The Synchronous Blog

A blog about reactive programming languages.

Posts Tagged ‘dynamic-allocation

Stress-testing Céu with SDL

with one comment

For a glimpse at the runtime costs of Céu, I created a simple application with SDL that animates 10000 rectangles on the screen.

My hypothesis is that lines of execution in Céu (a.k.a trails) are cheap and should be used carelessly, even for simple activities such as waiting for a single event and terminating. This way, the testing application in Céu has a peek of 40000 trails active at the same time, which are all visited by the scheduler on every frame.

I compare the implementation in Céu with a simpler application in C to serve as a benchmark.

The application in raw C is damn simple:

  • Declares a vector of 10000 rectangles.
  • Initializes each to 10×10 width/height, a random (x,y) position, and a random x velocity.
  • Animates them continuously, restarting from left when reaching the right side of the screen.
  • Achieved FPS = 120 .

The video shows the expected behavior, and the code, the full implementation.
(The video was adapted to 5000 rectangles to enable screencasting).

The application in Céu is considerably more complex (video and code):

  • The program has 5 lines of execution:
    1. On SDL_QUIT event: terminates the application.
    2. Every frame: clears the screen.
    3. Every 40ms: creates 30 new Rect organisms.
      • after 30s: kills all of them at once and restart the process
    4. Every frame: updates the screen.
    5. Every second: calculates the FPS.

    (technically they are 7 lines of execution because (3) and (5) are split in two)

  • The class Rect has 4 lines of execution:
    1. Every 200ms: varies the rectangle width, height, r, g, b (with a random epsilon).
    2. Every frame: animates the rectangle respecting its velocity, terminating when it reaches the end of the screen.
    3. Every frame: redraws the rectangle on screen.
    4. On termination: runs a simple finalizer (which maintains the number of organisms alive).
  • Achieved FPS = 107 . (when 10000 rectangles are on screen)

The main differences between the two implementations are marked in bold above and are summarized as follows:

  • The rectangles in Céu are allocated dynamically: in the peek, we have 750 rectangles being allocated (in the left side) and reclaimed (in the right side) every second.
  • Each rectangle on screen (10000 at most) has 4 active trails, all of which are visited by the scheduler every frame (4 x 10000 = 40000 lightweight threads active at the same time).
  • Finally, every 30 seconds the program restarts the spawning loop, killing all 10000 active organisms at once, running the associated finalizers, and reclaiming all memory.

Céu is a source-to-source compiler that generates single-threaded ansi-C code. This way, the generated code for the test application, although incomprehensible, can be compiled with gcc and tested in any computer that has SDL2:

  % gcc download_code.c -lSDL2
.

Conclusion

The simple SDL application in C is used as a benchmark to compare to a more complex application written in Céu, which shows a decrease around 10% in FPS.

Although the application in C could be re-implemented to achieve the same functionality (for a more strict comparison), the initial numbers are already satisfactory and make the point that the trail-oriented programming style of Céu is realistic, even in a highly dynamic scenario.

http://www.ceu-lang.org/

Written by francisco

June 6, 2013 at 12:04 am

Dynamic Applications in Céu (2/2)

with 3 comments

This is the follow up to the previous post on creating dynamic applications in Céu.

As introduced there, organisms are the main abstraction mechanism in Céu, reconciling objects and lines of execution in a single construct. The example below, extracted from the previous post, creates two instances of a class that prints the “Hello World!” message every second:

  class HelloWorld with
     var int id;
  do
     every 1s do
         _printf("[%d] Hello world!\n",
                  this.id);
     end
  end

  var HelloWorld hello1, hello2;
  hello1.id = 1;
  hello2.id = 2;
  await FOREVER;
.

One thing that arises in the code is how the organisms hello1 and hello2 are declared as local variables, using the same syntax of a normal variable declaration (e.g. “var int x=0”). This implies that an organism follows the standard scoping rules of conventional imperative languages, i.e., its memory is reclaimed when its enclosing block goes out of scope. Additionally, all of its lines of execution are seamlessly killed.

Céu supports three different ways to manage the life cycle of organisms automatically: local variables, anonymous allocations, and named allocations.

Local variables

The simplest way to manage organisms is to declare them as local variables, as shown in the previous example. As in the example, the two organisms don’t go out of scope, let’s include an explicit block to declare hello2 so that it goes out of scope after 5 seconds:

  var HelloWorld hello1;
  hello1.id = 1;
  do
     var HelloWorld hello2;
     hello2.id = 2;
     await 5s;
  end
  await FOREVER;
.

