preheating set based on the demand popularity and version list, and preheat the data blocks corresponding to the edge preheating set to the edge cache or operator cache of the corresponding region.
[0011] S3. Content delivery network nodes, edge cache, carrier cache, peer nodes and local cache are used as candidate download sources. The scheduling state is constructed based on the network status, node reputation, content popularity, playability priority and transmission cost of the data block to be downloaded. The linear upper confidence bound algorithm is used to determine the target download source or combination of target download sources and parallel weights corresponding to the data block to be downloaded. The scheduling parameters of the corresponding candidate download source are updated online based on the effective throughput, transmission cost, verification result, bad block result and timeout result after the data block is transmitted.
[0012] S4. Obtain the list of peer nodes through the P2P tracking server and establish a peer-to-peer data transmission connection. Use the data block passing the integrity check as the condition for inclusion of effective contribution. Update the node reputation according to the effective contribution of the peer nodes and provide the updated node reputation to step S3 as the subsequent data block scheduling status.
[0013] S5. The client obtains data blocks from the target download source determined in step S3, and performs integrity verification on the data blocks based on the block hash and Merkle root. Data blocks that fail the verification are re-obtained from other candidate download sources, and the verification result is fed back for node reputation update in step S4. At the same time, data blocks are scheduled according to playability priority. When the minimum set of playable data blocks is downloaded and passes the integrity verification, the client is allowed to start the game, and the remaining data blocks continue to be downloaded during game operation.
[0014] S6. Based on the measured transmission rate, transmission cost and service level of each candidate download source obtained in step S3, perform cost-experience joint optimization, adjust the source weight and unit price preference of each candidate download source, and feed the adjustment results back to the edge preheating in step S2 and the multi-source scheduling in step S3.
[0015] By adopting the above technical solution, step S1 reduces repeated downloads across versions and provides a unified block identifier for each stage; steps S2 and S3 enable high-demand data blocks to enter the near-end cache in advance and dynamically select sources based on real-time network conditions; steps S4 and S5 associate data block verification results with P2P node reputation and subsequent scheduling, and prioritize the completion of data blocks required for the first run; step S6 continues to adjust the warm-up and scheduling based on the measured rate and cost. This shortens the initial playability wait time, reduces the repeated transmission of abnormal data blocks, and lowers the proportion of high-priced content delivery network traffic.
[0016] In one embodiment of the present invention, in step S1, a rolling hash is used to traverse the game version file. When the rolling hash value meets the preset block division condition, the data block boundary is determined so that the average size of the data block is the preset average block size. The version list also records the file path, data block length, logical offset of the data block in the file, and data block order. The Merkle root is obtained by using the block hash of each data block as a leaf node and calculating the hash layer by layer on the hash concatenation result of adjacent child nodes.
[0017] By adopting the above technical solution, the data block boundaries are determined based on the content, so that data blocks that have not changed during local modifications can still be reused; combined with Merkle root for block-by-block verification, only the corresponding data block is pulled when a data block is abnormal.
[0018] In one embodiment of the present invention, the data block is encrypted using the key of the corresponding game during transmission and static storage, and the block hash is calculated based on the plaintext of the data block; each game corresponds to an independent content addressing storage namespace, and the data block is identified by a combination of game identifier, block hash and data block length, and cross-game data block reuse is performed only when different games have shared authorization.
[0019] By adopting the above technical solution, while maintaining a stable block hash for the same content, the scope of data block reuse is limited by game keys, independent namespaces, and shared authorization, making content reuse and data isolation compatible.
[0020] In one embodiment of the present invention, in step S2, a combination of region and operator is used as a regional network node, and the regional network nodes are connected by geographical adjacency or same operator relationship, and the edge weight is determined according to the historical round-trip delay between regional network nodes; the download volume, game reservation increment, server online peak, time features and version release flag of each regional network node in multiple historical time periods are input into a graph neural network for spatial feature aggregation, and then the obtained node spatial feature sequence is input into a time series model, and after fully connected regression, the demand heat matrix corresponding to the region, prediction time period and game version is output.
[0021] By adopting the above technical solution, the spatial characteristics are incorporated into the association of region, operator and historical network, and combined with time series data to predict the demand for game versions in each region, so as to provide a basis for the preheating of data blocks in the corresponding region.
[0022] In one embodiment of the present invention, the graph neural network is a two-layer GraphSAGE network, and the temporal model is a long short-term memory network; the regions with the highest demand popularity are selected and combined with game versions according to the demand popularity matrix, and corresponding data blocks are selected according to the playability priority in the version list, forming the edge preheating set under the cache capacity constraint of the edge cache; the deviation between the actual request volume and the predicted request volume is used as an online incremental update sample to update the subsequent demand popularity prediction results and adjust the subsequent edge preheating set.
[0023] By adopting the above technical solution, under the constraint of cache capacity, the preheating data block is determined by combining demand popularity and playability priority, and the subsequent prediction is corrected by using the actual request deviation. This can reduce invalid preheating and improve the near-end hit rate of first playable resources.
