Implements several concurrent data structures as well as some other thread-safe constructs. Supported targets include Neko and CPP so far.
Haxe
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
src
test
LICENSE
README.md
example.jpg
haxelib.json

README.md

haxe-concurrency is a library to provide high level thread-safe data structures for all targets that support shared memory concurrency. Also included is . Supported targets are neko and c++.

Data Structures

  • CopyOnWriteArray - A data structure for read-heavy storage. As the name suggests, the array is copied when a modification is performed.
  • ConcurrentMap - Provides a Map which is thread safe.

Performance

Here are some performance tests run on a win7-64bit i7 Dell Inspiron laptop (8 CPUs) under the neko target:

Comparison to Non-Concurrent Data Structures

  • Array Write MOD 1000: 83 ns
  • CopyOnWriteArray Write MOD 1000: 3487 ns
  • Array Read MOD 1000: 67 ns
  • CopyOnWriteArray Read MOD 1000: 130 ns
  • IntMap Write: 149 ns
  • ConcurrentIntMap Write: 342 ns
  • IntMap Read: 66 ns
  • ConcurrentIntMap Read: 236 ns

Throughput Speed Testing

  • Array Read 20M items: 1363 ms
  • CopyOnWriteArray Read 20M items (2 Threads): 1431 ms
  • CopyOnWriteArray Read 20M items (4 Threads): 813 ms
  • CopyOnWriteArray Read 20M items (8 Threads): 597 ms
  • CopyOnWriteArray Read 20M items (16 Threads): 369 ms
  • CopyOnWriteArray Read 20M items (32 Threads): 182 ms

Extras

  • neko.net.MultiThreadedServer - A fully concurrent server which is similar to neko.net.ThreadServer, except the application logic is concurrent as well.
  • sys.db.PooledConnection - A thread-safe abstraction layer for accessing databases. Connections are pooled and automatically restarted when a failure occurs.

PooledConnection Usage

Usage is very simple. Instead of creating your sys.db.Mysql.connect() or sys.db.Sqlite.read() connections just create a new PooledConnection and pass in a factory constructor as shown below:

//Create a pooled connection
var cnx = new sys.db.PooledConnection(function ():sys.db.Connection {
	return sys.db.Mysql.connect( { host:HOST, port:PORT, user:USER, pass:PASS, socket:null, database:DATABASE } );
}, NUM_CONNECTIONS);

//Use with SPOD!
//A point of caution - SPOD uses a global object cache by default so you will need to deal with this when using in a multithreaded application
sys.db.Manager.cnx = cnx;

Concurrent Application Debugger

You can use the Concurrent Application Debugger (CAD) to debug concurrent applications. Included in the cad package is cad.Debugger which will listen on a port you choose and output the current state of all threads within the application. Two formats are available: Full HTML and JSON. The HTML output is nice if you want a quick/pretty display from a browser. The JSON output is nicer for custom clients.

Here is an example:

cad.Debugger.listen(new sys.net.Host("0.0.0.0"), 9308, true);	//Last arg means use HTML output

CAD

To opt-in you need to make sure all Mutex/Locks/Deques are using the wrapper classes of cad.Mutex, cad.Lock, cad.Deque. This is the only modification required to your existing code base. You then compile your code with the '-D cad' compiler flag to turn on CAD.

The nice thing about this is that when you are done and don't want the overhead of CAD anymore you can just remove the compiler flag, and cad.* typedefs to the std classes. Also the Debugger will just ignore the listen() call.