Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2135

Last change on this file since 2135 was 2135, checked in by raasch, 8 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 58.1 KB

Line
1	#!/bin/bash
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 2135 2017-02-02 07:35:41Z raasch $
30	#
31	# 2134 2017-02-02 07:33:46Z raasch
32	# option -E added to msub commands on HLRN-III machines to allow output of more
33	# job informations in the job protocol files
34	#
35	# 1944 2016-06-15 06:29:00Z raasch
36	# adjustments for using HLRN ssh-keys
37	#
38	# 1940 2016-06-14 05:15:20Z raasch
39	# adjustments for lckiaps
40	#
41	# 1866 2016-04-15 06:50:59Z raasch
42	# adjusted for lcocean
43	#
44	# 1841 2016-04-07 19:14:06Z raasch
45	# script now running under bash
46	#
47	# 1701 2015-11-02 07:43:04Z maronga
48	# Bugfix: added missing init_cmds for lccrayh/lccrayb
49	#
50	# 1621 2015-07-17 11:39:33Z heinze
51	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
52	#
53	# 1575 2015-03-27 09:56:27Z raasch
54	# mpp2-queues added to lccrayh
55	#
56	# 1547 2015-01-29 15:09:12Z witha
57	# adjustments for ForWind computing cluster (lcflow)
58	#
59	# 1545 2015-01-29 06:52:23Z heinze
60	# local host name for blizzard further specified
61	#
62	# 1480 2014-10-17 14:41:49Z raasch
63	# adjustments for 2nd stage of HLRNIII
64	#
65	# 1468 2014-09-24 14:06:57Z maronga
66	# Typo removed (addres->address)
67	# Adjustments for lcxe6
68	#
69	# 1452 2014-08-22 09:41:06Z heinze
70	# local hosts for blizzard added
71	#
72	# 1450 2014-08-21 07:31:51Z heinze
73	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
74	#
75	# 1442 2014-07-28 07:09:10Z raasch
76	# HLRN-III (lccrayb/lccrayh) queues adjusted
77	#
78	# 1378 2014-04-28 06:04:58Z raasch
79	# -et option added for lctit
80	#
81	# 1350 2014-04-04 13:01:30Z maronga
82	# location of qsub updated for lcxe6
83	#
84	# 1289 2014-03-04 07:12:34Z raasch
85	# German comments translated to English
86	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
87	#
88	# 1279 2014-01-28 12:10:14Z raasch
89	# node calculation modified due to changes in mrun (tasks_per_node must not be
90	# an integral divisor of numprocs any more)
91	#
92	# 1274 2014-01-09 13:14:54Z heinze
93	# adjustments for lccrayh
94	#
95	# 1266 2013-12-11 12:07:34Z heinze
96	# further adjustments for lccrayb (use msub instead of qsub)
97	#
98	# 1264 2013-12-09 12:46:09Z fricke
99	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
100	#
101	# 1262 2013-12-09 10:57:20Z fricke
102	# further adjustments for lccrayb
103	#
104	# 1260 2013-12-04 12:48:04Z raasch
105	# jaboticaba admitted
106	#
107	# 1255 2013-11-07 14:43:35Z raasch
108	# further adjustments for lccrayb
109	#
110	# 1224 2013-09-16 07:27:23Z raasch
111	# first adjustments for lccrayb
112	#
113	# 1202 2013-07-10 16:22:07Z witha
114	# adjustments for Forwind cluster (lcflow)
115	#
116	# 1199 2013-07-05 14:52:22Z raasch
117	# adjustments for CSC Helsinki (lccrayf)
118	#
119	# use of cluster/express queue enabled (ibmh)
120	# vinessa added (imuk)
121	#
122	# 1103 2013-02-20 02:15:53Z raasch
123	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
124	# instead typeset -L/R),
125	# further adjustments for lckyuh
126	#
127	# 2013-02-10 01:47:43Z raasch
128	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
129	# and for Forwind cluster (lcflow)
130	#
131	# 1094 2013-02-03 01:52:12Z raasch
132	# new option -P for explicit setting of ssh/scp port,
133	# decalpha parts (yonsei) removed
134	#
135	# 2013-02-02 07:06:13Z raasch
136	# adjustments for Kyushu-University computing center (lckyut - tatara)
137	# old changelog messages removed
138	#
139	# 1046 2012-11-09 14:38:45Z maronga
140	# code put under GPL (PALM 3.9)
141	#
142	# 08/07/94 - Siggi - first version finished
143	# 29/06/94 - Siggi - script development started
144	#--------------------------------------------------------------------------------#
145	# subjob - script for automatic generation and submission of batch-job files
146	# for various batch queuing systems
147	#--------------------------------------------------------------------------------#
148
149
150	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
151	delete_dayfile=false
152	email_notification=none
153	group_number=none
154	locat=normal
155	no_default_queue=none
156	no_submit=false
157	job_catalog="~/job_queue"
158	job_name=none
159	local_user=$LOGNAME
160	node_usage=shared
161	numprocs=0
162	punkte="..........................................................."
