Improve the default parallel GC settings, and sanitise the flags (#3340)
[ghc.git] / docs / users_guide / runtime_control.xml
1 <?xml version="1.0" encoding="iso-8859-1"?>
2 <sect1 id="runtime-control">
3 <title>Running a compiled program</title>
4
5 <indexterm><primary>runtime control of Haskell programs</primary></indexterm>
6 <indexterm><primary>running, compiled program</primary></indexterm>
7 <indexterm><primary>RTS options</primary></indexterm>
8
9 <para>To make an executable program, the GHC system compiles your
10 code and then links it with a non-trivial runtime system (RTS),
11 which handles storage management, profiling, etc.</para>
12
13 <para>You have some control over the behaviour of the RTS, by giving
14 special command-line arguments to your program.</para>
15
16 <para>When your Haskell program starts up, its RTS extracts
17 command-line arguments bracketed between
18 <option>+RTS</option><indexterm><primary><option>+RTS</option></primary></indexterm>
19 and
20 <option>-RTS</option><indexterm><primary><option>-RTS</option></primary></indexterm>
21 as its own. For example:</para>
22
23 <screen>
24 % ./a.out -f +RTS -p -S -RTS -h foo bar
25 </screen>
26
27 <para>The RTS will snaffle <option>-p</option> <option>-S</option>
28 for itself, and the remaining arguments <literal>-f -h foo bar</literal>
29 will be handed to your program if/when it calls
30 <function>System.getArgs</function>.</para>
31
32 <para>No <option>-RTS</option> option is required if the
33 runtime-system options extend to the end of the command line, as in
34 this example:</para>
35
36 <screen>
37 % hls -ltr /usr/etc +RTS -A5m
38 </screen>
39
40 <para>If you absolutely positively want all the rest of the options
41 in a command line to go to the program (and not the RTS), use a
42 <option>&ndash;&ndash;RTS</option><indexterm><primary><option>--RTS</option></primary></indexterm>.</para>
43
44 <para>As always, for RTS options that take
45 <replaceable>size</replaceable>s: If the last character of
46 <replaceable>size</replaceable> is a K or k, multiply by 1000; if an
47 M or m, by 1,000,000; if a G or G, by 1,000,000,000. (And any
48 wraparound in the counters is <emphasis>your</emphasis>
49 fault!)</para>
50
51 <para>Giving a <literal>+RTS -f</literal>
52 <indexterm><primary><option>-f</option></primary><secondary>RTS option</secondary></indexterm> option
53 will print out the RTS options actually available in your program
54 (which vary, depending on how you compiled).</para>
55
56 <para>NOTE: since GHC is itself compiled by GHC, you can change RTS
57 options in the compiler using the normal
58 <literal>+RTS ... -RTS</literal>
59 combination. eg. to increase the maximum heap
60 size for a compilation to 128M, you would add
61 <literal>+RTS -M128m -RTS</literal>
62 to the command line.</para>
63
64 <sect2 id="rts-optinos-environment">
65 <title>Setting global RTS options</title>
66
67 <indexterm><primary>RTS options</primary><secondary>from the environment</secondary></indexterm>
68 <indexterm><primary>environment variable</primary><secondary>for
69 setting RTS options</secondary></indexterm>
70
71 <para>RTS options are also taken from the environment variable
72 <envar>GHCRTS</envar><indexterm><primary><envar>GHCRTS</envar></primary>
73 </indexterm>. For example, to set the maximum heap size
74 to 128M for all GHC-compiled programs (using an
75 <literal>sh</literal>-like shell):</para>
76
77 <screen>
78 GHCRTS='-M128m'
79 export GHCRTS
80 </screen>
81
82 <para>RTS options taken from the <envar>GHCRTS</envar> environment
83 variable can be overridden by options given on the command
84 line.</para>
85
86 </sect2>
87
88 <sect2 id="rts-options-misc">
89 <title>Miscellaneous RTS options</title>
90
91 <variablelist>
92 <varlistentry>
93 <term><option>-V<replaceable>secs</replaceable></option>
94 <indexterm><primary><option>-V</option></primary><secondary>RTS
95 option</secondary></indexterm></term>
96 <listitem>
97 <para>Sets the interval that the RTS clock ticks at. The
98 runtime uses a single timer signal to count ticks; this timer
99 signal is used to control the context switch timer (<xref
100 linkend="using-concurrent" />) and the heap profiling
101 timer <xref linkend="rts-options-heap-prof" />. Also, the
102 time profiler uses the RTS timer signal directly to record
103 time profiling samples.</para>
104
105 <para>Normally, setting the <option>-V</option> option
106 directly is not necessary: the resolution of the RTS timer is
107 adjusted automatically if a short interval is requested with
108 the <option>-C</option> or <option>-i</option> options.
109 However, setting <option>-V</option> is required in order to
110 increase the resolution of the time profiler.</para>
111
112 <para>Using a value of zero disables the RTS clock
113 completely, and has the effect of disabling timers that
114 depend on it: the context switch timer and the heap profiling
115 timer. Context switches will still happen, but
116 deterministically and at a rate much faster than normal.
117 Disabling the interval timer is useful for debugging, because
118 it eliminates a source of non-determinism at runtime.</para>
119 </listitem>
120 </varlistentry>
121
122 <varlistentry>
123 <term><option>--install-signal-handlers=<replaceable>yes|no</replaceable></option>
124 <indexterm><primary><option>--install-signal-handlers</option></primary><secondary>RTS
125 option</secondary></indexterm></term>
126 <listitem>
127 <para>If yes (the default), the RTS installs signal handlers to catch
128 things like ctrl-C. This option is primarily useful for when
129 you are using the Haskell code as a DLL, and want to set your
130 own signal handlers.</para>
131 </listitem>
132 </varlistentry>
133
134 <varlistentry>
135 <term><option>-xm<replaceable>address</replaceable></option>
136 <indexterm><primary><option>-xm</option></primary><secondary>RTS
137 option</secondary></indexterm></term>
138 <listitem>
139 <para>
140 WARNING: this option is for working around memory
141 allocation problems only. Do not use unless GHCi fails
142 with a message like &ldquo;<literal>failed to mmap() memory below 2Gb</literal>&rdquo;. If you need to use this option to get GHCi working
143 on your machine, please file a bug.