[0024] In one embodiment of the present invention, in step S3, the scheduling state includes the round-trip latency, available bandwidth, packet loss rate, node reputation, content popularity, playability priority, and real-time unit price of the candidate download source; the linear upper confidence bound algorithm is used to determine the download source based on the predicted revenue and exploration confidence terms of the candidate source arm, and a reward is formed based on the effective throughput, transmission cost, verification results, bad block results, and timeout results, and the candidate source arm parameters are updated block by block.
[0025] By adopting the above technical solution, multi-source scheduling can adjust the download source of subsequent data blocks based on the real-time quality, cost, and transmission results of the preceding data blocks.
[0026] In one embodiment of the present invention, peer nodes obtain a node list through a P2P tracking server and establish a data transmission connection through STUN, TURN, or ICE protocols; only data blocks that pass the block hash and Merkle check in step S5 are counted as valid contributions, and node reputation is updated according to the valid contributions; the client only receives data blocks whose block hashes match the official content addressing storage and version list.
[0027] By adopting the above technical solution, the reputation of peer nodes is associated with the actual effective upload results, and the reputation is fed back to subsequent multi-source scheduling.
[0028] In one embodiment of the present invention, in step S5, the version list sets playability priority based on the data blocks required for game startup and determines the minimum set of runnable data blocks; download sources are allocated to data blocks with high playability priority first; when all data blocks in the minimum set of runnable data blocks have been downloaded and passed integrity verification, the client is allowed to start the game, and the remaining data blocks continue to be downloaded during game operation.
[0029] By adopting the above technical solution, the data blocks required for the first run are completed first, eliminating the need to wait for the complete game package to be downloaded before starting the game.
[0030] A game intelligent distribution system includes: a central source station for storing game version files and raw data blocks;
[0031] The version management server is used to perform content-based variable-length chunking on game version files, calculate the block hash of each data block, construct content-addressable storage and Merkle hash tree, and generate version manifests.
[0032] The prediction server is used to predict demand based on historical download and operational data from multiple regional network nodes, and to determine the edge preheating set based on demand and version list.
[0033] Edge cache and carrier cache are used to cache data blocks corresponding to the edge preheating set according to the preheating instructions;
[0034] The scheduling server is used to select content distribution network nodes, edge caches, carrier caches, peer nodes and local caches as candidate download sources. It performs linear upper confidence bound multi-source scheduling based on the network status, node reputation, content popularity, data block playability priority and transmission cost of the candidate download sources, and adjusts the source weight and unit price preference of each candidate download source based on the measured transmission rate, transmission cost and service level.
[0035] A P2P tracking server is used to maintain a list of peer nodes and provide network address translation (NAT) traversal signaling.
[0036] The client is used to determine the data blocks to be downloaded based on the version list, obtain the data blocks according to the scheduling results, perform block hash and Merkel integrity verification on the received data blocks, re-obtain abnormal data blocks based on the verification results, and start the game after the minimum set of runnable data blocks has been downloaded and the verification has passed.
[0037] Among them, the data blocks that pass the integrity verification are used to count the effective contributions of peer nodes and update the node reputation. The updated node reputation is fed back to the scheduling server for subsequent data block scheduling. The actual transmission information and cost-experience joint optimization results of the scheduling server are fed back to the prediction server and the scheduling server for subsequent edge warm-up and multi-source scheduling.
[0038] By adopting the above technical solution, the central source station, version management server, prediction server, scheduling server, edge cache, carrier cache, P2P tracking server, client, and local cache respectively undertake version organization, demand prediction, data preheating, download source decision-making, near-end caching, peer node management, and client verification. Through block hashing and transmission feedback, information corresponding to the same data block is exchanged, enabling the prediction preheating, block-by-block scheduling, integrity verification, and reputation feedback processes to be continuously executed among distributed network nodes. Therefore, this system can implement the above closed-loop control in a real-world game platform environment where network source quality dynamically changes, and correspondingly achieves the technical effects of reducing the wait time for initial playable resource acquisition, reducing the repeated transmission of abnormal data blocks, and reducing reliance on high-cost content distribution network traffic when transmission conditions are met.
[0039] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the game intelligent distribution method described above.
[0040] As described above, the game intelligent distribution method and system of the present invention have the following beneficial effects:
[0041] Firstly, by reusing unchanged data blocks based on content-based variable-length chunks, high-demand data blocks are preheated to the near-end cache according to regional demand, and playability priority is incorporated into block-by-block multi-source scheduling. After the minimum set of runnable data blocks has been downloaded and verified, the client is allowed to start the game, so that download bandwidth is prioritized for the content required for the first run.
