CSE 130 Part 3

How can local procedure call misbehave?

Can read/write to out-of-scope local variables

Define soft modularity

Pass messages/data from one function to a second one by calling it

What is one error caused by soft modularity?

Propagation of errors

Define hard modularity

Pass messages/data over a network connection

Why is hard modularity safer from propagation of effects?

No shared memory or interpreter → failure in one system does not affect the other

What does LAMP stand for in a web server model?

Linux, Apache, MySQL, PHP

How would you alter LAMP to deal with static load/infrequent DB access?

Increase AP + load balancer, while keeping M minimal

How would you alter LAMP if you needed high reliability?

Use multiple M (redundancy)

What is a core issue with messages? What is one way we could address this?

They may not reach the other party, and we need a way to know what happened.

Use acknowledgements.

What are the three types of ____ once in RPC?

• Which one(s) uses unique id?

• Which requires a request to be idempotent?

At Least once: needs to be idempotent

At Most once: needs unique id (duplication in the network)

Exactly once: needs unique id

Why can we not guarantee 100% accurate implementation of exactly once RPC?

Set of unique ids/results cannot grow indefinitely

What is asynchronous RPC?

Client sends the request, but it doesn’t wait for the return. It continues with other works and later processes the results.

What is marshaling/serialization in RPC?

Agreeing that data on the two end systems have the same meaning (e.g. converting to ensure same endian-ness, size of int, pointers)

Is virtualization soft modularity or hard modualrity?

Neither, it’s in between

Three basic virtualization techniques and their definitions

1. Multiplexing: one physical object → multiple virtual objects

2. Aggregation: many physical objects → one virtual object

3. Emulation: physical object(s) → a new type of virtual object

Which of the three virtualization technique is “Sockets across a network”?

Multiplexing— one physical network interface → many sockets

Which of the three virtualization technique is “RAM disk”?

Emulation— OS uses RAM as if it’s a disk to store things

Which of the three virtualization technique is “RAID”?

Aggregation— combining multiple physical disks into one logical volume

Which of the three virtualization technique is “Content distribution network”?

Aggregation— many servers disguised as one server

Which of the three virtualization technique is “Logical volume manager“?

Multiplexing & Aggregation— Can combine multiple disks as one as well as partition one volume into many

Define virtual memory

Memory that processes can use as if they each have their private own memory

_______ allows two processes to read/write to same physical address

Inter-process communication

How can a bounded buffer act as a communication link?

Threads cannot communicate with each other directly— they communicate through the OS using bounded buffer

4 OS-provided functions for bounded buffers

1. allocate

2. deallocate

3. send

4. receive

Why should a bounded buffer be bounded?

Back pressure— sender should not send too much data at start

What is a problem with this code? (Find 2)

1. Busy wait (line 2)

2. Race condition (line 5)

Define race condition

Behavior of a system depends on timing or order of uncontrollable events

Define data race

Two unordered memory operations (one of which is write) act on the same variable

Is multiple reads on the same memory OK? (Does it have a race condition?)

Yes, it is OK

4 steps to fix race conditions

1. Critical regions

2. Mutual exclusion

3. Condition variables

4. Semaphores

Define critical region

Region that must be executed by only one thread at a time

What is mutual exclusion?

Ensuring only one thread is in its critical region at a time

4 conditions of mutual exclusion

1. No two threads may be simultaneously inside their critical regions

2. Any number of threads

3. No thread running outside its critical region may block another thread

4. No thread can wait forever to enter its critical region

Why should a while loop be included in a critical region?

If you lock after checking the loop condition, an OS may allow multiple threads to enter the loop

How do you prove possible outcomes?

1. Choose a “vocabulary” (e.g. Load, Add, Store)

2. Find an order of operations that shows the outcome

What is deadlock?

One thread acquires the lock and waits for a condition (e.g. push waits for the buffer to be pop()-ed) but another thread cannot fulfill that condition because of the lock

What is one inefficient way to resolve deadlock? What is another issue caused by this method?

Strategic release()/acquire() pairs, causing spinlock (wasting CPU resources)

Explain pthread_create(1,2,3,4)

1. Pointer to pthread_t

2. NULL (we don’t use in class)

3. Pointer to a thread function

4. Arguments to be passed to 3 as (void *)

Explain pthread_join(1,2)

pthread_join() waits/blocks until the thread function returns

1. pthread_t (not a pointer!)

2. Pointer to return val

Say we want to pass in a thread_id to the thread function. How do we do it?

Use intptr_t to cast int to void *

int thread_id = i; (using for loop)

intptr_t intptr = thread_id;

pthread_create(…, (void *) intptr);

What functions do we use to lock and unlock mutex?

pthread_mutex_lock(1) and pthread_mutex_unlock(1)

• 1: pointer to pthread_mutex_t

Three functions used to wake thread(s) and release lock using condition variable

1. wait(): release lock and reacquires it once released by another

2. signal(): wakes one thread waiting on a condition variable

3. broadcast(): wakes all threads waiting on a condition variable

Functions for creating and destroying a mutex lock

// static initialization
pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;

// dynamic initialization
pthread_mutex_t mutex;

int rc = pthread_mutex_init(&mutex, NULL);
assert(rc == 0);


int rc = pthread_mutex_destroy(&mutex, NULL);
assert(rc == 0);