spielwiese. (Posts about pthread_once.)https://spielwiese.fontein.de/tag/pthread_once.atom2024-01-05T07:10:12ZfelixNikolaa problem with pthread_once on an out-dated solaris installation.https://spielwiese.fontein.de/2012/01/22/a-problem-with-pthread_once-on-an-out-dated-solaris-installation/2012-01-22T19:52:02+01:002012-01-22T19:52:02+01:00felix<p>when adding a thread-specific allocator to a program of mine, to avoid terrible performance loss while using gmp and/or mpfr to do arbitrary precision integer respectively floating point arithmetic, i stumbled about a problem which seems to be fixed with newer solaris versions. in case anyone experiences a similar problem and cannot just update to a new enough solaris version, here’s some information on a dirty’n'quick fix for the problem.<br>
more precisely, i wanted to combine <a href="http://www.boost.org/doc/libs/1_35_0/doc/html/thread/thread_local_storage.html"><code>boost::thread_specific_pointer</code></a> (a portable implementation of <a href="https://en.wikipedia.org/wiki/Thread-local_storage">thread specific storage</a>, with <a href="http://g.oswego.edu/dl/html/malloc.html">dlmalloc</a>, to obtain an allocator which won’t block when used from different threads at once. if you use arbitrary precision arithmetic on a machine with many cores/cpus (say, 30 to 60), having a single blocking (via a <a href="https://en.wikipedia.org/wiki/Mutex">mutex</a>) allocator totally kills performance. for example, on our ultrasparc/solaris machine, running 29 threads (on 30 cpus) in parallel, only 20% of the system’s ressources were used effectively. if the machine would have only had 6 cpus, the program would have run at the same speed. quite a waste, isn’t it?<br>
anyway, combining thread local storage and a memory allocator solves this problem. in theory, at least. when i put the two things together, and ran my program with 30 threads, stlil only 60% of the 30 cpus processing power was used – the other 40% of the cycles were still spend waiting. (solaris has some excellent profiling tools on board. that’s why i like to use our slow old outdated solaris machine to profile, instead of our blazing fast newer big linux machine. in case anyone cares.) interestingly, on our linux machine, with 64 threads (running on 64 cores), the problem wasn’t there: 100% of the cycles went into computing, and essentially none into waiting.<br>
inspecting the problem closer with the sun studio analyzer, it turns out that the 40% waiting cycles are caused by <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span>, which is called by the internal boost method <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::detail::find_tss_data</code></span>. that method is called every time a <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::thread_specific_pointer<></code></span> is dereferenced. which in my program happens every time when the thread local allocator is fired up to allocate, reallocate or free a piece of memory. (more precisely, <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::detail::find_tss_data</code></span> calls <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::detail::get_current_thread_data</code></span>, which uses <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::call_once</code></span>, which in turn uses <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> in the <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread</code></span> implementation of <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::thread</code></span>, which is the implementation used on unixoid systems, such as solaris and linux.)<br>
in theory, <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> uses a <a href="https://en.wikipedia.org/wiki/Double-checked_locking">double-checked locking</a> mechanism to make sure that the function specified is ran exactly once during the execution of the wohle program. while searching online, i found the source of the <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread</code></span> implementation of a newer opensolaris from 2008 <a href="http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/lib/libc/port/threads/pthread.c#134">here</a>; it uses a double-checked locking with a <a href="https://en.wikipedia.org/wiki/Memory_barrier">memory barrier</a>, which should (at least in theory) turn it into a working solution (multi-threaded programming is far from being simple, both the compiler and the cpu can screw up your code by rearranging instructions in a deadly way).<br>
anyway, it seems that the <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> implementation on the soliaris installation on the machine i’m using just locks a mutex every time it is called. when you massively call the function from 30 threads at once, all running perfectly parallel on a machine with enough cpus, this gives a natural bottle-neck. to make sure it is <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> which causes the problem, i wrote the following test program:<br>