[0042] Secondly, by verifying the received data blocks one by one through block hash and Merkel check, in case of an anomaly, only the corresponding data block is reacquired and the anomaly result is fed back to the reputation of the download source, so that subsequent scheduling reduces the selection of the anomaly source, thereby avoiding the continuous transmission of bad blocks caused by the same anomaly source;
[0043] Third, the pre-warmed edge cache, carrier cache, peer nodes and local cache are used together with the content delivery network nodes as candidate download sources. Transmission cost, network quality, node reputation and playability priority are used together in the download source decision. When the service quality requirements are met, the chance of the near end or low cost source to undertake the transmission of data blocks is increased. When the network quality is insufficient, the content delivery network nodes supplement the transmission.
[0044] Fourth, by feeding back the actual request volume, transmission status and verification results to the subsequent scheduling and the warm-up decision of the next prediction cycle, the current distribution results can correct the download source of subsequent data blocks and the deployment location of subsequent caches. Thus, demand prediction, edge warm-up, multi-source scheduling, integrity verification and node reputation are not independent of each other, but form a feedback loop around the data block. Attached Figure Description
[0045] Figure 1 The diagram shown is a schematic representation of the deployment architecture of the game intelligent distribution system disclosed in an embodiment of the present invention.
[0046] Figure 2 The diagram shows a schematic of the new version release, edge warm-up, multi-source download, and feedback closed-loop process disclosed in this embodiment of the invention.
[0047] Figure 3 The diagram shown is a schematic representation of the demand heat prediction structure of the graph neural network and time series model disclosed in this embodiment of the invention.
[0048] Figure 4 The diagram shows the multi-source scheduling state, action, environment, and reward based on the linear upper confidence bound algorithm disclosed in this embodiment of the invention.
[0049] Figure 5 The diagram shown is a schematic of the unified block hash index disclosed in this embodiment of the invention, which integrates deduplication, scheduling, verification, and incentives.
[0050] Component designation explanation
[0051] 11. Central Source Server; 12. Version Management Server; 13. Prediction Server; 14. Scheduling Server; 15. Edge Cache; 16. Carrier Cache; 17. P2P Tracking Server; 18. Client; 19. Local Cache;
[0052] 310. Input graph; 320. GraphSAGE layer; 330. LSTM time series; 340. Fully connected regression layer; 350. Heat matrix output;
[0053] 410. Scheduling Status; 420. LinUCB Intelligent Scheduling Unit; 430. Scheduling Actions; 440. Download Environment; 450.
[0054] award;
[0055] 510. Deduplication; 520. Scheduling; 530. Verification; 540. Incentive processing. Detailed Implementation
[0056] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.
[0057] Please see Figures 1 to 5 It should be understood that the structures, proportions, sizes, etc., illustrated in the accompanying drawings are merely for illustrative purposes to aid those skilled in the art and to facilitate understanding and reading. They are not intended to limit the scope of the invention and therefore have no substantial technical significance. Any modifications to the structure, changes in proportions, or adjustments to size, without affecting the effectiveness and purpose of the invention, should still fall within the scope of the technical content disclosed in this invention. Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity and not intended to limit the scope of the invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention's implementation.
[0058] Example 1:
[0059] Please see Figures 1 to 5 This embodiment provides a game intelligent distribution system, including a central source station 11, a version management server 12, a prediction server 13, a scheduling server 14, an edge cache 15, a carrier cache 16, a P2P tracking server 17, a client 18, and a local cache 19. Each functional unit uses the block hash of a data block as a unified identifier, enabling content addressing, edge preheating, multi-source scheduling, P2P node governance, integrity verification, and cost optimization to be mapped to specific data blocks.
[0060] Central origin server 11 is used to store game version files, raw data blocks, and release resources corresponding to the game version. After the game developer releases a new version, central origin server 11 provides the corresponding version file to version management server 12; when edge cache 15, operator cache 16, or client 18 needs to obtain the corresponding data block, central origin server 11 can also act as a back-to-origin node to provide raw data blocks according to the block hash.
[0061] Version management server 12 communicates with central source station 11 to perform content-based variable-length chunking on the game version file, obtaining multiple data blocks, and calculating block hashes based on the plaintext content of each data block. In this embodiment, the block hash is represented using SHA-256 digest. Version management server 12 establishes a content-addressable storage index based on the block hash, and records the file path, data block length, data block order, and logical offset of the data blocks in the file, so that client 18 can restore the target version file according to the version list after downloading.
[0062] Version management server 12 sets playable priorities based on the data blocks required for game startup and determines the minimum set of data blocks that can run.
[0063] To facilitate block-by-block integrity verification by client 18, version management server 12 uses the block hash of each data block as the leaf node of the Merkle tree, calculates the hash layer by layer for the hash concatenation results of adjacent child nodes, and obtains the Merkle root corresponding to the game version; the version list records the Merkle check information corresponding to each data block and the Merkle root. After obtaining the data block, client 18 can determine whether the current data block corresponds to the official version list based on the block hash and Merkle check path, so that when a single data block is abnormal, only the corresponding data block needs to be re-obtained.
