Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1516

Last change on this file since 1516 was 1481, checked in by raasch, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 52.4 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2014 Leibniz Universitaet Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1481 2014-10-17 14:42:35Z boeske $
30	#
31	# 1480 2014-10-17 14:41:49Z raasch
32	# adjustments for 2nd stage of HLRNIII
33	#
34	# 1468 2014-09-24 14:06:57Z maronga
35	# Typo removed (addres->address)
36	# Adjustments for lcxe6
37	#
38	# 1452 2014-08-22 09:41:06Z heinze
39	# local hosts for blizzard added
40	#
41	# 1450 2014-08-21 07:31:51Z heinze
42	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
43	#
44	# 1442 2014-07-28 07:09:10Z raasch
45	# HLRN-III (lccrayb/lccrayh) queues adjusted
46	#
47	# 1378 2014-04-28 06:04:58Z raasch
48	# -et option added for lctit
49	#
50	# 1350 2014-04-04 13:01:30Z maronga
51	# location of qsub updated for lcxe6
52	#
53	# 1289 2014-03-04 07:12:34Z raasch
54	# German comments translated to English
55	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
56	#
57	# 1279 2014-01-28 12:10:14Z raasch
58	# node calculation modified due to changes in mrun (tasks_per_node must not be
59	# an integral divisor of numprocs any more)
60	#
61	# 1274 2014-01-09 13:14:54Z heinze
62	# adjustments for lccrayh
63	#
64	# 1266 2013-12-11 12:07:34Z heinze
65	# further adjustments for lccrayb (use msub instead of qsub)
66	#
67	# 1264 2013-12-09 12:46:09Z fricke
68	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
69	#
70	# 1262 2013-12-09 10:57:20Z fricke
71	# further adjustments for lccrayb
72	#
73	# 1260 2013-12-04 12:48:04Z raasch
74	# jaboticaba admitted
75	#
76	# 1255 2013-11-07 14:43:35Z raasch
77	# further adjustments for lccrayb
78	#
79	# 1224 2013-09-16 07:27:23Z raasch
80	# first adjustments for lccrayb
81	#
82	# 1202 2013-07-10 16:22:07Z witha
83	# adjustments for Forwind cluster (lcflow)
84	#
85	# 1199 2013-07-05 14:52:22Z raasch
86	# adjustments for CSC Helsinki (lccrayf)
87	#
88	# use of cluster/express queue enabled (ibmh)
89	# vinessa added (imuk)
90	#
91	# 1103 2013-02-20 02:15:53Z raasch
92	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
93	# instead typeset -L/R),
94	# further adjustments for lckyuh
95	#
96	# 2013-02-10 01:47:43Z raasch
97	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
98	# and for Forwind cluster (lcflow)
99	#
100	# 1094 2013-02-03 01:52:12Z raasch
101	# new option -P for explicit setting of ssh/scp port,
102	# decalpha parts (yonsei) removed
103	#
104	# 2013-02-02 07:06:13Z raasch
105	# adjustments for Kyushu-University computing center (lckyut - tatara)
106	# old changelog messages removed
107	#
108	# 1046 2012-11-09 14:38:45Z maronga
109	# code put under GPL (PALM 3.9)
110	#
111	# 08/07/94 - Siggi - first version finished
112	# 29/06/94 - Siggi - script development started
113	#--------------------------------------------------------------------------------#
114	# subjob - script for automatic generation and submission of batch-job files
115	# for various batch queuing systems
116	#--------------------------------------------------------------------------------#
117
118
119	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
120	delete_dayfile=false
121	email_notification=none
122	group_number=none
123	locat=normal
124	no_default_queue=none
125	no_submit=false
126	job_catalog="~/job_queue"
127	job_name=none
128	local_user=$LOGNAME
129	node_usage=shared
130	numprocs=0
131	punkte="..........................................................."
