source: doc/proposals/concurrency/text/results.tex@ 07c1e595

ADT aaron-thesis arm-eh ast-experimental cleanup-dtors deferred_resn demangler enum forall-pointer-decay jacob/cs343-translation jenkins-sandbox new-ast new-ast-unique-expr new-env no_list persistent-indexer pthread-emulation qualifiedEnum resolv-new with_gc
Last change on this file since 07c1e595 was 07c1e595, checked in by Thierry Delisle <tdelisle@…>, 8 years ago

Ran ispell on the thesis
  • Property mode set to 100644
File size: 10.7 KB
Line 
1% ======================================================================
2% ======================================================================
3\chapter{Performance results} \label{results}
4% ======================================================================
5% ======================================================================
6\section{Machine setup}
7Table \ref{tab:machine} shows the characteristics of the machine used to run the benchmarks. All tests where made on this machine.
8\begin{figure}[H]
9\begin{center}
10\begin{tabular}{| l | r | l | r |}
11\hline
12Architecture & x86\_64 & NUMA node(s) & 8 \\
13\hline
14CPU op-mode(s) & 32-bit, 64-bit & Model name & AMD Opteron\texttrademark Processor 6380 \\
15\hline
16Byte Order & Little Endian & CPU Freq & 2.5\si{\giga\hertz} \\
17\hline
18CPU(s) & 64 & L1d cache & \SI{16}{\kibi\byte} \\
19\hline
20Thread(s) per core & 2 & L1i cache & \SI{64}{\kibi\byte} \\
21\hline
22Core(s) per socket & 8 & L2 cache & \SI{2048}{\kibi\byte} \\
23\hline
24Socket(s) & 4 & L3 cache & \SI{6144}{\kibi\byte} \\
25\hline
26\hline
27Operating system & Ubuntu 16.04.3 LTS & Kernel & Linux 4.4.0-97-generic \\
28\hline
29Compiler & GCC 6.3.0 & Translator & CFA 1.0.0 \\
30\hline
31\end{tabular}
32\end{center}
33\caption{Machine setup used for the tests}
34\label{tab:machine}
35\end{figure}
36
37\section{Micro benchmarks}
38All 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 :
39\begin{pseudo}
40#define BENCH(run, result)
41 gettime();
42 run;
43 gettime();
44 result = (after - before) / N;
45\end{pseudo}
46The 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.
47
48\subsection{Context-switching}
49The 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, which is a resume/suspend cycle instead of a yield. Listing \ref{lst:ctx-switch} shows the code for coroutines and threads. All omitted tests are functionally identical to one of these tests. The results can be shown in table \ref{tab:ctx-switch}.
50\begin{figure}
51\begin{multicols}{2}
52\CFA Coroutines
53\begin{cfacode}
54coroutine GreatSuspender {};
55void main(GreatSuspender& this) {
56 while(true) { suspend(); }
57}
58int main() {
59 GreatSuspender s;
60 resume(s);
61 BENCH(
62 for(size_t i=0; i<n; i++) {
63 resume(s);
64 },
65 result
66 )
67 printf("%llu\n", result);
68}
69\end{cfacode}
70\columnbreak
71\CFA Threads
72\begin{cfacode}
73
74
75
76
77int main() {
78
79
80 BENCH(
81 for(size_t i=0; i<n; i++) {
82 yield();
83 },
84 result
85 )
86 printf("%llu\n", result);
87}
88\end{cfacode}
89\end{multicols}
90\caption{\CFA benchmark code used to measure context-switches for coroutines and threads.}
91\label{lst:ctx-switch}
92\end{figure}
93
94\begin{figure}
95\begin{center}
96\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] |}
97\cline{2-4}
98\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
99\hline
100Kernel Threads & 239 & 242.57 & 5.54 \\
101\CFA Coroutines & 38 & 38 & 0 \\
102\CFA Threads & 102 & 102.39 & 1.57 \\
103\uC Coroutines & 46 & 46.68 & 0.47 \\
104\uC Threads & 98 & 99.39 & 1.52 \\
105\hline
106\end{tabular}
107\end{center}
108\caption{Context Switch comparison. All numbers are in nanoseconds(\si{\nano\second})}
109\label{tab:ctx-switch}
110\end{figure}
111
112\subsection{Mutual-exclusion}
113The 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 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}.
