WO2021047665A1 - 终端之间连接状态的预测方法、装置和分析设备 - Google Patents

终端之间连接状态的预测方法、装置和分析设备 Download PDF

Info

Publication number
WO2021047665A1
WO2021047665A1 PCT/CN2020/114979 CN2020114979W WO2021047665A1 WO 2021047665 A1 WO2021047665 A1 WO 2021047665A1 CN 2020114979 W CN2020114979 W CN 2020114979W WO 2021047665 A1 WO2021047665 A1 WO 2021047665A1
Authority
WO
WIPO (PCT)
Prior art keywords
time period
unit
terminal
training
historical
Prior art date
Application number
PCT/CN2020/114979
Other languages
English (en)
French (fr)
Inventor
张�浩
谢于明
王苗苗
王仲宇
Original Assignee
华为技术有限公司
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 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP20862875.0A priority Critical patent/EP4024762A4/en
Priority to JP2022516077A priority patent/JP7354424B2/ja
Publication of WO2021047665A1 publication Critical patent/WO2021047665A1/zh
Priority to US17/692,569 priority patent/US20220200870A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/50—Testing arrangements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/147—Network analysis or design for predicting network behaviour
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G06F2009/45595—Network integration; Enabling network access in virtual machine instances
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/142—Network analysis or design using statistical or mathematical methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/02—Capturing of monitoring data
    • H04L43/022—Capturing of monitoring data by sampling
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0805—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability
    • H04L43/0811—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability by checking connectivity
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/16—Threshold monitoring

