WO2023016049A1 - 一种用于游戏的免下载运行方法及平台 - Google Patents

一种用于游戏的免下载运行方法及平台 Download PDF

Info

Publication number
WO2023016049A1
WO2023016049A1 PCT/CN2022/095065 CN2022095065W WO2023016049A1 WO 2023016049 A1 WO2023016049 A1 WO 2023016049A1 CN 2022095065 W CN2022095065 W CN 2022095065W WO 2023016049 A1 WO2023016049 A1 WO 2023016049A1
Authority
WO
WIPO (PCT)
Prior art keywords
data
game
download
compression
prediction
Prior art date
Application number
PCT/CN2022/095065
Other languages
English (en)
French (fr)
Inventor
许磊
赵庆鹏
Original Assignee
许磊
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 许磊 filed Critical 许磊
Priority to CN202280001452.4A priority Critical patent/CN115243771A/zh
Publication of WO2023016049A1 publication Critical patent/WO2023016049A1/zh

Links

Classifications

    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/30—Interconnection arrangements between game servers and game devices; Interconnection arrangements between game devices; Interconnection arrangements between game servers
    • A63F13/31—Communication aspects specific to video games, e.g. between several handheld game devices at close range
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/30—Interconnection arrangements between game servers and game devices; Interconnection arrangements between game devices; Interconnection arrangements between game servers
    • A63F13/35—Details of game servers
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/30—Interconnection arrangements between game servers and game devices; Interconnection arrangements between game devices; Interconnection arrangements between game servers
    • A63F13/35—Details of game servers
    • A63F13/355—Performing operations on behalf of clients with restricted processing capabilities, e.g. servers transform changing game scene into an encoded video stream for transmitting to a mobile phone or a thin client
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/45—Controlling the progress of the video game
    • A63F13/48—Starting a game, e.g. activating a game device or waiting for other players to join a multiplayer session
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/60—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor
    • A63F13/67—Generating or modifying game content before or while executing the game program, e.g. authoring tools specially adapted for game development or game-integrated level editor adaptively or by learning from player actions, e.g. skill level adjustment or by storing successful combat sequences for re-use
    • A—HUMAN NECESSITIES
    • A63—SPORTS; GAMES; AMUSEMENTS
    • A63F—CARD, BOARD, OR ROULETTE GAMES; INDOOR GAMES USING SMALL MOVING PLAYING BODIES; VIDEO GAMES; GAMES NOT OTHERWISE PROVIDED FOR
    • A63F13/00—Video games, i.e. games using an electronically generated display having two or more dimensions
    • A63F13/70—Game security or game management aspects
    • A63F13/77—Game security or game management aspects involving data related to game devices or game servers, e.g. configuration data, software version or amount of memory
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/131—Protocols for games, networked simulations or virtual reality
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/56—Provisioning of proxy services
    • H04L67/565—Conversion or adaptation of application format or content
    • H04L67/5651—Reducing the amount or size of exchanged application data
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L69/00—Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
    • H04L69/04—Protocols for data compression, e.g. ROHC

