source: doc/theses/colby_parsons_MMAth/benchmarks/actors/run@ ab81e3b

ADT ast-experimental
Last change on this file since ab81e3b was ab81e3b, checked in by caparson <caparson@…>, 3 years ago

updated run script and plotting script

  • Property mode set to 100755
File size: 16.8 KB
Line 
1#!/bin/bash -
2
3false=0; true=1
4
5# Usage: arch [ hostname ] returns hostname, cores, startcore
6#
7# Define machine architecture based on starting socket, CPUs (cores) per socket, number of
8# sockets, has hyperthreading.
9
10start=0
11
12arch() {
13 hostname=${1:-`hostname`} # return value
14 hashyper=${true} # assume machine has hyperthreads
15 if [ "${hostname}" = "plg2" ] ; then
16 startsocket=${start}
17 cps=16 # coresPerSocket
18 sockets=2
19 hashyper=${false} # has no hyperthreads
20 elif [ "${hostname}" = "nasus" ] ; then
21 startsocket=${start}
22 cps=64 # coresPerSocket
23 sockets=2
24 elif [ "${hostname}" = "pyke" ] ; then
25 startsocket=${start}
26 cps=24 # coresPerSocket
27 sockets=2
28 elif [ "${hostname}" = "jax" ] ; then
29 startsocket=${start}
30 cps=24 # coresPerSocket
31 sockets=4
32 else
33 echo "unsupported host" ${hostname}
34 exit 1
35 fi
36 cores=$(( ${cps} * ${sockets} ))
37 startcore=$(( ${startsocket} * ${cps} ))
38}
39
40# Usage: affinity (global cps, sockets, startsocket, hashyper, cores, startcore, wrap)
41# returns taskset argument
42#
43# This routine assumes hyperthreading has only 2 hyperthreads per core.
44#
45# If hyperthread scanning is used: processor units are assigned across the low-number hyperthreads
46# of the socket's cores. When the low-number hyperthreads are filled, the high-number hyperhtreads
47# are assigned across the socket's cores. Then the next socket is assigned.
48#
49# If hyperthread wrapping is used: processor units are assigned in low/high-number pairs of
50# hyperthreads across the socket's cores. Then the next socket is assigned.
51
52wrap=${false} # set to control hyperthread assignment across socket cores
53
54affinity() {
55 if [ ${wrap} -eq ${true} -a ${hashyper} -eq ${false} ] ; then
56 echo "architecture does not support hyperthreading for wrapping"
57 exit 1
58 fi
59 taskset="" # return value
60 set -- $(( ${1} - 1 )) # decrement $1
61 if [ ${1} -eq 0 ] ; then taskset="${startcore}-${startcore}"; return; fi
62 if [ ${1} -ge $(( ${cps} * ( ${sockets} - ${startsocket} ) * ( ${hashyper} + 1 ) )) ] ; then # error
63 echo "not enough cores $(( ${cores} * ${sockets} )) for $(( ${1} + 1 )) starting at ${startcore}"
64 exit 1
65 fi
66 if [ ${hashyper} -eq ${false} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi # no hyperthreads
67 start2=$(( ${startcore} + ${cores} ))
68 if [ ${wrap} -eq ${true} ] ; then # hyperthread wrapping
69 end1=$(( ${1} / 2 + ${startcore} ))
70 end2=$(( ${end1} + ${cores} ))
71 if [ $(( ${1} % 2 )) -eq 0 ] ; then
72 end2=$(( ${end2} - 1 ))