[0064] The prediction server 13 communicates with the version management server 12 and the scheduling server 14 to predict the demand for each game version within a preset time period based on historical download and operational data from multiple regional network nodes. Please refer to... Figure 3 In this embodiment, the combination of region and operator is used as the regional network node in the input graph 310. Graph edges are established based on geographical adjacency or same operator relationship, and edge weights are determined based on the historical round-trip delay between regional network nodes. The input features are aggregated spatially through the GraphSAGE layer 320, and then the heat matrix is output 350 through the LSTM time series layer 330 and the fully connected regression layer 340.
[0065] Prediction server 13 generates an edge preheating set E based on the demand heat matrix and version list. When generating edge preheating set E, it first determines the combination of regions, games, and versions with high demand heat. Then, under the cache capacity constraints of edge cache 15 or carrier cache 16, it prioritizes data blocks with higher playability and records the data blocks requiring preheating using block hashes. Scheduling server 14 generates preheating instructions based on edge preheating set E, causing the corresponding data blocks to be pre-entered into the edge cache 15 or carrier cache 16 of the corresponding region before the client 18 initiates an update request.
[0066] Edge cache 15 and carrier cache 16 are used to cache data blocks according to the preheating instructions and report data block identifiers, cache hit status, current load and available transmission capacity to the scheduling server 14 as candidate download sources for the corresponding areas.
[0067] P2P tracking server 17 maintains online peer nodes and the data block identifiers held by each peer node, and provides the client 18 with a list of peer nodes and STUN, TURN, or ICE network address translation (NAT) penetration signaling. P2P collaborative distribution uses the integrity verification result of client 18 as the basis for contribution statistics, counting only data blocks that pass verification as valid contributions, and updating node reputation based on valid contributions. The updated node reputation is then provided to scheduling server 14 for subsequent data block scheduling.
[0068] The scheduling server 14 is used to obtain candidate download sources corresponding to the data blocks to be downloaded by the client 18. The candidate download sources include at least one of the following: content delivery network nodes, edge cache 15, carrier cache 16, peer nodes, and local cache 19. The scheduling server 14 obtains the round-trip latency, available bandwidth, and packet loss rate for each candidate download source, and combines the node reputation, the demand popularity of the data blocks to be downloaded, the playability priority, and the transmission cost of the candidate download sources to form a scheduling state.
[0069] Please see Figure 4 The LinUCB intelligent scheduling unit 420 in the scheduling server 14 determines the scheduling action 430 based on the scheduling status 410, enabling the client 18 to obtain data blocks from the candidate download source in the download environment 440. After the data block transmission is completed, a reward 450 is generated based on the effective throughput, transmission cost, verification results, bad block results, and timeout results, and the parameters of the corresponding candidate source arm are updated.
[0070] Client 18 reads the version list, retrieves the local cache 19 based on the block hash, determines the data block to be downloaded, and obtains the data block according to the target download source determined by the scheduling server 14. Client 18 recalculates the block hash of the received data block and performs integrity verification in conjunction with the Merkle root; if the verification fails, it only re-acquires the corresponding abnormal data block and feeds back the verification result for P2P node reputation updates.
[0071] Client 18 also prioritizes acquiring the data blocks required for the first run based on the playability priority in the version list. Once the minimum set of playable data blocks has been downloaded and passed integrity verification, client 18 allows the game to launch, and the remaining data blocks continue to download during game execution.
[0072] Local cache 19 is set on the client 18's local machine or on a storage device on the same local area network as the client 18, and is used to store data blocks that have passed integrity verification. The client 18 queries local cache 19 based on the block hash. When local cache 19 has a data block corresponding to the version list, it directly uses that data block as a local candidate source; thus, unchanged data blocks can continue to be reused when the version is updated, reducing the amount of data transferred from remote sources.
[0073] Please see Figure 5 The block hash h forms a unified index for data blocks throughout the system: Deduplication 510 uses the block hash h to identify identical data blocks in local caches or different versions; Scheduling 520 uses the block hash h to query candidate download sources holding data blocks to be downloaded; Verification 530 recalculates the block hash of the received data block and verifies its integrity using Merkle check paths; Incentive 540 calculates effective contributions and updates node reputation based on the verification results. Therefore, the processing results of the same data block across different functional entities can be mapped using the block hash.
[0074] The scheduling server 14 also performs joint optimization based on the measured transmission rate, real-time billing cost, and service level execution cost-experience of each candidate download source. During peak bandwidth periods, it increases the scheduling ratio of low-cost sources such as edge cache 15, carrier cache 16, or P2P, while during off-peak periods, it is supplemented by the content delivery network. The optimized source weights and unit price preferences are fed back to the prediction server 13 for subsequent preheating and the scheduling server 14 for subsequent multi-source scheduling.
[0075] Through the above system architecture, content addressing and segmentation, demand prediction, edge preheating, multi-source scheduling, P2P reputation, integrity verification, download-while-playing, and cost optimization are interconnected according to the same data block identifier, so as to solve the problems of repeated downloads, waiting for the first playable, abnormal retransmission, and the high proportion of network traffic for high-priced content distribution.
