Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1451

Last change on this file since 1451 was 1451, checked in by heinze, 10 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 50.5 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 1451 2014-08-21 07:34:14Z heinze $
30	#
31	# 1450 2014-08-21 07:31:51Z heinze
32	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
33	#
34	# 1442 2014-07-28 07:09:10Z raasch
35	# HLRN-III (lccrayb/lccrayh) queues adjusted
36	#
37	# 1378 2014-04-28 06:04:58Z raasch
38	# -et option added for lctit
39	#
40	# 1350 2014-04-04 13:01:30Z maronga
41	# location of qsub updated for lcxe6
42	#
43	# 1289 2014-03-04 07:12:34Z raasch
44	# German comments translated to English
45	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
46	#
47	# 1279 2014-01-28 12:10:14Z raasch
48	# node calculation modified due to changes in mrun (tasks_per_node must not be
49	# an integral divisor of numprocs any more)
50	#
51	# 1274 2014-01-09 13:14:54Z heinze
52	# adjustments for lccrayh
53	#
54	# 1266 2013-12-11 12:07:34Z heinze
55	# further adjustments for lccrayb (use msub instead of qsub)
56	#
57	# 1264 2013-12-09 12:46:09Z fricke
58	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
59	#
60	# 1262 2013-12-09 10:57:20Z fricke
61	# further adjustments for lccrayb
62	#
63	# 1260 2013-12-04 12:48:04Z raasch
64	# jaboticaba admitted
65	#
66	# 1255 2013-11-07 14:43:35Z raasch
67	# further adjustments for lccrayb
68	#
69	# 1224 2013-09-16 07:27:23Z raasch
70	# first adjustments for lccrayb
71	#
72	# 1202 2013-07-10 16:22:07Z witha
73	# adjustments for Forwind cluster (lcflow)
74	#
75	# 1199 2013-07-05 14:52:22Z raasch
76	# adjustments for CSC Helsinki (lccrayf)
77	#
78	# use of cluster/express queue enabled (ibmh)
79	# vinessa added (imuk)
80	#
81	# 1103 2013-02-20 02:15:53Z raasch
82	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
83	# instead typeset -L/R),
84	# further adjustments for lckyuh
85	#
86	# 2013-02-10 01:47:43Z raasch
87	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
88	# and for Forwind cluster (lcflow)
89	#
90	# 1094 2013-02-03 01:52:12Z raasch
91	# new option -P for explicit setting of ssh/scp port,
92	# decalpha parts (yonsei) removed
93	#
94	# 2013-02-02 07:06:13Z raasch
95	# adjustments for Kyushu-University computing center (lckyut - tatara)
96	# old changelog messages removed
97	#
98	# 1046 2012-11-09 14:38:45Z maronga
99	# code put under GPL (PALM 3.9)
100	#
101	# 08/07/94 - Siggi - first version finished
102	# 29/06/94 - Siggi - script development started
103	#--------------------------------------------------------------------------------#
104	# subjob - script for automatic generation and submission of batch-job files
105	# for various batch queuing systems
106	#--------------------------------------------------------------------------------#
107
108
109	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
110	delete_dayfile=false
111	email_notification=none
112	group_number=none
113	locat=normal
114	no_default_queue=none
115	no_submit=false
116	job_catalog="~/job_queue"
117	job_name=none
118	local_user=$LOGNAME
119	node_usage=shared
120	numprocs=0
121	punkte="..........................................................."
122	submcom=qsub
123	queue=default
124	remote_host=none
125	remote_user=""
126	verify=true
127
128	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
129	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
130
131
132
133	# ERROR HANDLING
134	# IN CASE OF EXIT:
135	trap 'if [[ $locat != normal ]]
136	then
137	case $locat in
138	(option) printf "\n --> available optios can be displayed"
139	printf " by typing:"
140	printf "\n \"subjob ?\" \n";;
141	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
142	(*) printf "\n +++ unknown error"
143	printf "\n please inform S. Raasch!\n"
144	esac
145	[[ -f $job_to_send ]] && rm $job_to_send
146	printf "\n\n+++ SUBJOB killed \n\n"
147	fi' exit
148
149
150	# IN CASE OF TERMINAL-BREAK:
151	trap '[[ -f $job_to_send ]] && rm $job_to_send
152	printf "\n\n+++ SUBJOB killed \n\n"
153	exit
154	' 2
155
156
157
158
159	# DETERMINE NAME OF LOCAL HOST
160	local_host=$(hostname)
161
162
163
164	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
165	# IS ADMITTED AT ALL)
166	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
167	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
168	case $local_host in
169	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
170	(atmos) local_addres=172.20.25.35; local_host=lcide;;
171	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
172	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
173	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
174	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
175	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
176	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
177	(hlogin\|hxc) local_addres=130.75.4.1; local_host=lccrayh;;
178	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
179	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
180	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
181	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
182	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