73 fi
74 taskset="${startcore}-${end1},${start2}-${end2}"
75 else # hyperthread scanning
76 if [ ${1} -lt ${cps} ] ; then taskset="${startcore}-$(( ${1} + ${startcore} ))"; return; fi
77 filled=$(( ${1} / ( ${cps} * 2 ) * ${cps} ))
78 modulus=$(( ${1} % ( ${cps} * 2 ) )) # leftover cores added to saturated sockets
79 if [ ${modulus} -gt ${cps} ] ; then
80 taskset="${startcore}-$(( ${startcore} + ${filled} + ${cps} - 1 )),${start2}-$(( ${start2} + ${filled} + ${modulus} % ${cps} ))"
81 else
82 taskset="${startcore}-$(( ${startcore} + ${filled} + ${modulus} )),${start2}-$(( ${start2} + ${filled} - 1 ))"
83 fi
84 fi
85}
86
87# executor config
88batch='100'
89nRounds='400' #500
90
91# matrix config
92size='3072'
93
94# repeat config
95messages='100000'
96n_repeats='200' #200
97
98# balance config
99nOneRounds='2000'
100nMultiRounds='800'
101
102# static config
103n_static_sends='100000000'
104
105# dynamic config
106n_dynamic_sends='20000000'
107
108numtimes=5
109
110# bench_cores='1 2 4 8 16 24 32'
111# bench_cores='1 2 4 8 16 24 32 48'
112bench_cores='48'
113
114# toggle benchmarks
115executor=${false}
116matrix=${false}
117repeat=${false}
118balance=${false}
119static=${true}
120dynamic=${true}
121
122# names=('CFA-LV' 'CFA-NS' 'CFA-R')
123# var_flags=('-D__STEAL=1 -DSEARCH=1' '' '-D__STEAL=1 -DRAND=1')
124
125# names=('CFA-NS')
126# var_flags=('')
127
128names=('CFA')
129var_flags=('-D__STEAL=1 -DSEARCH=1')
130
131# names=('CFA' 'CFA-STAT')
132# var_flags=('-D__STEAL=1 -DSEARCH=1' '-D__STEAL=1 -DSTATS')
133
134# names=()
135# var_flags=()
136
137runCAF=${true}
138runUCPP=${true}
139runPROTO=${true}
140runAKKA=${true}
141# runCAF=${false}
142# runUCPP=${false}
143# runPROTO=${false}
144# runAKKA=${false}
145
146cfa=~/cfa-cc/driver/cfa
147
148# Helpers to minimize code duplication
149
150# repeats a command ${numtimes}
151preprint=''
152repeat_command() {
153 t=1
154 while [ ${t} -le ${numtimes} ] ; do
155 echo -n -e ${preprint}
156 "${@}"
157 t=`expr ${t} + 1`
158 done
159}
160
161# prints the leading info for a given run of a variant
162print_header() {
163 echo ${1}':'
164 echo -e "proc\ttime (s)"
165}
166
167# runs the current benchmark with provided args
168# only works for standard-run benchmarks (not Akka)
169# must split into pre and post args to be able to supply val of p
170pre_args=''
171post_args=''
172single_run() {
173 affinity ${1}
174 preprint="${1}\t"
175 repeat_command taskset -c ${taskset} ./a.${hostname} ${pre_args} ${1} ${post_args}
176}
177
178# runs the current bench for all processor vals
179# works for standard benchs that dont need to set a config file (not Akka or CAF)
180run_bench() {
181 for p in ${bench_cores} ; do
182 single_run ${p}
183 done
184}
185
186set_akka_parallelism() {
187 sed -i "s/parallelism-min = .*/parallelism-min = ${1}/g" application.conf