The organism hello2 runs in parallel with the do-end construct and is active while the block is in scope. After 5 seconds, the block goes out of scope and kills hello2, also reclaiming all its memory. As an outcome, the message “[2] Hello world!” stops to show up in the video.

The order of execution between blocks and organisms is determined: code inside a block executes before the organisms declared inside it. This way, the do-end has priority over hello1 (they are both in the top-level block), while await 5 has priority over hello2 (they are both inside the do-end). The order the messages appear in the video is correct, hello2 always awakes before hello1. Also, hello2 is killed before printing for the 5th time, because the await 5s has higher priority and terminates the block before hello2 has the chance to execute for the last time.

Regarding memory usage, a local organism has an associated slot in the “stack” of its enclosing block, which is calculated at compile time. Blocks in sequence can share memory slots. Local organisms do not incur overhead for calls to malloc/free.

Anonymous allocations

True dynamic applications often need to create an arbitrary number of entities while the application is running. Céu supports a spawn command to dynamically create and execute  a new organism:

  do
      var int i = 1;
      every 1s do
          spawn HelloWorld with
              this.id = i;
          end;
          i = i + 1;
      end
  end
.

We now spawn a new organism every second passing a different id in the constructor (the code between with-end). We can see in the video that every second a new organism starts printing its message on the screen. Again, the printing order is deterministic and never changes.

Note that the spawned organisms are anonymous, because there’s no way to refer to them after they are created. Anonymous organisms are useful when the interaction with the block that creates them happens only in the constructor, as in the example above.

Note also the we use an enclosing do-end apparently with no purpose in the code. However, in order to also provide seamless memory reclamation for anonymous organisms, a spawn statement must be enclosed by a do-end that defines the scope of its instances. This way, when the do-end block goes out of scope, all organisms are reclaimed in the same way local organisms are.

In the next example, we surround the previous code with a par/or that restarts the outer loop after 5 seconds:

  loop do
      par/or do
          await 5s;
      with
          do
              var int i = 1;
              every 1s do
                  spawn HelloWorld with
                      this.id = i;
                  end;
                  i = i + 1;
              end
          end
      end
  end
.

Now, after creating new instances during 5 seconds, the par/or terminates the do-end scope and all organisms are killed. The loop makes this pattern to execute continuously.

An anonymous organism can also be safely reclaimed when its body terminates, given that no one can refer and access its fields.

Named allocations

The most flexible way to deal with dynamic organisms is through the new statement, which not only spawns a new organism but also returns a reference to it:

  do
      var HelloWorld* hello = new HelloWorld;
      hello:id = 1;
      ... // some more code
  end
.

In the example, the returned reference is assigned to the variable hello, which is of type HelloWorld* (a pointer to the class). The organism can be manipulated through the colon operator (:), which is equivalent to the arrow operator in C (->).

A named organism is automatically reclaimed when the block holding the pointer it was first assigned goes out of scope. In the example, when the do-end block in which the hello pointer is declared goes out of scope, the referred instance is reclaimed.

For safety reasons, Céu does not allow a pointer to “escape” to an outer block. Without this precaution, a reference could last longer than the organism it points, yielding a dangling pointer in the program. In the following example, both the assignment to outer and the call to _cfunc are refused, given that their scope are broader the that of variable hello:

  var HelloWorld* outer;
  do
      var HelloWorld* hello = new HelloWorld;
      hello:id = 1;
      outer = hello;          // this assignment is refused at compile time
      _cfunc(hello);          // this call is refused at compile time
      ... // some more code
  end
.

In order to compile this code, we need to include finalizers to properly handle the organism going out of scope:

  var HelloWorld* outer;
  do
      var HelloWorld* hello = new HelloWorld;
      hello:id = 1;

      finalize
          outer = hello; // outer > hello
      with
          ...            // this code is executed just before do-end goes out of scope
      end

      _cfunc(hello)      // _cfunc > hello (_cfunc is a global function)
          finalize with
              ...        // this code is executed just before do-end goes out of scope
          end;
  end
.

A finalize block is tied to the scope of the dangerous pointer and gets executed automatically just before the associated block terminates. This way, programmers have means to handle the organism being reclaimed in a safe way.

Conclusion

Céu support three different ways to deal with dynamic allocation of organisms:

  • Local organisms should be used when the number of instances is known a priori.
  • Anonymous allocations should be used when the number of instances is arbitrary and the program only interacts with them at instantiation time.
  • Named allocations are the most flexible, but should only be used when the first two methods don’t apply.

In all cases, memory and trails reclamation is handled seamlessly, without programming efforts.