Definitions

  • the invention relates to the field of network technology, in particular to a download-free running method and platform for games.
  • Console games have developed from the PS1 era to the PS5 era.
  • the capacity of a large game program often exceeds 200GB. Therefore, how to reduce the cost of use for users has become a top priority.
  • the present invention provides a download-free running method and platform for games.
  • a data prediction model is established to predict in advance the loading sequence of each address segment when the game is executed. , so that the data segments needed in the future can be prepared in advance, and game users can be given a click-and-play game experience.
  • a download-free running method for games comprising the steps of:
  • the client starts the game through the virtual machine, it requests data input from the server, and the server predicts in real time the loading sequence of each address segment when the game is executed according to the trained prediction data model, and according to the corresponding relationship between the address segment data and the classified data, Prepare the data segments that are not needed in advance, and return them to the client in combination with block compression technology;
  • the client After receiving the prediction data segment provided by the server, the client uses different methods to decompress the blocks according to the type identification, and compresses and restores according to the compression format of different data types, so as to realize playing while playing.
  • a download-free running platform for games including a data collection identification and classification system, a big data analysis and machine learning system, and a prediction system;
  • the data collection identification and classification system is used to accurately and dynamically identify, identify and classify the dynamic data during game execution during the execution of the game through the virtual machine, and respectively perform lossy and lossless compression on the classified data; at the same time, record
  • the corresponding relationship between the address segment data accessed during game execution and the above-mentioned classification data is stored in the server's big data analysis and machine learning system;
  • Big data analysis and machine learning system use machine learning technology to analyze data for different data types to obtain the characteristics of different data types; and establish data prediction models;
  • Prediction system when the client starts the game through the virtual machine, it requests data input from the server, and the server predicts the loading sequence of each address segment in real time according to the trained data prediction model, and according to the corresponding relationship between the address segment data and the classification data, Prepare the data segments needed in the future in advance, and return them to the client in combination with block compression technology;
  • the client After the client receives the prediction data provided by the server, it uses different methods to decompress the blocks according to the type identification, and compresses and restores according to the compression formats of different data types, so as to realize playing while playing.
  • the data collection identification and classification system for each IO read record data for dynamic identification of data during game execution includes: time stamp, start address of IO read, length of requested data for IO read, and IO The type ID to read.
  • the data collection identification and classification system dynamically identifies data, it judges the data type read by this IO according to the flow direction of the dynamic data, and records the type identification, including texture data, audio data, video data and program data.
  • lossy compression includes: compressing video data into H264/5 format, compressing audio data into AAC format, compressing texture data into PNG format; lossless compression includes: compressing program data into 7z format.
  • the big data analysis and machine learning system uses machine learning technology to perform data analysis on different data types to obtain the characteristics of different data types, including the scattered characteristics of texture data and program data.
  • the big data analysis and machine learning system establishes a data prediction model based on a deep neural network, including input, prediction and output;
  • the input of the prediction model is the address read for the last N times, and the output is the target segment address combination to be read.
  • the prediction process is to predict the subsequent access address combination based on the address read for the last N times.
  • Block compression methods for different data types: 1) Program data is compressed in blocks of 256KB; 2) Texture data is compressed in blocks of 256 pixels x 256 pixels; 3) Audio data is compressed in units of 256KB Block compression; 4) Video data is compressed in blocks of 256KB.
  • the data prediction model is pre-trained, and the historical data of each IO reading record that is dynamically recognized during game execution is used to train the prediction model.
  • the present invention establishes a data prediction model through the method of dynamic data identification and classification, combined with big data analysis and machine learning technology, and predicts in advance the loading sequence of each address segment when the game is executed, so as to prepare in advance for other address segments in the future.
  • the required data segments can give game users a point-and-play game experience.
  • the present invention does not need source code, does not pick games, and is fully automatic without hiring manual testing and labeling.
  • the present invention can save 99% of the traffic cost of game distribution and game operation; in terms of experience, the experience of clicking and playing without waiting greatly reduces the cost of users trying games.
  • Fig. 1 is a flow chart of the download-free running method for games according to the present invention
  • Fig. 2 is a schematic diagram of the data collection identification and classification system of the download-free operating platform for games according to the present invention
  • Fig. 3 is a schematic diagram of a big data analysis and machine learning system for a download-free operating platform for games according to the present invention
  • Fig. 4 is a schematic diagram of a prediction system for a download-free operating platform for games according to the present invention.
  • Fig. 5 is a schematic diagram of prediction model training
  • Figure 6 is a schematic diagram of the prediction model.
  • the download-free running method for games according to the present invention comprises steps:
  • the dynamic texture data, audio data, video data and program data during the game execution are accurately and dynamically identified, marked and classified to obtain classified data; and Lossy and lossless compression of classified data respectively;
  • the player client When the player client starts the game through virtualization technology, it requests data input from the server, and the server predicts in real time the loading sequence of each address segment when the game is executed based on the trained prediction data model. Prepare the data segments it needs in the future, and return them to the client in combination with block compression technology;
  • the client After receiving the prediction data segment provided by the server, the client performs data processing such as block decompression, compression and restoration according to the identification, so as to realize that the game will not be stuck while playing.
  • the download-free operating platform for games of the present invention includes a data collection identification and classification system, a big data analysis and machine learning system, and a prediction system.
  • the data collection identification and classification system is implemented based on the software architecture.
  • the dynamic texture data (picture data), audio data, video data, and program data, etc. Perform accurate dynamic identification, labeling and classification, and perform lossy and lossless compression on classified data.