188 sed -i "s/parallelism-max = .*/parallelism-max = ${1}/g" application.conf
189}
190
191arch # get hostname
192
193# set up leading info for python script
194echo $numtimes
195echo $bench_cores
196
197for i in ${!names[@]}; do
198 echo -n ${names[$i]}" "
199done
200if [ ${runCAF} -eq ${true} ] ; then
201 echo -n 'CAF '
202fi # done CAF
203if [ ${runAKKA} -eq ${true} ] ; then
204 echo -n 'Akka '
205fi # done AKKA
206if [ ${runUCPP} -eq ${true} ] ; then
207 echo -n 'uC++ '
208fi # done UCPP
209if [ ${runPROTO} -eq ${true} ] ; then
210 echo -n 'ProtoActor '
211fi
212echo ""
213
214# done printing header info for output
215
216# cfa flags
217cfa_flags='-quiet -O3 -nodebug -DNDEBUG'
218
219# CAF flags
220prefixCAF='-O3 -Wall -std=c++17 -I/home/pabuhr/software/nasus/Actors/experiments/CAF/actor-framework/libcaf_core -I/home/pabuhr/software/nasus/Actors/experiments/CAF/actor-framework/libcaf_core/caf -I/home/pabuhr/software/nasus/Actors/experiments/CAF/actor-framework/build/libcaf_core -L/home/pabuhr/software/nasus/Actors/experiments/CAF/actor-framework/build/libcaf_core -L/home/pabuhr/software/nasus/Actors/experiments/CAF/actor-framework/build/libcaf_io'
221suffixCAF='-lcaf_io -lcaf_core -Wl,-rpath=CAF/actor-framework/build/libcaf_core'
222
223# AKKA flags
224sbtflags="--warn -J-Xmx32g"
225
226# UCPP flags
227UCPPflags="-quiet -g -Wall -Wextra -O3 -nodebug -DNDEBUG -multi"
228UCPP=~/ucpp/u++-7.0.0/bin/u++
229
230# run the benchmarks
231
232# /usr/bin/time -f "%Uu %Ss %Er %Mkb"
233if [ ${executor} -eq ${true} ] ; then
234 echo "executor"
235 pre_args="40000 100 ${nRounds}"
236 post_args="${batch}"
237 cd cfa # CFA RUN
238 for i in ${!names[@]}; do
239 print_header ${names[$i]}
240 ${cfa} ${cfa_flags} ${var_flags[$i]} executor.cfa -o a.${hostname} > /dev/null 2>&1
241 run_bench
242 rm a.${hostname}
243 done
244 cd - > /dev/null # done CFA
245 if [ ${runCAF} -eq ${true} ] ; then # CAF RUN
246 cd caf
247 print_header 'CAF'
248 g++ ${prefixCAF} CAFExecutor.cpp ${suffixCAF} -o a.${hostname} > /dev/null 2>&1
249 for p in ${bench_cores} ; do
250 sed -i "s/max-threads = .*/max-threads = ${p}/g" caf-application.conf # set CAF parallelism
251 single_run ${p}
252 done
253 rm a.${hostname}
254 # set back parallelism
255 sed -i "s/max-threads = .*/max-threads = 1/g" caf-application.conf
256 cd - > /dev/null
257 fi # done CAF
258 if [ ${runAKKA} -eq ${true} ] ; then # AKKA RUN
259 cd akka/Executor
260 rm -rf project target # random out of memory errors without this
261 print_header 'Akka'
262 for p in ${bench_cores} ; do
263 set_akka_parallelism ${p}
264 affinity ${p}
265 taskset -c ${taskset} sbt ${sbtflags} "run ${pre_args} ${p} ${post_args} ${numtimes}" 2>&1 | grep -v "SLF4J:" | grep -v "Slf4jLogger started" | sed -e "s/\x1b\[0J//"
266 sbt clean > /dev/null
267 done
268 # set back parallelism
269 set_akka_parallelism 1
270 cd - > /dev/null