132	submcom=qsub
133	queue=default
134	remote_host=none
135	remote_user=""
136	verify=true
137
138	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
139	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
140
141
142
143	# ERROR HANDLING
144	# IN CASE OF EXIT:
145	trap 'if [[ $locat != normal ]]
146	then
147	case $locat in
148	(option) printf "\n --> available optios can be displayed"
149	printf " by typing:"
150	printf "\n \"subjob ?\" \n";;
151	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
152	(*) printf "\n +++ unknown error"
153	printf "\n please inform S. Raasch!\n"
154	esac
155	[[ -f $job_to_send ]] && rm $job_to_send
156	printf "\n\n+++ SUBJOB killed \n\n"
157	fi' exit
158
159
160	# IN CASE OF TERMINAL-BREAK:
161	trap '[[ -f $job_to_send ]] && rm $job_to_send
162	printf "\n\n+++ SUBJOB killed \n\n"
163	exit
164	' 2
165
166
167
168
169	# DETERMINE NAME OF LOCAL HOST
170	local_host=$(hostname)
171
172
173
174	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
175	# IS ADMITTED AT ALL)
176	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
177	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
178	case $local_host in
179	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
180	(atmos) local_address=172.20.25.35; local_host=lcide;;
181	(austru) local_address=130.75.105.128; local_host=lcmuk;;
182	(autan) local_address=130.75.105.57; local_host=lcmuk;;
183	(bora) local_address=130.75.105.103; local_host=lcmuk;;
184	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
185	(blizzard1\|p*) local_address=136.172.40.15; local_host=ibmh;;
186	(blizzard2\|p*) local_address=136.172.40.16; local_host=ibmh;;
187	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
188	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
189	(breva) local_address=130.75.105.98; local_host=lcmuk;;
190	(buran) local_address=130.75.105.58; local_host=lcmuk;;
191	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
192	(climate*) local_address=165.132.26.68; local_host=lcyon;;
193	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
194	(cs*) local_address=136.172.44.131; local_host=nech;;
195	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
196	(flow01) local_address=10.141.255.71; local_host=lcflow;;
197	(flow02) local_address=10.141.255.72; local_host=lcflow;;
198	(node*) local_address=165.132.26.61 local_host=lck;;
199	# (node*) local_address=210.219.61.8 local_host=lckordi;;
200	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
201	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
202	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
203	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
204	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
205	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
206	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
207	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
208	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
209	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
210	(sno) local_address=130.75.105.113; local_host=lcmuk;;
211	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
212	(login*) local_address=118.128.66.223; local_host=lckiaps;;
213	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
214	(meller) local_address=134.106.74.155; local_host=lcfor;;
215	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
216	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
217	(nobel*) local_address=150.183.5.101; local_host=ibms;;
218	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
219	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
220	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
221	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
222	(pingui) local_address=134.106.74.118; local_host=lcfor;;
223	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
224	(rte*) local_address=133.5.185.60; local_host=lcrte;;
225	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
226	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
227	(solano) local_address=130.75.105.110; local_host=lcmuk;;
228	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
229	(t2a*) local_address=10.1.6.165; local_host=lctit;;
230	(urban*) local_address=147.46.30.151 local_host=lcsb;;
231	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
232	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
233	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
234	(*) printf "\n +++ \"$local_host\" unknown";
235	printf "\n please contact the PALM group at IMUK";
236	locat=parameter; exit;;
237	esac
238
239
240
241	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
242	remote_host=$local_host
243
244
245
246
247	# READ THE SHELLSCRIPT-OPTIONS
248	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
249	do
250	case $option in
251	(c) job_catalog=$OPTARG;;
252	(d) delete_dayfile=true;;
253	(D) no_submit=true;;
254	(e) email_notification=$OPTARG;;
255	(g) group_number=$OPTARG;;
256	(h) remote_host=$OPTARG;;
257	(m) memory=$OPTARG;;
258	(n) job_name=$OPTARG;;
259	(N) node_usage=$OPTARG;;
260	(O) threads_per_task=$OPTARG;;
261	(P) scp_port=$OPTARG;;
262	(q) no_default_queue=$OPTARG;;
263	(t) cputime=$OPTARG;;
264	(T) tasks_per_node=$OPTARG;;
265	(u) remote_user=$OPTARG;;
266	(v) verify=false;;
267	(X) numprocs=$OPTARG;;
268	(\?) printf "\n +++ Option $OPTARG unknown \n";
269	locat=option; exit;;
270	esac
271	done
272
273
274	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
275	(( to_shift = $OPTIND - 1 ))
276	shift $to_shift; file_to_send=$1
277
278
279	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
280	if [ "$1" = "?" ]
281	then
282	(printf "\n *** subjob can be called as follows:\n"
283	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
284	printf "\n Description of available options:\n"
285	printf "\n Option Description Default-Value"
286	printf "\n -c job-input- and output-catalog ~/job_queue"
287	printf "\n -d no job-protocol will be created ---"
288	printf "\n -D only the job-file will be created ---"
289	printf "\n -h execution host, available hosts: $remote_host"
290	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
291	printf "\n lckiaps, lctit, nech"
292	printf "\n -m memory demand per process in MByte ---"
293	printf "\n -n jobname <jobdatei>"
294	printf "\n -O threads per task (for OpenMP usage) 1"
295	printf "\n -P ssh/scp port default port"
296	printf "\n -q job-queue to be used default"
297	printf "\n -t allowed cpu-time in seconds ---"
298	printf "\n -T tasks per node (on parallel hosts) ---"
299	printf "\n -u username on execution host from .netrc"
300	printf "\n -v no prompt for confirmation ---"
301	printf "\n -X # of processors (on parallel hosts) 1"
302	printf "\n "
303	printf "\n The only possible positional parameter is <jobfile>:"
304	printf "\n The complete NQS-job must be provided here."