Definitions

  • This application relates to the field of computer network technology, and further relates to the application of artificial intelligence (AI) technology in the field of computer networks, and in particular to a method for predicting the connection status between terminals and a prediction for the connection status between virtual machines Device and a kind of analysis equipment.
  • AI artificial intelligence
  • a data center is a pool formed by several resources connected to each other using a communication network.
  • the resources include computing resources, storage resources, network resources, and so on. Because virtual machines have the advantages of low cost, agility, flexibility, and scalability, virtual machines have become an important computing resource in DC.
  • the Data Center Network (DCN) is used to interconnect the resources in the DC, and the DCN plays a key role in the DC. In order to cope with the growing demand for cloud computing, DCN needs to be scalable and efficient to connect hundreds of virtual machines, storage and other resources.
  • the virtual machines in the DC communicate with each other and cooperate to complete various services in the DC.
  • the connection status between the virtual machines indicates whether there is communication between the two virtual machines. When there is communication between two virtual machines, the connection state between the two virtual machines is connected. Conversely, when there is no communication between two virtual machines, the connection state between the two virtual machines is no connection.
  • connection state prediction technology between virtual machines (hereinafter referred to as "prediction technology" in this application) is one of the key technologies in DCN. This technology is widely used in many scenarios, such as failure impact analysis and configuration verification scenarios.
  • Failure impact analysis means that when a virtual machine fails, the connection state prediction technology between virtual machines is used to determine theoretically which other virtual machines are connected to the failed virtual machine, and then analyze the scope of the failure.
  • Configuration verification means that when the configuration of a virtual machine is about to be updated (here, the virtual machine that is about to undergo configuration update is recorded as VM 1), the connection state prediction technology between virtual machines is used to determine the assumption that there is no configuration update. In theory, which other virtual machines are The machine is connected to VM1.
  • the other is a forecasting method based on business continuity assumptions.
  • This prediction method predicts the connection status of a pair of virtual machines at the next time in the future based on the connection status of the pair of virtual machines at the previous time. For example, before 10:00 on January 11, 2015, the connection state of virtual machine VM 1 and virtual machine VM 2 at 9:00 on January 11, 2015 is taken as the predicted virtual machine VM 1 and virtual machine VM 2 Connection status at 10 o'clock on January 11, 2015.
  • the embodiment of the present application provides a method for predicting the connection state between terminals to improve the problem of low accuracy of related prediction technologies.
  • a method for predicting the connection state between terminals acquires the connection status of the test terminal pair corresponding to multiple unit moments in the first historical time period.
  • the test terminal pair is composed of a first terminal and a second terminal, and the first historical time period is before the current time.
  • a period of time, and the first historical period of time includes M consecutive unit moments, where M is a natural number greater than or equal to 2.
  • the analysis device determines the connection state of the test terminal pair corresponding to at least one unit time in the future time period according to the connection state of the test terminal pair corresponding to multiple unit moments in the first historical time period.
  • the future A time period is a time period after the current time, and the future time period includes Q consecutive unit times, the first unit time in the future time period and the last unit time in the first historical time period Is a continuous unit time, where Q is a natural number greater than or equal to 1.
  • the analysis device uses the connection state information of the test terminal pair at multiple unit moments in the history during the prediction process, instead of only using the test terminal pair in the history of a single unit.
  • the connection status information at all times helps to analyze and discover more useful information from the historical status information, thereby improving the accuracy of prediction.
  • the analysis device determines the connection state of the test terminal pair corresponding to at least one unit time in the future time period through the following steps.
  • the analysis device inputs the connection states of the test terminal pairs corresponding to multiple unit moments in the first historical time period into a prediction model and obtains the output result of the prediction model.
  • the prediction model is based on N training terminals Generated for connection states corresponding to multiple unit moments in a second historical time period, the second historical time period is a time period before the current time, and the second historical time period includes M+Q consecutive The unit time of, where N is a natural number greater than or equal to 1.
  • the analysis device determines the connection states respectively corresponding to at least one unit time of the test terminal pair in the future time period according to the output result.
  • the analysis equipment uses machine learning algorithms to fully utilize the long-term historical connection status information of a large number of trained terminal pairs to train a predictive model, which can extract general and dynamic trend information that can reflect the connection status of multiple terminal pairs in the same network scenario. Regular information, so as to make predictions more accurately.
  • the analysis device obtains the output result of the prediction model through the following steps.
  • the analysis device determines a first sample sequence according to the connection status of the test terminal pair corresponding to multiple unit moments in the first historical time period, and the first sample sequence includes M elements, so The value of each element of the M elements respectively corresponds to the connection state corresponding to each of the M consecutive unit times.
  • the analysis device inputs the first sample sequence into a prediction model and obtains an output result of the prediction model.
  • the output result is a prediction sequence.
  • the prediction sequence includes Q elements, and each of the Q elements The value of each element corresponds to the connection state corresponding to each of the Q consecutive unit moments.
  • the analysis device first obtains the first sample sequence reflecting the historical connection state trend of the test terminal, and then inputs the first sample sequence into the prediction model to obtain the prediction sequence as the output result. Predicting through sample sequences is an effective method of applying predictive models.
  • the value of one of the M elements or Q elements when the value of one of the M elements or Q elements is the first value, it indicates that the connection state of the corresponding unit time is connected, and the value of one of the M elements or Q elements is When the value is the second value, it indicates that the connection state of the corresponding unit time is disconnected, and the first value and the second value are different. It is a simple and efficient way to indicate the connection status through different element values in each sample sequence.
  • the analysis device trains the prediction model through the following steps.
  • the analysis device obtains the connection states respectively corresponding to multiple unit moments of the N training terminal pairs in the second historical time period.
  • the training sample sequence corresponding to the first training terminal pair includes M+Q elements, and the value of each of the M+Q elements corresponds to the first training terminal pair A connection state corresponding to each unit time in the M+Q consecutive unit times.
  • the N training sample sequences are used as the input of a machine learning algorithm to obtain the prediction model output by the machine learning algorithm.
  • the analysis device first obtains a training sample sequence reflecting the historical connection status trend of the training terminal, and then uses a machine learning algorithm to train a large number of training sample sequences to generate a prediction model. This provides an effective predictive model learning method.
  • the prediction model itself reflects the universal and dynamic trend information or regularity information of the connection status of multiple terminals in the same network scenario.
  • the analysis device obtains the connection states respectively corresponding to multiple unit moments of the test terminal pair in the first historical time period as described in the following steps.
  • the analysis device selects a first set of target entries from the entries corresponding to each of the stored multiple data streams, the first set of target entries including the recorded unit time belongs to the first historical time period, and the source IP address is the The IP address of the first terminal, the destination IP address is the entry of the second terminal, and the recorded unit time belongs to the first historical time period, the destination IP address is the IP address of the first terminal, and the source IP The address is the entry of the second terminal.
  • the analysis device determines that the connection state corresponding to the unit time recorded in the selected first group of target entries is connected, and determines that the selected first group of target entries is excluded from the first historical time period
  • the connection status corresponding to the unit time other than the unit time recorded in is disconnected, so that the connection status corresponding to the multiple unit times of the test terminal pair in the first historical time period is obtained.
  • the analysis device obtains the historical connection status information of the test terminal pair with the granularity of the time period between two adjacent unit moments, so as to facilitate subsequent analysis of the test terminal pair based on the historical connection status information of the test terminal pair.
  • the future connection status is predicted.
  • the analysis device obtains the connection states respectively corresponding to multiple unit moments of the N training terminal pairs in the second historical time period through the following steps.
  • the analysis device obtains N training terminal pairs.
  • the analysis device selects a training terminal pair from the N training terminal pairs, and executes the following processing steps on the selected training terminal pair until all the N training terminal pairs are processed.
  • the terminal pair is composed of a third terminal and a fourth terminal: the analysis device selects a second set of target entries from the entries corresponding to the stored multiple data streams, and the second set of target entries includes the recorded unit time belonging to the The second historical time period, the source IP address is the IP address of the third terminal, and the destination IP address is the entry of the fourth terminal, and the recorded unit time belongs to the second historical time period, and the destination IP address is the fourth terminal.
  • the terminal’s IP address and the source IP address is the entry of the third terminal; the analysis device determines that the connection status corresponding to the unit time recorded in the selected second target entry is connected, and determines the first 2.
  • the analysis device obtains the training terminal pair historical connection state information with the granularity of the time period between two adjacent unit moments, so as to facilitate subsequent training of the prediction model based on the training terminal pair historical connection state information.
  • the ratio between the number of positive samples and the number of negative samples in the N training sample sequences is greater than or equal to 0.5 and less than or equal to 2
  • the positive sample is a training sample sequence in which the connection state indicated by the value of the last element is connected
  • the negative sample is a training sample sequence in which the connection state indicated by the value of the last element is disconnected.
  • the training sample sequence that meets the above conditions is regarded as a balanced sample set.
  • the analysis equipment trains the prediction model based on the balanced sample set, and can obtain the prediction model with better prediction effect.
  • the analysis device obtains several items from the flow statistical information through the following steps.
  • the entry can be regarded as the raw data used to obtain the connection status of the test terminal pair or the training terminal pair.
  • the analysis device obtains multiple flow statistics information, each flow statistics information in the multiple flow statistics information corresponds to a data flow, and each flow statistics information includes the establishment time, closing time, and source IP address of the data flow And the destination IP address.
  • the analysis device performs time alignment processing on the statistical information of each stream based on the unit time, generates entries corresponding to multiple data streams, and saves the multiple data streams.
  • the entry corresponding to each data stream in the entries corresponding to each of the multiple data streams records a unit time, a source IP address, and a destination IP address.
  • the entry data generated in the above manner only retains the information related to the connection state in the flow statistical information, which reduces the amount of data and saves storage space compared with the flow statistical information.
  • time alignment processing is performed in the process of generating item data, which is beneficial to the subsequent improvement of processing efficiency.
  • the first terminal, the second terminal, the third terminal, and the fourth terminal in the first aspect or any one of the possible implementation manners of the foregoing first aspect are virtual machines, respectively.
  • the virtual machine is deployed in a data center connected by DCN.
  • the prediction method provided by the embodiment of the present application is suitable for predicting the connection state between two virtual machines in the DC.
  • a device for predicting a connection state between terminals has a function of implementing the method described in the first aspect or any one of the possible implementation manners of the foregoing aspects.
  • the function can be realized by hardware, or by hardware executing corresponding software.
  • the hardware or software includes one or more modules corresponding to the above-mentioned functions.
  • an analysis device in the third aspect, includes a memory and at least one processor. After the instructions are used to store instructions, after the instructions are read by the at least one processor, the analysis device executes the first aspect or the method in any one of the possible implementations of the first aspect. For details, refer to the above The detailed description will not be repeated here.
  • an embodiment of the present application provides a computer storage medium for storing computer software instructions used by the above analysis device, which includes the instructions used to execute the above first aspect or any one of the possible implementations of the above aspects. Designed procedures.
  • a computer program product containing instructions which when running on a computer, causes the computer to execute the method described in the first aspect or any one of the possible implementations of the first aspect.
  • an embodiment of the present application provides a chip including a memory and a processor, the memory is used to store computer instructions, and the processor is used to call and run the computer instructions from the memory to execute the first aspect and its first aspect.
  • the method in any possible implementation of the aspect.
  • FIG. 1 is a schematic diagram of an application scenario of an embodiment of the application
  • FIG. 2 is a flowchart of a method for predicting the connection state between terminals according to an embodiment of the application
  • FIG. 3 is a flowchart of a method for predicting the connection state between terminals based on a predictive model according to an embodiment of the application;
  • 4A is a schematic diagram of a training sample sequence corresponding to a first training virtual machine pair provided by an embodiment of the application;
  • 4B is a schematic diagram of N training sample sequences provided by an embodiment of the application.
  • 4C is a schematic diagram of a first sample sequence corresponding to a test virtual machine pair provided in an embodiment of the application;
  • FIG. 5 is a schematic diagram of a process in which the analysis device in an embodiment of the application inputs N training sample sequences into the MLP to obtain a prediction model;
  • FIG. 6 is a schematic structural diagram of an analysis device provided by an embodiment of the application.
  • FIG. 7 is a schematic structural diagram of an apparatus for predicting a connection state between terminals provided in an embodiment of the application.
  • an embodiment of the present invention proposes a method for predicting the connection state between terminals.
  • This method is based on the connection status of a terminal pair (in this embodiment, a pair of terminals consisting of two terminals is called a "terminal pair") at multiple unit moments in the historical time period, and uses artificial intelligence technology to extract useful information. For example, construct a mathematical model of the connection state of a terminal pair based on the long-term historical connection state information of a terminal, or construct a predictive model based on the long-term historical connection state information of several terminal pairs.
  • connection status corresponding to at least one unit time of the pair of terminals in the future time period is obtained according to the above useful information. This method makes full use of the terminal's long-term historical connection state information to predict, which helps to improve the accuracy of the prediction.
  • the prediction method provided by the embodiment of the present application is applicable to various network scenarios, such as a local area network within a company, a government department, or a school, and DCN.
  • the terminals constituting the terminal pair are personal computers, notebook computers, mobile terminals, wearable devices, or virtual machines.
  • the two terminals constituting the terminal pair are devices of the same type, for example, both are personal computers or both are virtual machines.
  • the two terminals constituting the terminal pair are devices of different types.
  • one terminal in the terminal pair is a personal computer and the other terminal is a virtual machine
  • one terminal in the terminal pair is a mobile terminal and the other terminal is a virtual machine. machine.
  • the DCN scenario is mainly used as an example to describe the prediction method provided in the embodiment of the present application.
  • the DCN scenario is characterized by a large number of virtual machines that provide computing resources.
  • a virtual machine is a logical computer device with complete software and hardware system functions simulated by virtualization technology.
  • the host is the basis for the implementation of virtualization technology, that is, the computer equipment that provides actual hardware resources for the virtualization technology.
  • the virtualization technology is implemented by virtualization software
  • the virtualization software after the virtualization software is installed on the host machine, one or more virtual machines can be generated according to the configuration based on the hardware resources of the host machine. Therefore, the host can also be regarded as the hardware platform on which the virtual machine runs.
  • a terminal pair refers to a virtual machine pair, that is, a pair of virtual machines composed of two virtual machines. Since the implementation principle of the prediction method is basically similar in different scenarios, it will not be illustrated one by one.
  • Fig. 1 is a schematic diagram of an application scenario of an embodiment of the present application in a DCN scenario.
  • the DCN includes multiple hosts, denoted as host 1, host 2, host 3, host 4, and host 5.
  • host 1 runs VM 1a and VM 1b
  • host 2 runs VM 2a
  • host 3 runs VM 3a, VM 3b, and VM 3c
  • host 4 runs VM 4a and VM 4b
  • host 5 runs VM 5a and VM 5b.
  • Packet forwarding equipment includes various switches, such as Layer 2 switches or Layer 3 switches, and so on.
  • the two-layer switch works at the data link layer, and can identify the MAC address information in the data packet, and according to the identified MAC address, look up the address table containing the correspondence between the MAC address and the port number to realize forwarding.
  • the three-layer switch works at the network layer and realizes forwarding through the three-layer switching technology.
  • Layer 3 switching technology is a technology that combines routing technology and switching technology into one. After the Layer 3 switch routes the first data flow, the Layer 3 switch will generate a MAC address and IP address mapping table.
  • switches S 1, S 2, S 3, and S 4 in Figure 1 are Layer 2 switches
  • S 5 and S 6 are Layer 3 switches.
  • the host computer As shown by the dotted line in FIG. 1, there is a physical connection between the host computer and the message forwarding device, for example, a connection via an Ethernet.
  • the virtual machines communicate with each other through the physical connection between the message forwarding device and the host machine as the virtual machine running platform.
  • the application scenario shown in Figure 1 also includes an analysis device.