114
115\begin{figure}
116\begin{cfacode}
117monitor M {};
118void __attribute__((noinline)) call( M & mutex m /*, m2, m3, m4*/ ) {}
119
120int main() {
121 M m/*, m2, m3, m4*/;
122 BENCH(
123 for(size_t i=0; i<n; i++) {
124 call(m/*, m2, m3, m4*/);
125 },
126 result
127 )
128 printf("%llu\n", result);
129}
130\end{cfacode}
131\caption{\CFA benchmark code used to measure mutex routines.}
132\label{lst:mutex}
133\end{figure}
134
135\begin{figure}
136\begin{center}
137\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] |}
138\cline{2-4}
139\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
140\hline
141C routine & 2 & 2 & 0 \\
142Pthreads Mutex Lock & 31 & 31.86 & 0.99 \\
143\uC \code{monitor} member routine & 30 & 30 & 0 \\
144\CFA \code{mutex} routine, 1 argument & 46 & 46.14 & 0.74 \\
145\CFA \code{mutex} routine, 2 argument & 82 & 83 & 1.93 \\
146\CFA \code{mutex} routine, 4 argument & 165 & 161.15 & 54.04 \\
147\hline
148\end{tabular}
149\end{center}
150\caption{Mutex routine comparison. All numbers are in nanoseconds(\si{\nano\second})}
151\label{tab:mutex}
152\end{figure}
153
154\subsection{Internal scheduling}
155The Internal scheduling benchmark measures the cost of waiting on and signalling a condition variable. Listing \ref{lst:int-sched} shows the code for \CFA. The results can be shown in table \ref{tab:int-sched}. As with all other benchmarks, all omitted tests are functionally identical to one of these tests.
156
157\begin{figure}
158\begin{cfacode}
159volatile int go = 0;
160condition c;
161monitor M {};
162M m1;
163
164void __attribute__((noinline)) do_call( M & mutex a1 ) { signal(c); }
165
166thread T {};
167void ^?{}( T & mutex this ) {}
168void main( T & this ) {
169 while(go == 0) { yield(); }
170 while(go == 1) { do_call(m1); }
171}
172int __attribute__((noinline)) do_wait( M & mutex a1 ) {
173 go = 1;
174 BENCH(
175 for(size_t i=0; i<n; i++) {
176 wait(c);
177 },
178 result
179 )
180 printf("%llu\n", result);
181 go = 0;
182 return 0;
183}
184int main() {
185 T t;
186 return do_wait(m1);
187}
188\end{cfacode}
189\caption{Benchmark code for internal scheduling}
190\label{lst:int-sched}
191\end{figure}
192
193\begin{figure}
194\begin{center}
195\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] |}
196\cline{2-4}
197\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
198\hline
199\uC \code{signal} & 322 & 322.57 & 2.77 \\
200\CFA \code{signal}, 1 \code{monitor} & 1145 & 1163.64 & 27.52 \\
201\CFA \code{signal}, 2 \code{monitor} & 1531 & 1550.75 & 32.77 \\
202\CFA \code{signal}, 4 \code{monitor} & 2288.5 & 2326.86 & 54.73 \\
203\hline
204\end{tabular}
205\end{center}
206\caption{Internal scheduling comparison. All numbers are in nanoseconds(\si{\nano\second})}
207\label{tab:int-sched}
208\end{figure}
209
210\subsection{External scheduling}
211The 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. The results can be shown in table \ref{tab:ext-sched}. As with all other benchmarks, all omitted tests are functionally identical to one of these tests.