163	submcom=qsub
164	queue=default
165	remote_host=none
166	remote_user=""
167	verify=true
168
169	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
170	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
171
172
173
174	# ERROR HANDLING
175	# IN CASE OF EXIT:
176	trap 'if [[ $locat != normal ]]
177	then
178	case $locat in
179	(option) printf "\n --> available optios can be displayed"
180	printf " by typing:"
181	printf "\n \"subjob ?\" \n";;
182	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
183	(*) printf "\n +++ unknown error"
184	printf "\n please inform S. Raasch!\n"
185	esac
186	[[ -f $job_to_send ]] && rm $job_to_send
187	printf "\n\n+++ SUBJOB killed \n\n"
188	fi' exit
189
190
191	# IN CASE OF TERMINAL-BREAK:
192	trap '[[ -f $job_to_send ]] && rm $job_to_send
193	printf "\n\n+++ SUBJOB killed \n\n"
194	exit
195	' 2
196
197
198	# DETERMINE NAME OF LOCAL HOST
199	local_host=$(hostname)
200
201	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
202	# IS ADMITTED AT ALL)
203	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
204	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
205	case $local_host in
206	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
207	(atmos) local_address=172.20.25.35; local_host=lcide;;
208	(austru) local_address=130.75.105.128; local_host=lcmuk;;
209	(autan) local_address=130.75.105.57; local_host=lcmuk;;
210	(bora) local_address=130.75.105.103; local_host=lcmuk;;
211	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
212	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
213	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
214	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
215	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
216	(breva) local_address=130.75.105.98; local_host=lcmuk;;
217	(buran) local_address=130.75.105.58; local_host=lcmuk;;
218	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
219	(climate*) local_address=165.132.26.68; local_host=lcyon;;
220	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
221	(cs*) local_address=136.172.44.131; local_host=nech;;
222	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
223	(flow01) local_address=10.141.255.71; local_host=lcflow;;
224	(flow02) local_address=10.141.255.72; local_host=lcflow;;
225	(node*) local_address=165.132.26.61 local_host=lck;;
226	# (node*) local_address=210.219.61.8 local_host=lckordi;;
227	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
228	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
229	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
230	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
231	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
232	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
233	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
234	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
235	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
236	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
237	(sno) local_address=130.75.105.113; local_host=lcmuk;;
238	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
239	(login*) local_address=118.128.66.201; local_host=lckiaps;;
240	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
241	(meller) local_address=134.106.74.155; local_host=lcfor;;
242	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
243	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
244	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
245	(nobel*) local_address=150.183.5.101; local_host=ibms;;
246	(ocean) local_address="ocean"; local_host=lcocean;;
247	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
248	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
249	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
250	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
251	(pingui) local_address=134.106.74.118; local_host=lcfor;;
252	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
253	(rte*) local_address=133.5.185.60; local_host=lcrte;;
254	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
255	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
256	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
257	(solano) local_address=130.75.105.110; local_host=lcmuk;;
258	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
259	(tc*) local_address="ocean"; local_host=lcocean;;
260	(t2a*) local_address=10.1.6.165; local_host=lctit;;
261	(urban*) local_address=147.46.30.151 local_host=lcsb;;
262	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
263	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
264	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
265	(*) printf "\n +++ \"$local_host\" unknown";
266	printf "\n please contact the PALM group at IMUK";
267	locat=parameter; exit;;
268	esac
269
270
271
272	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
273	remote_host=$local_host
274
275
276
277
278	# READ THE SHELLSCRIPT-OPTIONS