144 </para>
145
146 <para>
147 On 64-bit machines, the RTS needs to allocate memory in the
148 low 2Gb of the address space. Support for this across
149 different operating systems is patchy, and sometimes fails.
150 This option is there to give the RTS a hint about where it
151 should be able to allocate memory in the low 2Gb of the
152 address space. For example, <literal>+RTS -xm20000000
153 -RTS</literal> would hint that the RTS should allocate
154 starting at the 0.5Gb mark. The default is to use the OS's
155 built-in support for allocating memory in the low 2Gb if
156 available (e.g. <literal>mmap</literal>
157 with <literal>MAP_32BIT</literal> on Linux), or
158 otherwise <literal>-xm40000000</literal>.
159 </para>
160 </listitem>
161 </varlistentry>
162 </variablelist>
163 </sect2>
164
165 <sect2 id="rts-options-gc">
166 <title>RTS options to control the garbage collector</title>
167
168 <indexterm><primary>garbage collector</primary><secondary>options</secondary></indexterm>
169 <indexterm><primary>RTS options</primary><secondary>garbage collection</secondary></indexterm>
170
171 <para>There are several options to give you precise control over
172 garbage collection. Hopefully, you won't need any of these in
173 normal operation, but there are several things that can be tweaked
174 for maximum performance.</para>
175
176 <variablelist>
177
178 <varlistentry>
179 <term>
180 <option>-A</option><replaceable>size</replaceable>
181 <indexterm><primary><option>-A</option></primary><secondary>RTS option</secondary></indexterm>
182 <indexterm><primary>allocation area, size</primary></indexterm>
183 </term>
184 <listitem>
185 <para>&lsqb;Default: 512k&rsqb; Set the allocation area size
186 used by the garbage collector. The allocation area
187 (actually generation 0 step 0) is fixed and is never resized
188 (unless you use <option>-H</option>, below).</para>
189
190 <para>Increasing the allocation area size may or may not
191 give better performance (a bigger allocation area means
192 worse cache behaviour but fewer garbage collections and less
193 promotion).</para>
194
195 <para>With only 1 generation (<option>-G1</option>) the
196 <option>-A</option> option specifies the minimum allocation
197 area, since the actual size of the allocation area will be
198 resized according to the amount of data in the heap (see
199 <option>-F</option>, below).</para>
200 </listitem>
201 </varlistentry>
202
203 <varlistentry>
204 <term>
205 <option>-c</option>
206 <indexterm><primary><option>-c</option></primary><secondary>RTS option</secondary></indexterm>
207 <indexterm><primary>garbage collection</primary><secondary>compacting</secondary></indexterm>
208 <indexterm><primary>compacting garbage collection</primary></indexterm>
209 </term>
210 <listitem>
211 <para>Use a compacting algorithm for collecting the oldest
212 generation. By default, the oldest generation is collected
213 using a copying algorithm; this option causes it to be
214 compacted in-place instead. The compaction algorithm is
215 slower than the copying algorithm, but the savings in memory
216 use can be considerable.</para>
217
218 <para>For a given heap size (using the <option>-H</option>
219 option), compaction can in fact reduce the GC cost by
220 allowing fewer GCs to be performed. This is more likely
221 when the ratio of live data to heap size is high, say
222 &gt;30&percnt;.</para>
223
224 <para>NOTE: compaction doesn't currently work when a single
225 generation is requested using the <option>-G1</option>
226 option.</para>
227 </listitem>
228 </varlistentry>
229
230 <varlistentry>
231 <term><option>-c</option><replaceable>n</replaceable></term>
232
233 <listitem>
234 <para>&lsqb;Default: 30&rsqb; Automatically enable
235 compacting collection when the live data exceeds
236 <replaceable>n</replaceable>&percnt; of the maximum heap size
237 (see the <option>-M</option> option). Note that the maximum
238 heap size is unlimited by default, so this option has no
239 effect unless the maximum heap size is set with
240 <option>-M</option><replaceable>size</replaceable>. </para>
241 </listitem>
242 </varlistentry>
243
244 <varlistentry>
245 <term>
246 <option>-F</option><replaceable>factor</replaceable>
247 <indexterm><primary><option>-F</option></primary><secondary>RTS option</secondary></indexterm>
248 <indexterm><primary>heap size, factor</primary></indexterm>
249 </term>
250 <listitem>
251
252 <para>&lsqb;Default: 2&rsqb; This option controls the amount
253 of memory reserved for the older generations (and in the
254 case of a two space collector the size of the allocation
255 area) as a factor of the amount of live data. For example,
256 if there was 2M of live data in the oldest generation when
257 we last collected it, then by default we'll wait until it
258 grows to 4M before collecting it again.</para>
259
260 <para>The default seems to work well here. If you have
261 plenty of memory, it is usually better to use
262 <option>-H</option><replaceable>size</replaceable> than to
263 increase
264 <option>-F</option><replaceable>factor</replaceable>.</para>
265
266 <para>The <option>-F</option> setting will be automatically
267 reduced by the garbage collector when the maximum heap size
268 (the <option>-M</option><replaceable>size</replaceable>
269 setting) is approaching.</para>
270 </listitem>
271 </varlistentry>
272
273 <varlistentry>
274 <term>
275 <option>-G</option><replaceable>generations</replaceable>
276 <indexterm><primary><option>-G</option></primary><secondary>RTS option</secondary></indexterm>
277 <indexterm><primary>generations, number of</primary></indexterm>
278 </term>
279 <listitem>
280 <para>&lsqb;Default: 2&rsqb; Set the number of generations
281 used by the garbage collector. The default of 2 seems to be
282 good, but the garbage collector can support any number of
283 generations. Anything larger than about 4 is probably not a
284 good idea unless your program runs for a
285 <emphasis>long</emphasis> time, because the oldest
286 generation will hardly ever get collected.</para>
287
288 <para>Specifying 1 generation with <option>+RTS -G1</option>
289 gives you a simple 2-space collector, as you would expect.
