Concurrent network modular reading notes transfer

　　 At present, there are many network libraries , among libevent Is based on Reactor Realized , and boost.asio Is based on Proactor Realized .Reactor and Proactor The main difference between modes is the real operation （ As read / Write ） Who did it ,Reactor The application needs to read or write data by itself , And in the Proactor In the pattern , The application does not need to actually read / Writing process , The operating system will read the buffer or write the buffer to the real IO equipment , The application only needs to read from the buffer （ The operating system has been read for us ） Or write to the buffer （ The operating system will help us write ） that will do . stay Proactor In the pattern , After the user initiates the asynchronous operation, it returns , Let the operating system handle the request , Then wait for the callback to the completion event function to process the results of the asynchronous operation .

2.3 The driver （Proactor）

1. problem

      Processing multiple service requests asynchronously can usually improve the performance of event driven applications in distributed systems . After asynchronous service processing , The program must handle the corresponding completion event sent by the operating system to indicate the end of asynchronous calculation . To effectively support this asynchronous computing model , The following four mandatory conditions need to be solved ：

      1） To improve scalability and latency , An application must process multiple completion events at the same time while prohibiting time-consuming operations from excessively delaying the processing of other operations .

      2） To maximize throughput , It should be avoided CPU Any unnecessary context switching between , Synchronization and data mobility .

      3） Integrating new or improved services with existing completion event multiplexing and allocation mechanisms costs the least .

      4） Application code should try not to be affected by the complexity of multithreading and synchronization mechanism .

 

2. Solution

      Divide the application service into two parts ： Time consuming operations executed asynchronously and completion handlers that process the results after these operations are completed . Integrate the multiplex decomposition of completion events when asynchronous operations are completed and allocate them to the corresponding completion handlers that handle them . Separate the completion event multiplexing and allocation mechanism from the application related processing of completion events in the completion handler .

      Specifically, it means ： Introduce an asynchronous operation for each service provided by the application . These asynchronous operations actively initialize the processing of service requests through a handle and completion handler , The completion handler handles completion events that contain the results of these asynchronous operations . for example , Asynchronous operation can be called by the initiator in the application to accept the connection request sent from the remote application . Asynchronous operations are performed by asynchronous operation processors . After the operation is completed , The asynchronous operation processor inserts a completion event containing the operation result in the completion event queue .

      An asynchronous event demultiplexer called the initiator waits for the queue . After the asynchronous event demultiplexer deletes a completion event from the queue , The driver demultiplexes and allocates the event to the application related completion handler corresponding to the asynchronous operation , The completion handler then processes the results of the asynchronous operation , You may need to call other asynchronous operations , These asynchronous operations follow the above chain of activities .

 

3. structure

      The operating system provides a handle （handle） To identify entities （ Such as network connection ） Or open the file （ Can generate completion events ）, Can respond to external service requests （ Such as connection or data request from remote application ） And produce completion events , It can also respond to operations generated within the application （ Such as timeout or asynchrony I/O system call ） And produce completion events .

      Asynchronous operations （asynchronou operation） Indicates that in the implementation, such as through socket Handle the time-consuming operation to be used when reading and writing data asynchronously . After calling an asynchronous operation , Perform the asynchronous operation without blocking the caller's control thread . In this way, the caller can perform other operations . If an operation has to wait for another event to happen , Such as a connection request generated by a remote application , Then it will be delayed until the event arrives .

      Finish processing （Completion handler） Defines an interface consisting of one or more hook methods , You can use these methods to process the information returned in the completion event related to the application . The completion event is generated after the asynchronous operation is executed .

      

      After implementing the inherited hook method , Complete the processing procedure （concrete completion handler） A completion handler is specialized to define a specific application service . When the asynchronous operation related to the completion handler is completed , The hook method receives and processes the result information contained in the completion event . A concrete completion handler corresponds to a handle that can be used to call asynchronous operations .