183	(cs*) local_addres=136.172.44.131; local_host=nech;;
184	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
185	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
186	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
187	(node*) local_addres=165.132.26.61 local_host=lck;;
188	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
189	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
190	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
191	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
192	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
193	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
194	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
195	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
196	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
197	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
198	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
199	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
200	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
201	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
202	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
203	(meller) local_addres=134.106.74.155; local_host=lcfor;;
204	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
205	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
206	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
207	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
208	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
209	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
210	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
211	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
212	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
213	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
214	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
215	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
216	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
217	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
218	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
219	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
220	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
221	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
222	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
223	(*) printf "\n +++ \"$local_host\" unknown";
224	printf "\n please contact the PALM group at IMUK";
225	locat=parameter; exit;;
226	esac
227
228
229
230	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
231	remote_host=$local_host
232
233
234
235
236	# READ THE SHELLSCRIPT-OPTIONS
237	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
238	do
239	case $option in
240	(c) job_catalog=$OPTARG;;
241	(d) delete_dayfile=true;;
242	(D) no_submit=true;;
243	(e) email_notification=$OPTARG;;
244	(g) group_number=$OPTARG;;
245	(h) remote_host=$OPTARG;;
246	(m) memory=$OPTARG;;
247	(n) job_name=$OPTARG;;
248	(N) node_usage=$OPTARG;;
249	(O) threads_per_task=$OPTARG;;
250	(P) scp_port=$OPTARG;;
251	(q) no_default_queue=$OPTARG;;
252	(t) cputime=$OPTARG;;
253	(T) tasks_per_node=$OPTARG;;
254	(u) remote_user=$OPTARG;;
255	(v) verify=false;;
256	(X) numprocs=$OPTARG;;
257	(\?) printf "\n +++ Option $OPTARG unknown \n";
258	locat=option; exit;;
259	esac
260	done
261
262
263	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
264	(( to_shift = $OPTIND - 1 ))
265	shift $to_shift; file_to_send=$1
266
267
268	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
269	if [ "$1" = "?" ]
270	then
271	(printf "\n *** subjob can be called as follows:\n"
272	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
273	printf "\n Description of available options:\n"
274	printf "\n Option Description Default-Value"
275	printf "\n -c job-input- and output-catalog ~/job_queue"
276	printf "\n -d no job-protocol will be created ---"
277	printf "\n -D only the job-file will be created ---"
278	printf "\n -h execution host, available hosts: $remote_host"
279	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
280	printf "\n lckiaps, lctit, nech"
281	printf "\n -m memory demand per process in MByte ---"
282	printf "\n -n jobname <jobdatei>"
283	printf "\n -O threads per task (for OpenMP usage) 1"
284	printf "\n -P ssh/scp port default port"
285	printf "\n -q job-queue to be used default"
286	printf "\n -t allowed cpu-time in seconds ---"
287	printf "\n -T tasks per node (on parallel hosts) ---"
288	printf "\n -u username on execution host from .netrc"
289	printf "\n -v no prompt for confirmation ---"
290	printf "\n -X # of processors (on parallel hosts) 1"
291	printf "\n "
292	printf "\n The only possible positional parameter is <jobfile>:"
293	printf "\n The complete NQS-job must be provided here."
294	printf "\n <jobfile>=? creates this outline\n\n") \| more
295	exit
296	fi
297
298
299
300	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
301	if [[ "$file_to_send" = "" ]]
302	then
303	printf "\n +++ job-file missing"
304	locat=parameter; exit
305	else
306	if [[ -f $file_to_send ]]
307	then
308	true
309	else
310	printf "\n +++ job-file: "
311	printf "\n $file_to_send"
312	printf "\n does not exist"
313	locat=parameter; exit
314	fi
315	fi
316
317
318
319	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