305	printf "\n <jobfile>=? creates this outline\n\n") \| more
306	exit
307	fi
308
309
310
311	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
312	if [[ "$file_to_send" = "" ]]
313	then
314	printf "\n +++ job-file missing"
315	locat=parameter; exit
316	else
317	if [[ -f $file_to_send ]]
318	then
319	true
320	else
321	printf "\n +++ job-file: "
322	printf "\n $file_to_send"
323	printf "\n does not exist"
324	locat=parameter; exit
325	fi
326	fi
327
328
329
330	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
331	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
332	if [[ $job_name = none ]]
333	then
334	job_name=$file_to_send
335	fi
336	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
337	then
338	printf "\n +++ job-file name: "
339	printf "\n $job_name"
340	printf "\n must not contain \"/\"-characters"
341	locat=parameter; exit
342	fi
343
344
345
346
347	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
348	# OR IF NO HOST HAS BEEN GIVEN
349	if [[ $remote_host = none ]]
350	then
351	printf "\n +++ host missing"
352	locat=option; exit
353	else
354	case $remote_host in
355	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
356	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
357	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
358	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
359	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
360	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
361	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
362	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
363	(lcflow) remote_address=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
364	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
365	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
366	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
367	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
368	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
369	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
370	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
371	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
372	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
373	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
374	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
375	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
376	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
377	locat=parameter; exit;;
378	esac
379	fi
380
381
382	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
383	if [[ $no_default_queue != none ]]
384	then
385	error=false
386	ndq=$no_default_queue
387	case $remote_host in
388	(ibm) case $ndq in
389	(p690_express\|p690_standard\|p690_long) error=false;;
390	(*) error=true;;
391	esac;;
392	(ibmh) case $ndq in
393	(cluster\|express) error=false;;
394	(*) error=true;;
395	esac;;
396	(ibmkisti) case $ndq in
397	(class.32plus\|class.1-2\|class.2-32) error=false;;
398	(*) error=true;;
399	esac;;
400	(ibmku) case $ndq in
401	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
402	(*) error=true;;
403	esac;;
404	(ibms) case $ndq in
405	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
406	(*) error=true;;
407	esac;;
408	(lccrayb) case $ndq in
409	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
410	(*) error=true;;
411	esac;;
412	(lccrayh) case $ndq in
413	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
414	(*) error=true;;
415	esac;;
416	(lccrayf) case $ndq in
417	(usup\|test*\|small\|large) error=false;;
418	(*) error=true;;
419	esac;;
420	(lcflow) case $ndq in
421	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
422	(*) error=true;;
423	esac;;
424	(lckiaps) case $ndq in
425	(express\|normal) error=false;;
426	(*) error=true;;
427	esac;;
428	(lckyoto) case $ndq in
429	(eh\|ph) error=false;;
430	(*) error=true;;
431	esac;;
432	(lckyuh) case $ndq in
433	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
434	(*) error=true;;
435	esac;;
436	(lckyut) case $ndq in
437	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
438	(*) error=true;;
439	esac;;
440	(lctit) case $ndq in
441	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
442	(*) error=true;;
443	esac;;
444	(t3eb) case $ndq in
445	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
446	(*) error=true;;
447	esac;;
448	(t3eh) case $ndq in
449	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
450	(*) error=true;;
451	esac;;
452	(t3ej2\|t3ej5) case $ndq in
453	(low\|normal\|high) error=false;;
454	(*) error=true;;
455	esac;;
456	(t3es) case $ndq in
457	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
458	(*) error=true;;
459	esac;;
460	esac
461	if [[ $error = true ]]
462	then
463	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
464	locat=parameter; exit
465	else
466	queue=$no_default_queue
467	fi
468	fi
469
470
471
472	# CHECK THE CPU-TIME
473	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
474	done=false
475	while [[ $done = false ]]
476	do
477	if (( $cputime <= 0 ))
478	then
479	printf "\n +++ wrong cpu-time or cpu-time missing"
480	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
481	printf "\n >>> "
482	read cputime 1>/dev/null 2>&1
483	else
484	done=true
485	fi
486	done
487	if [[ $remote_host = nech ]]
488	then
489	if (( tasks_per_node != 0 ))
490	then
491	(( cputime = cputime * tasks_per_node ))
492	elif [[ $numprocs != 0 ]]
493	then
494	(( cputime = cputime * numprocs ))
495	fi
496	fi
497	(( stunden = cputime / 3600 ))
498	(( resttime = cputime - stunden * 3600 ))
499	(( minuten = resttime / 60 ))