      

 

      Asynchronous operation processor （asychronous operation processor） Call an asynchronous operation of a handle and run to the end , The asynchronous operation processor is usually implemented by the operating system kernel . After the asynchronous operation is completed , The asynchronous operation processor generates the corresponding completion event . According to the handle targeted by the operation , The asynchronous operation processor inserts the completion event into the completion event queue corresponding to the handle （completion event queue） in , The queue holds completion events to be multiplexed to the corresponding completion handler .

      

 

      Asynchronous event demultiplexer （asynchronous event demultiplexer） It's a function , It waits for the completion event that will be inserted into the completion event queue after the asynchronous operation is completed . Then the asynchronous event resolver function deletes one or more completion events from the queue , And return it to the caller .

      The driver （proactor） Provide event loops for application processes or threads . In this event cycle , The driver invokes an asynchronous event demultiplexer , Wait for the completion event to happen . When the event arrives , The asynchronous event resolver returns . Then the driver multiplexes the event to the corresponding completion program , And assign the appropriate hook method to the handler , To handle the results of the completion event .

      

 

      Start the program （initiator） It is an entity inside the application , It calls the asynchronous operation of the asynchronous operation processor . Usually, the startup program processes the result of the asynchronous operation called , under these circumstances , It plays the role of completing the handler .

      

 

4. Realization

      The participants in the active mode can be divided into two layers ：

      · Multichannel decomposition / Allocate infrastructure layer components . This layer follows the general , Application independent policies perform asynchronous operations . It also demultiplexes and assigns completion events of asynchronous operations to corresponding completion handlers .

      · Components of the application layer . This layer defines asynchronous operations and specific completion handlers that execute service processing related to applications .

 

1） Divide application services into asynchronous operations and completion handlers . In order to realize the driver mode , The activation of asynchronous operations through handles in application services must be separated from the processing of the results of these operations . The result of this activity is a set of asynchronous operations , A set of completion handlers , And the association between a set of asynchronous operations and completion handlers .

2） Define the completion handler interface . The interface in the completion handler consists of one or more hook methods . These hook methods represent the completion processing of application related completion events , These completion events are generated after the asynchronous operation is completed .

  2.1） Define a type that passes the results of asynchronous operations . When the asynchronous operation is completed or cancelled , The result of the completion event must be passed to the completion handler . These results indicate whether the asynchronous operation succeeded or failed , And the number of bytes successfully transferred .

  2.2） Determine the type of assignment target . There are two types of completion handlers that can be associated with handles , And as the target of the driver allocation mechanism ： Object and function pointers .

  2.3） Define the completion handler allocation interface policy . Single method allocation interface strategy and multi method allocation interface strategy .

3） Implement asynchronous operation processor .

  3.1） Define asynchronous operation interface . Different parameters can be passed to asynchronous operations , Such as handle 、 Data buffer 、 Buffer length and information required to perform completion processing after the operation is completed . When designing the program interface of the starter to call the asynchronous operation of the asynchronous operation processor , There are two problems to be solved ：

  · Try to increase portability and flexibility .

  · Effectively , Simply handle multiple completion handlers , Driver and completion event queue .

  3.2） Select the asynchronous operation processing mechanism . When the initiator calls an asynchronous operation , The asynchronous operation processor performs the operation without blocking the control thread of the initiator . An asynchronous operation processor provides a mechanism for managing asynchronous completion identification and asynchronous execution of operations . When the operation ends, it also generates a completion event , And insert the event into the appropriate completion event queue .

          Some asynchronous operation processors allow the initiator to cancel the asynchronous operation , But the completion of the event still needs to happen . such , The completion handler can correctly reuse asynchronous completion tags and other resources .

4） Define the driver interface . The application uses the driver interface to call an event loop , Delete the completion event from the completion event queue , Demultiplex the completion event to the corresponding completion handler and assign the corresponding hook method .

5） Implement the driver interface .

  5.1） Develop the driver implementation layer .

  5.2） Select the completion event queue and asynchronous event demultiplexer mechanism .