290 In a 2-space collector, the <option>-A</option> option (see
291 above) specifies the <emphasis>minimum</emphasis> allocation
292 area size, since the allocation area will grow with the
293 amount of live data in the heap. In a multi-generational
294 collector the allocation area is a fixed size (unless you
295 use the <option>-H</option> option, see below).</para>
296 </listitem>
297 </varlistentry>
298
299 <varlistentry>
300 <term>
301 <option>-qg<optional><replaceable>gen</replaceable></optional></option>
302 <indexterm><primary><option>-qg</option><secondary>RTS
303 option</secondary></primary></indexterm>
304 </term>
305 <listitem>
306 <para>&lsqb;New in GHC 6.12.1&rsqb; &lsqb;Default: 0&rsqb;
307 Use parallel GC in
308 generation <replaceable>gen</replaceable> and higher.
309 Omitting <replaceable>gen</replaceable> turns off the
310 parallel GC completely, reverting to sequential GC.</para>
311
312 <para>The default parallel GC settings are usually suitable
313 for parallel programs (i.e. those
314 using <literal>par</literal>, Strategies, or with multiple
315 threads). However, it is sometimes beneficial to enable
316 the parallel GC for a single-threaded sequential program
317 too, especially if the program has a large amount of heap
318 data and GC is a significant fraction of runtime. To use
319 the parallel GC in a sequential program, enable the
320 parallel runtime with a suitable <literal>-N</literal>
321 option, and additionally it might be beneficial to
322 restrict parallel GC to the old generation
323 with <literal>-qg1</literal>.</para>
324 </listitem>
325 </varlistentry>
326
327 <varlistentry>
328 <term>
329 <option>-qb<optional><replaceable>gen</replaceable></optional></option>
330 <indexterm><primary><option>-qb</option><secondary>RTS
331 option</secondary></primary></indexterm>
332 </term>
333 <listitem>
334 <para>
335 &lsqb;New in GHC 6.12.1&rsqb; &lsqb;Default: 1&rsqb; Use
336 load-balancing in the parallel GC in
337 generation <replaceable>gen</replaceable> and higher.
338 Omitting <replaceable>gen</replaceable> disables
339 load-balancing entirely.</para>
340
341 <para>
342 Load-balancing shares out the work of GC between the
343 available cores. This is a good idea when the heap is
344 large and we need to parallelise the GC work, however it
345 is also pessimal for the short young-generation
346 collections in a parallel program, because it can harm
347 locality by moving data from the cache of the CPU where is
348 it being used to the cache of another CPU. Hence the
349 default is to do load-balancing only in the
350 old-generation. In fact, for a parallel program it is
351 sometimes beneficial to disable load-balancing entirely
352 with <literal>-qb</literal>.
353 </para>
354 </listitem>
355 </varlistentry>
356
357 <varlistentry>
358 <term>
359 <option>-H</option><replaceable>size</replaceable>
360 <indexterm><primary><option>-H</option></primary><secondary>RTS option</secondary></indexterm>
361 <indexterm><primary>heap size, suggested</primary></indexterm>
362 </term>
363 <listitem>
364 <para>&lsqb;Default: 0&rsqb; This option provides a
365 &ldquo;suggested heap size&rdquo; for the garbage collector. The
366 garbage collector will use about this much memory until the
367 program residency grows and the heap size needs to be
368 expanded to retain reasonable performance.</para>
369
370 <para>By default, the heap will start small, and grow and
371 shrink as necessary. This can be bad for performance, so if
372 you have plenty of memory it's worthwhile supplying a big
373 <option>-H</option><replaceable>size</replaceable>. For
374 improving GC performance, using
375 <option>-H</option><replaceable>size</replaceable> is
376 usually a better bet than
377 <option>-A</option><replaceable>size</replaceable>.</para>
378 </listitem>
379 </varlistentry>
380
381 <varlistentry>
382 <term>
383 <option>-I</option><replaceable>seconds</replaceable>
384 <indexterm><primary><option>-I</option></primary>
385 <secondary>RTS option</secondary>
386 </indexterm>
387 <indexterm><primary>idle GC</primary>
388 </indexterm>
389 </term>
390 <listitem>
391 <para>(default: 0.3) In the threaded and SMP versions of the RTS (see
392 <option>-threaded</option>, <xref linkend="options-linker" />), a
393 major GC is automatically performed if the runtime has been idle
394 (no Haskell computation has been running) for a period of time.