500	(( sekunden = resttime - minuten * 60 ))
501	timestring=${stunden}:${minuten}:${sekunden}
502
503
504
505	# CHECK THE MEMORY DEMAND
506	done=false
507	while [[ $done = false ]]
508	do
509	if (( memory <= 0 ))
510	then
511	printf "\n +++ wrong memory demand or memory demand missing"
512	printf "\n >>> Please type memory in MByte per process as INTEGER:"
513	printf "\n >>> "
514	read memory 1>/dev/null 2>&1
515	else
516	done=true
517	fi
518	done
519
520	if [[ $remote_host = nech ]]
521	then
522	if (( tasks_per_node != 0 ))
523	then
524	(( Memory = memory * tasks_per_node / 1000 ))
525	elif [[ $numprocs != 0 ]]
526	then
527	(( Memory = memory * numprocs / 1000 ))
528	else
529	(( Memory = memory / 1000 ))
530	fi
531	elif [[ $remote_host = lctit ]]
532	then
533	(( Memory = memory * tasks_per_node / 1000 ))
534	fi
535
536
537	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
538	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
539	then
540	(( memory = memory * threads_per_task ))
541	fi
542
543
544	# CALCULATE NUMBER OF REQUIRED NODES
545	if (( tasks_per_node != 0 ))
546	then
547	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
548	fi
549
550
551	# CALCULATE NUMBER OF PROCESSES PER NODE
552	(( processes_per_node = tasks_per_node * threads_per_task ))
553
554
555	# CALCULATE NUMBER OF MPI TASKS
556	(( mpi_tasks = numprocs / threads_per_task ))
557
558
559	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
560	if [[ "$scp_port" != "" ]]
561	then
562	PORTOPT="-P $scp_port"
563	SSH_PORTOPT="-p $scp_port"
564	fi
565
566
567	# HEADER-OUTPUT
568	if [[ $verify = true ]]
569	then
570	printf "\n\n"
571	printf "#--------------------------------------------------------------# \n"
572	spalte1=SUBJOB;spalte2=$(date)
573	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
574	printf "\| \| \n"
575	printf "\| values of parameters/options: \| \n"
576	spalte1=$(echo local_host$punkte \| cut -c-20)
577	spalte2=$punkte$local_host
578	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
579	spalte1=$(echo remote_host$punkte \| cut -c-20)
580	spalte2=$punkte$remote_host
581	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
582	spalte1=$(echo queue$punkte \| cut -c-20)
583	spalte2=$punkte$queue
584	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
585	spalte1=$(echo memory$punkte \| cut -c-20)
586	spalte2="$punkte$memory mb"
587	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
588	spalte1=$(echo cputime$punkte \| cut -c-20)
589	spalte2="$punkte$cputime sec"
590	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
591	spalte1=$(echo job_name$punkte \| cut -c-20)
592	spalte2="$punkte$job_name"
593	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
594	printf "#--------------------------------------------------------------# \n\n"
595
596
597	# QUERY CHECK
598	antwort="dummy"
599	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
600	do
601	read antwort?" >>> continue (y/n) ? "
602	done
603	if [[ $antwort = n \|\| $antwort = N ]]
604	then
605	locat=verify; exit
606	fi
607	printf "\n"
608	fi
609
610
611
612	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
613	identifier=$RANDOM
614	job_on_remhost=${job_name}_${identifier}_$local_host
615	job_to_send=job_to_send_$identifier
616	if [[ $delete_dayfile = false ]]
617	then
618	remote_dayfile=${local_host}_${job_name}_result_$identifier
619	local_dayfile=${remote_host}_${job_name}
620	else
621	remote_dayfile=/dev/null
622	fi
623
624
625
626	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
627	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
628	then
629
630	# GENERAL LOADLEVELER SETTINGS
631	execute_in_shell="#!/bin/ksh"
632	use_shell="# @ shell = /bin/ksh"
633	consumable_memory="ConsumableMemory($memory mb)"
634	class="# @ class = $queue"
635	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
636	network_to_use="# @ network.mpi = sn_all,shared,us"
637	data_limit="# @ data_limit = 1.76gb"
638	image_size="# @ image_size = 50"
639	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
640
641	if [[ $email_notification = none ]]
642	then
643	notify_user=""
644	else
645	notify_user="# @ notify_user = $email_notification"
646	if [[ $delete_dayfile = true ]]
647	then
648	notification='# @ notification = never'
649	fi
650	fi
651
652	if [[ $remote_host = ibmh ]]
653	then
654	data_limit=""
655	network_to_use=""
656	class="# @ class = $queue"
657	environment=""
658	rset="# @ rset = RSET_MCM_AFFINITY"
659	task_affinity="# @ task_affinity = core(1)"
660	elif [[ $remote_host = ibmkisti ]]
661	then
662	network_to_use="# @ network.MPI = sn_all,shared,US"
663	wall_clock_limit="# @ wall_clock_limit = $timestring"
664	if [[ $threads_per_task = 1 ]]
665	then
666	rset="# @ rset = RSET_MCM_AFFINITY"
667	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
668	fi
669	environment=""
670	use_shell=""
671	data_limit=""
672	image_size=""
673	elif [[ $remote_host = ibmku ]]
674	then
675	execute_in_shell="#!/usr/bin/ksh"
676	use_shell="# @ shell = /usr/bin/ksh"
677	consumable_memory=""
678	environment=""
679	network_to_use="# @ network.mpi = sn_all,shared,us"
680	data_limit=""
681	image_size=""
682	elif [[ $remote_host = ibms ]]
683	then
684	network_to_use="# @ network.mpi = csss,shared,us"
685	fi
686
687	cat > $job_to_send << %%END%%
688	$execute_in_shell
689	$use_shell
690
691	# @ job_type = parallel
692	# @ job_name = $job_name
693	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
694	# @ output = $remote_dayfile
695	# @ error = $remote_dayfile
696	$wall_clock_limit
697	$image_size
