1 tube side
1.1 Definition of pipe process
A Monitor is a set of procedures that operate on a data structure about a shared resource. A process can only indirectly access data structures in a pipe by calling procedures in the pipe.
1.2 Two basic problems to be solved in pipe process
1.2.1 mutex
The pipe is mutually exclusive, so when one process calls the pipe, no other process can call the pipe. This is mainly designed to ensure data integrity. The mutual exclusion of pipe procedures is guaranteed by the compiler.
1.2.2 synchronization
Condition variables and wait/wake operations are set in the tube to solve synchronization problems. When a process or thread is waiting on a condition variable, it should release the use of the pipe first. It can also wake up the process or thread waiting on a condition variable by sending a signal.
1.3 Problems encountered in the application of pipe process
We already know that a process can wake up other processes waiting on a condition variable. If P enters the pipe and wakes up Q, if P and Q operate on data in the pipe at the same time, the data integrity in the pipe will be broken. How can we avoid this problem? There are three solutions:
- P process waiting Q (Hoare process)
- Q process waits for P process to continue (MESA process)
- The specified wake up operation is the last executable operation in the pipe. This means that when any process wakes up another process in the pipe, that process must exit the pipe.
2 Hoare tube
2.1 Core idea of Hoare pipeline
The Hoare tube can be illustrated with the following diagram:
As shown in the figure above, when a process P fromEntry waiting queueIt enters Hoare pipe procedure and performs various operations on resources (at this time, all other pipe procedures can only wait outside the pipe procedure). When a certain process n is executed, condition CN is found insufficient and cannot be continuedProcess P waits on the condition variable cn via the wait operation, process P gives up the mutex and opens the door for other processes to enter. Suppose Q enters, and after a number of operations, the condition variable cn is satisfied. At this point, Q wakes up PROCESS P through the signal operation, which, according to the semantics of the Hoare procedure, should be executed by P while Q waits. The Hoare procedure creates a special section within the procedure for all such processes to enter the wait state by waking up other processesEmergency waiting queue.
The above is a detailed introduction to the working process of pipe process, followed by two supplementary instructions for pipe process:
- A pipe is mutually exclusive, so when a process attempts to enter an already occupied pipe, it should wait at the pipe’s entrance. For this purpose, an entrance waiting queue is set up at the entrance of the pipeline.
- If process P wakes up process Q, process P waits for process Q to execute. If process Q wakes up process R, process Q waits for process R to execute. Processes P and Q enter the emergency waiting queue, and so on. Therefore, there may be multiple waiting processes inside the pipe, but at most one process can operate on the pipe at any time.
An emergency wait queue is a process wait queue created inside a pipe and has a higher priority than an entry wait queue.
2.2 Realization of Hoare process condition variables
A condition variable is actually a special type of variable that is specified and used inside a pipe. The wait and signal operations can be performed on the condition variable C. Table 2-1 Implementation mechanism of condition variables
The operation name | Implementation mechanisms |
---|---|
Wait (c) | If the emergency wait queue is not empty, the first wait is woken up; Otherwise, the mutex is released, and the process that does this goes to the end of the C chain |
Signal (c) | If the C-chain is empty, it is equivalent to a null operation, and the process performing this operation continues; Otherwise, the first wait is awakened and the process performing this operation enters the end of the emergency wait queue |
2.3 Application of Hoare tubing
This section focuses on using pipe procedures to solve producer/consumer problems. The logic is the same as PV operations using semaphores, as shown in the pseudocode in Listing 2-1. Listing 2-1 Hoare procedures for solving producer-consumer problems
monitor ProducerConsumer{
/* Define two condition variables, full means the producer stops producing when the buffer is full and empty means the consumer stops consuming when the buffer is empty
condition full,empty;
/* Number of non-empty buffers */
integer count;
procedure insert(item :integer);
begin
/* When the buffer is full, the producer process aborts execution and enters the condition variable full */
if(count==N) then wait(full); insert_item(item); count++;/* Wake up the consumer process waiting on the condition variable empty */ when the buffer data amount is 1
it(count==1) then signal(empty);
end;
function remove:integer;
begin
/* When the buffer is empty, the consumer process aborts execution and enters the condition variable empty */
if(count==0) then wait(empty); remove=remove_item; count--;/* Wake up the producer process waiting on the condition variable full when the buffer data amount is n-1
if(count==N- 1) then signal(full);
end;
end monitor;
}
// The producer process
procedure producer;
begin
while(TRUE){
item=produce_item;
ProducerConsumer.insert(item);
}
end;
// Consumer process
procedure consumer;
begin
while(TRUE){
item=ProducerConsumer.remove;
consume_item(item);
}
end;
Copy the code
The idea of using semaphores to solve producer-consumer problems can be found in the blog: [Operating systems] Synchronous mutex (I) : An introduction to synchronous mutex and an in-depth understanding of semaphores
3 the multi-ethnic tube
Section 2 of this paper focuses on the introduction of Hoare process. The idea of Hoare is that when P process wakes up Q process, Q process gets the execution right of CPU when P process switches, and then P process switches to CPU again when the condition variable is met. It can be seen from this process that Hoare process actually has two additional process switches. To solve this problem the MESA process was introduced.