  • the data during game execution is dynamically identified, the data is classified according to texture data, audio data, video data and program data, and the type of classified data is identified.
  • the data read from the game file will be preprocessed, and the data will be sent to each subsequent processing unit according to the data type, such as audio processing unit, video decoding unit, graphics processing unit, etc.
  • the data collection identification and classification system of the present invention dynamically identifies data, it judges the data type read by this IO according to the flow direction of the data, and records the type identification, including but not limited to texture, audio, video, model Mesh, Rendering scripts, program logic code, etc.
  • Texture is one or several two-dimensional graphics that represent the details of the surface of an object, also known as texture map (texture map). mapping). When the texture is mapped to the surface of the object in a specific way, it can make the object look more realistic.
  • the following data is recorded dynamically: time stamp, start address of IO reading, length of requested data for IO reading, and IO reading
  • the retrieved type is identified and uploaded to the server for data analysis and prediction model establishment.
  • Lossy compression video data compression to H264/5 format; audio data compression to AAC format; texture data compression to PNG format.
  • Lossless compression program data is compressed into 7z format.
  • the marked texture, image, audio, video, model, and other data are called game material data, which is editable, that is, operations such as modification, enhancement, and downgrade will not affect the game engine’s ability to edit it. Process and proceed as normal. And because of the lossy compression, usually the compression rate can be several times to dozens of times.
  • Texture is one or several two-dimensional graphics that represent the details of the surface of an object, also known as texture map (texture map). mapping). When the texture is mapped to the surface of the object in a specific way, it can make the object look more realistic. For texture data, you can perform operations such as replacement (Mod), compression, and enhanced resolution.
  • the game engine only requires the format of the texture data, and does not check the content of the texture format, but only renders the screen according to the texture data.
  • the marked program data including the logic model of the game
  • the decompressed data must be completely consistent with the original data, otherwise the game engine cannot process it normally.
  • the data collection identification and classification system during the execution of the game through the virtual machine, records the corresponding relationship between the address segment data accessed during the execution process and the above classification data, and stores it in the big data analysis and machine learning system of the server.
  • the big data analysis and machine learning system utilizes machine learning technology to analyze and process data for the characteristics of different data types to obtain the characteristics of different data types, including texture data and program data. Disperse features; and build data predictive models.
  • linear prediction is mainly used, and it can be loaded in advance according to the order.
  • Data such as textures, models, and program data present features such as multi-point reading and random reading.
  • deep learning technology is required for nonlinear data model training and prediction.
  • a data prediction model is established, and a prediction model based on a deep neural network is established, including input, prediction and output.
  • the deep learning method based on neural network is a very general method that has undergone a lot of practice in recent years and achieved good results.
  • the deep neural network can be divided into three layers, the input layer, the hidden layer and the output layer.
  • the first layer is the input layer
  • the last layer is the output layer
  • the number of layers in the middle is hidden layer.
  • the present invention can design multiple hidden layers according to calculation needs.
  • data practice shows that for this scenario, three hidden layers are used, and the number of neurons in each hidden layer is two-thirds of the number of corresponding file indexes.
  • the input of the prediction model is the segment address combination read for the last N times
  • the output is the target segment address combination to be read.
  • the prediction process is to predict the subsequent access address data combination based on the segment address combination read for the last N times.
  • the present invention adopts the prediction of the loading order of the addresses of the data segments, so that the data segments needed in the future are prepared in advance, and the game user can be given a click-and-play game experience, and the prediction speed is faster and the calculation amount is smaller.
  • the data prediction model is pre-trained. As shown in Figure 5, during the normal operation of the game, the historical data recorded for each IO read: time stamp, the start address of IO read, as the data input for training the prediction model. Wherein, a feedback mechanism of the prediction model can be provided according to the time stamp, that is, subsequent access address combinations.
  • the specific input of the final derived product prediction model is the segment address combination read for the last N times, and the output is a target segment address combination (representing the data address to be read by the game ).
  • the prediction system is designed based on the player client.
  • the client starts the game through virtualization technology, it requests data input from the server.
  • the server predicts the loading of each address segment in real time according to the trained data prediction model. sequence, and before accessing the data, according to the corresponding relationship between the address segment data and the classification data, prepare the data segment it needs in the future in advance, and return it to the client in combination with block compression technology.
  • the prediction request data input is the segment address combination read by the current game for the last N times.
  • Different block compression methods are used for different types of classified data: 1) program data is compressed in blocks of 256KB; 2) texture data is compressed in blocks of 256 pixels x 256 pixels; 3) audio data is in AAC format, The secondary compression is performed in units of 256KB; 4) The video data is compressed twice in units of 256KB according to the H264/5 format.
  • the player client After the player client receives the forecast data provided by the server, it uses different methods to decompress the blocks according to the type identification, and compresses and restores according to the lossy or lossless compression format of different data types, so that the game will not be stuck while playing. .
  • the execution subject of the client may be a mobile phone, a notebook computer, a tablet computer, a palmtop computer, a PAD, a desktop computer, etc., which are not limited in this application.
  • the beneficial effect of the present invention is that, compared with the prior art, the present invention establishes a data prediction model through the method of dynamic data identification and classification, combined with big data analysis and machine learning technology, and predicts in advance the loading sequence of each address segment when the game is executed , so that the data segments needed in the future can be prepared in advance, and game users can be given a click-and-play game experience.
  • the present invention does not need source code, does not pick games, and is fully automatic without hiring manual testing and labeling.
  • the present invention can save 99% of the traffic cost of game distribution and game operation; in terms of experience, the experience of clicking and playing without waiting greatly reduces the cost of users trying games.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Computer Security & Cryptography (AREA)
  • Business, Economics & Management (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Two-Way Televisions, Distribution Of Moving Picture Or The Like (AREA)