In practice, given that organisms have lines of execution and can react to the environment by themselves, anonymous organisms should be preferred over named organisms in order to avoid dealing with references explicitly.

In the next post, I’ll show a simple evaluation of the runtime costs of organisms in Céu.

http://www.ceu-lang.org/

Written by francisco

June 5, 2013 at 8:51 pm

Dynamic Applications in Céu (1/2)

with 4 comments

The basic prerequisite to build dynamic applications is language support to deal with abstractions and code reuse. Programming languages provide a multitude of abstraction mechanisms, from simple abstract data types, to OO classes. Regarding an abstraction, an effective mechanism should provide means to deal with at least the following points:

  • Hide its internal implementation details.
  • Expose a uniform programming interface to manipulate it.
  • Control its life cycle.

As an example, to build an ADT in C, one can define a struct, hide it with a typedef, expose functions to manipulate it, and control instances with local variables or malloc/free. Classes extend ADTs with richer mechanisms such as inheritance and polymorphism. Furthermore, the life cycle of an object is typically controlled automatically through a garbage collector.

Céu organisms

Abstractions in Céu are created through organisms, which basically reconcile threads and objects into a single concept:

  • An organism has intrinsic execution, being able to react to the environment on its own.
  • An organism exposes properties and actions in order to interact with other organisms during its life cycle.

Like an object, an organism exposes properties and methods (events in Céu) that can be accessed and invoked (emitted in Céu) by other instances. Like a thread, an organism has its own line(s) of execution, with persistent local variables and execution state.
In contrast, an object method call typically shares the same execution context with its calling method. Likewise, a thread does not expose fields or methods.

An example

The program below defines the class HelloWorld and executes two instances of it:

  class HelloWorld with
     var int id;   // organism interface
  do               // organism body
     every 1s do
         _printf("[%d] Hello world!\n",
                  this.id);
     end
  end

  var HelloWorld hello1, hello2;
  hello1.id = 1;
  hello2.id = 2;
  await FOREVER;
.

The behavior can be visualized in the video on the right. The top-level code creates two instances of the class HelloWorld, initializes the exposed id fields, and then awaits forever. As organisms have “life”, the two instances react to the environment autonomously, printing the “Hello world!” message every second.

Note in the example that organisms are simply declared as normal variables, which are automatically spawned by the language runtime to execute in parallel with its enclosing block.

In the following variation, we add the event stop in the class interface and include another line of execution in the organism body:

  class HelloWorld with
     var   int  id;
     event void stop;
  do
     par/or do
         every 1s do
             _printf("[%d] Hello world!\n",
                      this.id);
         end
     with
         await this.stop;
     end
  end

  var HelloWorld hello1, hello2;
  hello1.id = 1;
  hello2.id = 2;

  await 3s500ms;
  emit hello1.stop;
  hello2.id = 5;
  await 2s;
  emit hello2.stop;

  await FOREVER;
.

Now, besides printing the message every second, each organism also waits for the event stop in parallel. The par/or construct splits the running line of execution in two, rejoining when any of them terminate. (Céu also provides the par/and construct.)

After the top-level code instantiates the two organisms, it waits 3s500ms before taking the actions in sequence. At this point, the program has 5 active lines of execution: 1 in the top-level and 2 for each of the instances. Each organism prints its message 3 times before the top-level awakes from 3s500ms.

Then, the top-level emits the stop event to the first organism, which awakes and terminates. It also changes the id of the second organism and waits more 2s. During this period the second organism prints its message 2 times more (now with the id 5).

Note that although the first organism terminated its body, its reference hello1 is still visible. This way, the organism is still alive and its fields can be accessed normally (but now resembling a “dead” C struct).

Execution model

Lines of execution in Céu are known as trails and differ from threads in the very fundamental characteristic of how they are scheduled.

Céu is a synchronous language based on Esterel, in which lines of execution advance together with a unique global notion of time.
In practical terms, this means that Céu can provide seamless lock-free shared-memory concurrency. It also means that programs are deterministic and have reproducible execution. As a tradeoff, concurrency in Céu is not suitable for algorithmic-intensive activities as there is no automatic preemption among trails.

In contrast, asynchronous models have time independence among lines of execution, but either require synchronization primitives to acquire shared resources (e.g. locks and semaphores in pthreads), or completely forbid shared access in favor of message passing (e.g processes and channels in actor-based languages). In both cases, ensuring deterministic execution requires considerable programming efforts.

The post entitled “The case for synchronous concurrency” illustrates these differences in practical terms with an example.