395 The amount of idle time which must pass before a GC is performed is
396 set by the <option>-I</option><replaceable>seconds</replaceable>
397 option. Specifying <option>-I0</option> disables the idle GC.</para>
398
399 <para>For an interactive application, it is probably a good idea to
400 use the idle GC, because this will allow finalizers to run and
401 deadlocked threads to be detected in the idle time when no Haskell
402 computation is happening. Also, it will mean that a GC is less
403 likely to happen when the application is busy, and so
404 responsiveness may be improved. However, if the amount of live data in
405 the heap is particularly large, then the idle GC can cause a
406 significant delay, and too small an interval could adversely affect
407 interactive responsiveness.</para>
408
409 <para>This is an experimental feature, please let us know if it
410 causes problems and/or could benefit from further tuning.</para>
411 </listitem>
412 </varlistentry>
413
414 <varlistentry>
415 <term>
416 <option>-k</option><replaceable>size</replaceable>
417 <indexterm><primary><option>-k</option></primary><secondary>RTS option</secondary></indexterm>
418 <indexterm><primary>stack, minimum size</primary></indexterm>
419 </term>
420 <listitem>
421 <para>&lsqb;Default: 1k&rsqb; Set the initial stack size for
422 new threads. Thread stacks (including the main thread's
423 stack) live on the heap, and grow as required. The default
424 value is good for concurrent applications with lots of small
425 threads; if your program doesn't fit this model then
426 increasing this option may help performance.</para>
427
428 <para>The main thread is normally started with a slightly
429 larger heap to cut down on unnecessary stack growth while
430 the program is starting up.</para>
431 </listitem>
432 </varlistentry>
433
434 <varlistentry>
435 <term>
436 <option>-K</option><replaceable>size</replaceable>
437 <indexterm><primary><option>-K</option></primary><secondary>RTS option</secondary></indexterm>
438 <indexterm><primary>stack, maximum size</primary></indexterm>
439 </term>
440 <listitem>
441 <para>&lsqb;Default: 8M&rsqb; Set the maximum stack size for
442 an individual thread to <replaceable>size</replaceable>
443 bytes. This option is there purely to stop the program
444 eating up all the available memory in the machine if it gets
445 into an infinite loop.</para>
446 </listitem>
447 </varlistentry>
448
449 <varlistentry>
450 <term>
451 <option>-m</option><replaceable>n</replaceable>
452 <indexterm><primary><option>-m</option></primary><secondary>RTS option</secondary></indexterm>
453 <indexterm><primary>heap, minimum free</primary></indexterm>
454 </term>
455 <listitem>
456 <para>Minimum &percnt; <replaceable>n</replaceable> of heap
457 which must be available for allocation. The default is
458 3&percnt;.</para>
459 </listitem>
460 </varlistentry>
461
462 <varlistentry>
463 <term>
464 <option>-M</option><replaceable>size</replaceable>
465 <indexterm><primary><option>-M</option></primary><secondary>RTS option</secondary></indexterm>
466 <indexterm><primary>heap size, maximum</primary></indexterm>
467 </term>
468 <listitem>
469 <para>&lsqb;Default: unlimited&rsqb; Set the maximum heap size to
470 <replaceable>size</replaceable> bytes. The heap normally
471 grows and shrinks according to the memory requirements of
472 the program. The only reason for having this option is to
473 stop the heap growing without bound and filling up all the
474 available swap space, which at the least will result in the
475 program being summarily killed by the operating
476 system.</para>
477
478 <para>The maximum heap size also affects other garbage
479 collection parameters: when the amount of live data in the
480 heap exceeds a certain fraction of the maximum heap size,
481 compacting collection will be automatically enabled for the
482 oldest generation, and the <option>-F</option> parameter
483 will be reduced in order to avoid exceeding the maximum heap
484 size.</para>
485 </listitem>
486 </varlistentry>
487
488 <varlistentry>
489 <term>
490 <option>-t</option><optional><replaceable>file</replaceable></optional>
491 <indexterm><primary><option>-t</option></primary><secondary>RTS option</secondary></indexterm>
492 </term>
493 <term>
494 <option>-s</option><optional><replaceable>file</replaceable></optional>
495 <indexterm><primary><option>-s</option></primary><secondary>RTS option</secondary></indexterm>
496 </term>
497 <term>
498 <option>-S</option><optional><replaceable>file</replaceable></optional>
499 <indexterm><primary><option>-S</option></primary><secondary>RTS option</secondary></indexterm>
500 </term>
501 <term>
502 <option>--machine-readable</option>
503 <indexterm><primary><option>--machine-readable</option></primary><secondary>RTS option</secondary></indexterm>
504 </term>
505 <listitem>
506 <para>These options produce runtime-system statistics, such
507 as the amount of time spent executing the program and in the
508 garbage collector, the amount of memory allocated, the
509 maximum size of the heap, and so on. The three
510 variants give different levels of detail:
511 <option>-t</option> produces a single line of output in the
512 same format as GHC's <option>-Rghc-timing</option> option,
513 <option>-s</option> produces a more detailed summary at the
514 end of the program, and <option>-S</option> additionally
515 produces information about each and every garbage
516 collection.</para>
517
518 <para>The output is placed in
519 <replaceable>file</replaceable>. If
520 <replaceable>file</replaceable> is omitted, then the output
521 is sent to <constant>stderr</constant>.</para>
522
523 <para>
524 If you use the <literal>-t</literal> flag then, when your
525 program finishes, you will see something like this:
526 </para>
527
528 <programlisting>
529 &lt;&lt;ghc: 36169392 bytes, 69 GCs, 603392/1065272 avg/max bytes residency (2 samples), 3M in use, 0.00 INIT (0.00 elapsed), 0.02 MUT (0.02 elapsed), 0.07 GC (0.07 elapsed) :ghc&gt;&gt;
530 </programlisting>
531
532 <para>
533 This tells you:
534 </para>
535
536 <itemizedlist>
537 <listitem>
538 <para>
539 The total number of bytes allocated by the program over the
540 whole run.