  ·FIFO Multichannel decomposition . This type of asynchronous event resolver function waits for the completion event corresponding to any asynchronous operation associated with the completion event queue , Delete these events according to the insertion order of these completion events .

  · Selective multiplexing . This type of asynchronous event multiplexer function selectively waits for a specific subset of events to complete , These completion events must be explicitly passed when the function is called .

  5.3） Determine how to multiplex completion events to completion handlers . An effective way to multiplex completion events to completion handlers , The concise strategy is to use the asynchronous completion tag pattern . Using this strategy , Once the program is started, it calls the asynchronous operation , The asynchronous operation processor will receive information to guide the subsequent completion of processing .

         When the asynchronous operation is completed , The asynchronous operation processor generates the corresponding completion event , Associate completion events with asynchronous completion tags , And insert the modified completion event into the corresponding completion event queue . After the asynchronous event demultiplexer removes the completion event from its completion event queue , The driver implementation can be used as an asynchronous completion marker for completion events in constant time O(1) Inner multiplexed to the specified completion handler .

  5.4） Determine how to assign hook methods to the specified completion handler .

  5.5） Define the specific driver implementation .

6） Determine the number of drivers in the application .

7） Implement the specific completion handler .

  7.1） Determine the strategy for maintaining the state of the specific completion handler .

  7.2） Select the mechanism to configure the specific completion handler with the handle .

  7.3） Realize the function of completing the handler .

8） Realize the startup program .

 

5. Conclusion

advantage ：

      1） Transaction separation .

      2） Portability .

      3） Encapsulation of concurrency mechanism . One advantage of separating the driver from the asynchronous operation processor is that the application can configure the driver with different concurrency strategies , Without affecting other application components and services .

      4） Separate threading from concurrency control .

      5） performance .

      6） Simplify application synchronization .

Insufficient ：

      1） The scope of application is limited .

      2） Programming 、 Debugging and testing complexity .

      3） Asynchronously scheduled 、 Control and cancel the operation being performed

 

Receiver - The connector （Acceptor - Connector）

1. problem

      In connection oriented networked systems , Applications usually contain a considerable number of configuration agents for establishing connections and initializing Services . Most of these configuration codes are independent of the processing of data exchanged between connection transmission endpoints by the service . Therefore, close coupling between configuration code and service processing code is not ideal , Because this cannot solve the following four mandatory conditions ：

      1） It should be easy to change the connection role , To support different application features .

      2） It should be easy to add services , Types of service implementations and communication protocols , It does not affect the existing connection establishment and service initialization configuration code .

      3） The change of connection establishment and service initialization strategy is less frequent than that of the communication protocol and service implemented by the application .

      4） For large-scale networked systems , It should be possible to use advanced operating system features , Such as asynchronous connection mechanism to reduce the connection establishment delay .

 

2. Solution

      Connect and initialize peer services in networked applications , Separate from the processing performed by these services after connection and initialization .

      To be specific ： Encapsulate application services in peer service handlers . Each service handler implements half of the end-to-end services in networked applications . Use two factories to connect and initialize peer services ： Receptors and connectors . The two factories cooperate with each other , Create a full association between two peer service handlers and their two connected transport endpoints , Each transport endpoint is encapsulated in a transport handle .

      The receiver factory represents the relevant peer service handler , Passively establish a connection after the connection request event sent by the remote peer service handler arrives . In turn, , The connector factory actively establishes a connection with the specified remote peer service handler on behalf of the peer service handler .

      After the connection is established , The receiver and connector factory initialize the corresponding peer service handler , Send them their respective transfer handles . The peer service handler then uses the transport handle to exchange data through the connected transport endpoint , Conduct application related processing . Generally speaking , After connection and initialization , The service handler does not interact with the receiver and sender .

 

3. structure

      The transmission endpoint of passive mode is a factory , It listens for connection requests , Accept connection requests and establish transfer handles , The transport endpoint of the new connection is encapsulated in the transport handle . You can exchange data by reading and writing the corresponding transmission handle through the connected transmission endpoint .