Abstract

本发明公开了一种用于游戏的免下载运行方法及平台，对游戏执行时的动态数据进行准确的动态识别、标识和分类，对分类数据分别进行有损和无损压缩；记录游戏执行过程中访问的地址段数据和上述分类数据的对应关系；针对不同数据类型分别进行数据分析得到不同数据类型的特征，并建立数据预测模型；根据训练好的预测数据模型，实时预测游戏执行时各地址段的加载顺序，根据地址段数据和分类数据的对应关系，提前准备好未来需要的数据段，并结合分块压缩技术返回给客户端；客户端在收到服务器提供的预测数据段后，根据标识进行分块解压缩并压缩还原，实现边下边玩。本发明能够实现让游戏即点即玩，不用等待下载安装的免下载运行方法。

Description

一种用于游戏的免下载运行方法及平台 技术领域
本发明涉及网络技术领域，尤其涉及一种用于游戏的免下载运行方法及平台。
 背景技术
随着计算机技术的高速发展，游戏程序容量爆炸式增长，主机游戏从PS1时代发展到PS5时代，一个大型游戏程序的容量往往已经超过200GB，所以如何降低用户的使用成本，成为当务之急。
 技术问题
现有的游戏客户端下载耗用大量时间且占用网速，不满足目前网络玩家需求。最好能有一种技术可以实现让游戏即点即玩，不用等待安装，不用等待下载，将GB级别的大型游戏，做到和页游一样的游戏启动体验。
目前在虚拟系统建设的基础上，启动优化用户游戏启动体验的一系列工作，但是由于大型游戏的容量较大，现有网络传输速率较低，用户的网络环境复杂，用户通常需要等待几分钟乃至几十分钟才能开始体验游戏，严重影响了用户体验。
 技术解决方案
本发明提供一种用于游戏的免下载运行方法及平台，通过动态数据标识和分类的方法，结合大数据分析和机器学习技术，建立数据预测模型，提前预测游戏执行时各地址段的加载顺序，从而提前准备好未来其需要的数据段，能够给与游戏用户即点即玩的游戏体验。
为实现本发明的目的，本发明所采用的技术方案是：
一种用于游戏的免下载运行方法，所述方法包括步骤：
（1）在通过虚拟机执行游戏的过程中，对游戏执行时的动态数据进行准确的动态识别、标识和分类，并对分类数据分别进行有损和无损压缩；
（2）在通过虚拟机执行游戏的过程中，记录执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器中；
（3）利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征；并建立数据预测模型；
（4）客户端通过虚拟机启动游戏时，向服务器请求数据输入，服务器根据训练好的预测数据模型，实时预测游戏执行时各地址段的加载顺序，根据地址段数据和分类数据的对应关系，提前准备好未其需要的数据段，并结合分块压缩技术返回给客户端；
（5）客户端在收到服务器提供的预测数据段后，根据类型标识采用不同的方法进行分块解压缩并根据不同数据类型的压缩格式进行压缩还原，实现边下边玩。
一种用于游戏的免下载运行平台，包括数据收集标识和分类系统、大数据分析和机器学习系统、预测系统；
数据收集标识和分类系统，用于在通过虚拟机执行游戏的过程中，对游戏执行时的动态数据进行准确的动态识别、标识和分类，对分类数据分别进行有损和无损压缩；同时，记录游戏执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器的大数据分析和机器学习系统中；
大数据分析和机器学习系统，利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征；并建立数据预测模型；
预测系统，客户端通过虚拟机启动游戏时，向服务器请求数据输入，服务器根据训练好的数据预测模型，实时预测游戏执行时各地址段的加载顺序，根据地址段数据和分类数据的对应关系，提前准备好未来需要的数据段，并结合分块压缩技术返回给客户端；
客户端收到服务器提供的预测数据后，根据类型标识采用不同的方法进行分块解压缩并根据不同数据类型的压缩格式进行压缩还原，实现边下边玩。
进一步地，数据收集标识和分类系统，对于游戏执行时的数据进行动态识别的每次IO读取记录数据包括：时间戳、IO读取的起始地址、IO读取的请求数据长度、以及IO读取的类型标识。
进一步地，数据收集标识和分类系统，对数据进行动态识别时，是根据动态数据的流向，来判断本次IO读取的数据类型，并记录类型标识，包括纹理数据、音频数据、视频数据及程序数据。
进一步地，有损压缩包括：视频数据压缩为H264/5格式、音频数据压缩为AAC格式、纹理数据压缩为PNG格式；无损压缩包括：程序数据压缩成7z格式。
进一步地，大数据分析和机器学习系统，利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征，包括纹理数据和程序数据的分散特征。
进一步地，大数据分析和机器学习系统，建立基于深度神经网络的数据预测模型，包括输入、预测和输出；
预测模型的输入为最近N次读取的地址，输出为即将读取的目标段地址组合，预测流程是根据最近N次读取的地址预测后续的访问地址组合。
进一步地，对不同数据类型的分块压缩方式：1）程序数据按256KB为单位进行分块压缩；2）纹理数据按照256像素 x 256像素进行分块压缩；3）音频数据按256KB为单位进行分块压缩；4）视频数据按256KB为单位进行分块压缩。