</p><div class="code-c++"><pre class="code literal-block"><span></span><span class="linenos"> 1</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><pthread.h></span>
<span class="linenos"> 2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><iostream></span>
<span class="linenos"> 3</span>
<span class="linenos"> 4</span><span class="k">static</span><span class="w"> </span><span class="n">pthread_once_t</span><span class="w"> </span><span class="n">onceControl</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">PTHREAD_ONCE_INIT</span><span class="p">;</span>
<span class="linenos"> 5</span><span class="k">static</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">nocalls</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
<span class="linenos"> 6</span>
<span class="linenos"> 7</span><span class="k">extern</span><span class="w"> </span><span class="s">"C"</span><span class="w"> </span><span class="kt">void</span><span class="w"> </span><span class="n">onceRoutine</span><span class="p">(</span><span class="kt">void</span><span class="p">)</span>
<span class="linenos"> 8</span><span class="p">{</span>
<span class="linenos"> 9</span><span class="w"> </span><span class="n">std</span><span class="o">::</span><span class="n">cout</span><span class="w"> </span><span class="o"><<</span><span class="w"> </span><span class="s">"onceRoutine()</span><span class="se">\n</span><span class="s">"</span><span class="p">;</span>
<span class="linenos">10</span><span class="w"> </span><span class="n">nocalls</span><span class="o">++</span><span class="p">;</span>
<span class="linenos">11</span><span class="p">}</span>
<span class="linenos">12</span>
<span class="linenos">13</span><span class="k">extern</span><span class="w"> </span><span class="s">"C"</span><span class="w"> </span><span class="kt">void</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="n">thethread</span><span class="p">(</span><span class="kt">void</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="n">x</span><span class="p">)</span>
<span class="linenos">14</span><span class="p">{</span>
<span class="linenos">15</span><span class="w"> </span><span class="k">for</span><span class="w"> </span><span class="p">(</span><span class="kt">unsigned</span><span class="w"> </span><span class="n">i</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span><span class="w"> </span><span class="n">i</span><span class="w"> </span><span class="o"><</span><span class="w"> </span><span class="mi">10000000</span><span class="p">;</span><span class="w"> </span><span class="o">++</span><span class="n">i</span><span class="p">)</span>
<span class="linenos">16</span><span class="w"> </span><span class="n">pthread_once</span><span class="p">(</span><span class="o">&</span><span class="n">onceControl</span><span class="p">,</span><span class="w"> </span><span class="n">onceRoutine</span><span class="p">);</span>
<span class="linenos">17</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="nb">NULL</span><span class="p">;</span>
<span class="linenos">18</span><span class="p">}</span>
<span class="linenos">19</span>
<span class="linenos">20</span><span class="kt">int</span><span class="w"> </span><span class="n">main</span><span class="p">()</span>
<span class="linenos">21</span><span class="p">{</span>
<span class="linenos">22</span><span class="w"> </span><span class="k">const</span><span class="w"> </span><span class="kt">int</span><span class="w"> </span><span class="n">nothreads</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">30</span><span class="p">;</span>
<span class="linenos">23</span><span class="w"> </span><span class="n">pthread_t</span><span class="w"> </span><span class="n">threads</span><span class="p">[</span><span class="n">nothreads</span><span class="p">];</span>
<span class="linenos">24</span>
<span class="linenos">25</span><span class="w"> </span><span class="k">for</span><span class="w"> </span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">i</span><span class="o">=</span><span class="mi">0</span><span class="p">;</span><span class="w"> </span><span class="n">i</span><span class="w"> </span><span class="o"><</span><span class="w"> </span><span class="n">nothreads</span><span class="p">;</span><span class="w"> </span><span class="o">++</span><span class="n">i</span><span class="p">)</span>
<span class="linenos">26</span><span class="w"> </span><span class="n">pthread_create</span><span class="p">(</span><span class="o">&</span><span class="n">threads</span><span class="p">[</span><span class="n">i</span><span class="p">],</span><span class="w"> </span><span class="nb">NULL</span><span class="p">,</span><span class="w"> </span><span class="n">thethread</span><span class="p">,</span><span class="w"> </span><span class="nb">NULL</span><span class="p">);</span>
<span class="linenos">27</span>
<span class="linenos">28</span><span class="w"> </span><span class="k">for</span><span class="w"> </span><span class="p">(</span><span class="kt">int</span><span class="w"> </span><span class="n">i</span><span class="o">=</span><span class="mi">0</span><span class="p">;</span><span class="w"> </span><span class="n">i</span><span class="w"> </span><span class="o"><</span><span class="w"> </span><span class="n">nothreads</span><span class="p">;</span><span class="w"> </span><span class="o">++</span><span class="n">i</span><span class="p">)</span>