[0076] Example 2:
[0077] Please see Figure 2 This embodiment provides a method for distributing game resources using the game intelligent distribution system described in Embodiment 1. This method uses block hashes as unified data block identifiers and specifically includes the following steps:
[0078] Step S1: The version management server 12 uses a content-based variable-length chunking method to chunk the game version file, calculates the SHA-256 block hash for each data block, constructs content-addressable storage based on the block hash, and constructs a Merkle hash tree based on the block hash of each data block.
[0079] Specifically, rolling hashes are used to determine block breakpoints along the file data stream, resulting in an average data block size of approximately 4MB. The version manifest records the file path, data block length, data block order, logical offset, block hash, and playable priority. Data blocks with the same block hash can be reused across different versions.
[0080] The Merkle hash tree uses the block hash of the data block as the leaf node, calculates the hash of the hash concatenation result of the adjacent child nodes layer by layer to obtain the Merkle root, and sends the Merkle root to the client 18 along with the version list.
[0081] Step S2: The prediction server 13 predicts the demand for each game version within the next T time periods based on historical download data, game pre-registration volume, peak online activity in the region / server, time characteristics, and version release markers from multiple regional network nodes, and determines the edge preheating set E based on the demand and version list.
[0082] Specifically, the combination of region and operator is used as the regional network node, the graph edges are established based on geographical adjacency or same operator relationship, and the edge weights are determined based on the historical average round-trip delay; the node features are aggregated into spatial features by the GraphSAGE network and then input into the LSTM time series model, and the demand heat matrix H is obtained by fully connected regression.
[0083] Based on the demand heat matrix H, select high-heat regions, game and version combinations, and under the constraint of cache capacity, prioritize the selection of corresponding high-heat data blocks to form an edge preheating set E, and preheat the data blocks to the edge cache 15 or carrier cache 16 of the corresponding region.
[0084] in:
[0085] Node V: Each graph node is defined as "Province × Operator" (approximately 93 nodes across the entire network), representing the smallest geographic / network unit that can be independently warmed up.
[0086] Edge E: Connects nodes that are "geographically adjacent" or "on the same operator"; Edge weight w_e = normalized (1 / historical average RTT), the smaller the RTT, the greater the weight (or degenerates to 1 / geographical distance);
[0087] The node feature x_v ∈ R^d (d=32) includes: download volume sequence for the past N=24 time periods (hourly), game pre-registration increment, peak online time in the region / server, time period / weekly one-hot, and version release flag;
[0088] Spatial encoding: The neighborhood is aggregated using two layers of GraphSAGE (hidden dimension 64) to obtain the node embedding e_v (i.e., "the feature of this region after being influenced by its surroundings").
[0089] Temporal coding: The time series of e_v is fed into an LSTM (64 hidden dimensions, 2 layers; or it can be replaced by a Temporal Transformer) to predict the demand popularity h_v,t ∈ R^G (G = number of game versions) for each game version in the future T=6 time periods.
[0090] Input / Output: Input = Historical multidimensional sequence + graph structure; Output = Popularity matrix H ∈ R^{|V|×T×G} (one popularity value for each node, each time period, and each game version);
[0091] Training / Inference: Supervised regression, loss L = MSE(H_pred, H_true) + λ·L2; online incremental updates after deployment to adapt to sudden surges in popularity;
[0092] How popularity becomes a preheating task (key control relationship): For the top-K (region, game, version) hot blocks in H, issue a "preheating instruction" to push the corresponding blocks to the edge cache / carrier cache of that region in advance - that is, "content arrives before demand arrives".
[0093] Step S3: Abstract the content distribution network nodes, edge cache 15, carrier cache 16, peer nodes and local cache 19 into a candidate download source resource pool, and perform multi-source fusion adaptive scheduling based on the real-time network status, node reputation, content popularity, playable data block priority and real-time unit price of each candidate download source.
[0094] For the data block b to be downloaded, its scheduling state is represented as follows:
[0095] s_b = [Latency l_i of each candidate source, available bandwidth b_i, packet loss rate p_i, node reputation rep_i, content heat(b), block playability priority prio(b), real-time unit price of each source c_i]; the scheduling action is represented as: a_b = the source combination and parallel weight selected for this data block.
[0096] w = (w_CDN, w_edge, w_P2P) and satisfies Σw = 1.
[0097] Scheduling server 14 uses LinUCB for decision-making, taking state s_b as context x, and maintaining parameters (A_a, b_a) for each candidate source arm a, according to:
[0098] a*=argmax_a(θ̂_aᵀx+κ·√(xᵀA_a⁻¹x))
[0099] Candidate sources are selected, where κ is the exploration coefficient.
[0100] After each data block is transmitted, it is done according to:
[0101] r_b=α·throughput(b)−β·Σ(c_i·w_i)+γ·verifyOK(b)−δ·(badBlock(b)+timeout(b))
[0102] Calculate the reward and update (A_a, b_a) of the corresponding candidate source arm based on the actual test context and the reward.