       

 

      The service handler defines half of the end-to-end services in the network system . In this end-to-end service, a specific service handler usually has both the client role and the server role . In the case of peer-to-peer use , It can play these two roles at the same time . A service handler provides an activation hook method , Used to initialize after connecting to the service handler of its peer . in addition , The service handler also includes a transport handle that encloses the transport endpoint , Such as data mode socket handle . Once connected , The service handler uses the transport handle to exchange data with the peer service handler through their connection to the transport endpoint .

       

 

      The receiver is a factory , Its functions include passively establishing the connected transmission endpoint , Create and initialize related transport handles and service handlers . One receiver provides two methods ： Connection initialization and connection completion , These two methods realize the above functions with the help of passive mode transmission endpoint .

      When calling the initialization method , The receiver binds the passive mode transmission endpoint to a transmission address , Like a TCP The port number and IP The host address , It passively listens for connection requests .

      When the connection request arrives , The connection completion method of the receiver is as follows 3 A step ：

      · First , It uses the passive mode transport endpoint to create a connected transport endpoint , And encapsulate the endpoint in the transport handle .

      · second , It creates a service handler , Used to process data requests made by peer service handlers through connected transport endpoints .

      · Third , It stores the transfer handle in the corresponding service handler , Then call the activation hook method of the service handler , This method lets the service handler complete initialization by itself .

 

      The connector is also a factory , The functions it realizes include the transmission endpoint that actively establishes the connection , Initialize its corresponding transport handle and service handler . It provides 2 A way ： Connection start and connection completion methods .

      Pass the connection start method to an existing service handler , This method establishes a transmission endpoint for it and the receiver . As introduced earlier , The receiver must be listening for connection requests on a transport address .

      Separate the connection start method and completion method of the connector , This enables the connector to support transparent synchronous and asynchronous connection establishment .

      · For synchronization , The connector that activates the connection request blocks its caller , Until the transmission endpoint is connected . in the future , The connector directly calls the activation hook method of the service handler .

      · For asynchronous cases , Run connection requests asynchronously , The activation method of the connector returns immediately . Only after the connector is notified that the transport endpoint has completed the connection asynchronously , The connection completion method activates the service handler .

       

      Whether connecting the transmission endpoint synchronously or asynchronously , Upload and input the endpoint in the connection , Both the receiver and the connector initialize a service handler by calling its activation hook method . in the future , Service handlers generally do not need to interact with connector and receiver factories .

 

      The allocator is responsible for demultiplexing different types of service requests （ Such as connection request and data request ） Indication event of .

      

      · For receptors , The allocator demultiplexes connection indication events received from one or more transport handles that encapsulate the transport endpoint . Multiple receivers can register the same allocator , The allocator listens for them for connection requests from peer connectors .

      · For connectors , The allocator demultiplexes completion events that arrive in response to asynchronously activated connections . To deal with this situation , The connector registers itself with an allocator , To receive these connection completion events , Then the allocator runs its event loop . When the completion event arrives , It notifies the corresponding connector . Then the connector calls the activation hook method of the specified service handler , Let the service handler initialize itself . therefore , An allocator and connector can activate and complete connections asynchronously on behalf of multiple service handlers .

            Be careful , Asynchronous connection establishment does not require the use of allocators , Because the thread that starts the connection will be blocked by waiting for the connection completion event . therefore , This thread can directly activate the service handler .

 

      Specific service handlers define some end-to-end services related to applications , This part of the service is activated by a specific receiver or connector . The specific receiver uses the specific service handler . Transport endpoints and transport handles used by these service handlers to instantiate generic receivers . Similarly , Specific connectors also instantiate generic connectors .

 

4. Realization

      Receiver - The components in the connector pattern can be decomposed into the following three layers ：

      · Multichannel decomposition / Allocate infrastructure layer components . This layer implements a general application independent event allocation strategy .

      · Connect management components . This layer implements general application independent connection and initialization Services .

