Project of the course CS-453

Your goal is to implement a Software Transactional Memory (STM).

The real goal is of course to get you a hands-on experience with actual concurrent programming.

What prior knowledge do I need?

Only basic C or C++ knowledge.

How to use the atomic libraries and the memory models of C11 and C++11 will be taught in the weekly project sessions.

Useful resources

• C/C++ reference

  • C11 atomic

  • C++11 atomic

• Preshing on Programming - Stellar resources and facts about concurrent programming

What is a STM?

• This course.

• The Art of Multiprocessor Programming - Chapter 18.

Some implementations out there

• TinySTM

• LibLTX

• stmmap

You may read and inspire from existing STM libraries, but it must be your own code that carries out transactions.

Grading

• Correctness is a must.

  The executed transactions must satisfy three properties (see below). No correctness = no passing grade for the project.

• Throughput is to go from a passing grade to the maximum grade.

  Comparison of your implementation to the reference one. Metric: #transaction commits per second.

Evaluation machine

• SuperMicro 4 sockets server
• 4 Processors: AMD Opteron 12-core Processor 6172 at 2.1 Ghz. Total: 48 cores
• 32 GB of RAM
• Ubuntu 16.04.5 LTS (GNU/Linux 4.4.0-128-generic x86_64)
• g++ and gcc versions 5.4.0 20160609
• clang and clang++ versions 3.8.0-2ubuntu4

How to write my own STM?

1. Clone/download this repository.

2. Make a local copy of/rename the template directory with your SCIPER number (e.g. $ cp -r template 123456). Be aware make is sensitive to spaces in the path/names.

3. Complete or completely rewrite (your copy of) the template with your own code.

   1. Complete/rewrite your code; only the interface should be kept identical.

   2. Compile and test locally with: path/to/directory/grading$ make build-libs run.

   3. Send your code to the measurement server for testing on the evaluation machine.

   4. Repeat any of the previous steps until you are satisfied with correctness and performance.

Sending your code for measurement

1. Zip your modified copy of the template directory.

2. Send your code for evaluation with the submit.py script. You can specify for which step your submission is.

usage: submit.py [-h] --uuid UUID [--host HOST] [--port PORT] [--step STEP] zippath

positional arguments:
  zippath      Path to a zip file containing your library code

optional arguments:
  -h, --help   show this help message and exit
  --uuid UUID  Secret user unique identifier
  --host HOST  Server hostname
  --port PORT  Server TCP port
  --step STEP  Which step of the project your submission fulfills

You can submit code as often as you want until the deadline.

Only the (correct) submission with the highest speedup will be stored persistently and considered for grading, so no need to resubmit your best implementation just before the deadline. The TAs may still ask you to provide a copy of your best implementation though (possibly after the deadline), so do not delete anything before the end of the semester.

List of structures and functions to implement

The student may write her/his library either in C11 (and above) or C++11 (and above).

First iteration (deadline 2018/11/23 23:59:59):

• The structure of your shared memory region (i.e. struct region in reference/tm.c)

• shared_t tm_create(size_t, size_t)

• void tm_destroy(shared_t)

• void* tm_start(shared_t)

• size_t tm_size(shared_t)

• size_t tm_align(shared_t)

• tx_t tm_begin(shared_t, bool)

• bool tm_end(shared_t, tx_t)

• bool tm_read(shared_t, tx_t, void const*, size_t, void*)

• bool tm_write(shared_t, tx_t, void const*, size_t, void*)

Second iteration, adding dynamic memory management (deadline 2018/12/20 23:59:59):

• alloc_t tm_alloc(shared_t, tx_t, size_t, void**)

• bool tm_free(shared_t, tx_t, void*)

See the interface specification below for more information.

Interface specification

Overview and properties

To use this Software Transactional Memory (STM) library, the user (e.g. the grading tool) first creates a new shared memory region. A shared memory region is a non-empty set of shared memory segments. Shared memory region creation and destruction are respectively managed by tm_create and tm_destroy. The content of the shared memory region is only accessed from inside a transaction, and solely by the use of the functions mentioned below.

A transaction consists of a sequence of tm_read, tm_write, tm_alloc, tm_free operations in a shared memory region, enclosed between a call to tm_begin and a call to tm_end (as well as any number of non-transactional operations in private memory). A transaction is executed on one and only one shared memory region. A transaction either commits its speculative updates to the shared memory region when tm_end is reached, or aborts its execution (discarding its speculative updates) at any time (see the reference). When a transaction is aborted, the user (i.e. the grading tool for this project) is responsible for retrying the same transaction (i.e. going back to the same tm_begin call site).

Transactions executed on the same shared region must satisfy three properties:

• Atomicity

  All speculative memory updates of a transaction are either committed or discarded as a unit.

• Consistency

  A (pending) transaction observes its own modifications, e.g. a read following one or more writes observes the last value written in program order. Transactions appear to have been committed one at a time.

• Isolation

  No speculative memory update is visible outside of their transaction, until their transaction commits.

Reference

Create (i.e. allocate + init) a new shared memory region, with one first, non-free-able allocated segment of the requested size and alignment.

• shared_t tm_create(size_t size, size_t align);

Parameter	Description
size	Size of the first shared segment of memory to allocate (in bytes), must be a positive multiple of the alignment
align	Alignment (in bytes, must be a power of 2) that the shared memory region must support

Return: Opaque shared memory region handle, invalid_shared on failure.

NB: the requested alignment in that function will be the alignment assumed in every subsequent memory operation.

NB: the first allocated segment must be initialized with 0.

 

Destroy (i.e. clean-up + free) a given shared memory region.