541 </para>
542 </listitem>
543 <listitem>
544 <para>
545 The total number of garbage collections performed.
546 </para>
547 </listitem>
548 <listitem>
549 <para>
550 The average and maximum "residency", which is the amount of
551 live data in bytes. The runtime can only determine the
552 amount of live data during a major GC, which is why the
553 number of samples corresponds to the number of major GCs
554 (and is usually relatively small). To get a better picture
555 of the heap profile of your program, use
556 the <option>-hT</option> RTS option
557 (<xref linkend="rts-profiling" />).
558 </para>
559 </listitem>
560 <listitem>
561 <para>
562 The peak memory the RTS has allocated from the OS.
563 </para>
564 </listitem>
565 <listitem>
566 <para>
567 The amount of CPU time and elapsed wall clock time while
568 initialising the runtime system (INIT), running the program
569 itself (MUT, the mutator), and garbage collecting (GC).
570 </para>
571 </listitem>
572 </itemizedlist>
573
574 <para>
575 You can also get this in a more future-proof, machine readable
576 format, with <literal>-t --machine-readable</literal>:
577 </para>
578
579 <programlisting>
580 [("bytes allocated", "36169392")
581 ,("num_GCs", "69")
582 ,("average_bytes_used", "603392")
583 ,("max_bytes_used", "1065272")
584 ,("num_byte_usage_samples", "2")
585 ,("peak_megabytes_allocated", "3")
586 ,("init_cpu_seconds", "0.00")
587 ,("init_wall_seconds", "0.00")
588 ,("mutator_cpu_seconds", "0.02")
589 ,("mutator_wall_seconds", "0.02")
590 ,("GC_cpu_seconds", "0.07")
591 ,("GC_wall_seconds", "0.07")
592 ]
593 </programlisting>
594
595 <para>
596 If you use the <literal>-s</literal> flag then, when your
597 program finishes, you will see something like this (the exact
598 details will vary depending on what sort of RTS you have, e.g.
599 you will only see profiling data if your RTS is compiled for
600 profiling):
601 </para>
602
603 <programlisting>
604 36,169,392 bytes allocated in the heap
605 4,057,632 bytes copied during GC
606 1,065,272 bytes maximum residency (2 sample(s))
607 54,312 bytes maximum slop
608 3 MB total memory in use (0 MB lost due to fragmentation)
609
610 Generation 0: 67 collections, 0 parallel, 0.04s, 0.03s elapsed
611 Generation 1: 2 collections, 0 parallel, 0.03s, 0.04s elapsed
612
613 SPARKS: 359207 (557 converted, 149591 pruned)
614
615 INIT time 0.00s ( 0.00s elapsed)
616 MUT time 0.01s ( 0.02s elapsed)
617 GC time 0.07s ( 0.07s elapsed)
618 EXIT time 0.00s ( 0.00s elapsed)
619 Total time 0.08s ( 0.09s elapsed)
620
621 %GC time 89.5% (75.3% elapsed)
622
623 Alloc rate 4,520,608,923 bytes per MUT second
624
625 Productivity 10.5% of total user, 9.1% of total elapsed
626 </programlisting>
627
628 <itemizedlist>
629 <listitem>
630 <para>
631 The "bytes allocated in the heap" is the total bytes allocated
632 by the program over the whole run.
633 </para>
634 </listitem>
635 <listitem>
636 <para>
637 GHC uses a copying garbage collector by default. "bytes copied
638 during GC" tells you how many bytes it had to copy during
639 garbage collection.
640 </para>
641 </listitem>
642 <listitem>
643 <para>
644 The maximum space actually used by your program is the
645 "bytes maximum residency" figure. This is only checked during
646 major garbage collections, so it is only an approximation;
647 the number of samples tells you how many times it is checked.
648 </para>
649 </listitem>
650 <listitem>
651 <para>
652 The "bytes maximum slop" tells you the most space that is ever
653 wasted due to the way GHC allocates memory in blocks. Slop is
654 memory at the end of a block that was wasted. There's no way
655 to control this; we just like to see how much memory is being
656 lost this way.
657 </para>
658 </listitem>
659 <listitem>
660 <para>
661 The "total memory in use" tells you the peak memory the RTS has
662 allocated from the OS.
663 </para>
664 </listitem>
665 <listitem>
666 <para>
667 Next there is information about the garbage collections done.
668 For each generation it says how many garbage collections were
669 done, how many of those collections were done in parallel,
670 the total CPU time used for garbage collecting that generation,
671 and the total wall clock time elapsed while garbage collecting
672 that generation.
673 </para>
674 </listitem>
675 <listitem>
676 <para>The <literal>SPARKS</literal> statistic refers to the
677 use of <literal>Control.Parallel.par</literal> and related
678 functionality in the program. Each spark represents a call
679 to <literal>par</literal>; a spark is "converted" when it is
680 executed in parallel; and a spark is "pruned" when it is
681 found to be already evaluated and is discarded from the pool
682 by the garbage collector. Any remaining sparks are
683 discarded at the end of execution, so "converted" plus
684 "pruned" does not necessarily add up to the total.</para>
685 </listitem>
686 <listitem>
687 <para>
688 Next there is the CPU time and wall clock time elapsed broken
689 down by what the runtime system was doing at the time.
690 INIT is the runtime system initialisation.
691 MUT is the mutator time, i.e. the time spent actually running
692 your code.
693 GC is the time spent doing garbage collection.
694 RP is the time spent doing retainer profiling.
695 PROF is the time spent doing other profiling.
696 EXIT is the runtime system shutdown time.
697 And finally, Total is, of course, the total.
698 </para>
699 <para>
700 %GC time tells you what percentage GC is of Total.
701 "Alloc rate" tells you the "bytes allocated in the heap" divided
702 by the MUT CPU time.
703 "Productivity" tells you what percentage of the Total CPU and wall
704 clock elapsed times are spent in the mutator (MUT).