  • the analysis device communicates with the data source device to obtain multiple stream statistics.
  • the data source device includes a message forwarding device and a host computer.
  • Each flow statistics information corresponds to a data flow, and each flow statistics information includes the establishment time, source IP address, and destination IP address of the data flow.
  • a data stream refers to a series of messages from a source computer to a destination.
  • the destination can be another computer, a group of computers or a broadcast domain.
  • the data source device mirrors the traffic transmitted through the network interface of the device, and sends the mirrored traffic to the analysis device, and the analysis device simply analyzes the mirrored traffic to obtain the flow statistics information.
  • Simple analysis includes filtering out Synchronize Sequence Numbers (SYN) packets from all traffic and extracting source and destination IP addresses from SYN packets, according to the sending time of SYN packets and the extracted source IP The address and destination IP address generate flow statistics. This method does not consume too much processing resources of the data source device, and has low hardware requirements for the data source device, and is suitable for the case where the data source device is a switch or a host.
  • SYN Synchronize Sequence Numbers
  • the data source device simply analyzes the packet transmitted through the network interface of the device to obtain the flow statistics information, and sends the flow statistics information to the analysis device.
  • the data source device directly sending the mirrored message since the data volume of the flow statistics information is smaller than the data volume of the mirrored message, network transmission resources can be saved. Since this method has certain requirements on the processing capability of the data source device, it is relatively suitable for the case where the data source device is the host machine.
  • Table 1 is an example of flow statistical information received by the analysis device, where each row represents a piece of flow statistical information.
  • different data sources may use different formats and encoding methods to record the establishment time, source IP address, and destination IP address of the data stream. For example, use binary, decimal, or hexadecimal to record address information.
  • the analysis device first performs format conversion on the received original stream statistical information, and normalizes it to stream statistical information in a unified format. It is understandable that the IP address in the flow statistics information is to distinguish different virtual machines. For ease of understanding and description, the virtual machine identifier is used in this embodiment to replace the IP address.
  • the analysis device performs time alignment processing on the statistical information of each stream based on the unit time according to a predetermined time alignment rule, generates entries corresponding to multiple data streams, and saves the corresponding entries of the multiple data streams. Entry.
  • the time granularity used can be determined by the administrator based on various factors such as the storage space, processing resources, DCN network scale, and analysis purpose of the analysis device. Set up. Time alignment processing can not only reduce the amount of data to save storage space, but also help improve the efficiency of subsequent analysis.
  • the predetermined time alignment rule can be flexibly set.
  • the granularity during time alignment can be set according to requirements, such as 1 hour, half an hour, 10 minutes, 1 minute, and so on. It is assumed that in this embodiment of the present application, when the analysis device performs time alignment processing on the received multiple stream statistical information, the time granularity adopted is 1 hour. In other words, the entry obtained after the alignment process is based on the unit time of 1 hour.
  • a time alignment rule is to treat the time between two unit moments as the unit time that is the earlier of the two unit moments, for example, “2015-1-10 11:23:00” is processed It is "2015-1-10 11:00:00".
  • Another time alignment rule is to treat the time between two unit moments as the closest unit moment of the two unit moments. For example, “2015-1-10 11:55:00” is processed It is "2015-1-10 12:00:00".
  • the analysis equipment After time alignment processing, the analysis equipment obtains the entries shown in Table 2 and saves these entries for subsequent use.
  • Unit time identification with hour as granularity Source IP address Destination IP address 2015-1-10 11:00:00 VM 1a VM 1b 2015-1-10 12:00:00 VM 1a VM 1b 2015-1-10 13:00:00 VM 2a VM 3a 2015-1-10 9:00:00 VM 1a VM 2a ... ... ... ...
  • the analysis device analyzes the trend or regularity information of the connection state of the virtual machine pair based on the entries corresponding to the stored multiple data streams, or constructs a prediction model through artificial intelligence technology.
  • Artificial intelligence technology refers to the technology that makes machines manufactured by humans behave similar to human intelligence. Judging from existing research, artificial intelligence technology includes machine learning algorithms.
  • FIG. 2 is a flowchart of a method for predicting the connection state between terminals according to an embodiment of the present application.
  • Figure 2 mainly describes the method from the perspective of analysis equipment.
  • the analysis device in FIG. 2 is the analysis device in FIG. 1.
  • Step 21 The analysis device obtains the connection states of the test terminal pair corresponding to multiple unit moments in the first historical time period.
  • the test terminal pair is composed of a first terminal and a second terminal, the first historical time period is a time period before the current time, and the first historical time period includes M consecutive unit times, where M is A natural number greater than or equal to 2.
  • the current time is 9:20 on January 11, 2015
  • the current prediction task of the analysis device is to predict between the virtual machines VM 1a and VM 2a in the scenario shown in Figure 1 2015-1-11 10:00 Connection status.
  • the administrator can input prediction tasks through the input device connected to the input and output interface of the analysis device. That is, the test virtual machine pair in this embodiment is (VM 1a-VM 2a).
  • the first historical time period contains 3 unit times, namely 2015-1-11 7:00, 2015-1-11 8:00 and 2015-1-11 9:00.
  • the analysis equipment first obtains the test virtual machine pair (VM 1a-VM 2a) corresponding to 7:00 on 2015-1-11, 8:00 on 2015-1-11 and 10:00 on 2015-1-11 respectively Connection status.
  • the analysis device adopts step 21a and step 21b to obtain the connection states of the test terminal pair corresponding to multiple unit moments in the first historical time period.
  • Step 21a The analysis device selects a first set of target entries from the entries corresponding to each of the stored multiple data streams.
  • the first set of target entries includes the recorded unit time belonging to the first historical time period, and the source IP address is The IP address of the first terminal, the destination IP address is the entry of the second terminal, and the recorded unit time belongs to the first historical time period, the destination IP address is the IP address of the first terminal, and the source The IP address is the entry of the second terminal.
  • Step 21b The analysis device determines that the connection status corresponding to the unit time recorded in the selected first group of target entries is connected, and determines that the selected first group of targets are excluded from the first historical time period. The connection status corresponding to the unit time other than the unit time recorded in the entry is disconnected, so that the connection status corresponding to the multiple unit time of the test terminal pair in the first historical time period is obtained.
  • the analysis device selects items satisfying any one of the following two conditions from the items shown in Table 2 to form the first group of target items.
  • the unit time is 7:00 on January 11, 2015, 8:00 on January 11, 2015 or 10:00 on January 11, 2015, and the source IP address is VM 1a, and the destination IP address is VM 2a.
  • the unit time is 7:00 on January 11, 2015, 8:00 on January 11, 2015 or 10:00 on January 11, 2015, and the source IP address is VM 2a, and the destination IP address is VM 1a.
  • the analysis device determines that the connection status of the test virtual machine pair (VM 1a-VM 2a) at 2015-1-10 8:00:00 is connected, and the connection status at 2015-1-10 9:00:00 is connected , The connection status at 7:00:00 2015-1-10 is no connection.
  • Step 22 The analysis device determines the connection state corresponding to at least one unit time of the test terminal pair in the future time period according to the connection state of the test terminal pair corresponding to multiple unit times in the first historical time period.
  • the future time period is a time period after the current time, and the future time period includes Q consecutive unit times, the first unit time in the future time period and the last time in the first historical time period
  • a unit time is a continuous unit time, where Q is a natural number greater than or equal to 1.
  • the analysis device first obtains the test virtual machine as the test object from the data source device for multiple unit moments in history. Connection status information.
  • the analysis device predicts the connection state of the test virtual machine to the future time period based on the connection state information of the test virtual machine to multiple unit moments in the history. Since the connection status information of the test virtual machine to multiple unit moments in history is used in the prediction process, instead of the connection status information of the test virtual machine to a single unit moment in history, it is helpful to analyze and find more useful information from the historical state information. Information, such as more detailed and specific rules or trend information, to improve forecast accuracy.
  • the analysis device uses the prediction method provided in the embodiment of the present application to obtain the prediction result (that is, the connection state of the test terminal pair corresponding to at least one unit time in the future time period), the analysis device applies the prediction result to different In scenarios, such as failure impact analysis and configuration verification scenarios, the accuracy of failure impact analysis and configuration verification can be further improved.
  • connection status of the test terminal to the future time period based on the connection status information of the test terminal to multiple unit moments in the history
  • implementation schemes include, but are not limited to, methods of constructing mathematical models and predictive models based on historical connection state information including connection states at multiple unit moments.
  • the following embodiments of the present application take a mathematical model or a prediction model as an example to describe the prediction method provided by the embodiments of the present application.
  • the analysis device determines the mathematical model of the connection state of the test terminal pair according to the connection states respectively corresponding to the multiple unit moments of the test terminal pair in the first historical time period.
  • the further analysis device predicts the connection state of the test terminal corresponding to at least one unit time in the future time period according to the mathematical model.
  • the analysis device pre-stores matching rules for multiple mathematical models.
  • the analysis device matches the connection status of the test virtual machine pair (VM 1a-VM 2a) at multiple unit moments in the first historical time period with the matching rules one by one, thereby determining the test virtual machine pair (VM 1a-VM 2a)
  • the mathematical model that the historical connection status information conforms to.
  • the analysis device can also use other mechanisms to learn a mathematical model that matches the historical connection state information of the test virtual machine pair (VM 1a-VM 2a).
  • Example 1 Example 1
  • Example 2 Example 2
  • the analysis device determines the connection state of the test virtual machine pair (VM 1a-VM 2a) according to the connection state corresponding to each hour of the test virtual machine pair (VM 1a-VM 2a) in the past 24 hours as "the connection state is After being connected for 2 hours, switch to no connection for 3 hours, then switch to connected and last for 2 hours, reciprocating in turn.” As shown in Table 4. For the sake of brevity, the connection status is indicated by a value of 0 or 1 in Table 4. 0 means no connection and 1 means there is connection.
  • the analysis device determines the connection status of the test virtual machine pair (VM 1a-VM 2a) in the future time period 2015-1-11 0:00:00-24:00:00 as shown in Table 5.
  • the analysis device determines the connection state of the test virtual machine pair (VM 1a-VM 2a) according to the corresponding connection state of the test virtual machine pair (VM 1a-VM 2a) in the past 24 hours, and determines that the mathematical model of the connection state of the test virtual machine pair (VM 1a-VM 2a) is “the connection state is After there is a connection for n hours, it will be switched to no connection for n hours, and then it will be connected. n starts from 1 and increments by 1". As shown in Table 6. For the sake of brevity, the connection status is indicated by a value of 0 or 1 in Table 6, where 0 means no connection and 1 means there is connection.
  • the analysis device determines the connection status of the test virtual machine pair (VM 1a-VM 2a) in the future time period 2015-1-11 0:00:00-11:00:00 according to the above mathematical model, as shown in Table 7.
  • the analysis device first obtains the test virtual machine as the test object from the data source device to multiple consecutive units in history Connection status information at the moment.
  • the analysis device uses the connection state information of the test virtual machine to multiple consecutive unit moments in history as the test basis to predict the connection state of the test virtual machine to the future time period. Since the prediction is based on the connection state information of the test virtual machine to multiple consecutive unit moments in history, rather than the connection state information of the test virtual machine to a single unit moment in history, it is helpful to analyze and find the connection state of the test virtual machine. Long-term historical trends to improve forecast accuracy.
  • the analysis device determines the connection state of the test terminal pair corresponding to at least one unit time in the future time period according to the prediction model.
  • the prediction model is trained by the analysis equipment through the machine learning algorithm based on the historical connection status information of a large number of terminal pairs in the network scene where the test terminal pair as the prediction target is located. Among them, a large number of terminal pairs used to train the prediction model are used as the prediction target.
  • the test terminal pairs are in the same network scenario.
  • the terminal pair used for predicting model training is called a training terminal pair in this embodiment.
  • the training terminal pair may or may not include a prediction terminal pair, which is not limited here.
  • the analysis device generates a prediction model based on the connection states respectively corresponding to multiple unit moments in the second historical time period by the N training terminals.
  • the prediction model is generated based on the connection states of the N training terminal pairs corresponding to multiple unit moments in the second historical time period.
  • the second historical time period is a time period before the current time, and the second historical time period includes M+Q consecutive unit moments, where N is a natural number greater than or equal to 1, usually when N takes a value of 100 A more satisfactory effect can be achieved when it reaches 10,000 level. Within a reasonable value range, the larger the number of N, the more accurate the prediction result.
  • the analysis device inputs the connection states of the test terminal pairs corresponding to the multiple unit moments in the first historical time period into the prediction model and obtains the output result of the prediction model.
  • Predicted demand refers to M and Q, that is, "according to the connection state of the test terminal pair in the first historical time period including M unit moments, predict the connection state of the test terminal pair in the future time period including Q unit moments".
  • the administrator can input the predicted demand through the input device connected to the input interface of the analysis device.
  • the input of the prediction model is the connection state of the test terminal pair corresponding to multiple unit moments in the first historical time period.
  • the output is the connection status corresponding to at least one unit of the test terminal pair in the future time period.
  • the future time period will be 2015-1-11 9:20 to 2015-1-11 10:20.
  • the future time period contains 1 unit time, namely 10:00 on January 11, 2015.
  • the analysis device inputs the connection state of the test virtual machine pair (VM 1a-VM 2a) determined in step 21 into the prediction model at each unit time within 3 hours before the current time, that is, the "test virtual machine pair (VM 1a-VM 2a)
  • the connection status of 2015-1-10 7:00:00 is no connection
  • the connection status of 2015-1-10 8:00:00 is connected
  • the connection status of 2015-1-10 9:00:00 is The information of "connected” is input to the prediction model.
  • the prediction model outputs the connection status at 10:00 on January 11, 2015 as connected.
  • the analysis device determines that the connection status of the test virtual machine at 10:00:00 2015-1-10 that has not yet arrived is linked.
  • the analysis device obtains historical connection status information of a large number of training virtual machine pairs, and the historical connection status information includes multiple unit moments. The corresponding connection status.
  • the analysis device further combines the forecasting demand to generate a forecasting model based on the historical connection state information of a large number of training virtual machine pairs.
  • the analysis device inputs the connection status of the test virtual machine pair at multiple unit moments in the historical time period into the prediction model, and determines the test virtual machine according to the output of the prediction model For the connection status corresponding to at least one unit time in the future time period.
  • a large amount of historical connection state information includes, on the one hand, historical connection state information of a large number of virtual machine pairs, on the other hand, historical connection state information includes connection state information corresponding to at least two unit moments.
  • the connection state of the test virtual machine at the same time in the day before is used as the prediction scheme of the connection state of the pair of virtual machines at the same time in the future, and the connection state of the test virtual machine at the previous moment is predicted to be the connection of the virtual machine at the next moment in the future
  • the prediction method of the embodiment of the present application performs prediction based on a large amount of historical connection state information, which reduces errors caused by accidental factors and improves prediction accuracy.
  • Fig. 3 is a flowchart of a method for predicting the connection state between terminals based on a predictive model provided by an embodiment of the present application.
  • the analysis device in FIG. 3 is the analysis device in FIG. 1 or FIG. 2.
  • the sub-process composed of steps 31 to 33 in the process shown in FIG. 3 mainly describes the process of generating a predictive model by the analysis equipment, and the sub-process composed of steps 34 to 36 mainly describes the analysis equipment based on the predictive model, and the test terminal pair The process of predicting the connection status.
  • the analysis device after the analysis device generates the prediction model, it can predict the connection status of multiple test terminal pairs based on the prediction model, instead of regenerating the prediction model based on the historical connection status information of the training terminal pair for each prediction.
  • Step 31 The analysis device obtains the connection states of the N training terminal pairs corresponding to multiple unit moments in the second historical time period.
  • the second historical time period please refer to the previous introduction and will not be repeated here.
  • step 31 includes several sub-steps of step 311 to step 314.
  • Step 311 The analysis device obtains N training terminal pairs.
  • the analysis device adopts multiple methods to obtain the training terminal pair. For example, the analysis device reads the saved entries shown in Table 2, and obtains the training terminal pair according to the source IP address and the destination IP address in the entry. Alternatively, the analysis device may also obtain an IP address allocated to the terminal in the network through an address management device (for example, a Dynamic Host Configuration Protocol (DHCP) server), and then generate several terminal pairs through permutation and combination. The analysis device then selects N training terminal pairs from a number of terminal pairs generated by permutation and combination.
  • the selection methods include random selection, selection in a predetermined order, etc., which will not be described in detail here.
  • Step 312 The analysis device selects a training terminal pair from the N training terminal pairs, and performs processing steps 312a and 312b on the selected training terminal pair until all the N training terminal pairs are processed.
  • the selected training terminal pair is composed of a third terminal and a fourth terminal.