[0103] In step S4, during P2P collaborative distribution, client 18 obtains a list of peer nodes through P2P tracking server 17 and establishes a point-to-point data transmission connection through STUN, TURN, or ICE protocols; a node reputation mechanism based on effective contributions is established, with the data block that passes integrity verification in step S5 serving as the condition for counting effective contributions.
[0104] The contribution unit of a peer node is the number of valid uploaded block hashes. Node reputation is updated according to rep_i←rep_i+η·(valid contribution − average contribution), where η is the learning rate. Nodes that fail verification or are inefficient have their node reputation reduced. The updated node reputation rep_i is provided to step S3 as part of the subsequent data block scheduling status. The client only accepts data blocks whose block hashes match the official content addressing storage and version list.
[0105] In step S5, client 18 obtains data blocks from the target download source determined in step S3 and performs integrity verification on the received data blocks; at the same time, it schedules data blocks according to the playable priority in the version list to support downloading and playing simultaneously.
[0106] Specifically, client 18 recalculates the block hash of the received data block and verifies it by combining the Merkle check information with the Merkle root in the version list; data blocks that fail to be verified are pulled from other candidate download sources, only abnormal data blocks are retransmitted, and the verification result is fed back to step S4 for reputation update of the corresponding P2P node.
[0107] Version management server 12 determines the minimum set of playable data blocks based on the data blocks required for game startup and sets the playability priority of the data blocks. Scheduling server 14 prioritizes scheduling data blocks with higher playability priority. When the minimum set of playable data blocks has been downloaded and passed integrity verification, client 18 is allowed to start the game, and the remaining data blocks continue to download in the background.
[0108] In step S6, the scheduling server 14 obtains the source weight and unit price preference of each candidate download source based on the measured transmission rate, real-time billing cost and service level execution cost-experience joint optimization of each candidate download source obtained in step S3.
[0109] During peak bandwidth periods, the proportion of data block transmission handled by low-cost sources such as edge cache 15, carrier cache 16, and P2P is increased to reduce traffic to the high-priced content delivery network; during off-peak periods, the content delivery network replenishes the bandwidth. The source weights and unit price preferences are fed back to the edge preheating decision in step S2 and the multi-source scheduling in step S3.
[0110] Steps S2 to S6 form a feedback loop: the edge preheating set E provides near-end candidate download sources for step S3; the measured rate and cost of step S3 are fed back to step S6; the bad block verification result of step S5 is fed back to step S4 to update node reputation; the node reputation rep_i of step S4 re-enters step S3; the optimization result of step S6 further adjusts subsequent preheating and multi-source scheduling.
[0111] Through the above steps, incremental updates of content addressing, demand forecasting and edge warm-up, multi-source adaptive scheduling, P2P contribution reputation, integrity verification and download-while-playing, and cost-experience joint optimization are executed continuously according to the interaction relationship.
[0112] Steps S1 to S6 in this embodiment are executed in conjunction with the version management server 12, prediction server 13, scheduling server 14, P2P tracking server 17, client 18 and related cache nodes in embodiment one, thereby illustrating the main operation process of the system described in embodiment one.
[0113] Example 3:
[0114] This embodiment, based on Embodiments 1 and 2, further explains the content-addressed block segmentation, data block encryption, and cross-game reuse methods. When the version management server 12 uses rolling hashing for content-based variable-length block segmentation, it can set the target average block size, minimum block size, and maximum block size according to the game resource type; for example, a larger target average block size is set for larger resource packages, and a smaller target average block size is set for small resource files that need to be updated frequently. As long as the data block boundaries are still determined according to the content characteristics, they can be used to maintain the stability of data blocks in areas that have not changed under local modifications.
[0115] Data blocks can be encrypted using AES-256-GCM with the corresponding game's key during transmission and static storage, while the block hash is calculated based on the plaintext of the data block before encryption. Each game has an independent content-addressable storage namespace, and the data block identifier is a combination of the game identifier, block hash, and data block length; cross-game data block reuse is only performed when different games have shared licenses. When the combined identifiers are the same but the system determines that the data is still abnormal, byte comparison can be further performed to reduce the possibility of erroneous reuse caused by index anomalies.
[0116] Example 4:
[0117] This embodiment, based on Embodiments 1 and 2, further explains the specific parameters for demand heat prediction and edge warm-up. In this embodiment, regional network nodes are constructed according to provincial regions and operator combinations, with node feature dimension d=32, including download volume sequence of the past N=24 hours, game reservation increment, server online peak, time period / weekly features, and version release flag; GraphSAGE layer 320 adopts a two-layer network with a hidden dimension of 64, and LSTM time series 330 adopts a two-layer network with a hidden dimension of 64, used to predict demand heat in the future T=6 time periods.