271 fi # done AKKA
272 if [ ${runUCPP} -eq ${true} ] ; then # UCPP RUN
273 cd ucpp
274 print_header 'uC++'
275 ${UCPP} ${UCPPflags} uC++Executor.cc -o a.${hostname} > /dev/null 2>&1
276 run_bench
277 rm a.${hostname}
278 cd - > /dev/null
279 fi # done UCPP
280 if [ ${runPROTO} -eq ${true} ] ; then # PROTO RUN
281 print_header 'ProtoActor'
282 cd proto/Executor
283 go build -o a.${hostname} > /dev/null 2>&1
284 run_bench
285 rm a.${hostname}
286 cd - > /dev/null
287 fi # done PROTO
288 echo ""
289fi
290
291if [ ${matrix} -eq ${true} ] ; then
292 echo "matrix"
293 pre_args="${size} ${size} ${size}"
294 post_args=""
295 cd cfa
296 for i in ${!names[@]}; do
297 print_header ${names[$i]}
298 ${cfa} ${cfa_flags} ${var_flags[$i]} matrix.cfa -o a.${hostname} > /dev/null 2>&1
299 run_bench
300 rm a.${hostname}
301 done
302 cd - > /dev/null
303 if [ ${runCAF} -eq ${true} ] ; then # CAF RUN
304 cd caf
305 print_header 'CAF'
306 g++ ${prefixCAF} CAFMatrix.cpp ${suffixCAF} -o a.${hostname} > /dev/null 2>&1
307 for p in ${bench_cores} ; do
308 sed -i "s/max-threads = .*/max-threads = ${p}/g" caf-application.conf # set CAF parallelism
309 single_run ${p}
310 done
311 rm a.${hostname}
312
313 # set back parallelism
314 sed -i "s/max-threads = .*/max-threads = 1/g" caf-application.conf
315 cd - > /dev/null
316 fi # done CAF
317 if [ ${runAKKA} -eq ${true} ] ; then # AKKA RUN
318 cd akka/Matrix
319 rm -rf project target # random out of memory errors without this
320 print_header 'Akka'
321 for p in ${bench_cores} ; do
322 set_akka_parallelism ${p}
323 affinity ${p}
324 repeat_command taskset -c ${taskset} sbt ${sbtflags} "run ${pre_args} ${p}" 2>&1 | grep -v "SLF4J:" | grep -v "Slf4jLogger started" | sed -e "s/\x1b\[0J//"
325 sbt clean > /dev/null
326 done
327 # set back parallelism
328 set_akka_parallelism 1
329 cd - > /dev/null
330 fi # done AKKA
331 if [ ${runUCPP} -eq ${true} ] ; then # UCPP RUN
332 cd ucpp
333 print_header 'uC++'
334 ${UCPP} ${UCPPflags} uC++Matrix.cc -o a.${hostname} > /dev/null 2>&1
335 run_bench
336 rm a.${hostname}
337 cd - > /dev/null
338 fi # done UCPP
339 if [ ${runPROTO} -eq ${true} ] ; then # PROTO RUN
340 cd proto/Matrix
341 print_header 'ProtoActor'
342 go build -o a.${hostname} > /dev/null 2>&1
343 run_bench
344 rm a.${hostname}
345 cd - > /dev/null
346 fi # done PROTO
347 echo ""
348fi
349
350if [ ${repeat} -eq ${true} ] ; then
351 echo "repeat"
352 pre_args="${messages}"
353 post_args="${n_repeats}"
354 cd cfa
355 for i in ${!names[@]}; do
356 print_header ${names[$i]}
357 ${cfa} ${cfa_flags} ${var_flags[$i]} repeat.cfa -o a.${hostname} > /dev/null 2>&1
358 run_bench
359 rm a.${hostname}
360 done
361 cd - > /dev/null
362 if [ ${runCAF} -eq ${true} ] ; then # CAF RUN
363 cd caf
364 print_header 'CAF'
365 g++ ${prefixCAF} CAFRepeat.cpp ${suffixCAF} -o a.${hostname} > /dev/null 2>&1