      · Application layer components . This layer customizes the general strategies of the upper two layers , It uses subclass derivation , Object composition and / Or instantiation of parameterized types , To create concrete components that are used to establish connections , Exchange data and perform service related processing .

 

1） Realize multiplexing / Allocate infrastructure component layer .

      1.1） Select the transmission mechanism , These mechanisms include ：

            · Passive mode transport endpoint component .

            · Connected transport endpoint components .

            · Transmission address component .

            · Transfer handle component .

      1.2） Implement the allocation mechanism . These mechanisms consist of allocators and event handler components . The dispatcher is responsible for matching the request to the corresponding receiver , Connectors are associated with service handlers . The event handler defines the event handling interface provided by the service in the event driven application .

2） Implement the connection management component layer . This layer is responsible for creating service handlers , Actively or passively connect the service handler to the remote service handler , And activate the service handler after connecting . All components in this layer are common , And depends on specific IPC Mechanism , Specific service handlers , Specific receptors and specific connectors . There are three main components in the connection management layer ： Service Handler , Receptors and connectors .

      2.1） Define a common service handler interface . The service handler provides a common interface , It is by the client , Server or both in end-to-end services （ Client and server ） Jointly defined Services . This interface includes initializing the service handler , Execute the services it defines and maintain the IPC Mechanism and other methods .

      2.2） Define a general receiver interface . Implement a common strategy in the receiver component , This strategy is used to passively establish connections , The established initialization specific service handler , These specific service handlers use these connections to exchange data with peer service handlers . An initialization method is also defined in the receiver , For applications to call this method , It is used to notify other applications in the network of its passive mode transmission endpoint .

      2.3） Implement a general receiver method . The application initializes the receiver by calling its initialization method . When calling the initialization method, there should be a specified transmission address （ Such as local host IP Name and TCP Port number ） Parameters of . The receiver uses this address parameter to listen for connection requests initiated by the peer connector . It passes this address through the concrete receiver subclass of the receiver or through typed parameters to the concrete configured in the general receiver IPC Connection mechanism . then IPC The connection mechanism initializes the passive mode transmission endpoint of the receiver , Tell its address to the remote application that wants to establish a connection with the receiver .

      2.4） Define common connector interfaces . The connector component implements a common strategy for actively establishing connections and initializing service handlers , These service handlers are responsible for handling request and response events on the connection .

      2.5） Implement the general connector method . Connectors for applications connect() Method to start a connection , Specific connectors can use this template method , Modify the active connection strategy transparently without changing the connector interface and implementation . therefore ,connect() Leave the specific steps in the connection strategy to the hook method , Specific connectors can overload these hook methods for personalized operations .

3） Implement the components of the application layer . This layer implements specific service handlers , Specific receptors and specific connectors . 

 

5. Conclusion

advantage ：

      1） Reusability , Portability and scalability . Receiver - The connector pattern separates the mechanism of connecting and initializing service handlers from the service processing that follows . Application independent mechanisms in receivers and connectors are reusable components .

      2） Robustness, . Receiver - The connector pattern separates the service handler from the receiver . This separation ensures that passive mode transmission endpoints cannot be used for reading / Writing data . This increases type safety , It can eliminate a kind of error that often occurs when using weak type network programming interface .

      3） Efficiency . Receiver - Connector mode can actively establish connections with a large number of hosts asynchronously and efficiently in a wide area network with great delay .

 

Insufficient ：

      1） Increased indirection . Compared with directly using the underlying network programming interface , Receiver - Connector mode is not direct .

      2） Added complexity . For connecting to only one server , A simple client application that uses a single network programming interface to execute a service , Receiver - Connector patterns add unnecessary complexity . However, using a common receiver and connector wrapper appearance can keep developers from thinking about boring , Error prone and non portable low-level network programming mechanism , This simplifies the design of the application . 

http proxy server （3-4-7 Layer of the agent ）- Network event library common component 、 kernel kernel drive Camera drive tcpip Network protocol stack 、netfilter、bridge Seems to have seen ！！！！ But he that doeth good Don't ask future -- Height and weight 180 Fat man