  • Step 312a The analysis device selects a second set of target entries from the entries corresponding to each of the stored multiple data streams, the second set of target entries including the recorded unit time belongs to the second historical time period, and the source IP address is the first Three terminal IP addresses, the destination IP address is the entry of the fourth terminal, and the recorded unit time belongs to the second historical time period, the destination IP address is the IP address of the fourth terminal, and the source IP address is all Describe the entry of the third terminal.
  • Step 312b The analysis device determines that the connection status corresponding to the unit time recorded in the selected second target entry is connected, and determines that the selected second set of target entries are excluded from the second historical time period. The connection status corresponding to the unit time other than the unit time recorded in is disconnected, so that the connection status corresponding to the multiple unit times of the selected training terminal pair in the second historical time period is obtained.
  • Step 312a and step 312b are similar to step 21a and step 21b in FIG. 2 respectively. The description will not be repeated here.
  • the forecast demand is based on the connection status of each hour in the past 6 days, and the connection status of the next hour is predicted.
  • the connection status table shown in Table 8 a value of 0 or 1 is used to indicate different connection statuses, 0 means no connection, and 1 means there is a connection.
  • Table 8 is an example of the connection status corresponding to each unit moment of the selected training virtual machine pair in the second historical time period.
  • the analysis device performs step 312a and step 312b on all N training virtual machine pairs, and obtains N state information tables as shown in Table 8.
  • Step 32 The analysis device generates a training sample sequence corresponding to the first training terminal pair for the connection states corresponding to the first training terminal pair in the second historical time period at multiple unit moments in the N training terminal pairs, to By analogy, N training sample sequences are obtained.
  • the training sample sequence corresponding to the first training terminal pair includes M+Q elements, and the value of each of the M+Q elements corresponds to the The first training terminal pair corresponds to the connection state of each unit time in the M+Q consecutive unit times.
  • the analysis device For the first training virtual machine pair among the N training virtual machine pairs, the analysis device generates the first virtual machine pair based on the state information table similar to that shown in Table 8 corresponding to the first training virtual machine pair The corresponding training sample sequence.
  • the training sample sequence contains M+Q (145) elements, as shown in Figure 4A. Assuming that the value of the element is 0 or 1, 0 means no connection, and 1 means there is connection.
  • the analysis device uses the last element as the label in the training sample sequence, as shown in 42 in FIG. 4A.
  • the analysis device performs the foregoing step 32 for each training virtual machine pair in the N training virtual machine pairs, thereby generating N training sample sequences, as shown in FIG. 4B.
  • Step 33 The analysis device inputs the N training sample sequences as training samples into a machine learning algorithm to obtain the prediction model generated by the machine learning algorithm.
  • machine learning algorithms include but are not limited to neural networks, decision trees, random forests, support vector machines, and so on. Due to the numerous machine learning algorithms, it is difficult to describe the process of using each machine learning algorithm to generate a predictive model based on N training sample sequences.
  • the embodiment of the present application uses one of the machine learning algorithms to generate a prediction model as an example for illustration.
  • a multi-layer perceptron is taken as an example to describe the generation process of the prediction model in detail.
  • the basic computing unit of a neural network is a node, which is also called a neuron.
  • the node receives the input from the external input, and generates output after calculating the activation function.
  • Weight represents the strength of the connection between the output node and the receiving node. The size of the weight value will be automatically adjusted during the training process of the neural network until it stabilizes.
  • the weight value is the main object of training.
  • the activation function is denoted as f(), which is generally non-linear, and its main function is to add non-linear characteristics to the output of the neuron and enhance the learning ability of the neural network on the training samples.
  • FIG. 5 is a schematic diagram of the process of inputting N training sample sequences into the MLP to obtain the prediction model.
  • the MLP in Figure 5 includes an input layer (input layer) and an output layer (output layer).
  • the MLP further includes one or more hidden layers.
  • this embodiment uses the MLP including two hidden layers as an example for description.
  • the number of nodes contained in each hidden layer is configurable. For example, the first hidden layer contains 64 nodes, and the second hidden layer contains 16 nodes.
  • the number of nodes contained in the input layer of the MLP is the same as the number of elements contained in the sample part of the training sample sequence, and the number of output nodes is the same as the number of elements contained in the label of the training sample sequence. Since the number of elements contained in the sample part of the training sample sequence in this embodiment is 144, the number of nodes contained in the MLP input layer is 144; the number of elements contained in the label of the training sample sequence in this embodiment is 1, so the MLP output layer contains The number of nodes is 1.
  • each element of the sample part of the training sample sequence is input into the node corresponding to the MLP input layer.
  • the analysis device compares the value of the output layer node with the element value of the label of the training sample sequence. If the value of the output layer node is significantly different from the element value of the label of the training sample sequence, the MLP automatically adjusts the weight value through f().
  • the learning process of the prediction model is a process in which the MLP receives N training samples input by the analysis device, and adjusts the weight value according to the difference between the value of the output layer node and the element value of the label of the training sample sequence.
  • the N training sample sequences used by the analysis device to input the machine learning algorithm to generate the prediction model are balanced sample sets.
  • the balanced sample set refers to the number of positive samples and the number of negative samples in the N training sample sequence used to train and generate the prediction model. In other words, the ratio between the number of positive samples and the number of negative samples in the N training sample sequence is within a reasonable range.
  • the positive sample is a training sample sequence in which the connection state indicated by the value of the last element is connected
  • the negative sample is a training sample sequence in which the connection state indicated by the value of the last element is disconnected.
  • a reasonable range that can be implemented is a value from 0.5 to 2.
  • Step 34 The analysis device determines a first sample sequence according to the connection status of the test terminal pair corresponding to the multiple unit moments in the first historical time period.
  • the first sample sequence includes M elements, and The value of each element of the M elements respectively corresponds to the connection state corresponding to each of the M consecutive unit moments.
  • the method for the analysis device to determine the first sample sequence is basically similar to the method for generating the training sample sequence in step 32 of this process, and will not be further described here.
  • the generated first sample sequence is shown in FIG. 4C.
  • the first sample sequence contains M(144) elements.
  • Step 35 The analysis device inputs the first sample sequence into a prediction model and obtains an output result of the prediction model.
  • the output result of the prediction model is a prediction sequence
  • the prediction sequence includes Q elements
  • the value of each element in the Q elements corresponds to each unit in the Q consecutive unit moments.
  • Step 36 The analysis device determines the connection state of the test terminal pair corresponding to at least one unit time in the future time period according to the output result of the prediction model.
  • the analysis device determines that the connection status corresponding to the test virtual machine pair (VM 1a-VM 2a) in the next hour is connected.
  • the embodiment of the present application provides a detailed process of generating a prediction model and predicting the connection state between terminals based on the prediction model.
  • the sub-process composed of step 31 to step 33 in FIG. 3 introduces how to generate a prediction model based on the historical connection state information of the training terminal pair.
  • the analysis device first obtains the connection state information of a large number of training virtual machines to multiple consecutive unit moments in the history from the data source device, and generates the connection state information based on the obtained connection state information using a machine learning algorithm. Forecast model.
  • the trend information or law information reflected by the prediction model is more universal, reduces errors caused by accidental factors, and can further improve prediction accuracy.
  • the sub-process composed of step 34 to step 36 in FIG. 3 mainly describes the process of the analysis device predicting the connection state of the test terminal based on the prediction model. Based on the predictive model, a large amount of historical connection status information in DCN is fully utilized to predict the connection status. After testing on actual data, the accuracy of the prediction method provided in the embodiments of the present application can reach about 98%, which is significantly improved compared with the existing related technologies.
  • FIG. 6 is a schematic structural diagram of an analysis device provided by an embodiment of the present application.
  • the analysis device shown in FIG. 6 is the analysis device in the application scenario shown in FIG. 1 or the analysis device in the process shown in FIG. 2 or FIG. 3.
  • the analysis device includes at least one processor 61 and a memory 62.
  • the at least one processor 61 may be one or more CPUs, and the CPU may be a single-core CPU or a multi-core CPU.
  • the memory 62 includes but is not limited to random access memory (RAM), read only memory (ROM), erasable programmable read-only memory, EPROM or flash Memory), flash memory, or optical memory, etc.
  • RAM random access memory
  • ROM read only memory
  • EPROM erasable programmable read-only memory
  • flash memory or optical memory, etc.
  • the code of the operating system is stored in the memory 62.
  • the processor 61 implements the method in the foregoing embodiment by reading instructions stored in the memory 62, or the processor 61 may also implement the method in the foregoing embodiment by using internally stored instructions.
  • the processor 61 implements the method in the foregoing embodiment by reading the instructions stored in the memory 62
  • the memory 62 stores the instruction for implementing the method provided in the foregoing embodiment of the present application.
  • the analysis device After the program code stored in the memory 62 is read by the at least one processor 61, the analysis device performs the following operations: obtain the connection status of the test terminal pair corresponding to multiple unit moments in the first historical time period, and the test terminal For the first terminal and the second terminal, the first historical time period is a time period before the current time, and the first historical time period includes M consecutive unit times, where M is greater than or equal to 2 A natural number; according to the connection state of the test terminal pair corresponding to multiple unit moments in the first historical time period, determine the connection state of the test terminal pair corresponding to at least one unit time in the future time period, the The future time period is a time period after the current time, and the future time period includes Q consecutive unit times, the first unit time in the future time period and the last unit in the first historical time period Time is a continuous unit time, where Q is a natural number greater than or equal to 1.
  • the analysis device shown in FIG. 6 further includes a network interface 63.
  • the network interface 63 may be a wired interface, such as a Fiber Distributed Data Interface (FDDI) or a Gigabit Ethernet (GE) interface; the network interface 63 may also be a wireless interface.
  • the network interface 63 is used to receive mirrored traffic or multiple stream statistical information from the data source.
  • the memory 62 is used to store the mirrored traffic or multiple flow statistics received by the network interface 63.
  • the at least one processor 61 is configured to process the mirrored traffic or multiple flow statistical information, obtain several entries shown in Table 2 above, and save these entries to the memory 62.
  • the at least one processor 61 further executes the prediction method described in the foregoing method embodiment according to the entries stored in the memory 62.
  • the processor 61 implementing the foregoing functions please refer to the descriptions in the foregoing method embodiments, which are not repeated here.
  • the analysis device further includes a bus 64, and the aforementioned processor 61 and memory 62 are usually connected to each other through the bus 64, and may also be connected to each other in other ways.
  • the analysis device further includes an input and output interface 65, which is used to connect with the input device and receive the predicted demand input by the user through the input device.
  • Input devices include but are not limited to keyboards, touch screens, microphones, etc.
  • the input and output interface 65 is also used to connect with an output device and output the prediction result of the processor 61.
  • Output devices include but are not limited to displays, printers, etc.
  • the analysis device provided in the embodiments of the present application is used to execute the prediction methods provided in the foregoing method embodiments.
  • This analysis device uses the connection status information of the test virtual machine to multiple unit moments in history in the prediction process, rather than the connection status information of the test virtual machine to a single unit moment in history, which is helpful to analyze and find from the historical state information. More useful information to improve forecast accuracy.
  • FIG. 7 is a schematic structural diagram of an apparatus for predicting a connection state between terminals provided by an embodiment of the present application.
  • the device 70 for predicting the connection state between terminals includes an acquiring module 71 and a predicting module 72.
  • the obtaining module 71 is configured to obtain the connection status corresponding to multiple unit moments of the test terminal pair in the first historical time period.
  • the test terminal pair is composed of a first terminal and a second terminal, and the first historical time period is The time period before the current time and the first historical time period includes M consecutive unit moments, where M is a natural number greater than or equal to 2.
  • the prediction module 72 is configured to determine the respective connection states of the test terminal pair corresponding to at least one unit time in the future time period according to the connection states of the test terminal pair corresponding to multiple unit times in the first historical time period.
  • the future time period is a time period after the current time, and the future time period includes Q consecutive unit times, the first unit time in the future time period is the same as the first historical time period
  • the last unit time in is a continuous unit time, where Q is a natural number greater than or equal to 1.
  • the prediction module 72 includes a model testing unit 721 and a determination unit 722.
  • the model testing unit 721 is configured to input a prediction model and obtain an output result of the prediction model by the test terminal to the connection states corresponding to the multiple unit moments in the first historical time period.
  • the prediction model is The N training terminals are generated based on the connection states corresponding to the multiple unit moments in the second historical time period, where the second historical time period is a time period before the current time, and the second historical time period includes M+Q consecutive unit moments, where N is a natural number greater than or equal to 1;
  • the determining unit 722 is configured to determine, according to the output result, the connection state respectively corresponding to at least one unit time of the test terminal pair in the future time period.
  • the model testing unit 721 is configured to determine a first sample sequence according to the connection status of the test terminal pair corresponding to multiple unit moments in the first historical time period, and the first sample sequence Includes M elements in the M elements, and the value of each element in the M elements respectively corresponds to the connection state corresponding to each of the M consecutive unit moments; and inputting the first sample sequence Predicting model and obtaining the output result of the predicting model, the output result is a prediction sequence, the prediction sequence includes Q elements, the value of each element of the Q elements corresponds to the Q consecutive The connection state corresponding to each unit time in the unit time.
  • the prediction module 72 in FIG. 7 further includes a model learning unit 723, configured to perform the following steps before the model testing unit 721 inputs the first sample sequence into the prediction model: Obtain N training terminal pairs Connection states corresponding to multiple unit moments in the second historical period of time; for the first training terminal pair of the N training terminal pairs corresponding to multiple unit moments in the second historical period of time , Generating the training sample sequence corresponding to the first training terminal pair, and so on, so as to obtain N training sample sequences.
  • a model learning unit 723 configured to perform the following steps before the model testing unit 721 inputs the first sample sequence into the prediction model: Obtain N training terminal pairs Connection states corresponding to multiple unit moments in the second historical period of time; for the first training terminal pair of the N training terminal pairs corresponding to multiple unit moments in the second historical period of time , Generating the training sample sequence corresponding to the first training terminal pair, and so on, so as to obtain N training sample sequences.
  • the training sample sequence corresponding to the first training terminal pair includes M+Q elements, and the M+Q The value of each element of the elements corresponds to the connection state of the first training terminal pair corresponding to each of the M+Q consecutive unit moments; the N training sample sequences are taken as The machine learning algorithm is input to obtain the prediction model output by the machine learning algorithm.
  • the device embodiment described in FIG. 7 is merely illustrative.
  • the division of the modules is only a logical function division, and there may be other divisions in actual implementation, for example, multiple modules or components may be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the functional modules in the various embodiments of the present application may be integrated into one processing module, or each module may exist alone physically, or two or more modules may be integrated into one module.
  • the above-mentioned modules in FIG. 7 can be implemented in the form of hardware or software functional units. For example, when implemented by software, the acquisition module 71, the prediction module 72, the model testing unit 721, the determining unit 722, and the model learning unit 723 can be read by at least one processor 61 in FIG.
  • the acquisition module 71 is composed of the network interface 63 in FIG. 6 and part of the processing resources of the at least one processor 63 (for example, one of the multi-core processors). Core), and the prediction module 72 uses the remaining processing resources of at least one processor 63 in FIG. 6 (for example, other cores in a multi-core processor), or a Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA), or a programmable device such as a coprocessor.
  • the above functional modules can also be implemented by a combination of software and hardware.
  • the acquisition module 71 is implemented by a hardware programmable device
  • the prediction module 72 is a software functional module generated after the CPU reads the program code stored in the memory.
  • the computer may be implemented in whole or in part by software, hardware, firmware, or any combination thereof.
  • software it can be implemented in the form of a computer program product in whole or in part.
  • the computer program product includes one or more computer instructions.
  • the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.
  • the computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium.
  • the computer instructions may be transmitted from a website, computer, server, or data center.
  • the computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server or a data center integrated with one or more available media.
  • the usable medium may be a magnetic medium (for example, a floppy disk, a hard disk, and a magnetic tape), an optical medium (for example, a DVD), or a semiconductor medium (for example, a solid state disk (SSD)).