366 for p in ${bench_cores} ; do
367 sed -i "s/max-threads = .*/max-threads = ${p}/g" caf-application.conf # set CAF parallelism
368 single_run ${p}
369 done
370 rm a.${hostname}
371
372 # set back parallelism
373 sed -i "s/max-threads = .*/max-threads = 1/g" caf-application.conf
374 cd - > /dev/null
375 fi # done CAF
376 if [ ${runAKKA} -eq ${true} ] ; then # AKKA RUN
377 cd akka/Repeat
378 rm -rf project target # random out of memory errors without this
379 print_header 'Akka'
380 for p in ${bench_cores} ; do
381 set_akka_parallelism ${p}
382 affinity ${p}
383 repeat_command taskset -c ${taskset} sbt ${sbtflags} "run ${pre_args} ${p} ${post_args}" 2>&1 | grep -v "SLF4J:" | grep -v "Slf4jLogger started" | sed -e "s/\x1b\[0J//"
384 sbt clean > /dev/null
385 done
386 # set back parallelism
387 set_akka_parallelism 1
388 cd - > /dev/null
389 fi # done AKKA
390 if [ ${runUCPP} -eq ${true} ] ; then # UCPP RUN
391 cd ucpp
392 print_header 'uC++'
393 ${UCPP} ${UCPPflags} uC++Repeat.cc -o a.${hostname} > /dev/null 2>&1
394 run_bench
395 rm a.${hostname}
396 cd - > /dev/null
397 fi # done UCPP
398 if [ ${runPROTO} -eq ${true} ] ; then # PROTO RUN
399 print_header 'ProtoActor'
400 cd proto/Repeat
401 go build -o a.${hostname} > /dev/null 2>&1
402 run_bench
403 rm a.${hostname}
404 cd - > /dev/null
405 fi # done PROTO
406 echo ""
407fi
408
409
410if [ ${static} -eq ${true} ] ; then
411 echo "static"
412 preprint=''
413 cd cfa
414 for i in ${!names[@]}; do
415 echo ${names[$i]}
416 ${cfa} ${cfa_flags} ${var_flags[$i]} static.cfa -o a.${hostname} > /dev/null 2>&1
417 repeat_command taskset -c ${startcore} ./a.${hostname} ${n_static_sends}
418 rm a.${hostname}
419 done
420 cd - > /dev/null
421 if [ ${runCAF} -eq ${true} ] ; then # CAF RUN
422 cd caf
423 echo 'CAF:'
424 g++ ${prefixCAF} CAFSendStatic.cpp ${suffixCAF} -o a.${hostname} > /dev/null 2>&1
425 sed -i "s/max-threads = .*/max-threads = 1/g" caf-application.conf # set CAF parallelism
426 repeat_command taskset -c ${startcore} ./a.${hostname} 'd'
427 rm a.${hostname}
428 cd - > /dev/null
429 fi # done CAF
430 if [ ${runAKKA} -eq ${true} ] ; then # AKKA RUN
431 cd akka/SendStatic
432 rm -rf project target # random out of memory errors without this
433 echo 'Akka:'
434 set_akka_parallelism 1
435 taskset -c ${startcore} sbt ${sbtflags} "run ${n_static_sends} ${numtimes}" 2>&1 | grep -v "SLF4J:" | grep -v "Slf4jLogger started" | sed -e "s/\x1b\[0J//"
436 cd - > /dev/null
437 fi # done AKKA
438 if [ ${runUCPP} -eq ${true} ] ; then # UCPP RUN
439 cd ucpp
440 echo 'uC++:'
441 ${UCPP} ${UCPPflags} uC++SendStatic.cc -o a.${hostname} > /dev/null 2>&1
442 repeat_command taskset -c ${startcore} ./a.${hostname} ${n_static_sends}
443 rm a.${hostname}
444 cd - > /dev/null
445 fi # done UCPP
446 if [ ${runPROTO} -eq ${true} ] ; then # PROTO RUN
447 cd proto/SendStatic
448 echo 'ProtoActor:'