698	$class
699	$environment
700	$network_to_use
701	$data_limit
702	$rset
703	$mcm_affinity_options
704	$task_affinity
705	$notification
706	$notify_user
707
708	%%END%%
709
710	if (( nodes > 0 ))
711	then
712
713	if [[ $remote_host != ibmkisti ]]
714	then
715
716	cat >> $job_to_send << %%END%%
717	# @ node = $nodes
718	# @ tasks_per_node = $processes_per_node
719	# @ node_usage = $node_usage
720	# @ queue
721
722	%%END%%
723
724	else
725
726	cat >> $job_to_send << %%END%%
727	# @ total_tasks = $mpi_tasks
728	# @ blocking = unlimited
729	# @ queue
730
731	%%END%%
732
733	fi
734
735	else
736
737	cat >> $job_to_send << %%END%%
738	# @ blocking = unlimited
739	# @ total_tasks = $numprocs
740	# @ node_usage = $node_usage
741	# @ queue
742
743	%%END%%
744
745	fi
746
747	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
748	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
749	then
750	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
751	fi
752
753	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
754	then
755
756	cat > $job_to_send << %%END%%
757	#!/bin/ksh
758
759	# @ job_type = serial
760	# @ node_usage = $node_usage
761	# @ job_name = palm
762	# @ wall_clock_limit = ${timestring},$timestring
763	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
764	# @ output = $remote_dayfile
765	# @ error = $remote_dayfile
766	$class
767	$notification
768
769	# @ queue
770
771	%%END%%
772
773	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
774	then
775
776	if [[ "$feature" != "" ]]
777	then
778	featuredir="#PBS -l feature=$feature"
779	fi
780
781	if [[ $numprocs != 0 ]]
782	then
783	cat > $job_to_send << %%END%%
784	#!/bin/bash
785	#PBS -N $job_name
786	#PBS -l walltime=$timestring
787	#PBS -l nodes=$nodes:ppn=$processes_per_node
788	#PBS -o $remote_dayfile
789	#PBS -j oe
790	#PBS -q $queue
791	$featuredir
792
793	$module_calls
794
795	%%END%%
796
797	else
798
799	continue
800
801	fi
802
803	elif [[ $remote_host = lccrayf ]]
804	then
805
806	if [[ $numprocs != 0 ]]
807	then
808	cat > $job_to_send << %%END%%
809	#!/bin/bash -l
810	#SBATCH -J $job_name
811	#SBATCH -t $timestring
812	#SBATCH -N $nodes
813	#SBATCH --ntasks-per-node=$processes_per_node
814	#SBATCH -p $queue
815	#SBATCH -o $remote_dayfile
816	#SBATCH -e $remote_dayfile
817
818	$init_cmds
819	$module_calls
820
821	%%END%%
822
823	else
824	cat > $job_to_send << %%END%%
825	#!/bin/bash -l
826	#SBATCH -J $job_name
827	#SBATCH -t $timestring
828	#SBATCH -l ncpus=1
829	#SBATCH -l pmem=${memory}mb
830	#SBATCH -m abe
831	#SBATCH -o $remote_dayfile
832	#SBATCH -e $remote_dayfile
833
834	$init_cmds
835	$module_calls
836
837	%%END%%
838
839	fi
840
841	elif [[ $remote_host = lcflow ]]
842	then
843	if [ $memory -gt 1800 ]; then
844	use_himem=""
845	else
846	use_himem="#"
847	fi
848	if [[ $numprocs != 0 ]]
849	then
850	pe_set="#$ -pe impi41 $numprocs"
851	else
852	pe_set="#$ -pe impi41 1"
853	fi
854	if [[ $queue = default ]]
855	then
856	queue_set=""
857	else
858	queue_set="#$ -q $queue"
859	fi
860	[[ "$disc_space" = "" ]] && disc_space=50
861
862	cat > $job_to_send << %%END%%
863	#!/bin/bash
864	#$ -S /bin/bash
865	#$ -N $job_name
866	#$ -cwd
867	#$ -l h_rt=$timestring
868	#$ -l h_vmem=${memory}M
869	#$ -o $remote_dayfile
870	#$ -j y
871	$pe_set
872	#$ -R y
873	${use_himem}#$ -l highmem=true
874	#$ -l h_fsize=${disc_space}G
875	$queue_set
876
877	%%END%%
878
879	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
880	then
881
882	if [[ $numprocs != 0 ]]
883	then
884	cat > $job_to_send << %%END%%
885	#!/bin/ksh
886	#PBS -N $job_name
887	#PBS -l walltime=$timestring
888	#PBS -l ncpus=$numprocs
889	#PBS -l pmem=${memory}mb
890	#PBS -o $remote_dayfile
891	#PBS -l nodes=$nodes:ppn=${processes_per_node}
892	#PBS -j oe
893
894	mpd &
895
896	%%END%%
897
898	else
899	cat > $job_to_send << %%END%%
900	#!/bin/ksh
901	#PBS -N $job_name
902	#PBS -l walltime=$timestring
903	#PBS -l ncpus=1
904	#PBS -l pmem=${memory}mb
905	#PBS -o $remote_dayfile
906	#PBS -j oe
907
908	%%END%%
909
910	fi
911
912	elif [[ $remote_host = lckiaps ]]
913	then
914
915	if [[ $numprocs != 0 ]]
916	then
917	cat > $job_to_send << %%END%%
918	#!/bin/ksh
919	#PBS -N $job_name
920	#PBS -l walltime=$timestring
921	#PBS -l select=1:ncpus=$numprocs
922	#PBS -l pmem=${memory}mb
923	#PBS -q $queue
924	#PBS -o $remote_dayfile
925	#PBS -j oe
926	#PBS -V
927
928	%%END%%
929
930	else
931	cat > $job_to_send << %%END%%
932	#!/bin/ksh
933	#PBS -N $job_name
934	#PBS -l walltime=$timestring
935	#PBS -l ncpus=1
936	#PBS -l pmem=${memory}mb
937	#PBS -o $remote_dayfile
938	#PBS -j oe
939
940	%%END%%
941
942	fi
943
944	elif [[ $remote_host = lcyon ]]
945	then
946
947	if [[ $numprocs != 0 ]]
948	then
949	cat > $job_to_send << %%END%%
950	#!/bin/ksh
951	#PBS -N $job_name
952	#PBS -l walltime=$timestring
953	#PBS -l ncpus=$numprocs
954	#PBS -l pmem=${memory}mb
955	#PBS -o $remote_dayfile
956	#PBS -j oe
957
958	%%END%%
959
960	else
961	cat > $job_to_send << %%END%%
962	#!/bin/ksh
963	#PBS -N $job_name
964	#PBS -l walltime=$timestring
965	#PBS -l ncpus=1
966	#PBS -l pmem=${memory}mb
967	#PBS -o $remote_dayfile
968	#PBS -j oe
969
970	%%END%%
971
972	fi
973
974	elif [[ $remote_host = lcxe6 ]]
975	then
976
977	if [[ $numprocs != 0 ]]
978	then
979	cat > $job_to_send << %%END%%
980	#!/bin/ksh
981	#PBS -S /bin/ksh
982	#PBS -N $job_name
983	#PBS -A $project_account
984	#PBS -j oe
985	#PBS -l walltime=$timestring
986	#PBS -l mppwidth=${numprocs}
987	#PBS -l mppnppn=${processes_per_node}
988	#PBS -m abe
989	#PBS -o $remote_dayfile
990	$email_directive