[0118] The demand heat prediction model is trained using supervised regression, outputting a heat matrix H∈R^{|V|×T×G}, and is trained according to the loss function L=MSE(H_pred,H_true)+λ·L2, where H_pred is the predicted demand heat, H_true is the actual demand heat, and λ is the coefficient of the L2 regularization term. In this embodiment, λ can be taken as 1e-4.
[0119] In addition to the GraphSAGE network, spatial feature aggregation can also be achieved using GCN or Graph Attention; in addition to LSTM, temporal models can also use Temporal Transformer.
[0120] Example 5:
[0121] This embodiment further explains the parameters of steps S3 and S4 based on embodiments one and two.
[0122] The reward function for step S3 is:
[0123] r_b=α·throughput(b)−β·Σ(c_i·w_i)+γ·verifyOK(b)−δ·(badBlock(b)+timeout(b))
[0124] Where α=1.0, β=0.5, γ=0.3, δ=2.0; the exploration coefficient κ can be 1.0 during cold start and can be reduced to 0.2 after stabilization. The node reputation in step S4 is updated according to rep_i←rep_i+η·(effective contribution − mean contribution). The above parameters are used to illustrate an implementable configuration.
[0125] When the latency, packet loss rate, bad block rate, or transmission cost of candidate download sources changes, step S3 adjusts the download source for subsequent data blocks based on the online update results; when a P2P node continuously provides valid data blocks, its node reputation is increased, and when bad blocks or inefficient transmission occur, its node reputation is decreased. In other embodiments, multi-source scheduling can employ Thompson Sampling or policy gradient reinforcement learning algorithms, but network status, node reputation, playability priority, transmission cost, and verification feedback are still used as scheduling criteria.
[0126] Example 6:
[0127] This embodiment further explains the initial playability control and abnormal switching in step S5, based on embodiments one and two. A higher playability priority is set for the data blocks necessary for game startup, while other data blocks are set to a lower playability priority. The scheduling server 14 performs data block scheduling according to the playability priority, ensuring that the client 18 starts the game after the minimum set of playable data blocks has been downloaded and verified.
[0128] When a candidate download source is unavailable, data block verification fails, or P2P node transmission efficiency decreases, the scheduling server 14 reduces the scheduling weight of the corresponding candidate download source and switches the data block to be downloaded to another candidate download source; data blocks that fail verification are simply re-pulled.
[0129] In summary, Embodiment 1 illustrates the game intelligent distribution system and its main functional components; Embodiment 2 describes the main distribution methods corresponding to the system according to steps S1 to S6 of the disclosure document; and Embodiments 3 to 6 further illustrate implementation methods such as content addressing and security isolation, prediction parameters, multi-source scheduling, P2P reputation, and first-playability control. Therefore, this invention effectively overcomes the various shortcomings of the prior art and has high industrial application value.
[0130] The above embodiments are merely illustrative of the principles and effects of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of the present invention. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in the present invention should still be covered by the claims of the present invention.
Claims
1. A game intelligent distribution method, characterized in that, Includes the following steps: S1. Perform content-based variable-length chunking on the game version file to obtain multiple data blocks. Calculate the block hash for each data block. Construct content-addressable storage based on the block hash. Construct a Merkle hash tree based on the block hash of each data block. Generate a version list including data block logical offset, block hash, playable priority, and Merkle root. S2. Based on the historical download data and operation data of multiple regional network nodes, predict the demand popularity of each game version within a future preset time period, determine the edge preheating set based on the demand popularity and version list, and preheat the data blocks corresponding to the edge preheating set to the edge cache or operator cache of the corresponding region. S3. Content delivery network nodes, edge cache, carrier cache, peer nodes and local cache are used as candidate download sources. The scheduling state is constructed based on the network status, node reputation, content popularity, playability priority and transmission cost of the data block to be downloaded. The linear upper confidence bound algorithm is used to determine the target download source or combination of target download sources and parallel weights corresponding to the data block to be downloaded. The scheduling parameters of the corresponding candidate download source are updated online based on the effective throughput, transmission cost, verification result, bad block result and timeout result after the data block is transmitted. S4. Obtain the list of peer nodes through the P2P tracking server and establish a peer-to-peer data transmission connection. Use the data block passing the integrity check as the condition for inclusion of effective contribution. Update the node reputation according to the effective contribution of the peer nodes and provide the updated node reputation to step S3 as the subsequent data block scheduling status. S5. The client obtains data blocks from the target download source determined in step S3, and performs integrity verification on the data blocks based on the block hash and Merkle root. Data blocks that fail the verification are re-obtained from other candidate download sources, and the verification result is fed back for node reputation update in step S4. At the same time, data blocks are scheduled according to playability priority. When the minimum set of playable data blocks is downloaded and passes the integrity verification, the client is allowed to start the game, and the remaining data blocks continue to be downloaded during game operation. S6. Based on the measured transmission rate, transmission cost and service level of each candidate download source obtained in step S3, perform cost-experience joint optimization, adjust the source weight and unit price preference of each candidate download source, and feed the adjustment results back to the edge preheating in step S2 and the multi-source scheduling in step S3.