320	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
321	if [[ $job_name = none ]]
322	then
323	job_name=$file_to_send
324	fi
325	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
326	then
327	printf "\n +++ job-file name: "
328	printf "\n $job_name"
329	printf "\n must not contain \"/\"-characters"
330	locat=parameter; exit
331	fi
332
333
334
335
336	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
337	# OR IF NO HOST HAS BEEN GIVEN
338	if [[ $remote_host = none ]]
339	then
340	printf "\n +++ host missing"
341	locat=option; exit
342	else
343	case $remote_host in
344	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
345	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
346	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
347	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
348	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
349	(lccrayb) queue=mpp1testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
350	(lccrayh) queue=mpp1testq; remote_addres=130.75.4.1; submcom=/opt/moab/7.2.6/bin/msub;;
351	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
352	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
353	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
354	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
355	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
356	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
357	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
358	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
359	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
360	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
361	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
362	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
363	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
364	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
365	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
366	locat=parameter; exit;;
367	esac
368	fi
369
370
371	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
372	if [[ $no_default_queue != none ]]
373	then
374	error=false
375	ndq=$no_default_queue
376	case $remote_host in
377	(ibm) case $ndq in
378	(p690_express\|p690_standard\|p690_long) error=false;;
379	(*) error=true;;
380	esac;;
381	(ibmh) case $ndq in
382	(cluster\|express) error=false;;
383	(*) error=true;;
384	esac;;
385	(ibmkisti) case $ndq in
386	(class.32plus\|class.1-2\|class.2-32) error=false;;
387	(*) error=true;;
388	esac;;
389	(ibmku) case $ndq in
390	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
391	(*) error=true;;
392	esac;;
393	(ibms) case $ndq in
394	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
395	(*) error=true;;
396	esac;;
397	(lccrayb) case $ndq in
398	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
399	(*) error=true;;
400	esac;;
401	(lccrayh) case $ndq in
402	(dataq\|mpp1q\|mpp1testq\|smp1q\|smp1testq\|specialm1q) error=false;;
403	(*) error=true;;
404	esac;;
405	(lccrayf) case $ndq in
406	(usup\|test*\|small\|large) error=false;;
407	(*) error=true;;
408	esac;;
409	(lcflow) case $ndq in
410	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
411	(*) error=true;;
412	esac;;
413	(lckiaps) case $ndq in
414	(express\|normal) error=false;;
415	(*) error=true;;
416	esac;;
417	(lckyoto) case $ndq in
418	(eh\|ph) error=false;;
419	(*) error=true;;
420	esac;;
421	(lckyuh) case $ndq in
422	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
423	(*) error=true;;
424	esac;;
425	(lckyut) case $ndq in
426	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
427	(*) error=true;;
428	esac;;
429	(lctit) case $ndq in
430	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
431	(*) error=true;;
432	esac;;
433	(t3eb) case $ndq in
434	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
435	(*) error=true;;
436	esac;;
437	(t3eh) case $ndq in
438	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
439	(*) error=true;;
440	esac;;
441	(t3ej2\|t3ej5) case $ndq in
442	(low\|normal\|high) error=false;;
443	(*) error=true;;
444	esac;;
445	(t3es) case $ndq in
446	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
447	(*) error=true;;
448	esac;;
449	esac
450	if [[ $error = true ]]
451	then
452	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
453	locat=parameter; exit
454	else
455	queue=$no_default_queue
456	fi
457	fi
458
459
460
461	# CHECK THE CPU-TIME
462	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
463	done=false
464	while [[ $done = false ]]
465	do
466	if (( $cputime <= 0 ))
467	then
468	printf "\n +++ wrong cpu-time or cpu-time missing"
469	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
470	printf "\n >>> "
471	read cputime 1>/dev/null 2>&1
472	else
473	done=true
474	fi
475	done
476	if [[ $remote_host = nech ]]
477	then
478	if (( tasks_per_node != 0 ))
479	then
480	(( cputime = cputime * tasks_per_node ))
481	elif [[ $numprocs != 0 ]]
482	then
483	(( cputime = cputime * numprocs ))
484	fi
485	fi
486	(( stunden = cputime / 3600 ))
487	(( resttime = cputime - stunden * 3600 ))
488	(( minuten = resttime / 60 ))
489	(( sekunden = resttime - minuten * 60 ))
490	timestring=${stunden}:${minuten}:${sekunden}
491
492
493
494	# CHECK THE MEMORY DEMAND
495	done=false
496	while [[ $done = false ]]
