reshadman · August 29, 2015 14:03
diff --git a/Laravel_queue_work_benchmark b/Laravel_queue_work_benchmark
 Machine:
 Core i3 CPU 3Ghz
 RAM 6GB
 Hard Disk , SataII ~7000 rpm

 Logic:
 
 Laravel has  queue:work in daemon and non-daemon mode, the non-daemon mode takes a job from Queue processes it and then the framework is shut down.
 Test logic relies on multi processing + async request using non-daemon mode,

 I used http://supervisord.org/ as process mangaer for making me able to run multiple "queue:work" at once and restart them one by one when they are shut down. I did some try & error and made two Superviserd processes which each run 40 "queue:work" and then restart them one by one when they are shut down.

 In my machine I used "sleep(10)" instead of making real requests because of the lack of internet connection speed ( if I used my local machine for request backends It would effect the performance ) I considered that each request may tak 1-5 secs , when we are sending them in async mode the whole process for sending multiple requests takes 10 seconds and multipled by a safety factor we consider it 25 seconds , I pushed 1200 jobs to queue with the following code and above logic:

 https://gist.github.com/bigsinoos/2c4207c8b559e5a9b35a

 The above caused ~90% CPU Usage eternally (We have 80 queue:work process at once)

 The test results are:

 1200 times "queue:work" were run ( Number of our pushes) in 144 seconds an average of 8 run per second.

 In each run we waited 10 seconds to mock that we can send 25 async requests in this time. so we have mocked 1200 * 25 = 30000 request in 144 seconds which means 208 requests per sec.

 I also wrote an async code with the following to test in real mode and the result satisfied me with maximum possible connections based on my Internet speed ( about 3 request per sec):

 https://gist.github.com/bigsinoos/5b109bc3c26ce372276b

 The advantages of queue:work instead of queue:work --daemon is that the daemon mode is not really daemon it is shut down when it eats the memory limit (Behzad: the reason that we couldn't reach more than ~38000 jobs).
	Machine:
	Core i3 CPU 3Ghz
	RAM 6GB
	Hard Disk , SataII ~7000 rpm

	Logic:

	Laravel has queue:work in daemon and non-daemon mode, the non-daemon mode takes a job from Queue processes it and then the framework is shut down.
	Test logic relies on multi processing + async request using non-daemon mode,

	I used http://supervisord.org/ as process mangaer for making me able to run multiple "queue:work" at once and restart them one by one when they are shut down. I did some try & error and made two Superviserd processes which each run 40 "queue:work" and then restart them one by one when they are shut down.

	In my machine I used "sleep(10)" instead of making real requests because of the lack of internet connection speed ( if I used my local machine for request backends It would effect the performance ) I considered that each request may tak 1-5 secs , when we are sending them in async mode the whole process for sending multiple requests takes 10 seconds and multipled by a safety factor we consider it 25 seconds , I pushed 1200 jobs to queue with the following code and above logic:

	https://gist.github.com/bigsinoos/2c4207c8b559e5a9b35a

	The above caused ~90% CPU Usage eternally (We have 80 queue:work process at once)

	The test results are:

	1200 times "queue:work" were run ( Number of our pushes) in 144 seconds an average of 8 run per second.

	In each run we waited 10 seconds to mock that we can send 25 async requests in this time. so we have mocked 1200 * 25 = 30000 request in 144 seconds which means 208 requests per sec.

	I also wrote an async code with the following to test in real mode and the result satisfied me with maximum possible connections based on my Internet speed ( about 3 request per sec):

	https://gist.github.com/bigsinoos/5b109bc3c26ce372276b

	The advantages of queue:work instead of queue:work --daemon is that the daemon mode is not really daemon it is shut down when it eats the memory limit (Behzad: the reason that we couldn't reach more than ~38000 jobs).