Section 4: Threads Question 1 Using this figure, contrast the structure of the Single-Threaded Process (STP) model with the Multi-Threaded Process (MTP) model. 1. Describe differences between the Process Control Block in the STP and PCB / TCB in the MTP. 2. Describe the differences in how user stacks are maintained in each model. 3. Describe the differences in how processor state information (i.e. program counter, register contents, etc.) is maintained in each model. 4. Describe how OS allocated resources (files, semaphores, sockets, etc.) are managed in each model. Answer 1. In both models, a “process” is an address space maintaining the process image. The image contains the user address space (instructions and data (heap)) and control stack. In the single-threaded model, each process has only a single thread of execution and so there is little distinction between the process and its thread e.g. both process and single thread share a single process control block. In the multithreaded model, the process provides a context in which one or more threads execute. 2. In the STP model, there exist only a single thread so only a single user stack per process is needed. The MTP model has multiple threads each of which maintains its own instruction trace (thread of control) and each thread requires a separate user stack. 3. In the STP model there is only a single thread whose processor state information must be saved and restored when the process is context switched or interrupted. So in the STP, processor state can be maintained in the Process Control Block.In the MTP model, each thread maintains its own execution state. Each thread is individually context switched and interrupted so each thread’s processor state is maintained in its Thread Control Block. 4. In the MTP model, all threads share access to the resources allocated to the process as a whole. These resources are maintained in the single Process Control Block maintained by each process and shared by all threads. Question 2 How can the use of threads increase the performance of an application when run on a multiprocessor? Use the example of Worker Threads pattern to illustrate your answer. Answer Each thread represents a ‘thread of execution’ in the program. When executed on a multiprocessor, each thread can execute on its own processor in parallel with other threads increasing the overall processing throughput of the system. In the case of the worker threads pattern, if each processor is capable of processing N work items per second, then M processors can process M*N work items per second (in theory anyway). Question 3 What are the three advantages provided by Threads over Processes given in the slides? Answer 1. Threads can be created and destroyed more efficiently than processes. 2. N threads require fewer resources than the same number of processes. This means we can support N threads of execution in an application without the overhead of N processes. 3. Threads simplify the communication between concurrently executing instruction traces. Threads share access to the memory, files, sockets, and other resources allocated to their process while processes are isolated from each other. Question 4 What is the significance of blocking I/O (e.g. a network socket read) in the single vs. multi-threaded models? Use the example of the monitoring application presented in the slides. Answer When programming with a single-thread, blocking I/O will halt the execution of the entire process. If the monitoring application were implemented using a singlethread, that thread would be blocked waiting for an incoming message and would freeze the GUI i.e. the GUI would not update or respond to user input while the single thread blocks waiting for a new message. When programming with multiple threads, we design a process with threads that serve specific purposes. We can create a thread that blocks waiting for network messages and a second thread that maintains the GUI’s presentation. Question 5 Describe the three multithreading control strategies presented in the slides. Answer Note: For the exam, pay attention to the pseudo code associated with each answer. Worker Threads: If the problem involves processing individual units of work, multiple threads can be employed to process several work units in parallel (assuming a multiprocessor). See Slides. while (true) { workItem = FIFO.take(); process(workItem); } Task Scheduling: One or more tasks (i.e. some action that is required to maintain the system) are periodically scheduled (e.g. once every N seconds) and executed. A thread can be created which is programmed to alternatively sleep for N seconds (the wait period between task executions) and execute the task after the sleep period. while (true) { sleep(N); doTask() } Event Handling: Events are delivered to the system from many sources, each of which requires a specific response from the system. Each event handler can be executed in own thread. while (true) { message = socket.read(); // Blocking I/O Read process(message); } Note in this example the socket.read() operation is blocking i.e. the thread’s execution will block until a message arrives on the socket. Question 6 Describe the significance of Figure 4.3 A and B in terms of the time needed to execute the two RPC client calls with and without threads? Hint: Describe interms of the time the process spends blocked waiting for the response from the remote RPC server. Is a multi-core multiprocessor needed to obtain this increase in performance? Answer Figure 4.3a illustrates that without multiple threads, two blocking RPC calls must be executed sequentially taking 2x to execute both requests. Figure 4.3b illustrates that when we execute each blocking I/O request in a separate thread, we can see an overall improvement in the system’s performance. The second request can be executed in Thread 2 while the first request blocks the execution of thread one allowing both requests to be executed concurrently. Most interesting is that the increase in performance is obtained using a uniprocessor (NOT SMP). While thread one’s execution is blocked waiting for the server’s response, the second RPC call can be made in a second thread on the same (single) processor. Question 7 What does Amdahl’s Law tell us about scaling up an application’s performance using multiple processors (SMP)? Using the speedup formula (section 4.3), how much of a