X-Recipient:	archive-cygwin AT delorie DOT com
X-SWARE-Spam-Status:	No, hits=-0.9 required=5.0 tests=AWL,BAYES_00,DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,FREEMAIL_FROM,KAM_TIME,KHOP_RCVD_TRUST,RCVD_IN_DNSWL_LOW,RCVD_IN_HOSTKARMA_YE
X-Spam-Check-By:	sourceware.org
MIME-Version:	1.0
Date:	Thu, 9 Aug 2012 18:07:06 +1000
Message-ID:	<CAFY+UmUspDUbo4JaW9NDG1zf1oY2rgS5g3zO7q-qnby3tJqrSg@mail.gmail.com>
Subject:	App runs 8x slower on dual core machine (with test case to replicate issue)
From:	Zach Saw <zach DOT saw AT gmail DOT com>
To:	cygwin AT cygwin DOT com
Mailing-List:	contact cygwin-help AT cygwin DOT com; run by ezmlm
List-Id:	<cygwin.cygwin.com>
List-Subscribe:	<mailto:cygwin-subscribe AT cygwin DOT com>
List-Archive:	<http://sourceware.org/ml/cygwin/>
List-Post:	<mailto:cygwin AT cygwin DOT com>
List-Help:	<mailto:cygwin-help AT cygwin DOT com>, <http://sourceware.org/ml/#faqs>
Sender:	cygwin-owner AT cygwin DOT com
Mail-Followup-To:	cygwin AT cygwin DOT com
Delivered-To:	mailing list cygwin AT cygwin DOT com

Hi all,

While trying to compile my multithreaded app written for Linux on
Cygwin to run on Windows, I discovered the app would perform 8x slower
on a machine with the same specs. I then went on to triage the issue
and found that if I set CPU affinity of that process to 1 (i.e. single
core), I'd get it to speed up to almost the speed I'd get under Linux
(set to single core too).

I dug deeper and had my suspicion on a Cygwin bug. I suspected a
problem in its thread singalling (condvar). So to test my hypothesis,
I created a minimal test case to show case this issue. This minimal
test case compiled on MSVC++ too and the difference is staggering.
What you'll find is if you started the process with CPU Aff = 1,
you'll get it to run >8x as fast as the default.

On my machine, it took 4300ms to run in dual core mode, 460ms to run
when CPU Aff = 1.

Setup:
Cygwin
WinXP SP3
2GB RAM
Core2Duo 2.33GHz
All firewalls disabled
Virus / Malware scanners disabled
Boost 1.48.0 (as per official Cygwin installation)
[Only for test case app] Boost.Threadpool - http://threadpool.sourceforge.net

Code to replicate the issue (get Boost.Threadpool from above):

#include <boost/threadpool.hpp>
#include <iostream>
#include <time.h>

inline int GetTickCount()
{
    timespec t;
    clock_gettime(CLOCK_MONOTONIC, &t);
    return (((int)t.tv_sec) * 1000) + (t.tv_nsec / 1000000);
}

class Test
{
public:
    void Add()
    {
    }
    void Delete()
    {
    }
};

int main(int argc, char** argv)
{
    int start;
    {
        boost::threadpool::pool tp(50);
        Test test;

        start = GetTickCount();
        for (int i=0; i<100000; i++)
        {
            tp.schedule(boost::bind(&Test::Add, &test));
            tp.schedule(boost::bind(&Test::Delete, &test));
        }

        tp.wait();
    }

    int elapsed = GetTickCount() - start;
    std::cout << elapsed << std::endl;

    std::cin.get();

    return 0;
}


Cheers,
Zach

--
Problem reports:       http://cygwin.com/problems.html
FAQ:                   http://cygwin.com/faq/
Documentation:         http://cygwin.com/docs.html
Unsubscribe info:      http://cygwin.com/ml/#unsubscribe-simple

- Raw text -