进一步地，建立好数据预测模型之后，对数据预测模型进行预训练，采用游戏执行时进行动态识别的每次IO读取记录的历史数据进行预测模型的训练。
 有益效果
与现有技术相比，本发明通过动态数据标识和分类的方法，结合大数据分析和机器学习技术，建立数据预测模型，提前预测游戏执行时各地址段的加载顺序，从而提前准备好未来其需要的数据段，能够给与游戏用户即点即玩的游戏体验。
本发明无需源码，不挑游戏，且全自动无需雇人工测试和标注。本发明在成本上，可将游戏分发、游戏运行的流量成本节约99%；体验上，即点即玩，无需等待的体验，大大降低了用户试玩游戏的成本。
 附图说明
图1是本发明所述的用于游戏的免下载运行方法流程图；
图2是本发明所述的用于游戏的免下载运行平台的数据收集标识和分类系统示意图；
图3是本发明所述的用于游戏的免下载运行平台的大数据分析和机器学习系统示意图；
图4是本发明所述的用于游戏的免下载运行平台的预测系统示意图；
图5是预测模型训练示意图；
图6是预测模型示意图。
 本发明的实施方式
下面结合附图和实施例对本发明的技术方案作进一步的说明。以下实施例仅用于更加清楚地说明本发明的技术方案，而不能以此来限制本申请的保护范围。
如图1所示，本发明所述的用于游戏的免下载运行方法，包括步骤：
（1）根据虚拟化技术，在通过虚拟机执行游戏的过程中，对游戏执行时动态的纹理数据、音频数据、视频数据及程序数据进行准确的动态识别、标识和分类，得到分类数据；并对分类数据分别进行有损和无损压缩；
（2）在通过虚拟机执行游戏的过程中，记录执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器的大数据分析和机器学习系统中；
（3）对建立起来的大数据分析和机器学习系统，利用机器学习技术，针对不同数据类型的特征分别进行数据分析和处理，得到不同数据类型的特征；并基于地址段数据和分类数据的对应关系，以及不同数据类型的特征建立数据预测模型；建立好数据预测模型之后，对数据预测模型进行预训练；
（4）玩家客户端通过虚拟化技术启动游戏时，向服务器请求数据输入，服务器根据训练好的预测数据模型，实时预测游戏执行时各地址段的加载顺序，在尚未执行到访问数据前，提前准备好未来其需要的数据段，并结合分块压缩技术，返回给客户端；
（5）客户端在收到服务器提供的预测数据段后，根据标识进行分块解压缩、压缩还原等数据处理，实现边下边玩不卡顿。
本发明所述的用于游戏的免下载运行平台，包括数据收集标识和分类系统、大数据分析和机器学习系统、预测系统。
如图2所示，数据收集标识和分类系统，基于软件架构实现，在通过虚拟机执行游戏的过程中，对游戏执行时动态的纹理数据（图片数据）、音频数据、视频数据及程序数据等进行准确的动态识别、标识和分类，对分类数据分别进行有损和无损压缩。
将游戏执行时的数据进行动态识别，将数据按照纹理数据、音频数据、视频数据及程序数据进行分类，并对分类数据进行类型标识。
游戏在运行过程中，会把从游戏文件读取的数据，经过预处理后，根据数据类型把数据输送到各个后续处理单元，比如音频处理单元、视频解码单元、图形处理单元等。本发明数据收集标识和分类系统对数据进行动态识别时，是根据数据的流向，来判断本次IO读取的数据类型，并记录类型标识，包括但不限于纹理、音频、视频、模型Mesh、渲染脚本、程序逻辑代码等等。
比如纹理数据（Texture）：纹理是表示物体表面细节的一幅或几幅二维图形，也称纹理贴图（texture mapping）。当把纹理按照特定的方式映射到物体表面上的时候能使物体看上去更加真实。
在游戏的正常运行过程中，对于游戏执行时的数据进行动态识别的每次IO读取，记录如下数据：时间戳、IO读取的起始地址、IO读取的请求数据长度、以及IO读取的类型标识，并上传到服务器，用于数据分析和预测模型的建立。
有损压缩，视频数据压缩为H264/5格式；音频数据压缩为AAC格式；纹理数据压缩为PNG格式。无损压缩，程序数据压缩成7z格式。
对于标识出来的纹理、图像、音频、视频、模型等数据，称其为游戏素材类数据，具有可编辑性，即，对其进行修改、增强、降级等操作，不会影响游戏引擎对其的正常处理和进行。并且由于采用有损压缩，通常压缩率可以做到几倍到几十倍不等。
比如纹理数据（Texture）：纹理是表示物体表面细节的一幅或几幅二维图形，也称纹理贴图（texture mapping）。当把纹理按照特定的方式映射到物体表面上的时候能使物体看上去更加真实。对于纹理数据，可以进行替换（Mod）、压缩、增强分辨率等操作，游戏引擎只要求纹理数据的格式，不检查纹理格式的内容，只是根据纹理的数据进行画面渲染。
然而，对于标识出来的程序数据，包括游戏的逻辑模型，则不具备可编辑性，需要解压缩后的数据和原数据完全一致，不然游戏引擎无法正常处理。对于这类数据，我们采用无损的压缩算法，以保证解压缩后的数据和压缩前完全一致。
数据收集标识和分类系统，在通过虚拟机执行游戏的过程中，记录下执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器的大数据分析和机器学习系统中。
在游戏的正常运行过程中，对于每次IO读取，会记录如下数据：时间戳、IO读取的起始地址、IO读取的请求数据长度、以及IO读取的类型标识，然后上传到服务器，以表格形式存储，一个示例如表1。
 