705 </para>
706 </listitem>
707 </itemizedlist>
708
709 <para>
710 The <literal>-S</literal> flag, as well as giving the same
711 output as the <literal>-s</literal> flag, prints information
712 about each GC as it happens:
713 </para>
714
715 <programlisting>
716 Alloc Copied Live GC GC TOT TOT Page Flts
717 bytes bytes bytes user elap user elap
718 528496 47728 141512 0.01 0.02 0.02 0.02 0 0 (Gen: 1)
719 [...]
720 524944 175944 1726384 0.00 0.00 0.08 0.11 0 0 (Gen: 0)
721 </programlisting>
722
723 <para>
724 For each garbage collection, we print:
725 </para>
726
727 <itemizedlist>
728 <listitem>
729 <para>
730 How many bytes we allocated this garbage collection.
731 </para>
732 </listitem>
733 <listitem>
734 <para>
735 How many bytes we copied this garbage collection.
736 </para>
737 </listitem>
738 <listitem>
739 <para>
740 How many bytes are currently live.
741 </para>
742 </listitem>
743 <listitem>
744 <para>
745 How long this garbage collection took (CPU time and elapsed
746 wall clock time).
747 </para>
748 </listitem>
749 <listitem>
750 <para>
751 How long the program has been running (CPU time and elapsed
752 wall clock time).
753 </para>
754 </listitem>
755 <listitem>
756 <para>
757 How many page faults occured this garbage collection.
758 </para>
759 </listitem>
760 <listitem>
761 <para>
762 How many page faults occured since the end of the last garbage
763 collection.
764 </para>
765 </listitem>
766 <listitem>
767 <para>
768 Which generation is being garbage collected.
769 </para>
770 </listitem>
771 </itemizedlist>
772
773 </listitem>
774 </varlistentry>
775 </variablelist>
776
777 </sect2>
778
779 <sect2>
780 <title>RTS options for concurrency and parallelism</title>
781
782 <para>The RTS options related to concurrency are described in
783 <xref linkend="using-concurrent" />, and those for parallelism in
784 <xref linkend="parallel-options"/>.</para>
785 </sect2>
786
787 <sect2 id="rts-profiling">
788 <title>RTS options for profiling</title>
789
790 <para>Most profiling runtime options are only available when you
791 compile your program for profiling (see
792 <xref linkend="prof-compiler-options" />, and
793 <xref linkend="rts-options-heap-prof" /> for the runtime options).
794 However, there is one profiling option that is available
795 for ordinary non-profiled executables:</para>
796
797 <variablelist>
798 <varlistentry>
799 <term>
800 <option>-hT</option>
801 <indexterm><primary><option>-hT</option></primary><secondary>RTS
802 option</secondary></indexterm>
803 </term>
804 <listitem>
805 <para>Generates a basic heap profile, in the
806 file <literal><replaceable>prog</replaceable>.hp</literal>.
807 To produce the heap profile graph,
808 use <command>hp2ps</command> (see <xref linkend="hp2ps"
809 />). The basic heap profile is broken down by data
810 constructor, with other types of closures (functions, thunks,
811 etc.) grouped into broad categories
812 (e.g. <literal>FUN</literal>, <literal>THUNK</literal>). To
813 get a more detailed profile, use the full profiling
814 support (<xref linkend="profiling" />).</para>
815 </listitem>
816 </varlistentry>
817 </variablelist>
818 </sect2>
819
820 <sect2 id="rts-options-debugging">
821 <title>RTS options for hackers, debuggers, and over-interested
822 souls</title>
823
824 <indexterm><primary>RTS options, hacking/debugging</primary></indexterm>
825
826 <para>These RTS options might be used (a)&nbsp;to avoid a GHC bug,
827 (b)&nbsp;to see &ldquo;what's really happening&rdquo;, or
828 (c)&nbsp;because you feel like it. Not recommended for everyday
829 use!</para>
830
831 <variablelist>
832
833 <varlistentry>
834 <term>
835 <option>-B</option>
836 <indexterm><primary><option>-B</option></primary><secondary>RTS option</secondary></indexterm>
837 </term>
838 <listitem>
839 <para>Sound the bell at the start of each (major) garbage
840 collection.</para>
841
842 <para>Oddly enough, people really do use this option! Our
843 pal in Durham (England), Paul Callaghan, writes: &ldquo;Some
844 people here use it for a variety of
845 purposes&mdash;honestly!&mdash;e.g., confirmation that the
846 code/machine is doing something, infinite loop detection,
847 gauging cost of recently added code. Certain people can even
848 tell what stage &lsqb;the program&rsqb; is in by the beep
849 pattern. But the major use is for annoying others in the
850 same office&hellip;&rdquo;</para>
851 </listitem>
852 </varlistentry>
853
854 <varlistentry>
855 <term>
856 <option>-D</option><replaceable>num</replaceable>
857 <indexterm><primary>-D</primary><secondary>RTS option</secondary></indexterm>
858 </term>
859 <listitem>
860 <para>An RTS debugging flag; varying quantities of output
861 depending on which bits are set in
862 <replaceable>num</replaceable>. Only works if the RTS was
863 compiled with the <option>DEBUG</option> option.</para>
864 </listitem>
865 </varlistentry>
866
867 <varlistentry>
868 <term>
869 <option>-r</option><replaceable>file</replaceable>
870 <indexterm><primary><option>-r</option></primary><secondary>RTS option</secondary></indexterm>
871 <indexterm><primary>ticky ticky profiling</primary></indexterm>
872 <indexterm><primary>profiling</primary><secondary>ticky ticky</secondary></indexterm>
873 </term>
874 <listitem>
875 <para>Produce &ldquo;ticky-ticky&rdquo; statistics at the
876 end of the program run. The <replaceable>file</replaceable>
877 business works just like on the <option>-S</option> RTS
878 option (above).</para>
879
880 <para>&ldquo;Ticky-ticky&rdquo; statistics are counts of
881 various program actions (updates, enters, etc.) The program
882 must have been compiled using
883 <option>-ticky</option><indexterm><primary><option>-ticky</option></primary></indexterm>
884 (a.k.a. &ldquo;ticky-ticky profiling&rdquo;), and, for it to
885 be really useful, linked with suitable system libraries.