497	do
498	if (( memory <= 0 ))
499	then
500	printf "\n +++ wrong memory demand or memory demand missing"
501	printf "\n >>> Please type memory in MByte per process as INTEGER:"
502	printf "\n >>> "
503	read memory 1>/dev/null 2>&1
504	else
505	done=true
506	fi
507	done
508
509	if [[ $remote_host = nech ]]
510	then
511	if (( tasks_per_node != 0 ))
512	then
513	(( Memory = memory * tasks_per_node / 1000 ))
514	elif [[ $numprocs != 0 ]]
515	then
516	(( Memory = memory * numprocs / 1000 ))
517	else
518	(( Memory = memory / 1000 ))
519	fi
520	elif [[ $remote_host = lctit ]]
521	then
522	(( Memory = memory * tasks_per_node / 1000 ))
523	fi
524
525
526	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
527	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
528	then
529	(( memory = memory * threads_per_task ))
530	fi
531
532
533	# CALCULATE NUMBER OF REQUIRED NODES
534	if (( tasks_per_node != 0 ))
535	then
536	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
537	fi
538
539
540	# CALCULATE NUMBER OF PROCESSES PER NODE
541	(( processes_per_node = tasks_per_node * threads_per_task ))
542
543
544	# CALCULATE NUMBER OF MPI TASKS
545	(( mpi_tasks = numprocs / threads_per_task ))
546
547
548	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
549	if [[ "$scp_port" != "" ]]
550	then
551	PORTOPT="-P $scp_port"
552	SSH_PORTOPT="-p $scp_port"
553	fi
554
555
556	# HEADER-OUTPUT
557	if [[ $verify = true ]]
558	then
559	printf "\n\n"
560	printf "#--------------------------------------------------------------# \n"
561	spalte1=SUBJOB;spalte2=$(date)
562	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
563	printf "\| \| \n"
564	printf "\| values of parameters/options: \| \n"
565	spalte1=$(echo local_host$punkte \| cut -c-20)
566	spalte2=$punkte$local_host
567	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
568	spalte1=$(echo remote_host$punkte \| cut -c-20)
569	spalte2=$punkte$remote_host
570	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
571	spalte1=$(echo queue$punkte \| cut -c-20)
572	spalte2=$punkte$queue
573	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
574	spalte1=$(echo memory$punkte \| cut -c-20)
575	spalte2="$punkte$memory mb"
576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
577	spalte1=$(echo cputime$punkte \| cut -c-20)
578	spalte2="$punkte$cputime sec"
579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
580	spalte1=$(echo job_name$punkte \| cut -c-20)
581	spalte2="$punkte$job_name"
582	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
583	printf "#--------------------------------------------------------------# \n\n"
584
585
586	# QUERY CHECK
587	antwort="dummy"
588	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
589	do
590	read antwort?" >>> continue (y/n) ? "
591	done
592	if [[ $antwort = n \|\| $antwort = N ]]
593	then
594	locat=verify; exit
595	fi
596	printf "\n"
597	fi
598
599
600
601	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
602	identifier=$RANDOM
603	job_on_remhost=${job_name}_${identifier}_$local_host
604	job_to_send=job_to_send_$identifier
605	if [[ $delete_dayfile = false ]]
606	then
607	remote_dayfile=${local_host}_${job_name}_result_$identifier
608	local_dayfile=${remote_host}_${job_name}
609	else
610	remote_dayfile=/dev/null
611	fi
612
613
614
615	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
616	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
617	then
618
619	# GENERAL LOADLEVELER SETTINGS
620	execute_in_shell="#!/bin/ksh"
621	use_shell="# @ shell = /bin/ksh"
622	consumable_memory="ConsumableMemory($memory mb)"
623	class="# @ class = $queue"
624	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
625	network_to_use="# @ network.mpi = sn_all,shared,us"
626	data_limit="# @ data_limit = 1.76gb"
627	image_size="# @ image_size = 50"
628	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
629
630	if [[ $email_notification = none ]]
631	then
632	notify_user=""
633	else
634	notify_user="# @ notify_user = $email_notification"
635	if [[ $delete_dayfile = true ]]
636	then
637	notification='# @ notification = never'
638	fi
639	fi
640
641	if [[ $remote_host = ibmh ]]
642	then
643	data_limit=""
644	network_to_use=""
645	class="# @ class = $queue"
646	environment=""
647	rset="# @ rset = RSET_MCM_AFFINITY"
648	task_affinity="# @ task_affinity = core(1)"
649	elif [[ $remote_host = ibmkisti ]]
650	then
651	network_to_use="# @ network.MPI = sn_all,shared,US"
652	wall_clock_limit="# @ wall_clock_limit = $timestring"
653	if [[ $threads_per_task = 1 ]]
654	then
655	rset="# @ rset = RSET_MCM_AFFINITY"
656	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
657	fi
658	environment=""
659	use_shell=""
660	data_limit=""
661	image_size=""
662	elif [[ $remote_host = ibmku ]]
663	then
664	execute_in_shell="#!/usr/bin/ksh"
665	use_shell="# @ shell = /usr/bin/ksh"
666	consumable_memory=""
667	environment=""
668	network_to_use="# @ network.mpi = sn_all,shared,us"
669	data_limit=""
670	image_size=""
671	elif [[ $remote_host = ibms ]]
672	then
673	network_to_use="# @ network.mpi = csss,shared,us"
674	fi
675
676	cat > $job_to_send << %%END%%
677	$execute_in_shell
678	$use_shell
679
680	# @ job_type = parallel
681	# @ job_name = $job_name