991
992	$init_cmds
993	$module_calls
994
995	%%END%%
996
997	else
998	cat > $job_to_send << %%END%%
999	#!/bin/ksh
1000	#PBS -S /bin/ksh
1001	#PBS -N $job_name
1002	#PBS -A $project_account
1003	#PBS -j oe
1004	#PBS -l walltime=$timestring
1005	#PBS -l ncpus=1
1006	#PBS -l pmem=${memory}mb
1007	#PBS -m abe
1008	$email_directive
1009	#PBS -o $remote_dayfile
1010
1011	$init_cmds
1012	$module_calls
1013
1014	%%END%%
1015
1016	fi
1017
1018	elif [[ $remote_host = lckyoto ]]
1019	then
1020
1021	cat > $job_to_send << %%END%%
1022	#!/bin/ksh
1023	# @\$-o $remote_dayfile
1024	# @\$-eo -oi
1025	# @\$-lP 16
1026	# @\$-lp 1
1027	# @\$-lm 28gb -llm unlimited -ls unlimited
1028	# @\$-q $queue
1029	# @\$-Pvn abs_pack
1030	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1031	#. /thin/local/etc/setprofile/intel-11.0.sh
1032	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1033	. ~/.myprofile
1034	#. /home2/t/t51254/palm/current_version/myprofile
1035	#. /thin/apps/pgi/mpi.sh
1036	#
1037	env
1038	#
1039	set -x
1040
1041	%%END%%
1042
1043	elif [[ $remote_host = lcxt5m ]]
1044	then
1045
1046	if [[ $numprocs != 0 ]]
1047	then
1048	cat > $job_to_send << %%END%%
1049	#!/bin/ksh
1050	#PBS -S /bin/ksh
1051	#PBS -N $job_name
1052	#PBS -j oe
1053	#PBS -l walltime=$timestring
1054	#PBS -l mppwidth=${numprocs}
1055	#PBS -l mppnppn=${processes_per_node}
1056	#PBS -m abe
1057	#PBS -o $remote_dayfile
1058
1059	$init_cmds
1060	$module_calls
1061
1062	%%END%%
1063
1064	else
1065	cat > $job_to_send << %%END%%
1066	#!/bin/ksh
1067	#PBS -S /bin/ksh
1068	#PBS -N $job_name
1069	#PBS -j oe
1070	#PBS -l walltime=$timestring
1071	#PBS -l ncpus=1
1072	#PBS -l pmem=${memory}mb
1073	#PBS -m abe
1074	#PBS -o $remote_dayfile
1075
1076	$init_cmds
1077	$module_calls
1078
1079	%%END%%
1080
1081	fi
1082
1083	elif [[ $remote_host = lckyuh ]]
1084	then
1085	cat > $job_to_send << %%END%%
1086	#!/bin/bash
1087	#PJM -L "rscgrp=$queue"
1088	#PJM -L "node=$nodes"
1089	#PJM --mpi "proc=$numprocs"
1090	#PJM -L "elapse=$timestring"
1091	#PJM -o $remote_dayfile
1092	#PJM -j
1093	#PJM -X
1094	#PJM --no-stging
1095
1096	export LANG=en_US.UTF-8
1097	%%END%%
1098
1099	elif [[ $remote_host = lckyut ]]
1100	then
1101	cat > $job_to_send << %%END%%
1102	#!/bin/bash
1103	#PJM -L "rscgrp=$queue"
1104	#PJM -L "vnode=$numprocs"
1105	#PJM -L "vnode-core=1"
1106	#PJM -L "elapse=$timestring"
1107	#PJM --mpi proc=$numprocs
1108	#PJM -o $remote_dayfile
1109	#PJM -j
1110	#PJM -X
1111	#PJM --no-stging
1112
1113	export LANG=en_US.UTF-8
1114	%%END%%
1115
1116	elif [[ $remote_host = nech ]]
1117	then
1118
1119	if (( nodes > 1 ))
1120	then
1121	cat > $job_to_send << %%END%%
1122	#!/bin/ksh
1123	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1124	#PBS -l ${qsubmem}=${Memory}gb
1125	#PBS -b $nodes
1126	#PBS -o $remote_dayfile
1127	#PBS -N palm
1128	#PBS -j o
1129	#PBS -T mpisx
1130
1131	%%END%%
1132
1133	elif [[ $numprocs != 0 ]]
1134	then
1135	cat > $job_to_send << %%END%%
1136	#!/bin/ksh
1137	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1138	#PBS -l ${qsubmem}=${Memory}gb
1139	#PBS -o $remote_dayfile
1140	#PBS -N palm
1141	#PBS -j o
1142
1143	%%END%%
1144
1145	else
1146	cat > $job_to_send << %%END%%
1147	#!/bin/ksh
1148	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1149	#PBS -o $remote_dayfile
1150	#PBS -j o
1151
1152	%%END%%
1153
1154	fi
1155
1156	elif [[ $remote_host = lctit ]]
1157	then
1158	cat > $job_to_send << %%END%%
1159	#!/bin/ksh
1160	$init_cmds
1161	$module_calls
1162
1163	%%END%%
1164
1165	# SET OPTIONS FOR SUBMIT-COMMAND
1166	if [[ $tasks_per_node != $processes_per_node ]]
1167	then
1168	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -et 1 -q $queue "
1169	else
1170	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -et 1 -q $queue "
1171	fi
1172
1173	else
1174
1175	cat > $job_to_send << %%END%%
1176	# @\$-q ${queue}
1177	# @\$-l${qsubtime} $timestring
1178	# @\$-l${qsubmem} ${memory}mb
1179	# @\$-o $remote_dayfile
1180	# @\$-eo
1181
1182	%%END%%
1183
1184	fi
1185
1186
1187	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1188	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1189	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1190	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1191	then
1192	echo "set +vx" >> $job_to_send
1193	echo "trap '" >> $job_to_send
1194	echo "set +vx" >> $job_to_send
1195	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lccrayb \|\| $remote_host = lccrayh \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* \|\| $remote_host = lcxe6 ]]
1196	then
1197	if [[ $remote_host = ibmh ]]
1198	then
1199	return_queue=c1
1200	elif [[ $remote_host = ibmkisti ]]
1201	then
1202	return_queue=class.1-2
1203	elif [[ $remote_host = ibmku ]]
1204	then
1205	return_queue=sdbg2
1206	elif [[ $remote_host = ibms ]]
1207	then
1208	return_queue=p_normal
1209	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1210	then
1211	return_queue=dataq
1212	elif [[ $remote_host = lcxe6 ]]
1213	then
1214	return_queue=debug
1215	elif [[ $remote_host = lckiaps ]]
1216	then
1217	return_queue=express
1218	elif [[ $remote_host = lckyuh ]]
1219	then
1220	return_queue=cx-single
1221	elif [[ $remote_host = lckyut ]]
1222	then
1223	return_queue=cx-single
1224	else
1225	return_queue=unknown
1226	fi
1227
1228	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1229	then
1230
1231	if [[ $remote_host = ibmku ]]
1232	then
1233	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1234	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1235	else