<span class="linenos">29</span><span class="w"> </span><span class="p">{</span>
<span class="linenos">30</span><span class="w"> </span><span class="kt">void</span><span class="w"> </span><span class="o">*</span><span class="w"> </span><span class="n">status</span><span class="p">;</span>
<span class="linenos">31</span><span class="w"> </span><span class="n">pthread_join</span><span class="p">(</span><span class="n">threads</span><span class="p">[</span><span class="n">i</span><span class="p">],</span><span class="w"> </span><span class="o">&</span><span class="n">status</span><span class="p">);</span>
<span class="linenos">32</span><span class="w"> </span><span class="p">}</span>
<span class="linenos">33</span>
<span class="linenos">34</span><span class="w"> </span><span class="k">if</span><span class="w"> </span><span class="p">(</span><span class="n">nocalls</span><span class="w"> </span><span class="o">!=</span><span class="w"> </span><span class="mi">1</span><span class="p">)</span>
<span class="linenos">35</span><span class="w"> </span><span class="n">std</span><span class="o">::</span><span class="n">cout</span><span class="w"> </span><span class="o"><<</span><span class="w"> </span><span class="s">"pthread_once() screwed up totally!</span><span class="se">\n</span><span class="s">"</span><span class="p">;</span>
<span class="linenos">36</span><span class="w"> </span><span class="k">else</span>
<span class="linenos">37</span><span class="w"> </span><span class="n">std</span><span class="o">::</span><span class="n">cout</span><span class="w"> </span><span class="o"><<</span><span class="w"> </span><span class="s">"pthread_once() seems to be doing what it promises</span><span class="se">\n</span><span class="s">"</span><span class="p">;</span>
<span class="linenos">38</span><span class="w"> </span><span class="k">return</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span>
<span class="linenos">39</span><span class="p">}</span>
</pre></div><br>
i compiled the program with <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>CC -m64 -fast -xarch=native64 -xchip=native -xcache=native -mt -lpthread oncetest.cpp -o oncetest</code></span> and ran it with <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>time</code></span>. the result:<br>
<div class="code-unformatted"><pre class="code literal-block"><span></span><span class="linenos">1</span>real 16m9.541s
<span class="linenos">2</span>user 201m1.476s
<span class="linenos">3</span>sys 0m18.499s
</pre></div><br>
compiling the same program under linux and running it there (with enough cores in the machine) yielded<br>
<div class="code-unformatted"><pre class="code literal-block"><span></span><span class="linenos">1</span>real 0m0.243s
<span class="linenos">2</span>user 0m1.640s
<span class="linenos">3</span>sys 0m0.060s
</pre></div><br>
quite a difference, isn’t it? the solaris machine is slower, so a few seconds total time would be ok, but 16 minutes?! inspecting the running program on solaris with <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>prstat -Lmp <pid></code></span> shows the amount of waiting involved…<br>
to solve this problem, at least for me, with this old solaris verison running, i took the code of <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> from the above <a href="http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/lib/libc/port/threads/pthread.c#134">link</a> – namely the includes<br>
<div class="code-c++"><pre class="code literal-block"><span></span><span class="linenos">1</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><atomic.h></span>
<span class="linenos">2</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><thread.h></span>
<span class="linenos">3</span><span class="cp">#include</span><span class="w"> </span><span class="cpf"><errno.h></span>
</pre></div><br>
copied the lines 38 to 46 from the link, and the lines 157 to 179 from the link into <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost_directory/libs/thread/src/pthread/once.cpp</code></span>, renamed <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>pthread_once</code></span> to <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>my_pthread_once</code></span> in the code i copied and in the boost source file i added the lines to, and re-compiled boost. then, i re-ran my program, and suddenly, there was no more waiting (at least, not for mutexes :-) ). and the <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>oncetest</code></span> from above, rewritten using <span class="code-unformatted inline-code"><code class="code literal-block"><span></span>boost::once_call</code></span>, yielded:<br>
<div class="code-unformatted"><pre class="code literal-block"><span></span><span class="linenos">1</span>real 0m0.928s
<span class="linenos">2</span>user 0m20.181s
<span class="linenos">3</span>sys 0m0.036s
</pre></div><br>
perfect!