682	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
683	# @ output = $remote_dayfile
684	# @ error = $remote_dayfile
685	$wall_clock_limit
686	$image_size
687	$class
688	$environment
689	$network_to_use
690	$data_limit
691	$rset
692	$mcm_affinity_options
693	$task_affinity
694	$notification
695	$notify_user
696
697	%%END%%
698
699	if (( nodes > 0 ))
700	then
701
702	if [[ $remote_host != ibmkisti ]]
703	then
704
705	cat >> $job_to_send << %%END%%
706	# @ node = $nodes
707	# @ tasks_per_node = $processes_per_node
708	# @ node_usage = $node_usage
709	# @ queue
710
711	%%END%%
712
713	else
714
715	cat >> $job_to_send << %%END%%
716	# @ total_tasks = $mpi_tasks
717	# @ blocking = unlimited
718	# @ queue
719
720	%%END%%
721
722	fi
723
724	else
725
726	cat >> $job_to_send << %%END%%
727	# @ blocking = unlimited
728	# @ total_tasks = $numprocs
729	# @ node_usage = $node_usage
730	# @ queue
731
732	%%END%%
733
734	fi
735
736	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
737	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
738	then
739	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
740	fi
741
742	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
743	then
744
745	cat > $job_to_send << %%END%%
746	#!/bin/ksh
747
748	# @ job_type = serial
749	# @ node_usage = $node_usage
750	# @ job_name = palm
751	# @ wall_clock_limit = ${timestring},$timestring
752	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
753	# @ output = $remote_dayfile
754	# @ error = $remote_dayfile
755	$class
756	$notification
757
758	# @ queue
759
760	%%END%%
761
762	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
763	then
764
765	if [[ $numprocs != 0 ]]
766	then
767	cat > $job_to_send << %%END%%
768	#!/bin/bash
769	#PBS -N $job_name
770	#PBS -l walltime=$timestring
771	#PBS -l nodes=$nodes:ppn=$processes_per_node
772	#PBS -o $remote_dayfile
773	#PBS -j oe
774	#PBS -q $queue
775
776	$module_calls
777
778	%%END%%
779
780	else
781
782	continue
783
784	fi
785
786	elif [[ $remote_host = lccrayf ]]
787	then
788
789	if [[ $numprocs != 0 ]]
790	then
791	cat > $job_to_send << %%END%%
792	#!/bin/bash -l
793	#SBATCH -J $job_name
794	#SBATCH -t $timestring
795	#SBATCH -N $nodes
796	#SBATCH --ntasks-per-node=$processes_per_node
797	#SBATCH -p $queue
798	#SBATCH -o $remote_dayfile
799	#SBATCH -e $remote_dayfile
800
801	$init_cmds
802	$module_calls
803
804	%%END%%
805
806	else
807	cat > $job_to_send << %%END%%
808	#!/bin/bash -l
809	#SBATCH -J $job_name
810	#SBATCH -t $timestring
811	#SBATCH -l ncpus=1
812	#SBATCH -l pmem=${memory}mb
813	#SBATCH -m abe
814	#SBATCH -o $remote_dayfile
815	#SBATCH -e $remote_dayfile
816
817	$init_cmds
818	$module_calls
819
820	%%END%%
821
822	fi
823
824	elif [[ $remote_host = lcflow ]]
825	then
826	if [ $memory -gt 1800 ]; then
827	use_himem=""
828	else
829	use_himem="#"
830	fi
831	if [[ $numprocs != 0 ]]
832	then
833	pe_set="#$ -pe impi41 $numprocs"
834	else
835	pe_set="#$ -pe impi41 1"
836	fi
837	if [[ $queue = default ]]
838	then
839	queue_set=""
840	else
841	queue_set="#$ -q $queue"
842	fi
843	[[ "$disc_space" = "" ]] && disc_space=50
844
845	cat > $job_to_send << %%END%%
846	#!/bin/bash
847	#$ -S /bin/bash
848	#$ -N $job_name
849	#$ -cwd
850	#$ -l h_rt=$timestring
851	#$ -l h_vmem=${memory}M
852	#$ -o $remote_dayfile
853	#$ -j y
854	$pe_set
855	#$ -R y
856	${use_himem}#$ -l highmem=true
857	#$ -l h_fsize=${disc_space}G
858	$queue_set
859
860	%%END%%
861
862	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
863	then
864
865	if [[ $numprocs != 0 ]]
866	then
867	cat > $job_to_send << %%END%%
868	#!/bin/ksh
869	#PBS -N $job_name
870	#PBS -l walltime=$timestring
871	#PBS -l ncpus=$numprocs
872	#PBS -l pmem=${memory}mb
873	#PBS -o $remote_dayfile
874	#PBS -l nodes=$nodes:ppn=${processes_per_node}
875	#PBS -j oe
876
877	mpd &
878
879	%%END%%
880
881	else
882	cat > $job_to_send << %%END%%
883	#!/bin/ksh
884	#PBS -N $job_name
885	#PBS -l walltime=$timestring
886	#PBS -l ncpus=1
887	#PBS -l pmem=${memory}mb
888	#PBS -o $remote_dayfile
889	#PBS -j oe
890
891	%%END%%
892
893	fi
894
895	elif [[ $remote_host = lckiaps ]]
896	then
897
898	if [[ $numprocs != 0 ]]
899	then
900	cat > $job_to_send << %%END%%
901	#!/bin/ksh
902	#PBS -N $job_name
903	#PBS -l walltime=$timestring
904	#PBS -l select=1:ncpus=$numprocs
905	#PBS -l pmem=${memory}mb
906	#PBS -q $queue
907	#PBS -o $remote_dayfile
908	#PBS -j oe
909	#PBS -V
910
911	%%END%%
912
913	else
914	cat > $job_to_send << %%END%%
915	#!/bin/ksh
916	#PBS -N $job_name
917	#PBS -l walltime=$timestring
918	#PBS -l ncpus=1
919	#PBS -l pmem=${memory}mb
920	#PBS -o $remote_dayfile
921	#PBS -j oe
922
923	%%END%%
924
925	fi
926
927	elif [[ $remote_host = lcyon ]]
928	then
929
930	if [[ $numprocs != 0 ]]
931	then
932	cat > $job_to_send << %%END%%
933	#!/bin/ksh
934	#PBS -N $job_name
935	#PBS -l walltime=$timestring
936	#PBS -l ncpus=$numprocs
937	#PBS -l pmem=${memory}mb
938	#PBS -o $remote_dayfile
939	#PBS -j oe
940
941	%%END%%
942
943	else
944	cat > $job_to_send << %%END%%
945	#!/bin/ksh
946	#PBS -N $job_name
947	#PBS -l walltime=$timestring
948	#PBS -l ncpus=1
949	#PBS -l pmem=${memory}mb
950	#PBS -o $remote_dayfile
951	#PBS -j oe
952
953	%%END%%
954
955	fi
956
957	elif [[ $remote_host = lcxe6 ]]
958	then
959