1236	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1237	fi
1238	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1239	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1240	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1241	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1242	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1243	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1244	if [[ $host != "ibmh" ]]
1245	then
1246	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1247	fi
1248	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1249	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1250
1251	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1252	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1253
1254	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1255	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
1256	if [[ $remote_host = ibmku ]]
1257	then
1258	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1259	fi
1260	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1261
1262	elif [[ $remote_host = nech ]]
1263	then
1264	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1265	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1266	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1267	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1268	echo "#PBS -j o" >> $job_to_send
1269	echo " " >> $job_to_send
1270	echo "set -x" >> $job_to_send
1271	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1272	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1273	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1274	echo "%%END%%" >> $job_to_send
1275
1276	elif [[ $remote_host = lckyuh ]]
1277	then
1278	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1279	echo "#!/bin/bash" >> $job_to_send
1280	echo "#PJM -L \"node=1\"" >> $job_to_send
1281	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1282	echo "#PJM --no-stging" >> $job_to_send
1283	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1284	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1285	echo "#PJM -j" >> $job_to_send
1286	echo " " >> $job_to_send
1287	echo "export LANG=en_US.UTF-8" >> $job_to_send
1288	echo "set -x" >> $job_to_send
1289	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1290	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1291	echo "%%END%%" >> $job_to_send
1292
1293	elif [[ $remote_host = lckyut ]]
1294	then
1295	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1296	echo "#!/bin/bash" >> $job_to_send
1297	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1298	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1299	echo "#PJM --no-stging" >> $job_to_send
1300	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1301	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1302	echo "#PJM -j" >> $job_to_send
1303	echo " " >> $job_to_send
1304	echo "export LANG=en_US.UTF-8" >> $job_to_send
1305	echo "set -x" >> $job_to_send
1306	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1307	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1308	echo "%%END%%" >> $job_to_send
1309
1310	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1311	then
1312	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1313	echo "#!/bin/bash" >> $job_to_send
1314	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1315	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1316	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1317	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1318	echo "#PBS -j oe" >> $job_to_send
1319	echo " " >> $job_to_send
1320	echo "set -x" >> $job_to_send
1321	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1322	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1323	echo "%%END%%" >> $job_to_send
1324
1325	elif [[ $remote_host = lcflow ]]
1326	then
1327	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1328	echo "#!/bin/bash" >> $job_to_send
1329	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1330	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1331	echo "SGEPREFIX -cwd" >> $job_to_send
1332	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1333	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1334	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1335	echo "SGEPREFIX -j y" >> $job_to_send
1336	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1337	echo " " >> $job_to_send
1338	echo "set -x" >> $job_to_send
1339	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1340	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1341	echo "" >> $job_to_send
1342	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1343	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1344	echo "rm -f scpjob.${identifier}" >> $job_to_send
1345	echo "%%END%%" >> $job_to_send
1346	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1347	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1348	elif [[ $remote_host = lcxe6 ]]
1349	then
1350	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1351	echo "#!/bin/ksh" >> $job_to_send
1352	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1353	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1354	echo "#PBS -A $project_account" >> $job_to_send
1355	echo "#PBS -l mppwidth=1" >> $job_to_send
1356	echo "#PBS -l mppnppn=1" >> $job_to_send
1357	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1358	echo "#PBS -j oe" >> $job_to_send
1359	echo " " >> $job_to_send
1360	echo "set -x" >> $job_to_send
1361	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1362	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1363	echo "%%END%%" >> $job_to_send