• void tm_destroy(shared_t shared);

Parameter	Description
shared	Shared memory region handle to destroy

NB: no concurrent call for the same shared memory region.

NB: it is guaranteed that when this function is called the associated shared memory region has not been destroyed yet.

NB: it is guaranteed that no transaction is running on the shared memory region when this function is called.

NB: the first allocated segment, along with all the segments that were allocated with tm_alloc but not freed with tm_free at the time of the call, must be freed by this function.

 

Return the start address of the first allocated segment in the shared memory region.

• void* tm_start(shared_t shared);

Parameter	Description
shared	Shared memory region to query

Return: Start address of the first allocated segment

NB: this function can be called concurrently.

NB: the returned address must be aligned on the shared region alignment.

NB: this function never fails: it must always return the address of the first allocated segment, which is not free-able.

NB: the start address returned must not be NULL.

 

Return the size (in bytes) of the first allocated segment in the shared memory region.

• size_t tm_size(shared_t shared);

Parameter	Description
shared	Shared memory region to query

Return: First allocated segment size (in bytes)

NB: this function can be called concurrently.

NB: the returned size must be aligned on the shared region alignment.

NB: this function never fails: it must always return the size of the first allocated segment, which is not free-able.

 

Return the alignment (in bytes) of the memory accesses on given shared memory region.

• size_t tm_align(shared_t shared);

Parameter	Description
shared	Shared memory region to query

Return: Alignment used globally (in bytes)

NB: this function can be called concurrently.

 

Begin a new transaction on the given shared memory region.

• tx_t tm_begin(shared_t shared, bool is_ro);

Parameter	Description
shared	Shared memory region to start a transaction on
is_ro	Whether the transaction will be read-only

Return: Opaque transaction identifier, invalid_tx on failure

NB: this function can be called concurrently.

NB: there is no concept of nested transactions, i.e. one transaction started in another transaction.

NB: if is_ro is set to true, only tm_read can be called in the begun transaction.

 

End the given transaction.

• bool tm_end(shared_t shared, tx_t tx);

Parameter	Description
shared	Shared memory region associated with the transaction
tx	Transaction to end

Return: Whether the whole transaction committed

NB: this function can be called concurrently, concurrent calls must be made with at least a different shared parameter or a different tx parameter.

NB: this function will not be called by the user (e.g. the grading tool) when any of tm_read, tm_write, tm_alloc, tm_free notifies that the transaction was aborted.

 

Read operation in the given transaction, source in the shared region and target in a private region.

• bool tm_read(shared_t shared, tx_t tx, void const* source, size_t size, void* target);

Parameter	Description
shared	Shared memory region associated with the transaction
tx	Transaction to use
source	Source (aligned) start address (in shared memory)
size	Length to copy (in bytes)
target	Target (aligned) start address (in private memory)

Return: Whether the whole transaction can continue

NB: this function can be called concurrently, concurrent calls must be made with at least a different shared parameter or a different tx parameter.

NB: the private buffer target can only be dereferenced for the duration of the call.

NB: the length size must be a positive multiple of the shared memory region's alignment, otherwise the behavior is undefined.

NB: the length of the buffers source and target must be at least size, otherwise the behavior is undefined.

NB: the source and target addresses must be a positive multiple of the shared memory region's alignment, otherwise the behavior is undefined.

 

Write operation in the given transaction, source in a private region and target in the shared region.

• bool tm_write(shared_t shared, tx_t tx, void const* source, size_t size, void* target);

Parameter	Description
shared	Shared memory region associated with the transaction
tx	Transaction to use
source	Source (aligned) start address (in private memory)
size	Length to copy (in bytes)
target	Target (aligned) start address (in shared memory)

Return: Whether the whole transaction can continue

NB: this function can be called concurrently, concurrent calls must be made with at least a different shared parameter or a different tx parameter.

NB: the private buffer source can only be dereferenced for the duration of the call.

NB: the length size must be a positive multiple of the shared memory region's alignment, otherwise the behavior is undefined.

NB: the length of the buffers source and target must be at least size, otherwise the behavior is undefined.

NB: the source and target addresses must be a positive multiple of the shared memory region's alignment, otherwise the behavior is undefined.

 

Memory allocation in the given transaction.

• alloc_t tm_alloc(shared_t shared, tx_t tx, size_t size, void** target);

Parameter	Description
shared	Shared memory region associated with the transaction
tx	Transaction to use
size	Allocation requested size (in bytes)
target	Pointer in private memory receiving the address of the first byte of the newly allocated, aligned segment

Return: One of: success_alloc (allocation was successful and transaction can continue), abort_alloc (transaction was aborted) and nomem_alloc (memory allocation failed)

NB: this function can be called concurrently, concurrent calls must be made with at least a different shared parameter or a different tx parameter.

NB: the pointer target can only be dereferenced for the duration of the call.

NB: the value of *target is defined only if success_alloc was returned, and undefined otherwise.

NB: the value of *target after the call if success_alloc was returned must not be NULL.

NB: when nomem_alloc is returned, the transaction is not aborted.

NB: the allocated segment must be initialized with 0.

NB: only tm_free must be used to free the allocated segment.

NB: the length size must be a positive multiple of the shared memory region's alignment, otherwise the behavior is undefined.

 

Memory freeing in the given transaction.

• bool tm_free(shared_t shared, tx_t tx, void* target);

Parameter	Description
shared	Shared memory region associated with the transaction
tx	Transaction to use
target	Address of the first byte of the previously allocated segment (with tm_alloc only) to deallocate

Return: Whether the whole transaction can continue

NB: this function can be called concurrently, concurrent calls must be made with at least a different shared parameter or a different tx parameter.

NB: this function must not be called with target as the first allocated segment (the address returned by tm_start).