960	if [[ $numprocs != 0 ]]
961	then
962	cat > $job_to_send << %%END%%
963	#!/bin/ksh
964	#PBS -S /bin/ksh
965	#PBS -N $job_name
966	#PBS -A $project_account
967	#PBS -j oe
968	#PBS -l walltime=$timestring
969	#PBS -l mppwidth=${numprocs}
970	#PBS -l mppnppn=${processes_per_node}
971	#PBS -m abe
972	#PBS -o $remote_dayfile
973	$email_directive
974
975	$init_cmds
976	$module_calls
977
978	%%END%%
979
980	else
981	cat > $job_to_send << %%END%%
982	#!/bin/ksh
983	#PBS -S /bin/ksh
984	#PBS -N $job_name
985	#PBS -A $project_account
986	#PBS -j oe
987	#PBS -l walltime=$timestring
988	#PBS -l ncpus=1
989	#PBS -l pmem=${memory}mb
990	#PBS -m abe
991	$email_directive
992	#PBS -o $remote_dayfile
993
994	$init_cmds
995	$module_calls
996
997	%%END%%
998
999	fi
1000
1001	elif [[ $remote_host = lckyoto ]]
1002	then
1003
1004	cat > $job_to_send << %%END%%
1005	#!/bin/ksh
1006	# @\$-o $remote_dayfile
1007	# @\$-eo -oi
1008	# @\$-lP 16
1009	# @\$-lp 1
1010	# @\$-lm 28gb -llm unlimited -ls unlimited
1011	# @\$-q $queue
1012	# @\$-Pvn abs_pack
1013	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1014	#. /thin/local/etc/setprofile/intel-11.0.sh
1015	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1016	. ~/.myprofile
1017	#. /home2/t/t51254/palm/current_version/myprofile
1018	#. /thin/apps/pgi/mpi.sh
1019	#
1020	env
1021	#
1022	set -x
1023
1024	%%END%%
1025
1026	elif [[ $remote_host = lcxt5m ]]
1027	then
1028
1029	if [[ $numprocs != 0 ]]
1030	then
1031	cat > $job_to_send << %%END%%
1032	#!/bin/ksh
1033	#PBS -S /bin/ksh
1034	#PBS -N $job_name
1035	#PBS -j oe
1036	#PBS -l walltime=$timestring
1037	#PBS -l mppwidth=${numprocs}
1038	#PBS -l mppnppn=${processes_per_node}
1039	#PBS -m abe
1040	#PBS -o $remote_dayfile
1041
1042	$init_cmds
1043	$module_calls
1044
1045	%%END%%
1046
1047	else
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 ncpus=1
1055	#PBS -l pmem=${memory}mb
1056	#PBS -m abe
1057	#PBS -o $remote_dayfile
1058
1059	$init_cmds
1060	$module_calls
1061
1062	%%END%%
1063
1064	fi
1065
1066	elif [[ $remote_host = lckyuh ]]
1067	then
1068	cat > $job_to_send << %%END%%
1069	#!/bin/bash
1070	#PJM -L "rscgrp=$queue"
1071	#PJM -L "node=$nodes"
1072	#PJM --mpi "proc=$numprocs"
1073	#PJM -L "elapse=$timestring"
1074	#PJM -o $remote_dayfile
1075	#PJM -j
1076	#PJM -X
1077	#PJM --no-stging
1078
1079	export LANG=en_US.UTF-8
1080	%%END%%
1081
1082	elif [[ $remote_host = lckyut ]]
1083	then
1084	cat > $job_to_send << %%END%%
1085	#!/bin/bash
1086	#PJM -L "rscgrp=$queue"
1087	#PJM -L "vnode=$numprocs"
1088	#PJM -L "vnode-core=1"
1089	#PJM -L "elapse=$timestring"
1090	#PJM --mpi proc=$numprocs
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 = nech ]]
1100	then
1101
1102	if (( nodes > 1 ))
1103	then
1104	cat > $job_to_send << %%END%%
1105	#!/bin/ksh
1106	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1107	#PBS -l ${qsubmem}=${Memory}gb
1108	#PBS -b $nodes
1109	#PBS -o $remote_dayfile
1110	#PBS -N palm
1111	#PBS -j o
1112	#PBS -T mpisx
1113
1114	%%END%%
1115
1116	elif [[ $numprocs != 0 ]]
1117	then
1118	cat > $job_to_send << %%END%%
1119	#!/bin/ksh
1120	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1121	#PBS -l ${qsubmem}=${Memory}gb
1122	#PBS -o $remote_dayfile
1123	#PBS -N palm
1124	#PBS -j o
1125
1126	%%END%%
1127
1128	else
1129	cat > $job_to_send << %%END%%
1130	#!/bin/ksh
1131	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1132	#PBS -o $remote_dayfile
1133	#PBS -j o
1134
1135	%%END%%
1136
1137	fi
1138
1139	elif [[ $remote_host = lctit ]]
1140	then
1141	cat > $job_to_send << %%END%%
1142	#!/bin/ksh
1143	$init_cmds
1144	$module_calls
1145
1146	%%END%%
1147
1148	# SET OPTIONS FOR SUBMIT-COMMAND
1149	if [[ $tasks_per_node != $processes_per_node ]]
1150	then
1151	submcom="$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 $queue "
1152	else
1153	submcom="$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 "
1154	fi
1155
1156	else
1157
1158	cat > $job_to_send << %%END%%
1159	# @\$-q ${queue}
1160	# @\$-l${qsubtime} $timestring
1161	# @\$-l${qsubmem} ${memory}mb
1162	# @\$-o $remote_dayfile
1163	# @\$-eo
1164
1165	%%END%%
1166
1167	fi
1168
1169
1170	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1171	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1172	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1173	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1174	then
1175	echo "set +vx" >> $job_to_send
1176	echo "trap '" >> $job_to_send
1177	echo "set +vx" >> $job_to_send
1178	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* ]]
1179	then
1180	if [[ $remote_host = ibmh ]]
1181	then
1182	return_queue=c1
1183	elif [[ $remote_host = ibmkisti ]]
1184	then
1185	return_queue=class.1-2
1186	elif [[ $remote_host = ibmku ]]
1187	then
1188	return_queue=sdbg2
1189	elif [[ $remote_host = ibms ]]
1190	then
1191	return_queue=p_normal
1192	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1193	then
1194	return_queue=dataq
1195	elif [[ $remote_host = lckiaps ]]
1196	then
1197	return_queue=express
1198	elif [[ $remote_host = lckyuh ]]
1199	then
1200	return_queue=cx-single
1201	elif [[ $remote_host = lckyut ]]
1202	then
1203	return_queue=cx-single
1204	else
1205	return_queue=unknown
1206	fi
1207
