source: doc/proposals/concurrency/text/results.tex @ cf966b5

% ======================================================================
% ======================================================================
\chapter{Performance results} \label{results}
% ======================================================================
% ======================================================================
\section{Machine setup}
Table \ref{tab:machine} shows the characteristics of the machine used to run the benchmarks. All tests where made on this machine.
\begin{table}[H]
\begin{center}
\begin{tabular}{| l | r | l | r |}
\hline
Architecture            & x86\_64                       & NUMA node(s)  & 8 \\
\hline
CPU op-mode(s)          & 32-bit, 64-bit                & Model name    & AMD Opteron\texttrademark  Processor 6380 \\
\hline
Byte Order                      & Little Endian                 & CPU Freq              & 2.5\si{\giga\hertz} \\
\hline
CPU(s)                  & 64                            & L1d cache     & \SI{16}{\kibi\byte} \\
\hline
Thread(s) per core      & 2                             & L1i cache     & \SI{64}{\kibi\byte} \\
\hline
Core(s) per socket      & 8                             & L2 cache              & \SI{2048}{\kibi\byte} \\
\hline
Socket(s)                       & 4                             & L3 cache              & \SI{6144}{\kibi\byte} \\
\hline
\hline
Operating system                & Ubuntu 16.04.3 LTS    & Kernel                & Linux 4.4.0-97-generic \\
\hline
Compiler                        & GCC 6.3.0             & Translator    & CFA 1.0.0 \\
\hline
Java version            & OpenJDK-9             & Go version    & 1.9.2 \\
\hline
\end{tabular}
\end{center}
\caption{Machine setup used for the tests}
\label{tab:machine}
\end{table}

\section{Micro benchmarks}
All benchmarks are run using the same harness to produce the results, seen as the \code{BENCH()} macro in the following examples. This macro uses the following logic to benchmark the code :
\begin{pseudo}
#define BENCH(run, result)
        before = gettime();
        run;
        after  = gettime();
        result = (after - before) / N;
\end{pseudo}
The method used to get time is \code{clock_gettime(CLOCK_THREAD_CPUTIME_ID);}. Each benchmark is using many iterations of a simple call to measure the cost of the call. The specific number of iteration depends on the specific benchmark.

\subsection{Context-switching}
The first interesting benchmark is to measure how long context-switches take. The simplest approach to do this is to yield on a thread, which executes a 2-step context switch. In order to make the comparison fair, coroutines also execute a 2-step context-switch (\gls{uthread} to \gls{kthread} then \gls{kthread} to \gls{uthread}), which is a resume/suspend cycle instead of a yield. Listing \ref{lst:ctx-switch} shows the code for coroutines and threads whith the results in table \ref{tab:ctx-switch}. All omitted tests are functionally identical to one of these tests.
\begin{figure}
\begin{multicols}{2}
\CFA Coroutines
\begin{cfacode}
coroutine GreatSuspender {};
void main(GreatSuspender& this) {
        while(true) { suspend(); }
}
int main() {
        GreatSuspender s;
        resume(s);
        BENCH(
                for(size_t i=0; i<n; i++) {
                        resume(s);
                },
                result
        )
        printf("%llu\n", result);
}
\end{cfacode}
\columnbreak
\CFA Threads
\begin{cfacode}




int main() {


        BENCH(
                for(size_t i=0; i<n; i++) {
                        yield();
                },
                result
        )
        printf("%llu\n", result);
}
\end{cfacode}
\end{multicols}
\begin{cfacode}[caption={\CFA benchmark code used to measure context-switches for coroutines and threads.},label={lst:ctx-switch}]
\end{cfacode}
\end{figure}

\begin{table}
\begin{center}
\begin{tabular}{| l | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] |}
\cline{2-4}
\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
\hline
Kernel Thread   & 239           & 242.57        & 5.54 \\
\CFA Coroutine  & 38            & 38            & 0    \\
\CFA Thread             & 102           & 102.39        & 1.57 \\
\uC Coroutine   & 46            & 46.68 & 0.47 \\
\uC Thread              & 98            & 99.39 & 1.52 \\
Goroutine               & 148           & 148.0 & 0 \\
Java Thread             & 271           & 271.0 & 0 \\
\hline
\end{tabular}
\end{center}
\caption{Context Switch comparison. All numbers are in nanoseconds(\si{\nano\second})}
\label{tab:ctx-switch}
\end{table}