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Signal Processing (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Environmental & Geological Engineering (AREA)
  • Debugging And Monitoring (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本申请公开了一种终端之间连接状态的预测方法、装置和分析设备。涉及人工智能技术在计算机网络领域中的应用。分析设备获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数。分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态。

Description

终端之间连接状态的预测方法、装置和分析设备 技术领域
本申请涉及计算机网络技术领域，进一步涉及人工智能(Artificial Intelligence，AI)技术在计算机网络领域中的应用，尤其涉及一种终端之间连接状态的预测方法、一种虚拟机之间连接状态的预测装置和一种分析设备。
背景技术
数据中心(data center,DC)是使用通信网络相互连接的若干资源形成的池(pool)，其中资源包括计算资源、存储资源和网络资源等等。由于虚拟机具有低成本、敏捷灵活、可扩展性好等方面的优势，因而虚拟机成为DC中重要的计算资源。数据中心网络(Data Center Network，DCN)用于将DC中的资源互联在一起，DCN在DC中起关键作用。为了应对增长的云计算的需求，DCN需要可扩展并高效地连接数以千百计的虚拟机、以及存储器等其他资源。
DC中的虚拟机之间相互通信，协同完成DC中的各种业务。虚拟机之间的连接状态指示两个虚拟机之间是否有通信。当两个虚拟机之间有通信时，这两个虚拟机之间的连接状态为有连接。反之，当两个虚拟机之间没有通信时，这两个虚拟机之间的连接状态为无连接。
虚拟机之间的连接状态预测技术(在本申请中后续将简称为“预测技术”)是DCN中的关键技术之一。该技术广泛应用于许多场景，例如故障影响分析和配置验证场景。故障影响分析是指当一个虚拟机发生故障后，通过虚拟机之间的连接状态预测技术确定理论上哪些其他虚拟机与该发生故障的虚拟机有连接，进而分析故障的影响范围。配置验证是指将要更新一个虚拟机的配置时(这里将即将进行配置更新的虚拟机记为VM 1)，通过虚拟机之间的连接状态预测技术确定假设没有进行配置更新，理论上哪些其他虚拟机与VM 1有连接。在对VM 1配置更新完成后，检测这些虚拟机与刚完成配置更新的VM 1是否有连接，来分析配置更新对VM 1与其他虚拟机之间连接的影响，避免错误配置影响业务通畅。
相关技术提出了几种预测方法。一种是基于业务规律性假设的预测方法。这种预测方法根据一对虚拟机前一天某一时刻的连接状态预测这对虚拟机未来一天同一时刻的连接状态。例如在2015年1月11日10点尚未到来前，将虚拟机VM 1与虚拟机VM 2在2015年1月10日10点的连接状态作为预测出的虚拟机VM 1与虚拟机VM 2在2015年1月11日10点的连接状态。
另一种是基于业务连续性假设的预测方法。这种预测方法根据一对虚拟机上一时刻的连接状态预测这对虚拟机未来下一个时刻的连接状态。例如在2015年1月11日10点尚未到来前，将虚拟机VM 1与虚拟机VM 2在2015年1月11日9点的连接状态作为预测出的虚拟机VM 1与虚拟机VM 2在2015年1月11日10点的连接状态。
然而，从具体实践结果来看，上面两种预测方法准确率不佳。
发明内容
本申请实施例提供一种终端之间连接状态的预测方法，用以改善相关预测技术的准确性不高的问题。
第一方面，提供了一种终端之间连接状态的预测方法。分析设备获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数。分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻对应的连接状态，所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
根据本申请实施例提供的终端之间连接状态的预测方法，由于分析设备在预测过程中使用了测试终端对历史上多个单位时刻的连接状态信息，而不是仅使用测试终端对历史上单一单位时刻的连接状态信息，有助于从历史状态信息中分析发现更多有用信息，从而提升预测准确性。
可选地，在一种可能的实现方式中，分析设备通过以下步骤确定测试终端对在未来时间段内至少一个单位时刻对应的连接状态。分析设备将所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，输入预测模型并获取所述预测模型的输出结果，所述预测模型是根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的，所述第二历史时间段是当前时间之前的时间段、且所述第二历史时间段中包括M+Q个连续的单位时刻，其中N为大于等于1的自然数。所述分析设备根据所述输出结果确定所述测试终端对在所述未来时间段内至少一个单位时刻分别对应的连接状态。分析设备利用机器学习算法，充分地利用大量训练终端对的长期历史连接状态信息训练出预测模型，能够提取出能够反映同一网络场景下多个终端对连接状态的普遍地、动态性的趋势信息或规律信息，从而更精准地进行预测。
可选地，在一种可能的实现方式中，分析设备过以下步骤获取所述预测模型的输出结果。所述分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定第一样本序列，所述第一样本序列中包括M个元素，所述M个元素中的每个元素的取值分别对应所述M个连续的单位时刻中每个单位时刻分别对应的连接状态。所述分析设备将所述第一样本序列输入预测模型并获取所述预测模型的输出结果，所述输出结果为预测序列，所述预测序列包括Q个元素，所述Q个元素中的每个元素的取值分别对应所述Q个连续的单位时刻中每个单位时刻分别对应的连接状态。分析设备首先获取反映测试终端对历史上连接状态趋势的第一样本序列，再将第一样本序列输入预测模型得到作为输出结果的预测序列。通过样本序列进行预测是一种有效的应用预测模型的方法。
可选地，所述M个元素或Q个元素中的一个元素的取值为第一值时指示对应单位时刻的连接状态为有连接，所述M个元素或Q个元素中的一个元素的取值为第二值时指示对应单位时刻的连接状态为无连接，所述第一值和所述第二值不同。在各个样本序列中通过不 同元素值指示连接状态是一种简单高效的连接状态表示方法。
可选地，在一种可能的实现方式中，分析设备通过以下步骤训练出预测模型。分析设备获得N个训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态。针对所述N个训练终端对中的第一训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，生成第一训练终端对对应的训练样本序列，以此类推，从而得到N个训练样本序列，所述第一训练终端对对应的训练样本序列中包括M+Q个元素，所述M+Q个元素中的每个元素的取值分别对应所述第一训练终端对在所述M+Q个连续的单位时刻中每个单位时刻分别对应的连接状态。将所述N个训练样本序列作为机器学习算法的输入，获得所述机器学习算法输出的所述预测模型。分析设备首先获取反映训练终端对历史上连接状态趋势的训练样本序列，再利用机器学习算法对大量训练样本序列进行训练从而生成预测模型。从而提供了一种有效的预测模型的学习方法。预测模型本身反映同一网络场景下多个终端对连接状态的普遍地、动态性的趋势信息或规律信息。
可选地，分析设备通过以下步骤所述获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态。所述分析设备从保存的多个数据流分别对应的条目中选择第一组目标条目，所述第一组目标条目包括记录的单位时刻属于所述第一历史时间段、源IP地址为所述第一终端的IP地址、且目的IP地址为所述第二终端的条目，以及记录的单位时刻属于所述第一历史时间段、目的IP地址为所述第一终端的IP地址、且源IP地址为所述第二终端的条目。所述分析设备确定所述选择出的第一组目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第一历史时间段内、除所述选择出的第一组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态。通过上述数据处理方法，分析设备获得了以两个相邻两个单位时刻之间的时间段为粒度的测试终端对历史连接状态信息，以便于后续基于测试终端对历史连接状态信息对测试终端对未来连接状态进行预测。
可选地，在一种可能的实现方式中，分析设备通过以下步骤获得N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态。所述分析设备获取N个训练终端对。所述分析设备从所述N个训练终端对中选择一个训练终端对，对选择出的训练终端对执行以下处理步骤，直到处理完全部所述N个训练终端对为止，所述选择出的训练终端对由第三终端和第四终端构成：所述分析设备从保存的多个数据流分别对应的条目中选择第二组目标条目，所述第二组目标条目包括记录的单位时刻属于所述第二历史时间段、源IP地址为第三终端的IP地址、且目的IP地址为第四终端的条目，以及记录的单位时刻属于所述第二历史时间段、目的IP地址为所述第四终端的IP地址、且源IP地址为所述第三终端的条目；所述分析设备确定所述选择出的第二目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第二历史时间段内、除所述选择出的第二组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述选择出的训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态。通过上述数据处理方法，分析设备获得了以两个相邻两个单位时刻之间的时间段为粒度的训练终端对历史连接状态信息，以便于后续基于训练终端对历史连接状态信息训练预测模型。
可选地，在一种可能的实现方式中，在Q＝1的情况下，所述N个训练样本序列中正样本的数量与负样本的数量之间的比值大于等于0.5、且小于等于2，所述正样本是最后一个 元素的取值指示的连接状态为有连接的训练样本序列，所述负样本中是最后一个元素的取值指示的连接状态为无连接的训练样本序列。满足上述条件的训练样本序列被视为平衡样本集。分析设备基于平衡样本集训练预测模型，能够获得预测效果更佳的预测模型。
可选地，在一种可能的实现方式中，分析设备通过以下步骤从流统计信息中获得若干条目。条目可以被视为用于获取测试终端对或训练终端对连接状态的原始数据。分析设备获取多条流统计信息，所述多条流统计信息中的每条流统计信息分别对应一个数据流，所述每条流统计信息中包括数据流的建立时间、关闭时间、源IP地址和目的IP地址。所述分析设备根据预定的时间对齐规则，对所述每条流统计信息进行以所述单位时刻为基准的时间对齐处理，生成多个数据流分别对应的条目并保存所述多个数据流分别对应的条目，所述多个数据流分别对应的条目中每个数据流对应的条目记录有单位时刻、源IP地址和目的IP地址。通过上述方式生成的条目数据，仅保留了流统计信息中与连接状态相关的信息，与流统计信息相比降低了数据量，节约了存储空间。另一方面，在生成条目数据的过程中进行了时间对齐处理，有利于后续提高处理效率。
可选地，第一方面或上述第一方面的任意一种可能的实现方式中所述第一终端、所述第二终端、所述第三终端和所述第四终端分别为虚拟机。进一步地，所述虚拟机部署于由DCN连接的数据中心中。本申请实施例提供的预测方法，适用于预测DC中两个虚拟机之间的连接状态。
第二方面，提供了一种终端之间连接状态的预测装置，该装置具有实现上述第一方面所述方法或上述方面的任意一种可能的实现方式的功能。所述功能可以通过硬件实现，也可以通过硬件执行相应的软件实现。所述硬件或软件包括一个或多个与上述功能相对应的模块。
第三方面，提供了一种分析设备。该分析设备包括存储器和至少一个处理器。所述用于存储指令，所述指令被所述至少一个处理器读取后，所述分析设备执行上述第一方面或第一方面的任意一种可能的实现方式中的方法，具体参见上面的详细描述，此处不再赘述。
第三方面，本申请实施例提供了一种计算机存储介质，用于储存为上述分析设备所用的计算机软件指令，其包含用于执行上述第一方面或上述方面的任意一种可能的实现方式所设计的程序。
第四方面，提供了一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上述第一方面或第一方面的任意一种可能的实现方式中所述的方法。
第五方面，本申请实施例提供了一种芯片，包括存储器和处理器，存储器用于存储计算机指令，处理器用于从存储器中调用并运行该计算机指令，以执行上述第一方面及其第一方面任意可能的实现方式中的方法。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作一简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例应用场景的示意图；
图2为本申请实施例提供的一种终端之间连接状态的预测方法的流程图；
图3为本申请实施例提供的一种基于预测模型的、终端之间连接状态的预测方法的流程图；
图4A为本申请实施例提供的第一训练虚拟机对对应的训练样本序列的示意图；
图4B为本申请实施例提供的N个训练样本序列的示意图；
图4C为本申请实施例提供的测试虚拟机对对应的第一样本序列的示意图；
图5为本申请实施例中分析设备将N个训练样本序列输入MLP，从而得到预测模型的过程示意图；
图6为本申请实施例提供的分析设备的结构示意图；
图7为本申请实施例提供的终端之间连接状态的预测装置的结构示意图。
具体实施方式
由于相关技术中几种预测方法准确性不佳，本发明实施例提出一种终端之间连接状态的预测方法。该方法基于终端对(在本实施例中，将由两台终端组成的一对终端称为“终端对”)在历史时间段中多个单位时刻的连接状态，通过人工智能技术来提取有用信息，例如根据一个终端对长期的历史连接状态信息构建这个终端对连接状态的数学模型、或者根据若干终端对长期的历史连接状态信息构建预测模型。在需要对一个终端对的连接状态进行预测时，根据上述有用信息获得这对终端在在未来时间段内至少一个单位时刻分别对应的连接状态。这种方法充分利用了终端对较长期的历史连接状态信息来进行预测，有助于提高预测准确性。
本申请实施例提供的预测方法适用于多种网络场景，例如公司、政府部门或者学校内部的局域网，以及DCN。根据具体应用场景的不同，可选地，组成终端对的终端是个人计算机、笔记本电脑、移动终端、可穿戴设备或者虚拟机。
可选地，构成终端对的两个终端是同类设备，例如均是个人计算机、或者均是虚拟机。可替换地，构成终端对的两个终端是不同类设备，例如终端对中的一个终端是个人计算机而另一个终端是虚拟机，或者终端对中的一个终端是移动终端而另一个终端是虚拟机。
在后面的实施例中，主要以DCN场景为例对本申请实施例提供的预测方法进行描述。DCN场景的特点是提供计算资源的是大量虚拟机。虚拟机是通过虚拟化技术模拟出的具有完整软硬件系统功能的逻辑计算机设备。宿主机是虚拟化技术实施基础，即为虚拟化技术提供实际硬件资源的计算机设备。例如当虚拟化技术通过虚拟化软件实现时，在宿主机上安装虚拟化软件后，可以基于宿主机的硬件资源，按照配置生成一个或多个虚拟机。因此宿主机也可以被视为虚拟机运行的硬件平台。在DCN场景中终端对是指虚拟机对，即由两个虚拟机组成的一对虚拟机。由于在不同场景中该预测方法的实现原理基本类似，因此不进行一一举例说明。
下面结合各个附图对本发明实施例技术方案的主要实现原理、具体实施方式及其对应能够达到的有益效果进行详细的阐述。
附图1是本申请实施例在DCN场景下的应用场景的示意图。DCN中包括多个宿主机(host)，记为host 1、host 2、host 3、host 4、和host 5。不同宿主机上运行有一个 或多个不同虚拟机。例如host 1上运行VM 1a和VM 1b，host 2上运行VM 2a，host 3上运行VM 3a、VM 3b和VM 3c，host 4上运行VM 4a和VM 4b，host 5上运行VM 5a和VM 5b。
附图1所示的场景中还包括多个报文转发设备。报文转发设备包括各种交换机，如二层交换机或者三层交换机等等。二层交换机工作在数据链路层，可以识别数据包中的MAC地址信息，根据识别出的MAC地址查找包含MAC地址与端口号对应关系的地址表实现转发。三层交换机工作在网络层，通过三层交换技术实现转发。三层交换技术是将路由技术与交换技术合二为一的技术。三层交换机在对第一个数据流进行路由后，三层交换机会产生一个MAC地址与IP地址的映射表。当同样的数据流再次通过三层交换机时，三层交换机根据上述映射表之间从二层进行转发而不是再次路由。例如附图1中的交换机S 1、S 2、S 3和S 4为二层交换机，S 5和S 6为三层交换机。
如附图1中的虚线所示，宿主机与报文转发设备之间存在物理连接，例如通过以太网连接。通过报文转发设备与作为虚拟机运行平台的宿主机之间的物理连接，虚拟机之间进行相互通信。