1208	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1209	then
1210
1211	if [[ $remote_host = ibmku ]]
1212	then
1213	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1214	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1215	else
1216	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1217	fi
1218	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1219	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1220	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1221	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1222	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1223	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1224	if [[ $host != "ibmh" ]]
1225	then
1226	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1227	fi
1228	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1229	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1230
1231	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1232	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1233
1234	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1235	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
1236	if [[ $remote_host = ibmku ]]
1237	then
1238	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1239	fi
1240	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1241
1242	elif [[ $remote_host = nech ]]
1243	then
1244	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1245	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1246	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1247	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1248	echo "#PBS -j o" >> $job_to_send
1249	echo " " >> $job_to_send
1250	echo "set -x" >> $job_to_send
1251	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1252	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1253	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1254	echo "%%END%%" >> $job_to_send
1255
1256	elif [[ $remote_host = lckyuh ]]
1257	then
1258	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1259	echo "#!/bin/bash" >> $job_to_send
1260	echo "#PJM -L \"node=1\"" >> $job_to_send
1261	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1262	echo "#PJM --no-stging" >> $job_to_send
1263	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1264	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1265	echo "#PJM -j" >> $job_to_send
1266	echo " " >> $job_to_send
1267	echo "export LANG=en_US.UTF-8" >> $job_to_send
1268	echo "set -x" >> $job_to_send
1269	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1270	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1271	echo "%%END%%" >> $job_to_send
1272
1273	elif [[ $remote_host = lckyut ]]
1274	then
1275	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1276	echo "#!/bin/bash" >> $job_to_send
1277	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1278	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1279	echo "#PJM --no-stging" >> $job_to_send
1280	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1281	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1282	echo "#PJM -j" >> $job_to_send
1283	echo " " >> $job_to_send
1284	echo "export LANG=en_US.UTF-8" >> $job_to_send
1285	echo "set -x" >> $job_to_send
1286	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1287	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1288	echo "%%END%%" >> $job_to_send
1289
1290	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1291	then
1292	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1293	echo "#!/bin/bash" >> $job_to_send
1294	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1295	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1296	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1297	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1298	echo "#PBS -j oe" >> $job_to_send
1299	echo " " >> $job_to_send
1300	echo "set -x" >> $job_to_send
1301	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1302	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1303	echo "%%END%%" >> $job_to_send
1304
1305	elif [[ $remote_host = lcflow ]]
1306	then
1307	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1308	echo "#!/bin/bash" >> $job_to_send
1309	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1310	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1311	echo "SGEPREFIX -cwd" >> $job_to_send
1312	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1313	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1314	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1315	echo "SGEPREFIX -j y" >> $job_to_send
1316	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1317	echo " " >> $job_to_send
1318	echo "set -x" >> $job_to_send
1319	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1320	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1321	echo "" >> $job_to_send
1322	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1323	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1324	echo "rm -f scpjob.${identifier}" >> $job_to_send
1325	echo "%%END%%" >> $job_to_send
1326	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1327	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1328
1329	else
1330
1331	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1332	echo "# @\\\$-q $return_queue" >> $job_to_send