\subsection{Mutual-exclusion}
The next interesting benchmark is to measure the overhead to enter/leave a critical-section. For monitors, the simplest approach is to measure how long it takes to enter and leave a monitor routine. Listing \ref{lst:mutex} shows the code for \CFA. To put the results in context, the cost of entering a non-inline function and the cost of acquiring and releasing a pthread mutex lock are also measured. The results can be shown in table \ref{tab:mutex}.

\begin{figure}
\begin{cfacode}[caption={\CFA benchmark code used to measure mutex routines.},label={lst:mutex}]
monitor M {};
void __attribute__((noinline)) call( M & mutex m /*, m2, m3, m4*/ ) {}

int main() {
        M m/*, m2, m3, m4*/;
        BENCH(
                for(size_t i=0; i<n; i++) {
                        call(m/*, m2, m3, m4*/);
                },
                result
        )
        printf("%llu\n", result);
}
\end{cfacode}
\end{figure}

\begin{table}
\begin{center}
\begin{tabular}{| l | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] |}
\cline{2-4}
\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
\hline
C routine                                               & 2             & 2             & 0      \\
FetchAdd + FetchSub                             & 2             & 2             & 0      \\
Pthreads Mutex Lock                             & 31            & 31.86 & 0.99   \\
\uC \code{monitor} member routine               & 30            & 30            & 0      \\
\CFA \code{mutex} routine, 1 argument   & 46            & 46.14 & 0.74   \\
\CFA \code{mutex} routine, 2 argument   & 82            & 83            & 1.93   \\
\CFA \code{mutex} routine, 4 argument   & 165           & 161.15        & 54.04  \\
Java synchronized routine                       & 165           & 161.15        & 54.04  \\
\hline
\end{tabular}
\end{center}
\caption{Mutex routine comparison. All numbers are in nanoseconds(\si{\nano\second})}
\label{tab:mutex}
\end{table}

\subsection{Internal scheduling}
The internal-scheduling benchmark measures the cost of waiting on and signalling a condition variable. Listing \ref{lst:int-sched} shows the code for \CFA, with results table \ref{tab:int-sched}. As with all other benchmarks, all omitted tests are functionally identical to one of these tests.

\begin{figure}
\begin{cfacode}[caption={Benchmark code for internal scheduling},label={lst:int-sched}]
volatile int go = 0;
condition c;
monitor M {};
M m1;

void __attribute__((noinline)) do_call( M & mutex a1 ) { signal(c); }

thread T {};
void ^?{}( T & mutex this ) {}
void main( T & this ) {
        while(go == 0) { yield(); }
        while(go == 1) { do_call(m1); }
}
int  __attribute__((noinline)) do_wait( M & mutex a1 ) {
        go = 1;
        BENCH(
                for(size_t i=0; i<n; i++) {
                        wait(c);
                },
                result
        )
        printf("%llu\n", result);
        go = 0;
        return 0;
}
int main() {
        T t;
        return do_wait(m1);
}
\end{cfacode}
\end{figure}

\begin{table}
\begin{center}
\begin{tabular}{| l | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] |}
\cline{2-4}
\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
\hline
\uC \code{signal}                                       & 322           & 322.57        & 2.77  \\
\CFA \code{signal}, 1 \code{monitor}    & 1145  & 1163.64       & 27.52 \\
\CFA \code{signal}, 2 \code{monitor}    & 1531  & 1550.75       & 32.77 \\
\CFA \code{signal}, 4 \code{monitor}    & 2288.5        & 2326.86       & 54.73 \\
Java \code{notify}                              & 2288.5        & 2326.86       & 54.73 \\
\hline
\end{tabular}
\end{center}
\caption{Internal scheduling comparison. All numbers are in nanoseconds(\si{\nano\second})}
\label{tab:int-sched}
\end{table}