3.1 Mechanism of MESA – Wait /notify
Hoare procedures are based on wait/ Signal, which wakes up a process and immediately relinquish the CPU, while MESA procedures are based on Wait/Notify. Notify means that when a process in the pipeline executes notify©, it causes the C conditional queue to be notified and the signaling process to continue execution. Three things to note about Notify:
- The result of notify is that processes at the head of the conditional queue resume execution when CPU is available in the future (unlike Hoare procedures, which do not immediately fetch CPU execution).
- However, since there is no guarantee that no other process has entered the pipe and modified the data resources before the process, the process rechecks the condition each time before acquiring the CPU for execution. So MESA procedures replace the if statement in Hoare procedures with a while loop.
- The effect of using the while loop to judge is that it results in at least one additional check for the condition variable. However, there are no additional process switches, and there are no restrictions on when waiting processes run after Notify.
- Improvements: Since the process is scheduled every time the condition check is performed by the while statement, it can be implementedA monitor timer is associated with each condition primitive, and processes that exceed a certain wait time are directly set to ready state, whether notified or not(No error is reported even if the condition variable is not met, because the while statement ensures that the program receiving the signal checks the relevant variable and continues to wait if the desired condition is not met.)
Listing 3-1 MESA procedures for solving producer-consumer problems
monitor ProducerConsumer{
/* Define two condition variables, full means the producer stops producing when the buffer is full and empty means the consumer stops consuming when the buffer is empty
condition full,empty;
/* Number of non-empty buffers */
integer count;
procedure insert(item :integer);
begin
/* When the buffer is full, the producer process aborts execution and enters the condition variable full */
while(count==N) then wait(full); insert_item(item); count++;/* Wake up the consumer process waiting on the condition variable empty */ when the buffer data amount is 1
cnotify(empty);
end;
function remove:integer;
begin
/* When the buffer is empty, the consumer process aborts execution and enters the condition variable empty */
while(count==0) then wait(empty); remove=remove_item; count--;/* Wake up the producer process waiting on the condition variable full when the buffer data amount is n-1
cnotify(full);
end;
end monitor;
}
Copy the code
3.3 MECHANISM of MESA – Wait/Broadcast
The notify function notifies only one process at a time. MESA also implements a mechanism for notifying multiple processes at a time, that is, broadcast.
Broadcast: Causes all processes waiting on this condition to be released and queued.
Broadcast usage scenarios:
- This is handy when a process does not know how many processes will be activated. Such as: In the producer/consumer problem, the insert and remove functions are assumed to work with variable-length characters. At this point, if a producer adds a batch of characters to the buffer, it doesn’t need to know how many characters each waiting consumer is prepared to consume. The broadcast mechanism allows all waiting processes to be notified and try again.
- Broadcasts can also be used when it is not difficult for a process to determine exactly which process will be activated.
4 Interprocess communication mechanism IPC
The mechanism of semaphores and tubes has been introduced respectively in the previous and previous sections of this paper. However, these two communication modes can only transmit simple information and do not support the transmission of large amounts of information. In addition, tubes are not suitable for users with multiple processors, so a new communication mechanism needs to be introduced.
4.1 Interprocess communication mode
4.1.1 Message Passing
The core idea of messaging is shown below:
Listing 4-1 implements the SEND primitive with the PV operation
/* Define the following four semaphores */
EmptyBuff: Empty message buffer, initial value N
// fullBuff: full message buffer with initial value 0
// mutex1: the initial value is 1
// mutex2: initial value is 1
send(destination,msg){
// Find the receiving process according to destination; If not, report an error and returnApply for empty buffer P(emptyBuff); P(mutex1); Empty buffer; V(mutex1); Copy the message from MSG to an empty buffer; P(mutex2); Attach the message buffer to the receiving process's message queue V(mutex2); V(fullBuff); }Copy the code
4.1.2 Shared Memory
As shown in the figure below, both processes 1 and 2 have their own memory address space. In the physical memory space, there is a shared memory space. Each process has a shared memory space mapped to the same physical memory (that is, shared memory). The Shared memory
There is also a reader-writer problem: multiple processes cannot write at the same time, but simultaneous reads are supported. If process 1 wants to write to shared memory, it only needs to write to the locally mapped memory space, which is equivalent to writing to shared memory.
4.1.3 Communication mode PIPE
Using aBuffered transport medium – memory or filesConnects two processes that communicate with each other. The schematic diagram is as follows:
The following three points should be paid attention to in pipeline communication mode:
- Write as a stream of characters
- First in, first out order is guaranteed
- Coordination capabilities that the pipeline communication mechanism must provide: *Mutual exclusion, synchronization, check whether the other process exists
In addition to these three communication methods, there are sockets, remote procedure calls, etc.
4.2 Linux Process Communication Mechanism
4.2.1 Atomic operation
It is indivisible and will not be interrupted by other tasks or events until it is completed; Often used to implement reference counting of resources
4.2.2 Barriers
1 the concept
A synchronization mechanism (aka fence, checkpoint); Used to coordinate a group of threads
2 Application Scenarios
A group of threads working together on a task requires all threads to reach a point of convergence before moving forward together.
In fact, communication mechanisms in Linux include: spinlocks, read/write spinlocks, semaphores, read/write semaphores, mutex, completion variables, sequential locks, and so on.
5 subtotal
In this paper, we introduce the concept of pipeline and the synchronous mutual exclusion mechanism, focusing on Hoare pipeline and MESA pipeline. This paper introduces the interprocess communication mechanisms — message passing, shared memory and pipes, and introduces two IPC mechanisms in Linux system.