1333	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1334	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1335	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1336	then
1337	echo "# @\$-l mpp_p=0" >> $job_to_send
1338	fi
1339	echo '# @\$-lF 10mb' >> $job_to_send
1340	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1341	echo '# @\\\$-eo' >> $job_to_send
1342	echo " " >> $job_to_send
1343	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1344	then
1345	echo "set +vx" >> $job_to_send
1346	echo ". .profile" >> $job_to_send
1347	fi
1348	echo "set -x" >> $job_to_send
1349	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1350	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1351	echo "%%END%%" >> $job_to_send
1352	fi
1353
1354	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1355	then
1356	echo "llsubmit scpjob.$identifier" >> $job_to_send
1357	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1358	then
1359	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1360	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1361	then
1362	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1363	elif [[ $remote_host = t3es ]]
1364	then
1365	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1366	elif [[ $remote_host = lckiaps ]]
1367	then
1368	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1369	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1370	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1371	elif [[ $remote_host = lckyu* ]]
1372	then
1373	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1374	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1375	elif [[ $remote_host = lcflow ]]
1376	then
1377	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1378	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1379	else
1380	echo "$submcom scpjob.$identifier" >> $job_to_send
1381	fi
1382	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1383	then
1384	echo "rm scpjob.$identifier" >> $job_to_send
1385	fi
1386	if [[ $remote_host = nech ]]
1387	then
1388	echo "cd -" >> $job_to_send
1389	fi
1390	else
1391	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1392	# ??? funktioniert das ÃŒberhaupt noch ???
1393	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1394	fi
1395	echo "set -x" >> $job_to_send
1396	echo " ' exit" >> $job_to_send
1397	echo "set -x" >> $job_to_send
1398	fi
1399
1400
1401
1402	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1403	cat $file_to_send >> $job_to_send
1404
1405	if [[ $remote_host = ibm ]]
1406	then
1407	echo " " >> $job_to_send
1408	echo "exit" >> $job_to_send
1409	fi
1410
1411	# REMOVE JOB-FILE
1412	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1413	then
1414	echo " " >> $job_to_send
1415	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1416	fi
1417
1418
1419	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1420	if [[ $no_submit = false ]]
1421	then
1422	if [[ $remote_host != $local_host ]]
1423	then
1424	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1425	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1426	then
1427	job_catalog_save=$job_catalog
1428	job_catalog=job_queue
1429	elif [[ $remote_host = nech ]]
1430	then
1431	job_catalog_save=$job_catalog
1432	job_catalog=/hpf/b/${remote_user}/job_queue
1433	fi
1434	if [[ $remote_host = nech ]]
1435	then
1436	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1437	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1438	else
1439	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1440	fi
1441	if [[ $? = 1 ]]
1442	then
1443	locat=scp; exit
1444	fi
1445	if [[ $remote_host = ibms ]]
1446	then
1447	job_catalog=$job_catalog_save
1448	fi
1449	[[ $verify = true ]] && printf "\n >>> finished\n"
1450	else
1451	eval job_catalog=$job_catalog
1452	cp $job_to_send ${job_catalog}/$job_on_remhost
1453	fi
1454
1455
1456
1457	# START NQS- / LOADLEVELER-JOB
1458	if [[ $remote_host != $local_host ]]
1459	then
1460	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1461
1462	if [[ $remote_host = ibmku ]]
1463	then
1464	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1465	elif [[ $remote_host = lcflow ]]
1466	then
1467	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1468	else
1469	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1470	fi
1471
1472	[[ $verify = true ]] && printf " >>> o.k.\n"
1473	else
1474	cd $job_catalog
1475	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1476	then
1477	eval $submcom $job_on_remhost
1478	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1479	then
1480	chmod u+x $job_on_remhost
1481	eval $submcom $job_on_remhost
1482	elif [[ $local_host = nech ]]
1483	then
1484	if [[ $queue = default ]]
1485	then
1486	eval $submcom $job_on_remhost
1487	else
1488	eval $submcom -q $queue $job_on_remhost
1489	fi
1490	else
1491	qsub $job_on_remhost
1492	fi
1493
1494	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1495	# AT THE END OF THE JOB
1496	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1497	then
1498	rm $job_on_remhost
1499	fi
1500	cd - > /dev/null
1501	fi
1502	fi
1503
1504
1505
1506	# FINAL ACTIONS
1507	if [[ $no_submit = false ]]
1508	then
1509	rm -f $job_to_send
1510	fi
1511	[[ $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 |