886 Not a trivial undertaking: consult the installation guide on
887 how to set things up for easy &ldquo;ticky-ticky&rdquo;
888 profiling. For more information, see <xref
889 linkend="ticky-ticky"/>.</para>
890 </listitem>
891 </varlistentry>
892
893 <varlistentry>
894 <term>
895 <option>-xc</option>
896 <indexterm><primary><option>-xc</option></primary><secondary>RTS option</secondary></indexterm>
897 </term>
898 <listitem>
899 <para>(Only available when the program is compiled for
900 profiling.) When an exception is raised in the program,
901 this option causes the current cost-centre-stack to be
902 dumped to <literal>stderr</literal>.</para>
903
904 <para>This can be particularly useful for debugging: if your
905 program is complaining about a <literal>head []</literal>
906 error and you haven't got a clue which bit of code is
907 causing it, compiling with <literal>-prof
908 -auto-all</literal> and running with <literal>+RTS -xc
909 -RTS</literal> will tell you exactly the call stack at the
910 point the error was raised.</para>
911
912 <para>The output contains one line for each exception raised
913 in the program (the program might raise and catch several
914 exceptions during its execution), where each line is of the
915 form:</para>
916
917 <screen>
918 &lt; cc<subscript>1</subscript>, ..., cc<subscript>n</subscript> &gt;
919 </screen>
920 <para>each <literal>cc</literal><subscript>i</subscript> is
921 a cost centre in the program (see <xref
922 linkend="cost-centres"/>), and the sequence represents the
923 &ldquo;call stack&rdquo; at the point the exception was
924 raised. The leftmost item is the innermost function in the
925 call stack, and the rightmost item is the outermost
926 function.</para>
927
928 </listitem>
929 </varlistentry>
930
931 <varlistentry>
932 <term>
933 <option>-Z</option>
934 <indexterm><primary><option>-Z</option></primary><secondary>RTS option</secondary></indexterm>
935 </term>
936 <listitem>
937 <para>Turn <emphasis>off</emphasis> &ldquo;update-frame
938 squeezing&rdquo; at garbage-collection time. (There's no
939 particularly good reason to turn it off, except to ensure
940 the accuracy of certain data collected regarding thunk entry
941 counts.)</para>
942 </listitem>
943 </varlistentry>
944 </variablelist>
945
946 </sect2>
947
948 <sect2 id="rts-hooks">
949 <title>&ldquo;Hooks&rdquo; to change RTS behaviour</title>
950
951 <indexterm><primary>hooks</primary><secondary>RTS</secondary></indexterm>
952 <indexterm><primary>RTS hooks</primary></indexterm>
953 <indexterm><primary>RTS behaviour, changing</primary></indexterm>
954
955 <para>GHC lets you exercise rudimentary control over the RTS
956 settings for any given program, by compiling in a
957 &ldquo;hook&rdquo; that is called by the run-time system. The RTS
958 contains stub definitions for all these hooks, but by writing your
959 own version and linking it on the GHC command line, you can
960 override the defaults.</para>
961
962 <para>Owing to the vagaries of DLL linking, these hooks don't work
963 under Windows when the program is built dynamically.</para>
964
965 <para>The hook <literal>ghc_rts_opts</literal><indexterm><primary><literal>ghc_rts_opts</literal></primary>
966 </indexterm>lets you set RTS
967 options permanently for a given program. A common use for this is
968 to give your program a default heap and/or stack size that is
969 greater than the default. For example, to set <literal>-H128m
970 -K1m</literal>, place the following definition in a C source
971 file:</para>
972
973 <programlisting>
974 char *ghc_rts_opts = "-H128m -K1m";
975 </programlisting>
976
977 <para>Compile the C file, and include the object file on the
978 command line when you link your Haskell program.</para>
979
980 <para>These flags are interpreted first, before any RTS flags from
981 the <literal>GHCRTS</literal> environment variable and any flags
982 on the command line.</para>
983
984 <para>You can also change the messages printed when the runtime
985 system &ldquo;blows up,&rdquo; e.g., on stack overflow. The hooks
986 for these are as follows:</para>
987
988 <variablelist>
989
990 <varlistentry>
991 <term>
992 <function>void OutOfHeapHook (unsigned long, unsigned long)</function>
993 <indexterm><primary><function>OutOfHeapHook</function></primary></indexterm>
994 </term>
995 <listitem>
996 <para>The heap-overflow message.</para>
997 </listitem>
998 </varlistentry>
999
1000 <varlistentry>
1001 <term>
1002 <function>void StackOverflowHook (long int)</function>
1003 <indexterm><primary><function>StackOverflowHook</function></primary></indexterm>
1004 </term>
1005 <listitem>
1006 <para>The stack-overflow message.</para>
1007 </listitem>
1008 </varlistentry>
1009
1010 <varlistentry>
1011 <term>
1012 <function>void MallocFailHook (long int)</function>
1013 <indexterm><primary><function>MallocFailHook</function></primary></indexterm>
1014 </term>
1015 <listitem>
1016 <para>The message printed if <function>malloc</function>
1017 fails.</para>
1018 </listitem>
1019 </varlistentry>
1020 </variablelist>
1021
1022 <para>For examples of the use of these hooks, see GHC's own
1023 versions in the file