\subsection{External scheduling}
The Internal scheduling benchmark measures the cost of the \code{waitfor} statement (\code{_Accept} in \uC). Listing \ref{lst:ext-sched} shows the code for \CFA, with results in table \ref{tab:ext-sched}. As with all other benchmarks, all omitted tests are functionally identical to one of these tests.

\begin{figure}
\begin{cfacode}[caption={Benchmark code for external scheduling},label={lst:ext-sched}]
volatile int go = 0;
monitor M {};
M m1;
thread T {};

void __attribute__((noinline)) do_call( M & mutex a1 ) {}

void ^?{}( T & mutex this ) {}
void main( T & this ) {
        while(go == 0) { yield(); }
        while(go == 1) { do_call(m1); }
}
int  __attribute__((noinline)) do_wait( M & mutex a1 ) {
        go = 1;
        BENCH(
                for(size_t i=0; i<n; i++) {
                        waitfor(call, a1);
                },
                result
        )
        printf("%llu\n", result);
        go = 0;
        return 0;
}
int main() {
        T t;
        return do_wait(m1);
}
\end{cfacode}
\end{figure}

\begin{table}
\begin{center}
\begin{tabular}{| l | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] |}
\cline{2-4}
\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
\hline
\uC \code{Accept}                                       & 349           & 339.32        & 3.14  \\
\CFA \code{waitfor}, 1 \code{monitor}   & 1155.5        & 1142.04       & 25.23 \\
\CFA \code{waitfor}, 2 \code{monitor}   & 1361  & 1376.75       & 28.81 \\
\CFA \code{waitfor}, 4 \code{monitor}   & 1941.5        & 1957.07       & 34.7  \\
\hline
\end{tabular}
\end{center}
\caption{External scheduling comparison. All numbers are in nanoseconds(\si{\nano\second})}
\label{tab:ext-sched}
\end{table}

\subsection{Object creation}
Finally, the last benchmark measurs the cost of creation for concurrent objects. Listing \ref{lst:creation} shows the code for pthreads and \CFA threads, with results shown in table \ref{tab:creation}. As with all other benchmarks, all omitted tests are functionally identical to one of these tests. The only note here is that the call-stacks of \CFA coroutines are lazily created, therefore without priming the coroutine, the creation cost is very low.

\begin{figure}
\begin{center}
pthread
\begin{ccode}
int main() {
        BENCH(
                for(size_t i=0; i<n; i++) {
                        pthread_t thread;
                        if(pthread_create(&thread,NULL,foo,NULL)<0) {
                                perror( "failure" );
                                return 1;
                        }

                        if(pthread_join(thread, NULL)<0) {
                                perror( "failure" );
                                return 1;
                        }
                },
                result
        )
        printf("%llu\n", result);
}
\end{ccode}



\CFA Threads
\begin{cfacode}
int main() {
        BENCH(
                for(size_t i=0; i<n; i++) {
                        MyThread m;
                },
                result
        )
        printf("%llu\n", result);
}
\end{cfacode}
\end{center}
\begin{cfacode}[caption={Benchmark code for pthreads and \CFA to measure object creation},label={lst:creation}]
\end{cfacode}
\end{figure}

\begin{table}
\begin{center}
\begin{tabular}{| l | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] | S[table-format=5.2,table-number-alignment=right] |}
\cline{2-4}
\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
\hline
Pthreads                        & 26974.5       & 26977 & 124.12 \\
\CFA Coroutine Lazy     & 5             & 5             & 0      \\
\CFA Coroutine Eager    & 335.0 & 357.67        & 34.2   \\
\CFA Thread                     & 1122.5        & 1109.86       & 36.54  \\
\uC Coroutine           & 106           & 107.04        & 1.61   \\
\uC Thread                      & 525.5 & 533.04        & 11.14  \\
Goroutine                       & 525.5 & 533.04        & 11.14  \\
Java Thread                     & 525.5 & 533.04        & 11.14  \\
\hline
\end{tabular}
\end{center}
\caption{Creation comparison. All numbers are in nanoseconds(\si{\nano\second})}
\label{tab:creation}
\end{table}