Figure dest_path_image001
如图3所示，大数据分析和机器学习系统，基于服务器设计，利用机器学习技术，针对不同数据类型的特征分别进行数据分析和处理，得到不同数据类型的特征，包括纹理数据和程序数据的分散特征；并建立数据预测模型。
对于音频数据和视频数据，因其数据格式本身的特性就是和时间轴呈线性相关，在预测模型中，主要是以线性预测为主，根据顺序提前加载即可。
而纹理、模型、程序数据等数据，呈现多点读取、随机读取等特征，预测模型中，需要用到深度学习技术进行非线性数据模型训练和预测。
基于地址段数据和分类数据的对应关系，以及不同数据类型的特征建立数据预测模型，建立基于深度神经网络的预测模型，包括输入、预测和输出。
基于神经网络的深度学习方法，是近年来经过大量实践，并取得很好成果的一种很通用的方法。基于深度神经网络的预测模型，深度神经网络可以分为三层，输入层，隐藏层和输出层，一般来说第一层是输入层，最后一层是输出层，而中间的层数都是隐藏层。本发明可根据计算需要设计多层隐藏层。
本实施例中，数据实践表明对于本场景，采用3层隐藏层，每个隐藏层神经元的数量是对应文件索引数量的三分之二。输入是最近N次读取的段地址组合（N=10），输出层节点数量为对应文件索引数量，然后根据概率排序选择前M个目标地址段（M=10）。
这里主要侧重介绍预测模型的输入输出，以及其运行的流程。预测模型的输入为最近N次读取的段地址组合，输出为即将读取的目标段地址组合，预测流程是根据最近N次读取的段地址组合预测后续的访问地址数据组合。本发明采用对数据段地址加载顺序的预测，从而提前准备好未来其需要的数据段，能够给与游戏用户即点即玩的游戏体验，其预测速度更快，计算量更小。
建立好深度神经网络数据预测模型之后，对数据预测模型进行预训练。如图5所示，在游戏的正常运行过程中，对于每次IO读取记录的历史数据：时间戳、IO读取的起始地址，作为对预测模型进行训练的数据输入。其中，根据时间戳能提供预测模型的反馈机制，即，后续的访问地址组合。
如图6所示，通过大数量的反复训练模型，最终导出的产品预测模型的具体输入为最近N次读取的段地址组合，输出为一个目标段地址组合（代表游戏即将读取的数据地址）。
如图4所示，预测系统，基于玩家客户端设计，客户端通过虚拟化技术启动游戏时，向服务器请求数据输入，服务器根据训练好的数据预测模型，实时预测游戏执行时各地址段的加载顺序，然后在尚未执行到访问数据前，根据地址段数据和分类数据的对应关系，提前准备好未来其需要的数据段，并结合分块压缩技术返回给客户端。
所述预测请求数据输入为目前游戏最近N次读取的段地址组合。
具体地，根据不同类型的分类数据采用不同的分块压缩方式，同时根据不同数据类型的有损或无损压缩格式进行压缩，对数据进行分块压缩之后返回给客户端。
对不同类型的分类数据采用不同的分块压缩方式：1）程序数据按256KB为单位进行分块压缩；2）纹理数据按照256像素 x 256像素进行分块压缩；3）音频数据按照AAC格式，按256KB为单位进行二次压缩；4）视频数据按照H264/5格式，按256KB为单位进行二次压缩。
玩家客户端在收到服务器提供的预测数据后，根据类型标识采用不同的方法进行分块解压缩，根据不同数据类型的有损或无损压缩格式进行压缩还原，从而做到边下边玩不卡顿。
客户端的执行主体可以是手机、笔记本电脑、平板电脑、掌上电脑、PAD、台式电脑等，本申请在此不作限定。
本发明的有益效果在于，与现有技术相比，本发明通过动态数据标识和分类的方法，结合大数据分析和机器学习技术，建立数据预测模型，提前预测游戏执行时各地址段的加载顺序，从而提前准备好未来其需要的数据段，能够给与游戏用户即点即玩的游戏体验。
本发明无需源码，不挑游戏，且全自动无需雇人工测试和标注。本发明在成本上，可将游戏分发、游戏运行的流量成本节约99%；体验上，即点即玩，无需等待的体验，大大降低了用户试玩游戏的成本。