附图1所示的应用场景中还包括分析设备。分析设备与数据源设备进行通信，从而获得多条流统计信息。可选地，数据源设备包括报文转发设备和宿主机。每条流统计信息分别对应一个数据流，所述每条流统计信息中包括数据流的建立时间、源IP地址和目的IP地址。在本申请实施例中，数据流是指从一个源计算机到一个目的方的一系列报文。目的方可以是另一个计算机，也可以是一组计算机或者广播域。
可选地，数据源设备将通过本设备的网络接口传输的流量镜像后，将镜像的流量发送给分析设备，分析设备对镜像的流量进行简单解析后得到流统计信息。简单的解析包括从全部流量中筛选出同步序列编号(Synchronize Sequence Numbers，SYN)报文以及从SYN报文中提取源IP地址和目的IP地址，根据SYN报文的发送时间、以及提取的源IP地址和目的IP地址生成流统计信息。这种方式不会过多消耗数据源设备的处理资源，对数据源设备硬件要求较低，适用于数据源设备是交换机或宿主机的情况。
可替换地，数据源设备对通过本设备的网络接口传输的报文进行简单解析后得到流统计信息，将流统计信息发送给分析设备。与数据源设备直接发送镜像的报文相比，由于流统计信息的数据量小于镜像的报文的数据量，可以节省网络传输资源。由于这种方式对数据源设备的处理能力有一定要求，因此相对适用于数据源设备是宿主机的情况。
表1是分析设备接收到的流统计信息的一种示例，其中每行表示一条流统计信息。可选地，不同数据源可以采用不同的格式、编码方式记录数据流的建立时间、源IP地址和目的IP地址。例如，用二进制、十进制或者十六进制记录地址信息。分析设备首先对接收到的原始流统计信息进行格式转换，归一化为统一格式的流统计信息。可以理解地，流统计信息中的IP地址是为了区分不同虚拟机，为了便于理解和描述，在本实施例中用虚拟机标识来替代IP地址。
表1
时间 源IP地址 目的IP地址
2015-1-10 11:23:00 VM 1a VM 1b
2015-1-10 11:45:00 VM 1a VM 1b
2015-1-10 11:55:00 VM 1a VM 1b
2015-1-10 11:02:00 VM 2a VM 3a
…… …… ……
分析设备对根据预定的时间对齐规则，对所述每条流统计信息进行以所述单位时刻为基准的时间对齐处理，生成多个数据流分别对应的条目并保存所述多个数据流分别对应的条目。
可选地，分析设备对收到的多条流统计信息进行时间对齐处理时，采用的时间粒度可以由管理员根据分析设备的存储空间、处理资源、DCN的网络规模，分析目的等各种因素设置。通过时间对齐处理不仅可以降低数据量以节省存储空间，还有助于提升后续分析效率。
可选地，预定的时间对齐规则可以灵活设定。在进行时间对齐处理时的粒度可以根据需求设置，如1小时、半小时、10分钟、1分钟等等。假定在本申请实施例中，分析设备对收到的多条流统计信息进行时间对齐处理时，采用的时间粒度为1小时。换句话说，对齐处理后得到的条目中是以1小时为单位时间。例如，一种时间对齐规则是将时间处于两个单位时刻之间的时间处理为两个单位时刻中靠前的单位时刻，举例来说，将“2015-1-10 11:23:00”处理为“2015-1-10 11:00:00”。另一种时间对齐规则是将时间处于两个单位时刻之间的时间处理为两个单位时刻中较为接近的一个单位时刻，举例来说，将“2015-1-10 11:55:00”处理为“2015-1-10 12:00:00”。
经时间对齐处理，分析设备获得表2所示的条目，并保存这些条目以备后续使用。
表2
以小时为粒度的单位时刻标识 源IP地址 目的IP地址
2015-1-10 11:00:00 VM 1a VM 1b
2015-1-10 12:00:00 VM 1a VM 1b
2015-1-10 13:00:00 VM 2a VM 3a
2015-1-10 9:00:00 VM 1a VM 2a
…… …… ……
进一步地，分析设备基于保存的多个数据流分别对应的条目，通过人工智能技术，分析出虚拟机对的连接状态的趋势或规律信息，或者构建预测模型。后面将结合各个实施例描述本申请实施例提供的一种终端之间连接状态的预测方法。人工智能技术是指使得由人制造出来的机器表现出类似于人类智能的技术。从现有研究来看，人工智能技术包括机器学习算法。
图2是本申请实施例提供的一种终端之间连接状态的预测方法的流程图。图2主要从分析设备的角度对该方法进行描述。可选地，图2中的分析设备为附图1中的分析设备。
步骤21，分析设备获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态。所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数。
例如，当前时间为2015-1-11 9:20，分析设备当前的预测任务是在预测附图1所示的场景中，预测虚拟机VM 1a与VM 2a之间2015-1-11 10:00的连接状态。管理员可以通过分析设备的输入输出接口连接的输入设备输入预测任务。即在本实施例中测试虚拟机对为 (VM 1a-VM 2a)。
假定第一历史时间段为当前时间之前3小时，即M＝3。那么第一历史时间段为2015-1-11 6:20至2015-1-11 9:20。第一历史时间段内包含3个单位时刻，即2015-1-11 7:00，2015-1-11 8:00和2015-1-11 9:00。
为完成上述预测任务，分析设备首先获得测试虚拟机对(VM 1a-VM 2a)在2015-1-11 7:00，2015-1-11 8:00和2015-1-11 10:00分别对应的连接状态。
可选地，分析设备采用步骤21a和步骤21b获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态。
步骤21a，分析设备从保存的多个数据流分别对应的条目中选择第一组目标条目，所述第一组目标条目包括记录的单位时刻属于所述第一历史时间段、源IP地址为所述第一终端的IP地址、且目的IP地址为所述第二终端的条目，以及记录的单位时刻属于所述第一历史时间段、目的IP地址为所述第一终端的IP地址、且源IP地址为所述第二终端的条目。
步骤21b，分析设备确定所述选择出的第一组目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第一历史时间段内、除所述选择出的第一组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态。
回到上面的实例，分析设备在表2所示的条目中选择满足以下两个条件中任一条件的条目组成第一组目标条目。
条件1：单位时刻为2015-1-11 7:00，2015-1-11 8:00或2015-1-11 10:00三者之一，且源IP地址为VM 1a，目的IP地址为VM 2a。
条件2：单位时刻为2015-1-11 7:00，2015-1-11 8:00或2015-1-11 10:00三者之一，且源IP地址为VM 2a，目的IP地址为VM 1a。
假定分析设备在表2所示的条目中筛选出第一组目标条目如表3所示。
表3
以小时为粒度的单位时刻标识 源IP地址 目的IP地址
2015-1-10 8:00:00 VM 1a VM 2a
2015-1-10 9:00:00 VM 2a VM 1a
由于表3所示的选择出的第一组目标条目中包括单位时刻标识2015-1-10 8:00:00和2015-1-10 9:00:00，不包含单位时刻2015-1-10 7:00:00。因此分析设备确定测试虚拟机对(VM 1a-VM 2a)在2015-1-10 8:00:00的连接状态为有连接，在2015-1-10 9:00:00的连接状态为有连接，在2015-1-10 7:00:00的连接状态为无连接。
步骤22，分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻对应的连接状态。所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
根据本申请实施例提供的终端之间连接状态的预测方法，在以DCN为例的应用场景中，分析设备首先从数据源设备中获取作为测试对象的测试虚拟机对历史上多个单位时刻的连 接状态信息。分析设备基于测试虚拟机对历史上多个单位时刻的连接状态信息为测试依据，预测测试虚拟机对未来时间段的连接状态。由于预测过程中使用了测试虚拟机对历史上多个单位时刻的连接状态信息，而不是测试虚拟机对历史上单一单位时刻的连接状态信息，有助于从历史状态信息中分析发现更多有用信息，例如更详细具体的规律或趋势信息，从而提升预测准确性。
可选地，分析设备采用本申请实施例提供的预测方法得到预测结果(即测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态)后，分析设备将该预测结果应用于不同的场景中，例如故障影响分析和配置验证场景中，能够进一步提高故障影响分析的准确性，以及配置验证的准确性。
在“基于测试终端对历史上多个单位时刻的连接状态信息为测试依据，预测测试终端对未来时间段的连接状态”这一整体思路下，在实施过程中有多种可能的实现方案。这些实现方案包括但不限于根据包含多个单位时刻的连接状态的历史连接状态信息构建数学模型和预测模型的方法。本申请后面的各实施例以数学模型或预测模型为例，对本申请实施例提供的预测方法进行描述。
一、基于数学模型，对测试终端对的连接状态进行预测
分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对连接状态的数学模型。进一步分析设备根据数学模型来预测所述测试终端对未来时间段中至少一个单位时刻对应的连接状态。
可选地，分析设备预先存储多种数学模型的匹配规则。分析设备将测试虚拟机对(VM 1a-VM 2a)在第一历史时间段内多个单位时刻分别对应的连接状态与匹配规则逐一进行匹配，从而确定测试虚拟机对(VM 1a-VM 2a)历史连接状态信息符合的数学模型。当然，分析设备也可以采用其他机制学习到测试虚拟机对(VM 1a-VM 2a)历史连接状态信息符合的数学模型。
下面两个例子(例1和例2)作为示例性说明。显然类似的数学模型还有很多，在这里难以一一列举。
例1
分析设备根据测试虚拟机对(VM 1a-VM 2a)在过去24小时中每一小时对应的连接状态，确定出测试虚拟机对(VM 1a-VM 2a)连接状态的数学模型为“连接状态为有连接持续2小时后，转换为无连接持续3小时，再转换为有连接持续2小时，依次往复”。如表4所示。为了简明起见，连接状态在表4中用0或1的数值指示，0表示无连接，1表示有连接。
表4
Figure PCTCN2020114979-appb-000001
Figure PCTCN2020114979-appb-000002
分析设备根据上述数学模型，确定测试虚拟机对(VM 1a-VM 2a)在未来时间段2015-1-11 0:00:00-24:00:00的连接状态如表5所示。
表5
Figure PCTCN2020114979-appb-000003
在本实例中，M＝24，Q＝24。
例2
分析设备根据测试虚拟机对(VM 1a-VM 2a)在过去24小时中每一小时对应的连接状态，确定出测试虚拟机对(VM 1a-VM 2a)连接状态的数学模型为“连接状态为有连接持续n小时后，转换为无连接持续n小时，再转换为有连接，n从1开始每次递加1”。如表6所示。为了简明起见，连接状态在表6中用0或1的数值指示，0表示无连接，1表示有连接。
表6
Figure PCTCN2020114979-appb-000004
分析设备根据上述数学模型，确定测试虚拟机对(VM 1a-VM 2a)在未来时间段2015-1-11 0:00:00-11:00:00的连接状态如表7所示。
表7
Figure PCTCN2020114979-appb-000005
Figure PCTCN2020114979-appb-000006
在本实施例中，M＝24，Q＝12。
根据本申请实施例提供的终端之间连接状态的预测方法，在以DCN为例的应用场景中，分析设备首先从数据源设备中获取作为测试对象的测试虚拟机对历史上多个连续的单位时刻的连接状态信息。分析设备以测试虚拟机对历史上多个连续的单位时刻的连接状态信息为测试依据，预测测试虚拟机对未来时间段的连接状态。由于预测的基础是测试虚拟机对历史上多个连续的单位时刻的连接状态信息，而不是测试虚拟机对历史上单一单位时刻的连接状态信息，有助于分析发现测试虚拟机对连接状态的长期历史趋势，从而提升预测准确性。
二、基于预测模型，对测试虚拟机对的连接状态进行预测
分析设备根据预测模型来确定测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态。预测模型是分析设备根据作为预测目标的测试终端对所处的网络场景中的大量终端对的历史连接状态信息通过机器学习算法训练出的，其中用于训练预测模型的大量终端对与作为预测目标的测试终端对处于同一网络场景中。用于进行预测模型训练的终端对在本实施例中被称为训练终端对。可选地，训练终端对可以包含预测终端对，也可以不包含预测终端对，在这里不进行限定。
具体地，分析设备根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的预测模型。换句话说，预测模型是根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的。所述第二历史时间段是当前时间之前的时间段、且所述第二历史时间段中包括M+Q个连续的单位时刻，其中N为大于等于1的自然数，通常当N取值为百万级的时候就可以达到较为满意的效果。在合理的取值范围内，N的数量越大，预测结果越准确。然后，分析设备将所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，输入预测模型并获取所述预测模型的输出结果。
可以理解的是，预测模型是根据预测需求训练出的。预测需求是指M和Q，即“根据测试终端对包含M个单位时刻的第一历史时间段内的连接状态，预测测试终端对包含Q个单位时刻的未来时间段的连接状态”。可选地，管理员可以通过分析设备的输入接口连接的输入设备输入预测需求。
预测模型的输入是测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态。输出是测试终端对在未来时间段内至少一个单位分别对应的连接状态。本申请将结合后续实施例对预测模型的生成过程进行详细说明。
仍以附图1所示场景为例，假定第一历史时间段是当前时间之前的3小时，即M＝3。未来时间段是当前时间之后1小时，即Q＝1。那么未来时间段为2015-1-11 9:20至2015-1-11 10:20。未来时间段内包含1个单位时刻，即2015-1-11 10:00。
分析设备将步骤21确定出的测试虚拟机对(VM 1a-VM 2a)在当前时间之前3小时内各单位时刻的连接状态输入预测模型，即将“测试虚拟机对(VM 1a-VM 2a)在2015-1-10 7:00:00的连接状态为无连接，在2015-1-10 8:00:00的连接状态为有连接，在2015-1-10  9:00:00的连接状态为有连接”这一信息输入预测模型。预测模型输出2015-1-11 10:00的连接状态为有连接。
分析设备根据预测模型的输出，确定测试虚拟机对在尚未到来的2015-1-10 10:00:00的连接状态为有链接。
根据本申请实施例提供的终端之间连接状态的预测方法，在以DCN为例的应用场景中，分析设备获取大量训练虚拟机对的历史连接状态信息，历史连接状态信息包含多个单位时刻分别对应的连接状态。分析设备进一步结合预测需求，根据大量训练虚拟机对的历史连接状态信息生成预测模型。在进行预测时，针对作为预测对象的测试虚拟机对，分析设备将测试虚拟机对在历史时间段内多个单位时刻分别对应的连接状态输入预测模型，并根据预测模型的输出确定测试虚拟机对在未来时间段内至少一个单位时刻分别对应的连接状态。本申请实施例中大量历史连接状态信息一方面包含大量虚拟机对的历史连接状态信息，另一方面历史连接状态信息包含至少两个单位时刻对应的连接状态信息。与将测试虚拟机前一天同一时刻的连接状态作为这对虚拟机未来一天同一时刻的连接状态的预测方案、以及将测试虚拟机上一时刻的连接状态预测这对虚拟机未来下一个时刻的连接状态的预测方案相比，本申请实施例的预测方法基于大量历史连接状态信息来进行预测，降低了偶然因素带来的误差，提升了预测准确性。
附图3是本申请实施例提供的一种基于预测模型的、终端之间连接状态的预测方法的流程图。可选地，图3中的分析设备为附图1或附图2中的分析设备。附图3所示的流程中步骤31至步骤33组成的子流程主要描述了分析设备生成预测模型的过程，步骤34至步骤36组成的子流程主要描述了分析设备基于预测模型，对测试终端对的连接状态进行预测的过程。显然，分析设备在生成预测模型之后，可以基于预测模型对多个测试终端对的连接状态进行预测，而无需每次预测时都根据训练终端对的历史连接状态信息重新生成预测模型。
步骤31，分析设备获得N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态。关于第二历史时间段的定义请参考前面的介绍，在这里不再重复。
可选地，步骤31中包括步骤311-步骤314几个子步骤。
步骤311，分析设备获取N个训练终端对。