1024 <filename>ghc/compiler/parser/hschooks.c</filename> in a GHC
1025 source tree.</para>
1026 </sect2>
1027
1028 <sect2>
1029 <title>Getting information about the RTS</title>
1030
1031 <indexterm><primary>RTS</primary></indexterm>
1032
1033 <para>It is possible to ask the RTS to give some information about
1034 itself. To do this, use the <option>--info</option> flag, e.g.</para>
1035 <screen>
1036 $ ./a.out +RTS --info
1037 [("GHC RTS", "YES")
1038 ,("GHC version", "6.7")
1039 ,("RTS way", "rts_p")
1040 ,("Host platform", "x86_64-unknown-linux")
1041 ,("Host architecture", "x86_64")
1042 ,("Host OS", "linux")
1043 ,("Host vendor", "unknown")
1044 ,("Build platform", "x86_64-unknown-linux")
1045 ,("Build architecture", "x86_64")
1046 ,("Build OS", "linux")
1047 ,("Build vendor", "unknown")
1048 ,("Target platform", "x86_64-unknown-linux")
1049 ,("Target architecture", "x86_64")
1050 ,("Target OS", "linux")
1051 ,("Target vendor", "unknown")
1052 ,("Word size", "64")
1053 ,("Compiler unregisterised", "NO")
1054 ,("Tables next to code", "YES")
1055 ]
1056 </screen>
1057 <para>The information is formatted such that it can be read as a
1058 of type <literal>[(String, String)]</literal>. Currently the following
1059 fields are present:</para>
1060
1061 <variablelist>
1062
1063 <varlistentry>
1064 <term><literal>GHC RTS</literal></term>
1065 <listitem>
1066 <para>Is this program linked against the GHC RTS? (always
1067 "YES").</para>
1068 </listitem>
1069 </varlistentry>
1070
1071 <varlistentry>
1072 <term><literal>GHC version</literal></term>
1073 <listitem>
1074 <para>The version of GHC used to compile this program.</para>
1075 </listitem>
1076 </varlistentry>
1077
1078 <varlistentry>
1079 <term><literal>RTS way</literal></term>
1080 <listitem>
1081 <para>The variant (&ldquo;way&rdquo;) of the runtime. The
1082 most common values are <literal>rts</literal> (vanilla),
1083 <literal>rts_thr</literal> (threaded runtime, i.e. linked using the
1084 <literal>-threaded</literal> option) and <literal>rts_p</literal>
1085 (profiling runtime, i.e. linked using the <literal>-prof</literal>
1086 option). Other variants include <literal>debug</literal>
1087 (linked using <literal>-debug</literal>),
1088 <literal>t</literal> (ticky-ticky profiling) and
1089 <literal>dyn</literal> (the RTS is
1090 linked in dynamically, i.e. a shared library, rather than statically
1091 linked into the executable itself). These can be combined,
1092 e.g. you might have <literal>rts_thr_debug_p</literal>.</para>
1093 </listitem>
1094 </varlistentry>
1095
1096 <varlistentry>
1097 <term>
1098 <literal>Target platform</literal>,
1099 <literal>Target architecture</literal>,
1100 <literal>Target OS</literal>,
1101 <literal>Target vendor</literal>
1102 </term>
1103 <listitem>
1104 <para>These are the platform the program is compiled to run on.</para>
1105 </listitem>
1106 </varlistentry>
1107
1108 <varlistentry>
1109 <term>
1110 <literal>Build platform</literal>,
1111 <literal>Build architecture</literal>,
1112 <literal>Build OS</literal>,
1113 <literal>Build vendor</literal>
1114 </term>
1115 <listitem>
1116 <para>These are the platform where the program was built
1117 on. (That is, the target platform of GHC itself.) Ordinarily
1118 this is identical to the target platform. (It could potentially
1119 be different if cross-compiling.)</para>
1120 </listitem>
1121 </varlistentry>
1122
1123 <varlistentry>
1124 <term>
1125 <literal>Host platform</literal>,
1126 <literal>Host architecture</literal>
1127 <literal>Host OS</literal>
1128 <literal>Host vendor</literal>
1129 </term>
1130 <listitem>
1131 <para>These are the platform where GHC itself was compiled.
1132 Again, this would normally be identical to the build and
1133 target platforms.</para>
1134 </listitem>
1135 </varlistentry>
1136
1137 <varlistentry>
1138 <term><literal>Word size</literal></term>
1139 <listitem>
1140 <para>Either <literal>"32"</literal> or <literal>"64"</literal>,
1141 reflecting the word size of the target platform.</para>
1142 </listitem>
1143 </varlistentry>
1144
1145 <varlistentry>
1146 <term><literal>Compiler unregistered</literal></term>
1147 <listitem>
1148 <para>Was this program compiled with an &ldquo;unregistered&rdquo;
1149 version of GHC? (I.e., a version of GHC that has no platform-specific
1150 optimisations compiled in, usually because this is a currently
1151 unsupported platform.) This value will usually be no, unless you're
1152 using an experimental build of GHC.</para>
1153 </listitem>
1154 </varlistentry>
1155
1156 <varlistentry>
1157 <term><literal>Tables next to code</literal></term>
1158 <listitem>
1159 <para>Putting info tables directly next to entry code is a useful
1160 performance optimisation that is not available on all platforms.
1161 This field tells you whether the program has been compiled with
1162 this optimisation. (Usually yes, except on unusual platforms.)</para>
1163 </listitem>
1164 </varlistentry>
1165
1166 </variablelist>
1167
1168 </sect2>
1169 </sect1>
1170
1171 <!-- Emacs stuff:
1172 ;;; Local Variables: ***
1173 ;;; mode: xml ***
1174 ;;; sgml-parent-document: ("users_guide.xml" "book" "chapter" "sect1") ***
1175 ;;; End: ***
1176 -->