Claims (9)

  1. 一种用于游戏的免下载运行方法，其特征在于，所述方法包括步骤：
    （1）在通过虚拟机执行游戏的过程中，对游戏执行时的动态数据进行准确的动态识别、标识和分类，并对分类数据分别进行有损和无损压缩；
    （2）在通过虚拟机执行游戏的过程中，记录执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器中；
    （3）利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征；并建立数据预测模型；
    （4）客户端通过虚拟机启动游戏时，向服务器请求数据输入，服务器根据训练好的预测数据模型，实时预测游戏执行时各地址段的加载顺序，根据地址段数据和分类数据的对应关系，提前准备好未其需要的数据段，并结合分块压缩技术返回给客户端；
    （5）客户端在收到服务器提供的预测数据段后，根据类型标识采用不同的方法进行分块解压缩并根据不同数据类型的压缩格式进行压缩还原，实现边下边玩。
  2. 一种用于游戏的免下载运行平台，其特征在于，包括数据收集标识和分类系统、大数据分析和机器学习系统、预测系统；
    数据收集标识和分类系统，用于在通过虚拟机执行游戏的过程中，对游戏执行时的动态数据进行准确的动态识别、标识和分类，对分类数据分别进行有损和无损压缩；同时，记录游戏执行过程中访问的地址段数据和上述分类数据的对应关系，并存储在服务器的大数据分析和机器学习系统中；
    大数据分析和机器学习系统，利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征；并建立数据预测模型；
    预测系统，客户端通过虚拟机启动游戏时，向服务器请求数据输入，服务器根据训练好的数据预测模型，实时预测游戏执行时各地址段的加载顺序，根据地址段数据和分类数据的对应关系，提前准备好未来需要的数据段，并结合分块压缩技术返回给客户端；
    客户端收到服务器提供的预测数据后，根据类型标识采用不同的方法进行分块解压缩并根据不同数据类型的压缩格式进行压缩还原，实现边下边玩。
  3. 根据权利要求2所述的用于游戏的免下载运行平台，其特征在于，
    数据收集标识和分类系统，对于游戏执行时的数据进行动态识别的每次IO读取记录数据包括：时间戳、IO读取的起始地址、IO读取的请求数据长度、以及IO读取的类型标识。
  4. 根据权利要求3所述的用于游戏的免下载运行平台，其特征在于，
    数据收集标识和分类系统，对数据进行动态识别时，是根据动态数据的流向，来判断本次IO读取的数据类型，并记录类型标识，包括纹理数据、音频数据、视频数据及程序数据。
  5. 根据权利要求4所述的用于游戏的免下载运行平台，其特征在于，
    有损压缩包括：视频数据压缩为H264/5格式、音频数据压缩为AAC格式、纹理数据压缩为PNG格式；无损压缩包括：程序数据压缩成7z格式。
  6. 根据权利要求2所述的用于游戏的免下载运行平台，其特征在于，
    大数据分析和机器学习系统，利用机器学习技术针对不同数据类型分别进行数据分析得到不同数据类型的特征，包括纹理数据和程序数据的分散特征。
  7. 根据权利要求3所述的用于游戏的免下载运行平台，其特征在于，
    大数据分析和机器学习系统，建立基于深度神经网络的数据预测模型，包括输入、预测和输出；
    预测模型的输入为最近N次读取的地址，输出为即将读取的目标段地址组合，预测流程是根据最近N次读取的地址预测后续的访问地址组合。
  8. 根据权利要求2所述的用于游戏的免下载运行平台，其特征在于，
    对不同数据类型的分块压缩方式：1）程序数据按256KB为单位进行分块压缩；2）纹理数据按照256像素 x 256像素进行分块压缩；3）音频数据按256KB为单位进行分块压缩；4）视频数据按256KB为单位进行分块压缩。
  9. 根据权利要求2所述的用于游戏的免下载运行平台，其特征在于，
    建立好数据预测模型之后，对数据预测模型进行预训练，采用游戏执行时进行动态识别的每次IO读取记录的历史数据进行预测模型的训练。
PCT/CN2022/095065 2021-08-13 2022-05-25 一种用于游戏的免下载运行方法及平台 WO2023016049A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202280001452.4A CN115243771A (zh) 2021-08-13 2022-05-25 一种用于游戏的免下载运行方法及平台

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110930821.5A CN113730902A (zh) 2021-08-13 2021-08-13 一种用于游戏的免下载运行方法
CN202110930821.5 2021-08-13