可选地，分析设备采用多种方法获取训练终端对。例如，分析设备读取保存的如表2所示的条目，根据条目中的源IP地址和目的IP地址获取训练终端对。可替换地，分析设备也可以通过地址管理设备(例如，动态主机配置协议(Dynamic Host Configuration Protocol，DHCP)服务器)获取网络中已分配给终端使用的IP地址，然后通过排列组合生成若干终端对。分析设备再从排列组合生成的若干终端对中选择出N个训练终端对。选择的方式包括随机选取，按预定顺序选取等等，在这里不再详述。
步骤312，分析设备从所述N个训练终端对中选择一个训练终端对，对选择出的训练终端对执行处理步骤312a，和步骤312b，直到处理完全部所述N个训练终端对为止，所述选择出的训练终端对由第三终端和第四终端构成。
步骤312a，分析设备从保存的多个数据流分别对应的条目中选择第二组目标条目，所述第二组目标条目包括记录的单位时刻属于所述第二历史时间段、源IP地址为第三终端的IP地址、且目的IP地址为第四终端的条目，以及记录的单位时刻属于所述第二历史时间段、 目的IP地址为所述第四终端的IP地址、且源IP地址为所述第三终端的条目。
步骤312b，分析设备确定所述选择出的第二目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第二历史时间段内、除所述选择出的第二组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述选择出的训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态。
步骤312a和步骤312b分别与附图2中步骤21a和步骤21b类似。在这里不再重复描述。
由于篇幅所限，在本实施例中给出一个较为简单实例进行说明。假定预测需求是“根据测试虚拟机对包含24*6个单位时刻的第一历史时间段内的连接状态，预测测试虚拟机对未来时间段的连接状态，未来时间段包含1个单位时刻”，即M＝24*6，Q＝1。形象的说，预测需求是根据过去6天每一小时的连接状态，预测未来1小时的连接状态。
分析设备对于选择出的训练虚拟机对，采用步骤312a和步骤312b得到的过去M+Q(24*6+1＝145)小时的连接状态如表8所示。为了简明起见，在如表8所示的连接状态表中用0或1的数值指示不同的连接状态，0表示无连接，1表示有连接。
表8
单位时刻标识 1 2 3 4 5 6 7 …… 145
连接状态 1 1 0 0 0 0 1   1
表8即为选择出的训练虚拟机对在所述第二历史时间段内每一单位时刻分别对应的连接状态的示例。分析设备对N个训练虚拟机对均执行步骤312a，和步骤312b，得到了N个如表8所示的状态信息表。
步骤32，分析设备针对所述N个训练终端对中的第一训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，生成第一训练终端对对应的训练样本序列，以此类推，从而得到N个训练样本序列，所述第一训练终端对对应的训练样本序列中包括M+Q个元素，所述M+Q个元素中的每个元素的取值分别对应所述第一训练终端对在所述M+Q个连续的单位时刻中每个单位时刻分别对应的连接状态。
继续上面的例子，针对N个训练虚拟机对中的第一训练虚拟机对，分析设备基于第一训练虚拟机对所对应的类似于表8所示的状态信息表，生成第一虚拟机对对应的训练样本序列。训练样本序列中包含M+Q(145)个元素，如图4A所示。假定元素的取值为0或1，0表示无连接，1表示有连接。分析设备将样本序列中的前M(24*6＝144)个元素作为训练样本序列中的样本部分，如图4A中41。分析设备将最后1个元素作为训练样本序列中的标签，如图4A中42。
分析设备针对N个训练虚拟机对中的每个训练虚拟机对均执行上述步骤32，从而生成N个个训练样本序列，如图4B所示。
步骤33，分析设备将所述N个训练样本序列作为训练样本输入机器学习算法，获得所述机器学习算法生成的所述预测模型。
可选地，机器学习算法包括但不限于神经网络、决策树、随机森林、支持向量机等等。由于机器学习算法众多，难以一一描述采用每种机器学习算法基于N个训练样本序列生成预测模型的过程。本申请实施例以应用其中一种机器学习算法生成预测模型为例进行举例 说明。
在本实施例以多层感知神经网络(Multi Layer Perceptron，MLP)为例对预测模型的生成过程进行详细描述。神经网络的基础计算单元是节点(node)，节点也被称为神经元(neuron)。节点接收来自外部输入的输入，经过计算激活函数(activation function)后产生输出。权重(weight)表示输出节点与接收节点之间联系的强弱，权重值的大小在神经网络的训练过程中会自动调整直到趋于稳定，权重值是训练的主要对象。激活函数记为f()，一般是非线性的，主要作用是为神经元的输出加入非线性特性，增强神经网络对训练样本的学习能力。
图5是将N个训练样本序列输入MLP，从而得到预测模型的过程示意图。图5中的MLP包含输入层(input layer)，输出层(output layer)。可选地，为了达到更佳的学习效果，MLP还包括一个或多个隐含层(hidden layer)。为简明起见，本实施例以MLP中包含2个隐含层为例进行说明。每个隐含层中包含的节点数目是可设定的，例如第一个隐含层包含64个节点，第二个隐含层包含16个节点。
MLP输入层中包含的节点数目与训练样本序列的样本部分中包含的元素数目相同，输出节点数目与训练样本序列的标签中包含的元素数目相同。由于本实施例中训练样本序列的样本部分包含的元素数目为144，因此MLP输入层包含的节点数目为144；本实施例中训练样本序列的标签包含的元素数目为1，因此MLP输出层包含的节点数目为1。
分析设备将一个训练样本序列输入MLP时，将训练样本序列样本部分的各元素分别输入MLP输入层对应的节点。分析设备将输出层节点的值与训练样本序列的标签的元素值进行对比，如果输出层节点的值与训练样本序列的标签的元素值差异较大，则MLP自动通过f()调整权重值。预测模型的学习过程是MLP接收分析设备输入的N个训练样本，并根据输出层节点的值与训练样本序列的标签的元素值之间差异调整权重值的过程。当MLP中的权重值自动调整到一个稳定的理想状态时，学习过程结束，此时如图5所示结构的MLP以及理想状态的权重值即为预测模型。
可选地，为了得到预测效果更佳的预测模型，分析设备输入机器学习算法用以生成预测模型的N个训练样本序列是平衡样本集。平衡样本集是指用以训练生成预测模型的N个训练样本序列中正样本的数量与负样本的数量大致相同、相差不大。换句话说N个训练样本序列中正样本的数量与负样本的数量之间的比值在合理范围内。其中正样本是最后一个元素的取值指示的连接状态为有连接的训练样本序列，所述负样本中是最后一个元素的取值指示的连接状态为无连接的训练样本序列。可选地，一种可实施的合理范围为0.5至2中的一个值。
步骤34，分析设备根据测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定第一样本序列，所述第一样本序列中包括M个元素，所述M个元素中的每个元素的取值分别对应所述M个连续的单位时刻中每个单位时刻分别对应的连接状态。
分析设备确定第一样本序列的方法与本流程步骤32中生成训练样本序列的方法基本类似，在这里不再展开描述，生成的第一样本序列如图4C所示。第一样本序列包含M(144)个元素。
步骤35，分析设备将所述第一样本序列输入预测模型并获取所述预测模型的输出结果。
在本实施例中预测模型的输出结果为预测序列，所述预测序列包括Q个元素，所述Q 个元素中的每个元素的取值分别对应所述Q个连续的单位时刻中每个单位时刻分别对应的连接状态。
例如，分析设备将图4C所示的第一样本序列输入预测模型后，预测模型输出的预测序列为“[1]”。由于在本实施例中是以Q＝1为例进行介绍的，因此预测序列包含1个元素。当Q取值为其他大于1的自然数时，预测序列包含更多元素。例如当Q＝3时，预测序列的形式为“[1,0,1]”。
步骤36，分析设备根据预测模型的输出结果确定所述测试终端对在所述未来时间段内至少一个单位时刻对应的连接状态。
可理解地，在预测模型输出的预测序列为“[1]”的情况下，分析设备确定测试虚拟机对(VM 1a-VM 2a)在所述未来1小时对应的连接状态为有连接。
本申请实施例提供了生成预测模型、以及基于预测模型对终端之间连接状态进行预测的详细过程。附图3步骤31至步骤33组成的子流程介绍了如何根据训练终端对的历史连接状态信息生成预测模型。在以DCN为例的应用场景中，分析设备首先从数据源设备中获取作为大量训练虚拟机对历史上多个连续的单位时刻的连接状态信息，在基于获取的连接状态信息利用机器学习算法生成预测模型。与前一实施例中的数学模型相比，预测模型反映的趋势信息或规律信息更具有普遍性，降低偶然因素带来的误差，能够进一步提升预测准确性。附图3步骤34至步骤36组成的子流程主要描述了分析设备基于预测模型，对测试终端对的连接状态进行预测的过程。基于预测模型，DCN中的大量历史连接状态信息被充分利用，用来预测连接状态。经过实际数据测试，本申请实施例提供的预测方法准确率可以达到98％左右，与现有相关技术相比，预测准确率有明显提升。
相应地，本申请实施例还提供了一种分析设备，用以实施上述实施例描述的预测方法。图6是本申请实施例提供的分析设备的结构示意图。可选地，图6所示的分析设备是图1所示应用场景中的分析设备、图2或图3所示流程中的分析设备。分析设备包括至少一个处理器61、和存储器62。
至少一个处理器61可以是一个或多个CPU，该CPU可以是单核CPU，也可以是多核CPU。
存储器62包括但不限于是随机存取存储器(random access memory，RAM)、只读存储器(Read only Memory，ROM)、可擦除可编程只读存储器(erasable programmable read-only memory，EPROM或者快闪存储器)、快闪存储器、或光存储器等。存储器62中保存有操作系统的代码。
可选地，处理器61通过读取存储器62中保存的指令实现上述实施例中的方法，或者，处理器61也可以通过内部存储的指令实现上述实施例中的方法。在处理器61通过读取存储器62中保存的指令实现上述实施例中的方法的情况下，存储器62中保存实现本申请上述实施例提供的方法的指令。
存储器62中存储的程序代码被所述至少一个处理器61读取后，分析设备执行以下操作：获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数；根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态，所述未来时间段是当前时间之 后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
可选地，附图6所示的分析设备还包括网络接口63。网络接口63可以是有线接口，例如光纤分布式数据接口(Fiber Distributed Data Interface，FDDI)、千兆以太网(Gigabit Ethernet，GE)接口；网络接口63也可以是无线接口。网络接口63用于接收来自于数据源的镜像的流量、或者多条流统计信息。
存储器62用于存储网络接口63接收到的镜像的流量、或者多条流统计信息。所述至少一个处理器61用于对所述镜像的流量、或者多条流统计信息进行处理后，获得上述表2所示的若干条目并将这些条目保存至存储器62。
所述至少一个处理器61进一步根据存储器62保存的这些条目来执行上述方法实施例所描述的预测方法。处理器61实现上述功能的更多细节请参考前面各个方法实施例中的描述，在这里不再重复。
可选地，分析设备还包括总线64，上述处理器61、存储器62通常通过总线64相互连接，也可以采用其他方式相互连接。
可选地，分析设备还包括输入输出接口65，输入输出接口65用于与输入设备连接，接收用户通过输入设备输入的预测需求。输入设备包括但不限于键盘、触摸屏、麦克风等等。输入输出接口65还用于与输出设备连接，输出处理器61的预测结果。输出设备包括但不限于显示器、打印机等等。
本申请实施例提供的分析设备用于执行上述各个方法实施例提供的预测方法。该分析设备由于预测过程中使用了测试虚拟机对历史上多个单位时刻的连接状态信息，而不是测试虚拟机对历史上单一单位时刻的连接状态信息，有助于从历史状态信息中分析发现更多有用信息，从而提升预测准确性。
图7是本申请实施例提供的一种终端之间连接状态的预测装置的结构示意图。该终端之间连接状态的预测装置70包括获取模块71、预测模块72。
获取模块71，用于获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数。
预测模块72，用于根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态，所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
可选地，预测模块72包括模型测试单元721和确定单元722。
模型测试单元721，用于将所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，输入预测模型并获取所述预测模型的输出结果，所述预测模型是根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的，所述第二历史时间段是当前时间之前的时间段、且所述第二历史时间段中包括M+Q个连续的单位时刻， 其中N为大于等于1的自然数；
确定单元722，用于根据所述输出结果确定所述测试终端对在所述未来时间段内至少一个单位时刻分别对应的连接状态。
可选地，模型测试单元721，用于根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定第一样本序列，所述第一样本序列中包括M个元素，所述M个元素中的每个元素的取值分别对应所述M个连续的单位时刻中每个单位时刻分别对应的连接状态；以及将所述第一样本序列输入预测模型并获取所述预测模型的输出结果，所述输出结果为预测序列，所述预测序列包括Q个元素，所述Q个元素中的每个元素的取值分别对应所述Q个连续的单位时刻中每个单位时刻分别对应的连接状态。
可选地，图7中的预测模块72还包括模型学习单元723，用于在所述模型测试单元721将所述第一样本序列输入预测模型之前，执行以下步骤：获得N个训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态；针对所述N个训练终端对中的第一训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，生成第一训练终端对对应的训练样本序列，以此类推，从而得到N个训练样本序列，所述第一训练终端对对应的训练样本序列中包括M+Q个元素，所述M+Q个元素中的每个元素的取值分别对应所述第一训练终端对在所述M+Q个连续的单位时刻中每个单位时刻分别对应的连接状态；将所述N个训练样本序列作为机器学习算法的输入，获得所述机器学习算法输出的所述预测模型。
附图7所描述的装置实施例仅仅是示意性的，例如，所述模块的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个模块或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。在本申请各个实施例中的各功能模块可以集成在一个处理模块中，也可以是各个模块单独物理存在，也可以两个或两个以上模块集成在一个模块中。附图7中上述各个模块既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。例如，采用软件实现时，上述获取模块71、预测模块72、以及模型测试单元721、确定单元722和模型学习单元723可以是由附图6中的至少一个处理器61读取存储器中存储的程序代码后，生成的软件功能模块来实现。图7中上述各个模块也可以由分析设备中的不同硬件分别实现，例如获取模块71由附图6中的网络接口63和至少一个处理器63中的一部分处理资源(例如多核处理器中的一个核)共同实现，而预测模块72由附图6中至少一个处理器63中的其余部分处理资源(例如多核处理器中的其他核)，或者采用现场可编程门阵列(Field－Programmable Gate Array，FPGA)、或协处理器等可编程器件来完成。显然上述功能模块也可以采用软件硬件相结合的方式来实现，例如获取模块71由硬件可编程器件实现，而预测模块72是由CPU读取存储器中存储的程序代码后，生成的软件功能模块。
附图7中获取模块71，预测模块72、以及预测模块中的各个单元实现上述功能的更多细节请参考前面各个方法实施例中的描述，在这里不再重复。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。 当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本发明实施例所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线(DSL))或无线(例如红外、无线、微波等)方式向另一个网站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质(例如固态硬盘Solid State Disk(SSD))等。
显然，本领域的技术人员可以对本申请进行各种改动和变型而不脱离本申请的范围。这样，倘若对本申请的这些修改和变型属于本申请权利要求的范围之内，则本申请也意图包括这些改动和变型在内。

