Computer Systems: A Programmer's Perspective (Global Edition): Taxonomy of Concurrency: Processes, I/O Multiplexing, and Threads

In the realm of computer systems, application-level concurrency is the deliberate overlapping of logical control flows to boost performance and responsiveness. It is a functional abstraction: a program is partitioned into independent tasks that can be interleaved or executed in parallel.

1. The Concurrency Taxonomy

Developers generally choose between three fundamental mechanisms to manage these concurrent flows:

Processes: High isolation with separate address spaces; requires kernel-mediated IPC.
I/O Multiplexing: A single flow manually switching between "ready" events (state machines).
Threads: Lightweight flows sharing a single virtual address space for easy data exchange.

2. Logical vs. Physical Execution

While all parallel programs are concurrent, not all concurrent programs are parallel. Parallelism is the physical execution of flows on separate hardware cores. Concurrency is the logical design that allows such execution to happen.

TERMINAL bash — 80x24

> Ready. Click "Run" to execute.

QUESTION 1

Which concurrency model offers the highest level of isolation between flows?

Threads

I/O Multiplexing

Processes

Coroutines

QUESTION 2

A program is 'Parallel' only if its concurrent flows...

...are managed by the kernel.

...execute simultaneously on separate processor cores.

...use shared memory for communication.

...are triggered by I/O events.

QUESTION 3

What is a primary disadvantage of I/O Multiplexing compared to Threads?

High context-switch overhead.

Inability to exploit multi-core parallelism.

Difficulty in sharing global data.

Requirement for complex IPC mechanisms.

QUESTION 4

Why do modern windowing systems use concurrency for UI responsiveness?

To ensure memory safety.

To allow the interface to respond to user input while background tasks execute.

To prevent the kernel from reaping child processes.

To bypass the need for a file descriptor table.

QUESTION 5

In the taxonomy of concurrency, which model models the application as a state machine?

Process-based

Thread-based

I/O Multiplexing

Sequential programming

Deep Dive: Concurrency Architecture & Synchronization

Analysis of Process Management and Buffer Safety

Examine the mechanics of a concurrent server using process-based concurrency and the synchronization requirements of shared buffers in producer-consumer systems.

Figure 12.5 demonstrates a concurrent server in which the parent process creates a child process to handle each new connection request. Trace the value of the reference counter for the associated file table for Figure 12.5.

Solution:
The reference counter for the connected file descriptor table entry evolves as follows:
1. Initial: The parent accepts a connection, creating connfd. Reference count = 1.
2. Fork: The parent calls fork(). The child inherits the descriptor table. Reference count = 2.
3. Parent Close: The parent calls Close(connfd). Reference count = 1.
4. Child Termination: The child handles the request and exits (or calls Close). Reference count = 0, allowing the kernel to reclaim the socket.

Let $p$ denote producers, $c$ consumers, and $n$ the buffer size. Is the mutex semaphore in sbuf_insert/sbuf_remove necessary if (Scenario 1) $p=1, c=1$ and (Scenario 2) $p=1, c=10$?

Solution:
Scenario 1 ($p=1, c=1$): In a basic circular buffer where the producer only modifies the 'rear' and the consumer only modifies the 'front', a mutex is not strictly necessary for correctness, though often used for simplicity.
Scenario 2 ($p=1, c=10$): Yes, the mutex is mandatory. Because multiple consumers are competing to update the 'front' index and decrement the item count, a race condition would occur without mutual exclusion.

If the parent process in Figure 12.5 failed to close its copy of the connected descriptor, what specific resource leak occurs?

Solution:
This results in a file descriptor leak. Even if the child process exits and closes its copy, the reference count in the kernel's file table never reaches zero because the parent (which runs forever) still holds a reference. Eventually, the server will hit its limit for open files and fail to accept new connections.