Publications (1)

Publication Number Publication Date
WO2023016049A1 true WO2023016049A1 (zh) 2023-02-16

Family

ID=78731074

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/095065 WO2023016049A1 (zh) 2021-08-13 2022-05-25 一种用于游戏的免下载运行方法及平台

Country Status (2)

Country Link
CN (1) CN113730902A (zh)
WO (1) WO2023016049A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113730902A (zh) * 2021-08-13 2021-12-03 许磊 一种用于游戏的免下载运行方法
CN114880107A (zh) * 2021-12-09 2022-08-09 许磊 一种高效低成本的云游戏系统
CN114791893B (zh) * 2021-12-15 2023-05-09 许磊 一种随机数据访问的串行化系统
CN114832373B (zh) * 2021-12-21 2023-03-07 许磊 一种纹理、音频和视频的压缩和重建方法
CN114840859B (zh) * 2021-12-21 2023-03-24 许磊 一种通用数据的流媒体化系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108122127A (zh) * 2016-11-29 2018-06-05 韩国电子通信研究院 预测在线游戏服务的运行结果的方法及装置
CN111249724A (zh) * 2018-12-03 2020-06-09 索尼互动娱乐有限责任公司 机器学习驱动式资源分配
CN111729305A (zh) * 2020-06-23 2020-10-02 网易（杭州）网络有限公司 地图场景预加载方法、模型训练方法、设备及存储介质
AU2020233716A1 (en) * 2006-11-10 2020-10-08 Aristocrat Technologies Australia Pty Limited Casino game download system and method of use
CN112463386A (zh) * 2020-12-08 2021-03-09 内蒙古大学 用于异构云环境下在线游戏应用的虚拟机管理方法及系统
CN113730902A (zh) * 2021-08-13 2021-12-03 许磊 一种用于游戏的免下载运行方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AU2020233716A1 (en) * 2006-11-10 2020-10-08 Aristocrat Technologies Australia Pty Limited Casino game download system and method of use
CN108122127A (zh) * 2016-11-29 2018-06-05 韩国电子通信研究院 预测在线游戏服务的运行结果的方法及装置
CN111249724A (zh) * 2018-12-03 2020-06-09 索尼互动娱乐有限责任公司 机器学习驱动式资源分配
CN111729305A (zh) * 2020-06-23 2020-10-02 网易（杭州）网络有限公司 地图场景预加载方法、模型训练方法、设备及存储介质
CN112463386A (zh) * 2020-12-08 2021-03-09 内蒙古大学 用于异构云环境下在线游戏应用的虚拟机管理方法及系统
CN113730902A (zh) * 2021-08-13 2021-12-03 许磊 一种用于游戏的免下载运行方法

Also Published As

Publication number Publication date
CN113730902A (zh) 2021-12-03

Similar Documents

Publication Publication Date Title
WO2023016049A1 (zh) 一种用于游戏的免下载运行方法及平台
US11712621B2 (en) Video clip classification using feature vectors of a trained image classifier
US7934058B2 (en) Predictive caching of assets to improve level load time on a game console
CN111902807B (zh) 生成硬件兼容压缩纹理的方法、计算机设备和介质
US9582919B2 (en) Automatic run-time identification of textures
CN105205014A (zh) 一种数据存储方法和装置
US20110084964A1 (en) Automatic Real-Time Shader Modification for Texture Fetch Instrumentation
US11586842B2 (en) System and method for machine learning based video quality assessment
US20110197151A1 (en) Graphics remoting using augmentation data
CN113069769B (zh) 云游戏界面显示方法、装置、电子设备以及存储介质
US10699443B2 (en) Reducing the search space for real time texture compression
US11282179B2 (en) System and method for machine learning based video quality assessment
JP2024511103A (ja) 近似値に基づいて画像又はビデオの品質を評価する方法及び装置、第１のモデルの訓練方法及び装置、電子機器、記憶媒体、並びにコンピュータプログラム
CN108664579A (zh) 海量数据实时筛选分析方法、系统及存储介质
CN114840859B (zh) 一种通用数据的流媒体化系统
CN114419473B (zh) 一种基于嵌入式设备的深度学习实时目标检测方法
CN115243771A (zh) 一种用于游戏的免下载运行方法及平台
CN114356566A (zh) 一种基于相关性的游戏数据流媒体化方法
CN114297022A (zh) 云环境异常检测方法、装置、电子设备和存储介质
CN112188215B (zh) 视频解码方法、装置、设备及存储介质
US20150178880A1 (en) Composition data dispatcher
CN114708467B (zh) 基于知识蒸馏的不良场景识别方法及系统及设备
CN116700838B (zh) 一种as语言接口函数、音视频处理方法和系统
CN112749724B (zh) 训练分类器、预测应用性能扩展性的方法、设备
US20160253832A1 (en) Scene image generator

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22855035

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE