Re: [Jython-users] Concurrency question

[Frank C. <fc...@pu...>]

Thank you for the tips. I'll change the script, run the tests and post the
results here for everyone's info.

On a related topic, I ran this script on Jython 2.1 on RedHat Linux 7.2. In
this environment is start_new_thread creating a new Java thread? A new
native Linux process?

-Frank

-- 
Frank Cohen, CEO, PushToTest, www.pushtotest.com, phone: 408 374 7426
Come to PushToTest for free open-source Active Security solutions that test,
monitor and automate Web Service systems for functionality, scalability and
performance.


> From: Ype Kingma <yk...@xs...>
> Date: Sun, 7 Apr 2002 16:48:30 +0100
> To: jyt...@li...
> Subject: Re: [Jython-users] Concurrency question
> 
> [posting the reply already sent to Frank]
> 
> Frank,
> 
> <snip>
> 
>> I wrote a script that creates 1001 threads. Each thread creates an HTTP
>> connection object and requests a random page from a selection of 10 possible
>> pages. The thread records the starting time and ending time. Once all 1001
>> agents have completed I have a simple evaluation function to count the
>> number of threads with start and end times that overlap.
>> 
>> I then calculate the total possible overlapping threads as each thread could
>> overlap every other but not itself. For 1001 threads that would be 1001 x
>> 1001 - 1001 = 1,001,000 possible overlaps.
>> 
>> So if I count a total of 991,246 overlaps out of a possible 1,001,000 then I
>> can say that 99.03% of the agents ran concurrently.
>> 
>> Does this method make sense?
> 
> Due to the symmettry of overlaps
> you could make the final computation roughly twice as fast
> by using the range(i,  ...) instead of range(1, ...) for j.
> You would then also have to divide the max. nr. of overlaps by two
> ( n * n - n == n * (n - 1) is always even.)
> 
> You might also consider using range(0, ...) for i,
> (python indexes lists from zero.)
> 
> The appends to stime and duration are not thread safe.
> You might collect the stats with something like:
> 
> stats = [] # init
> 
> stats.append((starttime, duration)) # append a tuple for each thread
> 
> and get them back afterwards by:
> 
> for i in range(len(stats)):
> st1, dr1 = stats[i]
> for j in range(i + 1, len(stats)):
> st2, dr2 = stats[j]
> ....       
> 
> I'd also add
> 
> assert agentcount >= 2
> 
> somewhere.
> 
> 
> Have fun,
> Ype
> 
> 
>> -Frank
>> 
>> --
>> 
>> # Test script written in TestMaker - an open-source utility for testing
>> # Web Services for functionality, scalability and performance. Download
>> # TestMaker for free at http://www.PushToTest.com
>> 
>> # The from statements import objects from the TestMaker objects (TOOL.)
>> 
>> print "Concurrency is a test to learn how many agents a machine may run
>> concurrently."
>> print
>> print "This version of Concurrency creates local threads that make HTTP"
>> print "requests to the remote Apache host. The threads records their start"
>> print "time and duration. The values are compared to determine the number"
>> print "of threads that overlapped."
>> print
>> print "(c) 2002 Frank Cohen. All rights reserved."
>> print "for details see http://www.PushToTest.com"
>> print
>> 
>> import thread
>> import time
>> import random
>> import sys
>> from com.pushtotest.tool.protocolhandler import Header
>> from com.pushtotest.tool.protocolhandler import HTTPHeader
>> from com.pushtotest.tool.protocolhandler import Body
>> from com.pushtotest.tool.protocolhandler import HTTPBody
>> from com.pushtotest.tool.protocolhandler import HTTPProtocol
>> from com.pushtotest.tool.response import Response
>> 
>> agentcount = 1000
>> min = 10.0        # shortest time a thread took to run
>> avg = 0.0         # average time
>> max = 0.0         # maximum time it took for a thread to run
>> count = 0.0       # count of threads reporting results
>> runtime = time.time()
>> stime = [ 0 ]
>> dur = [ 0 ]
>> 
>> # This thread is called and returned
>> 
>> def dominic( me ):
>> 
>> global min, avg, max, count
>> 
>> start = time.time();
>> 
>> # Create the needed objects to communicate with the host
>> httphandler = HTTPProtocol()
>> httphandler.setHost( "192.168.1.100" )
>> 
>> httphandler.setPath( "frank" + random.choice("0123456789") + ".html" )
>> 
>> # Request the document from the host
>> response = httphandler.connect()
>> 
>> stop = time.time()
>> 
>> duration = stop - start
>> 
>> if duration<0:
>> print "gotcha"
>> 
>> if duration<min:
>> min=duration
>> 
>> if duration>max:
>> max=duration
>> 
>> avg = avg + duration
>> count = count + 1
>> 
>> stime.append( start )
>> dur.append( duration )
>> 
>> print "Initializing ", agentcount, "threads"
>> 
>> # reset the values
>> min = 10.0
>> avg = 0.0
>> max = 0.0
>> count = 1.0
>> 
>> for j in range( agentcount ):
>> 
>> #start_new_thread is a handy feature of Python
>> #it calls a function in a new thread, when the
>> #function is complete the thread terminates silently
>> 
>> thread.start_new_thread( dominic, ( 0, ) )
>> 
>> print "running"
>> 
>> while count < ( agentcount-100 ):
>> time.sleep(1)
>> print "Min = %9.6f Max = %9.6f Avg = %9.6f count=%d duration=%d" % (min,
>> max, avg/count, count, (time.time()-runtime) )
>> 
>> ccount = 0
>> 
>> print "calculating concurrent threads"
>> 
>> for i in range( 1, len( stime ) ):
>> for j in range( 1, len( stime ) ):
>> if i != j:
>> start1 = stime[i]
>> end1 = start1 + dur[i]
>> start2 = stime[j]
>> end2 = start2 + dur[j]
>> if ( ( ( start2 > start1 ) and (  end1 <= end2 ) ) or ( ( start1
>>> start2 ) and ( end2 <= end1 ) ) ):
>> ccount = ccount + 1
>> 
>> maxoverlaps = (len(stime)*len(stime))-len(stime)
>> curper = ( float( ccount) / maxoverlaps ) * 100
>> print "Total number of agents counted: ", len(stime)
>> print "Overlap counts: ", ccount, " out of a total possible overlaps ",
>> maxoverlaps
>> print "percentage of agents that ran concurrently: %3.2f" % curper
>> print "Total duration: ", time.time() - runtime
>> 
>> 
>> 
>> --
>> Frank Cohen, CEO, PushToTest, www.pushtotest.com, phone: 408 374 7426
>> Come to PushToTest for free open-source Active Security solutions that test,
>> monitor and automate Web Service systems for functionality, scalability and
>> performance.
>> 
>> 
>> 
>> _______________________________________________
>> Jython-users mailing list
>> Jyt...@li...
>> https://lists.sourceforge.net/lists/listinfo/jython-users
> 
> 
> -- 
> 
> _______________________________________________
> Jython-users mailing list
> Jyt...@li...
> https://lists.sourceforge.net/lists/listinfo/jython-users