1364	else
1365
1366	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1367	echo "# @\\\$-q $return_queue" >> $job_to_send
1368	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1369	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1370	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1371	then
1372	echo "# @\$-l mpp_p=0" >> $job_to_send
1373	fi
1374	echo '# @\$-lF 10mb' >> $job_to_send
1375	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1376	echo '# @\\\$-eo' >> $job_to_send
1377	echo " " >> $job_to_send
1378	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1379	then
1380	echo "set +vx" >> $job_to_send
1381	echo ". .profile" >> $job_to_send
1382	fi
1383	echo "set -x" >> $job_to_send
1384	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1385	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1386	echo "%%END%%" >> $job_to_send
1387	fi
1388
1389	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1390	then
1391	echo "llsubmit scpjob.$identifier" >> $job_to_send
1392	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1393	then
1394	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1395	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1396	then
1397	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1398	elif [[ $remote_host = t3es ]]
1399	then
1400	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1401	elif [[ $remote_host = lckiaps ]]
1402	then
1403	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1404	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1405	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1406	elif [[ $remote_host = lckyu* ]]
1407	then
1408	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1409	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1410	elif [[ $remote_host = lcflow ]]
1411	then
1412	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1413	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1414	else
1415	echo "$submcom scpjob.$identifier" >> $job_to_send
1416	fi
1417	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1418	then
1419	echo "rm scpjob.$identifier" >> $job_to_send
1420	fi
1421	if [[ $remote_host = nech ]]
1422	then
1423	echo "cd -" >> $job_to_send
1424	fi
1425	else
1426	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1427	# ??? funktioniert das ÃŒberhaupt noch ???
1428	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1429	fi
1430	echo "set -x" >> $job_to_send
1431	echo " ' exit" >> $job_to_send
1432	echo "set -x" >> $job_to_send
1433	fi
1434
1435
1436
1437	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1438	cat $file_to_send >> $job_to_send
1439
1440	if [[ $remote_host = ibm ]]
1441	then
1442	echo " " >> $job_to_send
1443	echo "exit" >> $job_to_send
1444	fi
1445
1446	# REMOVE JOB-FILE
1447	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1448	then
1449	echo " " >> $job_to_send
1450	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1451	fi
1452
1453
1454	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1455	if [[ $no_submit = false ]]
1456	then
1457	if [[ $remote_host != $local_host ]]
1458	then
1459	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1460	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1461	then
1462	job_catalog_save=$job_catalog
1463	job_catalog=job_queue
1464	elif [[ $remote_host = nech ]]
1465	then
1466	job_catalog_save=$job_catalog
1467	job_catalog=/hpf/b/${remote_user}/job_queue
1468	fi
1469	if [[ $remote_host = nech ]]
1470	then
1471	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1472	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1473	else
1474	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1475	fi
1476	if [[ $? = 1 ]]
1477	then
1478	locat=scp; exit
1479	fi
1480	if [[ $remote_host = ibms ]]
1481	then
1482	job_catalog=$job_catalog_save
1483	fi
1484	[[ $verify = true ]] && printf "\n >>> finished\n"
1485	else
1486	eval job_catalog=$job_catalog
1487	cp $job_to_send ${job_catalog}/$job_on_remhost
1488	fi
1489
1490
1491
1492	# START NQS- / LOADLEVELER-JOB
1493	if [[ $remote_host != $local_host ]]
1494	then
1495	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1496
1497	if [[ $remote_host = ibmku ]]
1498	then
1499	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1500	elif [[ $remote_host = lcflow ]]
1501	then
1502	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1503	else
1504	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1505	fi
1506
1507	[[ $verify = true ]] && printf " >>> o.k.\n"
1508	else
1509	cd $job_catalog
1510	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1511	then
1512	eval $submcom $job_on_remhost
1513	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1514	then
1515	chmod u+x $job_on_remhost
1516	eval $submcom $job_on_remhost
1517	elif [[ $local_host = nech ]]
1518	then
1519	if [[ $queue = default ]]
1520	then
1521	eval $submcom $job_on_remhost
1522	else
1523	eval $submcom -q $queue $job_on_remhost
1524	fi
1525	else
1526	qsub $job_on_remhost
1527	fi
1528
1529	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1530	# AT THE END OF THE JOB
1531	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1532	then
1533	rm $job_on_remhost
1534	fi
1535	cd - > /dev/null
1536	fi
1537	fi
1538
1539
1540
1541	# FINAL ACTIONS
1542	if [[ $no_submit = false ]]
1543	then
1544	rm -f $job_to_send
1545	fi
1546	[[ $verify = true ]] && printf "\n\n *** SUBJOB finished \n\n"

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

| Impressum | ©Leibniz Universität Hannover |