Claims (16)

  1. 一种终端之间连接状态的预测方法，其特征在于，包括：
    分析设备获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数；
    所述分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻对应的连接状态，所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
  2. 根据权利要求1所述的预测方法，其特征在于，确定测试终端对在未来时间段内至少一个单位时刻对应的连接状态，包括：
    所述分析设备将所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，输入预测模型并获取所述预测模型的输出结果，所述预测模型是根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的，所述第二历史时间段是当前时间之前的时间段、且所述第二历史时间段中包括M+Q个连续的单位时刻，其中N为大于等于1的自然数；
    所述分析设备根据所述输出结果确定所述测试终端对在所述未来时间段内至少一个单位时刻分别对应的连接状态。
  3. 根据权利要求2所述的预测方法，其特征在于，所述输入预测模型并获取所述预测模型的输出结果，包括：
    所述分析设备根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定第一样本序列，所述第一样本序列中包括M个元素，所述M个元素中的每个元素的取值分别对应所述M个连续的单位时刻中每个单位时刻分别对应的连接状态；
    所述分析设备将所述第一样本序列输入预测模型并获取所述预测模型的输出结果，所述输出结果为预测序列，所述预测序列包括Q个元素，所述Q个元素中的每个元素的取值分别对应所述Q个连续的单位时刻中每个单位时刻分别对应的连接状态。
  4. 根据权利要求3所述的预测方法，其特征在于，所述M个元素或Q个元素中的一个元素的取值为第一值时指示对应单位时刻的连接状态为有连接，所述M个元素或Q个元素中的一个元素的取值为第二值时指示对应单位时刻的连接状态为无连接，所述第一值和所述第二值不同。
  5. 根据权利要求3或4所述的预测方法，其特征在于，所述将所述第一样本序列输入预测模型之前，还包括：
    获得N个训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态；
    针对所述N个训练终端对中的第一训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，生成第一训练终端对对应的训练样本序列，以此类推，从而得到N个训练样本序列，所述第一训练终端对对应的训练样本序列中包括M+Q个元素，所述M+Q个元素中的每个元素的取值分别对应所述第一训练终端对在所述M+Q个连续的单位时刻中每个单位时刻分别对应的连接状态；
    将所述N个训练样本序列作为机器学习算法的输入，获得所述机器学习算法输出的所述预测模型。
  6. 根据权利要求1-5任一所述的预测方法，其特征在于，所述获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，包括：
    所述分析设备从保存的多个数据流分别对应的条目中选择第一组目标条目，所述第一组目标条目包括记录的单位时刻属于所述第一历史时间段、源IP地址为所述第一终端的IP地址、且目的IP地址为所述第二终端的条目，以及记录的单位时刻属于所述第一历史时间段、目的IP地址为所述第一终端的IP地址、且源IP地址为所述第二终端的条目；
    所述分析设备确定所述选择出的第一组目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第一历史时间段内、除所述选择出的第一组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态。
  7. 根据权利要求2-6任一所述的预测方法，其特征在于，获得N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，包括：
    所述分析设备获取N个训练终端对；
    所述分析设备从所述N个训练终端对中选择一个训练终端对，对选择出的训练终端对执行以下处理步骤，直到处理完全部所述N个训练终端对为止，所述选择出的训练终端对由第三终端和第四终端构成：
    所述分析设备从保存的多个数据流分别对应的条目中选择第二组目标条目，所述第二组目标条目包括记录的单位时刻属于所述第二历史时间段、源IP地址为第三终端的IP地址、且目的IP地址为第四终端的条目，以及记录的单位时刻属于所述第二历史时间段、目的IP地址为所述第四终端的IP地址、且源IP地址为所述第三终端的条目；
    所述分析设备确定所述选择出的第二目标条目中记录的单位时刻对应的连接状态为有连接，并确定所述第二历史时间段内、除所述选择出的第二组目标条目中记录的单位时刻之外的单位时刻对应的连接状态为无连接，从而得到所述选择出的训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态。
  8. 根据权利要求7所述的预测方法，其特征在于，在Q＝1的情况下，所述N个训练样本序列中正样本的数量与负样本的数量之间的比值大于等于0.5、且小于等于2，所述正样本是最后一个元素的取值指示的连接状态为有连接的训练样本序列，所述负样本中是最后一个元素的取值指示的连接状态为无连接的训练样本序列。
  9. 根据权利要求6或7所述的预测方法，其特征在于，还包括：
    所述分析设备获取多条流统计信息，所述多条流统计信息中的每条流统计信息分别对应一个数据流，所述每条流统计信息中包括数据流的建立时间、关闭时间、源IP地址和目的IP地址；
    所述分析设备根据预定的时间对齐规则，对所述每条流统计信息进行以所述单位时刻为基准的时间对齐处理，生成多个数据流分别对应的条目并保存所述多个数据流分别对应的条目，所述多个数据流分别对应的条目中每个数据流对应的条目记录有单位时刻、源IP地址和目的IP地址。
  10. 根据权利要求7-9任一所述的预测方法，其特征在于，所述第一终端、所述第二 终端、所述第三终端和所述第四终端分别为虚拟机。
  11. 根据权利要求10所述的预测方法，其特征在于，所述虚拟机部署于由数据中心网络连接的数据中心中。
  12. 一种终端之间连接状态的预测装置，其特征在于，包括：
    获取模块，用于获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数；
    预测模块，用于根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态，所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
  13. 根据权利要求12所述的预测装置，其特征在于，所述预测模块包括模型测试单元和确定单元，其中，
    所述模型测试单元，用于将所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，输入预测模型并获取所述预测模型的输出结果，所述预测模型是根据N个训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态生成的，所述第二历史时间段是当前时间之前的时间段、且所述第二历史时间段中包括M+Q个连续的单位时刻，其中N为大于等于1的自然数；
    所述确定单元，用于根据所述输出结果确定所述测试终端对在所述未来时间段内至少一个单位时刻分别对应的连接状态。
  14. 根据权利要求13所述的预测装置，其特征在于，
    所述模型测试单元，用于根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定第一样本序列，所述第一样本序列中包括M个元素，所述M个元素中的每个元素的取值分别对应所述M个连续的单位时刻中每个单位时刻分别对应的连接状态；以及将所述第一样本序列输入预测模型并获取所述预测模型的输出结果，所述输出结果为预测序列，所述预测序列包括Q个元素，所述Q个元素中的每个元素的取值分别对应所述Q个连续的单位时刻中每个单位时刻分别对应的连接状态。
  15. 根据权利要求14所述的预测装置，其特征在于，所述预测模块包括还包括：
    模型学习单元，用于在所述模型测试单元将所述第一样本序列输入预测模型之前，执行以下步骤：获得N个训练终端对在所述第二历史时间段内多个单位时刻分别对应的连接状态；针对所述N个训练终端对中的第一训练终端对在第二历史时间段内多个单位时刻分别对应的连接状态，生成第一训练终端对对应的训练样本序列，以此类推，从而得到N个训练样本序列，所述第一训练终端对对应的训练样本序列中包括M+Q个元素，所述M+Q个元素中的每个元素的取值分别对应所述第一训练终端对在所述M+Q个连续的单位时刻中每个单位时刻分别对应的连接状态；将所述N个训练样本序列作为机器学习算法的输入，获得所述机器学习算法输出的所述预测模型。
  16. 一种分析设备，其特征在于，包括存储器和与所述存储器连接的至少一个处理器，
    所述存储器用于存储指令，所述指令被所述至少一个处理器读取后，所述分析设备执行以下操作：
    获取测试终端对在第一历史时间段内多个单位时刻分别对应的连接状态，所述测试终端对由第一终端和第二终端构成，所述第一历史时间段是当前时间之前的时间段、且所述第一历史时间段内包括M个连续的单位时刻，其中M是大于等于2的自然数；根据所述测试终端对在所述第一历史时间段内多个单位时刻分别对应的连接状态，确定所述测试终端对在未来时间段内至少一个单位时刻分别对应的连接状态，所述未来时间段是当前时间之后的时间段、且所述未来时间段内包括Q个连续的单位时刻，所述未来时间段中的第一个单位时刻与所述第一历史时间段中最后一个单位时刻是连续的单位时刻，其中Q是大于等于1的自然数。
PCT/CN2020/114979 2019-09-12 2020-09-14 终端之间连接状态的预测方法、装置和分析设备 WO2021047665A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
EP20862875.0A EP4024762A4 (en) 2019-09-12 2020-09-14 METHOD AND DEVICE FOR PREDICTING THE CONNECTION STATE BETWEEN TERMINALS AND ANALYSIS DEVICE
JP2022516077A JP7354424B2 (ja) 2019-09-12 2020-09-14 端末間接続状態予測方法及び装置、及び分析デバイス
US17/692,569 US20220200870A1 (en) 2019-09-12 2022-03-11 Inter-terminal connection state prediction method and apparatus and analysis device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910866653.0 2019-09-12
CN201910866653.0A CN112491572B (zh) 2019-09-12 2019-09-12 终端之间连接状态的预测方法、装置和分析设备

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/692,569 Continuation US20220200870A1 (en) 2019-09-12 2022-03-11 Inter-terminal connection state prediction method and apparatus and analysis device

Publications (1)

Publication Number Publication Date
WO2021047665A1 true WO2021047665A1 (zh) 2021-03-18

Family

ID=74866560

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/114979 WO2021047665A1 (zh) 2019-09-12 2020-09-14 终端之间连接状态的预测方法、装置和分析设备

Country Status (5)

Country Link
US (1) US20220200870A1 (zh)
EP (1) EP4024762A4 (zh)
JP (1) JP7354424B2 (zh)
CN (1) CN112491572B (zh)
WO (1) WO2021047665A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115416160A (zh) * 2022-09-23 2022-12-02 湖南三一智能控制设备有限公司 搅拌筒转向识别方法、装置及搅拌车

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117041073B (zh) * 2023-09-05 2024-05-28 广州天懋信息系统股份有限公司 网络行为预测方法、系统、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105099953A (zh) * 2014-04-28 2015-11-25 华为技术有限公司 云数据中心虚拟网络的隔离方法与装置
CN105871751A (zh) * 2016-03-25 2016-08-17 中国科学院计算技术研究所 一种数据中心网络带宽保证方法及系统
US20180278331A1 (en) * 2015-11-30 2018-09-27 Huawei Technologies Co., Ltd. Data Center Network System and Signal Transmission System
CN109617765A (zh) * 2019-01-02 2019-04-12 中国联合网络通信集团有限公司 一种物联网连接质量的预测方法和装置
CN109992367A (zh) * 2017-12-29 2019-07-09 广东欧珀移动通信有限公司 应用处理方法和装置、电子设备、计算机可读存储介质

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP2161665A4 (en) * 2007-06-22 2012-12-12 Nec Corp DATA PROCESSING METHOD FOR PORTABLE COMMUNICATION TERMINAL AND PORTABLE COMMUNICATION TERMINAL
JP5476880B2 (ja) * 2009-09-14 2014-04-23 ソニー株式会社 情報提供装置及び情報提供方法、コンピューター・プログラム、並びに無線通信装置
EP2768256B1 (en) * 2013-02-15 2017-05-31 Telefonaktiebolaget LM Ericsson (publ) Prediction of quality of service of a possible future connection of a device to a wireless network
JP5922825B1 (ja) * 2015-04-21 2016-05-24 日本電信電話株式会社 トラヒック推定装置、方法及びプログラム
JP2017220040A (ja) * 2016-06-08 2017-12-14 株式会社日立製作所 ネットワーク装置、コネクション制御方法、及び計算機システム
US11190965B2 (en) * 2016-10-31 2021-11-30 Veniam, Inc. Systems and methods for predictive connection selection in a network of moving things, for example including autonomous vehicles
US10855550B2 (en) * 2016-11-16 2020-12-01 Cisco Technology, Inc. Network traffic prediction using long short term memory neural networks
CN106954207B (zh) * 2017-04-25 2018-06-05 腾讯科技（深圳）有限公司 一种获取目标终端的帐号属性值的方法及装置
CN107577522B (zh) * 2017-09-30 2020-04-21 Oppo广东移动通信有限公司 应用控制方法、装置、存储介质以及电子设备
JP7017162B2 (ja) * 2017-11-06 2022-02-08 日本電気株式会社 稼働状況予測システム、稼働状況予測方法および稼働状況予測プログラム
US10484891B2 (en) * 2017-12-23 2019-11-19 Fortinet, Inc. Generating recommendations for achieving optimal cellular connectivity based on connectivity details and current and predicted future events

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105099953A (zh) * 2014-04-28 2015-11-25 华为技术有限公司 云数据中心虚拟网络的隔离方法与装置
US20180278331A1 (en) * 2015-11-30 2018-09-27 Huawei Technologies Co., Ltd. Data Center Network System and Signal Transmission System
CN105871751A (zh) * 2016-03-25 2016-08-17 中国科学院计算技术研究所 一种数据中心网络带宽保证方法及系统
CN109992367A (zh) * 2017-12-29 2019-07-09 广东欧珀移动通信有限公司 应用处理方法和装置、电子设备、计算机可读存储介质
CN109617765A (zh) * 2019-01-02 2019-04-12 中国联合网络通信集团有限公司 一种物联网连接质量的预测方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115416160A (zh) * 2022-09-23 2022-12-02 湖南三一智能控制设备有限公司 搅拌筒转向识别方法、装置及搅拌车
CN115416160B (zh) * 2022-09-23 2024-01-23 湖南三一智能控制设备有限公司 搅拌筒转向识别方法、装置及搅拌车

Also Published As

Publication number Publication date
CN112491572B (zh) 2022-01-21
JP7354424B2 (ja) 2023-10-02
EP4024762A4 (en) 2023-07-05
CN112491572A (zh) 2021-03-12
EP4024762A1 (en) 2022-07-06
US20220200870A1 (en) 2022-06-23
JP2022547582A (ja) 2022-11-14

Similar Documents

Publication Publication Date Title
US11088929B2 (en) Predicting application and network performance
Verma et al. A survey on network methodologies for real-time analytics of massive IoT data and open research issues
CN110351118B (zh) 根因告警决策网络构建方法、装置和存储介质
Banerjee et al. Everything as a service: Powering the new information economy
CN108462594B (zh) 虚拟专有网络及规则表生成方法、装置及路由方法
US20160091913A1 (en) Smart power management in switches and routers
WO2021047665A1 (zh) 终端之间连接状态的预测方法、装置和分析设备
Rajagopal et al. FedSDM: Federated learning based smart decision making module for ECG data in IoT integrated Edge-Fog-Cloud computing environments
CN113485792B (zh) 一种kubernetes集群内Pod调度方法、终端设备及存储介质
Zhong et al. An efficient SDN load balancing scheme based on variance analysis for massive mobile users
CN115358487A (zh) 面向电力数据共享的联邦学习聚合优化系统及方法
Yi et al. A multi-criteria decision approach for minimizing the influence of VNF migration
US10148516B2 (en) Inter-networking device link provisioning system
Kanwal et al. Head node selection algorithm in cloud computing data center
Yuan et al. A DRL-Based Container Placement Scheme with Auxiliary Tasks.
CN113934767A (zh) 一种数据处理的方法及装置、计算机设备和存储介质
WO2022110974A1 (zh) 数据分析模型的训练方法、装置及存储介质
Aziz et al. Content-Aware Network Traffic Prediction Framework for Quality of Service-Aware Dynamic Network Resource Management
US11403200B2 (en) Provisioning resources for monitoring hosts based on defined functionalities of hosts
Santos et al. SPIDER: An availability‐aware framework for the service function chain placement in distributed scenarios
US10680878B2 (en) Network-enabled devices
Hong et al. Retracted: Artificial intelligence point‐to‐point signal communication network optimization based on ubiquitous clouds
CN114900441B (zh) 网络性能预测方法，性能预测模型训练方法及相关装置
Seddiki et al. Sustainability-based framework for virtual machines migration among cloud data centers
CN111106974A (zh) 一种测试无损网络性能的方法和装置

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: 20862875

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2022516077

Country of ref document: JP

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2020862875

Country of ref document: EP

Effective date: 20220329