Project3

This page for documentation Project3: Wait-Free Snapshot

Review

Implementation of Wait-Free Sanpshot

The latest C++ standard features

• If statement with initializer
• Generic lambda-capture initializers
• valarray

Classes

The following classes are implemented. Click on the link to see my implementation.

• atomic::Snapshot
• atomic::Snapshot::StampedSnap
• atomic::Register
• arg::Parser
• thread::Pool
• util::Random

See atomic::Snapshot, major implementation of wait-free atomic snapshot. This class implements the specifications required by the assignment. Main thread just push snapshot random update to thread::Pool.

The following code is the main content of the main thread.

size_t n = parser.get<size_t>("N");
size_t t = parser.get<size_t>("T", 60);

// main thread scope
{
	atomic::Snapshot<int> snapshot(n);		// snapshot instance
	util::Random<int> random;				// util::Random generator
	thread::Pool pool(n);					// thread::Pool for multi thread
	std::queue<std::future<void>> tasks;		// thread::Pool tasks
		
	// time guard to run only the set time  
	std::thread time_guard([&pool, &t]() {
		std::this_thread::sleep_for(std::chrono::seconds(t));
		pool.terminate();
	});

	// push tasks to thread::Pool until set time
	while (!pool.is_stop()) {
		pool.push([&pool, &snapshot, &random](size_t tid) {
			snapshot.update(tid, random.next());
		});
	}
		
	time_guard.join();
}

parser instance parse input argument thread_number and task time as optional. snapshot is Wait-Free Snapshot implementation instance, random is distributed random generator and pool is thread pool to run multiple tasks as parallel. time_guard ensures that the thread::Pool will only work for a specified amount of time. Below while loop update snapshot randomly.

Performance

Checked in many clients. Here is my clients specs.

• Macbook pro 13" 2016: i5-5257U 2.7GHz 2C4T, g++ 5.4 based on LLVM version 9.0.0
• Linux Server: i7-7700k 4.2GHz 4C8T, g++ 5.4
• ML Server: i7-5930K 3.5GHz 6C12T (and many GPUs, but no effect) g++ 5.4
• SCS Lab Server: E5-2697 2.7GHz 24C48T g++ 5.4

And each results is... (all result executed while 60 seconds)

I do not understand why SCSLab Server's return low performance than others.

Review

I expect more performance on many-core-server. But not, I can't understand why. (Maybe, Single core speed make difference... maybe...)

Lower cores is more faster because the number of threads increases, the chance of getting a clean cut is reduced.

Waiting contributes!

you can comments and make merge requests to fix my codes. 😄