The synchronous model of Céu is presented in more depth in these videos.
The videos also show organisms in action together with the SDL graphical library.

Conclusion

Céu organisms reconcile objects and threads in a single abstraction mechanism.

Classes specify the behavior of organisms, hiding implementation details and exposing an interface in which they can be manipulated by other organisms.

In the next post, I’ll show how Céu can control the life cycle of organisms with lexical scope in three different ways: local variables, named allocation, and anonymous allocation.

http://www.ceu-lang.org/

Written by francisco

May 22, 2013 at 6:59 pm

Dynamic Memory Management in Embedded Systems

with 4 comments

Dynamic functionality in embedded systems is usually discouraged due to resource constraints.
However, some types of applications inherently require memory allocation.

As an example, protocols in sensor networks typically forward messages through nodes at a non-deterministic rate, given that the number of neighbors and transmission periods can vary.
Hence, many protocols require dynamic memory management to hold receiving messages until they are successfully forwarded.

A simple FIFO queue might not be always optimal because forwarding a message may involve multiple steps with delays (e.g. transmission acknowledgments).
In such scenario, the protocol would rather handle multiple messages at the same time, raising the possibility of a message received later be discarded first.

Unfortunately, out-of-the-box dynamic memory schemes, such as malloc/free, are not suitable for embedded systems which have quite different requirements in comparison to standard desktop systems.

Follows a list of issues concerning memory management schemes in embedded systems:

  1. Memory corruption
    Many embedded systems lack memory protection, and continuous allocations in the heap may end up corrupting the stack (and vice versa).
  2. Run-time overhead
    Memory management requires extra run-time bookkeeping. Also, in the context of embedded systems, a predictable execution model can be even more important than the fastest scheme on the average.
  3. Metadata overhead
    Metadata used by the memory manager can spend precious bytes (e.g. linked lists of free blocks).
  4. Memory fragmentation
    For constrained memory platforms, unusable holes between and inside allocated blocks (external and internal fragmentation, respectively) can waste a big percentage of available memory.
  5. Unreproducible execution
    Successive executions of the same program may allocate memory in different ways, possibly leading to different outcomes (e.g. an allocation fail).
  6. Deallocation hazards
    Properly deallocating memory is far from trivial. A missed deallocation leads to a memory leak that wastes memory, while deallocating a memory block still in use leads to a dangling pointer that will eventually crash the application.

As the C standard is loose about these issues, out-of-the-box malloc/free can perform bad in all items.
Furthermore, deallocation hazards are inherent in schemes that require an explicit free operation.

Garbage collected systems eliminate deallocation hazards, but may incur unacceptable run-time overheads.

Memory Pools

Embedded systems usually rely on memory pools to manage dynamic memory.
In the context of sensor networks, both TinyOS and Coniki OSes offer and promote the use of memory pools (through Pool and MEMB, respectively).

A memory pool allocates N predefined fixed-sized blocks of memory that can be used by the application.
Most of the raised concerns are alleviated with this scheme:

  1. Because the allocation is static, the maximum amount of memory is known at compile time, reducing considerably the risk of memory corruption.
  2. The run-time overhead is minimal as implementations use simple arrays to hold the memory blocks. For instance, in the TinyOS implementation both allocation and deallocation are O(1).
  3. TinyOS’ Pools use an auxiliary vector of size N to hold pointers for free blocks.
  4. Regardless of different allocation patterns in applications, memory pools will always guarantee the minimal N of memory blocks. Hence, external fragmentation is non-existent. Internal fragmentation, however, can be an issue and is discussed below.
  5. Given that the memory operations are simple and handle fixed-size blocks, the execution is always deterministic and predictable.
  6. Memory pools are still manipulated through malloc/free-like operations. Hence, all challenges to properly deallocate memory still hold.

Internal fragmentation occurs when an allocated memory block is bigger than the requested size.
Given that memory pools can only handle fixed-size blocks, any allocation that requests smaller blocks will contain internal fragmentation (allocation of bigger blocks always fail).
That said, embedded applications are usually simple and contain only one or two different object units that require dynamic allocation.
This way, an application will use a different memory pool for each kind of object, thus also eliminating internal fragmentation.

Conclusion

Memory pools are the way to go for dynamic allocation in embedded systems.

They offer memory compactness, efficient and small operations, and predictable execution.

However, programming dynamic applications is still hard and error prone, given that missing and wrong deallocations may lead to memory leaks and subtle crashes.

In the next post I will show how Céu offers safer and higher level mechanisms for dynamic applications, while using memory pools transparently under the hoods.

Written by francisco

May 20, 2013 at 7:45 pm