2. The game intelligent distribution method according to claim 1, characterized in that: In step S1, a rolling hash is used to traverse the game version file. When the rolling hash value meets the preset block division condition, the data block boundary is determined so that the average size of the data block is the preset average block size. The version list also records the file path, data block length, logical offset of the data block in the file, and data block order. The Merkle root is obtained by using the block hash of each data block as a leaf node and calculating the hash layer by layer on the hash concatenation result of adjacent child nodes.
3. The game intelligent distribution method according to claim 2, characterized in that: The data blocks are encrypted using the corresponding game's key during transmission and static storage, and the block hash is calculated based on the plaintext of the data block. Each game corresponds to an independent content addressing storage namespace, and the data blocks are identified by a combination of game identifier, block hash, and data block length. Cross-game data block reuse is only performed when different games have shared authorization.
4. The game intelligent distribution method according to claim 1, characterized in that: In step S2, a combination of region and operator is used as a regional network node. The regional network nodes are connected by geographical adjacency or same operator relationship, and the edge weight is determined according to the historical round-trip delay between regional network nodes. The download volume, game reservation increment, server online peak, time features and version release flags of each regional network node in multiple historical time periods are input into a graph neural network for spatial feature aggregation. The obtained node spatial feature sequence is then input into a time series model, and after fully connected regression, the demand heat matrix corresponding to the region, prediction time period and game version is output.
5. The game intelligent distribution method according to claim 4, characterized in that: The graph neural network is a two-layer GraphSAGE network, and the temporal model is a long short-term memory network. Based on the demand popularity matrix, regions with high demand popularity are selected and combined with game versions. According to the playability priority in the version list, corresponding data blocks are selected to form the edge preheating set under the cache capacity constraint of the edge cache. The deviation between the actual request volume and the predicted request volume is used as an online incremental update sample to update the subsequent demand popularity prediction results and adjust the subsequent edge preheating set.
6. The game intelligent distribution method according to claim 1, characterized in that: In step S3, the scheduling status includes the round-trip latency, available bandwidth, packet loss rate, node reputation, content popularity, playability priority, and real-time unit price of the candidate download sources. The linear upper confidence bound algorithm is used to determine the download source based on the predicted revenue and exploration confidence terms of the candidate source arm, and to form a reward based on the effective throughput, transmission cost, verification results, bad block results, and timeout results, and to update the candidate source arm parameters block by block.
7. The game intelligent distribution method according to claim 1, characterized in that: In step S4, the peer nodes obtain the node list through the P2P tracking server and establish a data transmission connection through the STUN, TURN or ICE protocol. Only data blocks that pass the block hash and Merkel verification in step S5 are counted as valid contributions, and node reputation is updated based on valid contributions; the client only receives data blocks whose block hashes match the official content addressing storage and version list.
8. The game intelligent distribution method according to claim 1, characterized in that: In step S5, the version list sets playability priority based on the data blocks required for game startup and determines the minimum set of runnable data blocks; download sources are allocated to data blocks with high playability priority first; when all data blocks in the minimum set of runnable data blocks have been downloaded and passed integrity verification, the client is allowed to start the game, and the remaining data blocks continue to be downloaded during game operation.
9. A game intelligent distribution system, characterized in that: include: The central source site is used to store game version files and raw data blocks; The version management server is used to perform content-based variable-length chunking on game version files, calculate the block hash of each data block, construct content-addressable storage and Merkle hash tree, and generate version manifests. The prediction server is used to predict demand based on historical download and operational data from multiple regional network nodes, and to determine the edge preheating set based on demand and version list. Edge cache and carrier cache are used to cache data blocks corresponding to the edge preheating set according to the preheating instructions; The scheduling server is used to select content distribution network nodes, edge caches, carrier caches, peer nodes and local caches as candidate download sources. It performs linear upper confidence bound multi-source scheduling based on the network status, node reputation, content popularity, data block playability priority and transmission cost of the candidate download sources, and adjusts the source weight and unit price preference of each candidate download source based on the measured transmission rate, transmission cost and service level. A P2P tracking server is used to maintain a list of peer nodes and provide network address translation (NAT) traversal signaling. The client is used to determine the data blocks to be downloaded based on the version list, obtain the data blocks according to the scheduling results, perform block hash and Merkel integrity verification on the received data blocks, re-obtain abnormal data blocks based on the verification results, and start the game after the minimum set of runnable data blocks has been downloaded and the verification has passed. Among them, the data blocks that pass the integrity verification are used to count the effective contributions of peer nodes and update the node reputation. The updated node reputation is fed back to the scheduling server for subsequent data block scheduling. The actual transmission information and cost-experience joint optimization results of the scheduling server are fed back to the prediction server and the scheduling server for subsequent edge preheating and multi-source scheduling.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the game intelligent distribution method according to any one of claims 1 to 8.