449 go build -o a.${hostname} > /dev/null 2>&1
450 repeat_command taskset -c ${startcore} ./a.${hostname} ${n_static_sends}
451 rm a.${hostname}
452 cd - > /dev/null
453 fi # done PROTO
454 echo ""
455fi
456
457if [ ${dynamic} -eq ${true} ] ; then
458 echo "dynamic"
459 cd cfa
460 for i in ${!names[@]}; do
461 echo ${names[$i]}
462 ${cfa} ${cfa_flags} ${var_flags[$i]} dynamic.cfa -o a.${hostname} > /dev/null 2>&1
463 repeat_command taskset -c ${startcore} ./a.${hostname} ${n_dynamic_sends}
464 rm a.${hostname}
465 done
466 cd - > /dev/null
467 if [ ${runCAF} -eq ${true} ] ; then # CAF RUN
468 cd caf
469 echo 'CAF:'
470 g++ ${prefixCAF} CAFSendDynamic.cpp ${suffixCAF} -o a.${hostname} > /dev/null 2>&1
471 sed -i "s/max-threads = .*/max-threads = 1/g" caf-application.conf # set CAF parallelism
472 repeat_command taskset -c ${startcore} ./a.${hostname} 'd'
473 rm a.${hostname}
474 cd - > /dev/null
475 fi # done CAF
476 if [ ${runAKKA} -eq ${true} ] ; then # AKKA RUN
477 cd akka/SendDynamic
478 rm -rf project target # random out of memory errors without this
479 echo 'Akka:'
480 set_akka_parallelism 1
481 taskset -c ${startcore} sbt ${sbtflags} "run d ${numtimes}" 2>&1 | grep -v "SLF4J:" | grep -v "Slf4jLogger started" | sed -e "s/\x1b\[0J//"
482 cd - > /dev/null
483 fi # done AKKA
484 if [ ${runUCPP} -eq ${true} ] ; then # UCPP RUN
485 cd ucpp
486 echo 'uC++:'
487 ${UCPP} ${UCPPflags} uC++SendDynamic.cc -o a.${hostname} > /dev/null 2>&1
488 repeat_command taskset -c ${startcore} ./a.${hostname} ${n_dynamic_sends}
489 rm a.${hostname}
490 cd - > /dev/null
491 fi # done UCPP
492 if [ ${runPROTO} -eq ${true} ] ; then # PROTO RUN
493 cd proto/SendDynamic
494 echo 'ProtoActor:'
495 go build -o a.${hostname} > /dev/null 2>&1
496 repeat_command taskset -c ${startcore} ./a.${hostname} 'd'
497 rm a.${hostname}
498 cd - > /dev/null
499 fi # done PROTO
500 echo ""
501fi
502
503
504if [ ${balance} -eq ${true} ] ; then
505 cd cfa
506 echo "balance_one"
507 for i in ${!names[@]}; do
508 echo ${names[$i]}':'
509 echo -e "proc\ttime (s)"
510 ${cfa} ${cfa_flags} ${var_flags[$i]} balance.cfa -o a.${hostname} > /dev/null 2>&1
511 for p in ${bench_cores} ; do
512 affinity ${p}
513 preprint="${p}\t"
514 repeat_command taskset -c ${taskset} ./a.${hostname} 32 32 ${nOneRounds} ${p}
515 done
516 rm a.${hostname}
517 done
518 echo ""
519 echo "balance_multi"
520 for i in ${!names[@]}; do
521 echo ${names[$i]}':'
522 echo -e "proc\ttime (s)"
523 ${cfa} ${cfa_flags} ${var_flags[$i]} -DMULTI balance.cfa -o a.${hostname} > /dev/null 2>&1
524 for p in ${bench_cores} ; do
525 affinity ${p}
526 preprint="${p}\t"
527 repeat_command taskset -c ${taskset} ./a.${hostname} 32 32 ${nMultiRounds} ${p}
528 done
529 rm a.${hostname}
530 done
531 echo ""
532 cd - > /dev/null
533fi
534
Note: See TracBrowser for help on using the repository browser.