212
213\begin{figure}
214\begin{cfacode}
215volatile int go = 0;
216monitor M {};
217M m1;
218thread T {};
219
220void __attribute__((noinline)) do_call( M & mutex a1 ) {}
221
222void ^?{}( T & mutex this ) {}
223void main( T & this ) {
224 while(go == 0) { yield(); }
225 while(go == 1) { do_call(m1); }
226}
227int __attribute__((noinline)) do_wait( M & mutex a1 ) {
228 go = 1;
229 BENCH(
230 for(size_t i=0; i<n; i++) {
231 waitfor(call, a1);
232 },
233 result
234 )
235 printf("%llu\n", result);
236 go = 0;
237 return 0;
238}
239int main() {
240 T t;
241 return do_wait(m1);
242}
243\end{cfacode}
244\caption{Benchmark code for external scheduling}
245\label{lst:ext-sched}
246\end{figure}
247
248\begin{figure}
249\begin{center}
250\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] |}
251\cline{2-4}
252\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
253\hline
254\uC \code{Accept} & 349 & 339.32 & 3.14 \\
255\CFA \code{waitfor}, 1 \code{monitor} & 1155.5 & 1142.04 & 25.23 \\
256\CFA \code{waitfor}, 2 \code{monitor} & 1361 & 1376.75 & 28.81 \\
257\CFA \code{waitfor}, 4 \code{monitor} & 1941.5 & 1957.07 & 34.7 \\
258\hline
259\end{tabular}
260\end{center}
261\caption{External scheduling comparison. All numbers are in nanoseconds(\si{\nano\second})}
262\label{tab:ext-sched}
263\end{figure}
264
265\subsection{Object creation}
266Finally, the last benchmark measured is the cost of creation for concurrent objects. Listing \ref{lst:creation} shows the code for pthreads and \CFA threads. The results can be 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.
267
268\begin{figure}
269\begin{multicols}{2}
270pthread
271\begin{cfacode}
272int main() {
273 BENCH(
274 for(size_t i=0; i<n; i++) {
275 pthread_t thread;
276 if(pthread_create(
277 &thread,
278 NULL,
279 foo,
280 NULL
281 ) < 0) {
282 perror( "failure" );
283 return 1;
284 }
285
286 if(pthread_join(
287 thread,
288 NULL
289 ) < 0) {
290 perror( "failure" );
291 return 1;
292 }
293 },
294 result
295 )
296 printf("%llu\n", result);
297}
298\end{cfacode}
299\columnbreak
300\CFA Threads
301\begin{cfacode}
302int main() {
303 BENCH(
304 for(size_t i=0; i<n; i++) {
305 MyThread m;
306 },
307 result
308 )
309
310 printf("%llu\n", result);
311}
312\end{cfacode}
313\end{multicols}
314\caption{Benchmark code for pthreads and \CFA to measure object creation}
315\label{lst:creation}
316\end{figure}
317
318\begin{figure}
319\begin{center}
320\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] |}
321\cline{2-4}
322\multicolumn{1}{c |}{} & \multicolumn{1}{c |}{ Median } &\multicolumn{1}{c |}{ Average } & \multicolumn{1}{c |}{ Standard Deviation} \\
323\hline
324Pthreads & 26974.5 & 26977 & 124.12 \\
325\CFA Coroutines Lazy & 5 & 5 & 0 \\
326\CFA Coroutines Eager & 335.0 & 357.67 & 34.2 \\
327\CFA Threads & 1122.5 & 1109.86 & 36.54 \\
328\uC Coroutines & 106 & 107.04 & 1.61 \\
329\uC Threads & 525.5 & 533.04 & 11.14 \\
330\hline
331\end{tabular}
332\end{center}
333\caption{Creation comparison. All numbers are in nanoseconds(\si{\nano\second})}
334\label{tab:creation}
335\end{figure}
Note: See TracBrowser for help on using the repository browser.