279	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
280	do
281	case $option in
282	(c) job_catalog=$OPTARG;;
283	(d) delete_dayfile=true;;
284	(D) no_submit=true;;
285	(e) email_notification=$OPTARG;;
286	(g) group_number=$OPTARG;;
287	(h) remote_host=$OPTARG;;
288	(m) memory=$OPTARG;;
289	(n) job_name=$OPTARG;;
290	(N) node_usage=$OPTARG;;
291	(O) threads_per_task=$OPTARG;;
292	(P) scp_port=$OPTARG;;
293	(q) no_default_queue=$OPTARG;;
294	(t) cputime=$OPTARG;;
295	(T) tasks_per_node=$OPTARG;;
296	(u) remote_user=$OPTARG;;
297	(v) verify=false;;
298	(X) numprocs=$OPTARG;;
299	(\?) printf "\n +++ Option $OPTARG unknown \n";
300	locat=option; exit;;
301	esac
302	done
303
304
305	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
306	(( to_shift = $OPTIND - 1 ))
307	shift $to_shift; file_to_send=$1
308
309
310	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
311	if [ "$1" = "?" ]
312	then
313	(printf "\n *** subjob can be called as follows:\n"
314	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
315	printf "\n Description of available options:\n"
316	printf "\n Option Description Default-Value"
317	printf "\n -c job-input- and output-catalog ~/job_queue"
318	printf "\n -d no job-protocol will be created ---"
319	printf "\n -D only the job-file will be created ---"
320	printf "\n -h execution host, available hosts: $remote_host"
321	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
322	printf "\n lckiaps, lctit, nech"
323	printf "\n -m memory demand per process in MByte ---"
324	printf "\n -n jobname <jobdatei>"
325	printf "\n -O threads per task (for OpenMP usage) 1"
326	printf "\n -P ssh/scp port default port"
327	printf "\n -q job-queue to be used default"
328	printf "\n -t allowed cpu-time in seconds ---"
329	printf "\n -T tasks per node (on parallel hosts) ---"
330	printf "\n -u username on execution host from .netrc"
331	printf "\n -v no prompt for confirmation ---"
332	printf "\n -X # of processors (on parallel hosts) 1"
333	printf "\n "
334	printf "\n The only possible positional parameter is <jobfile>:"
335	printf "\n The complete NQS-job must be provided here."
336	printf "\n <jobfile>=? creates this outline\n\n") \| more
337	exit
338	fi
339
340
341
342	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
343	if [[ "$file_to_send" = "" ]]
344	then
345	printf "\n +++ job-file missing"
346	locat=parameter; exit
347	else
348	if [[ -f $file_to_send ]]
349	then
350	true
351	else
352	printf "\n +++ job-file: "
353	printf "\n $file_to_send"
354	printf "\n does not exist"
355	locat=parameter; exit
356	fi
357	fi
358
359
360
361	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
362	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
363	if [[ $job_name = none ]]
364	then
365	job_name=$file_to_send
366	fi
367	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
368	then
369	printf "\n +++ job-file name: "
370	printf "\n $job_name"
371	printf "\n must not contain \"/\"-characters"
372	locat=parameter; exit
373	fi
374
375
376
377
378	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
379	# OR IF NO HOST HAS BEEN GIVEN
380	if [[ $remote_host = none ]]
381	then
382	printf "\n +++ host missing"
383	locat=option; exit
384	else
385	case $remote_host in
386	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
387	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
388	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
389	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
390	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
391	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
392	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
393	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
394	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
395	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
396	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
397	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
398	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
399	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
400	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
401	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
402	(lcocean) remote_address="ocean"; submcom=qsub;;
403	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
404	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
405	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
406	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
407	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
408	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
409	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
410	locat=parameter; exit;;
411	esac
412	fi
413
414
415	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
416	if [[ $no_default_queue != none ]]
417	then
418	error=false
419	ndq=$no_default_queue
420	case $remote_host in
421	(ibm) case $ndq in
422	(p690_express\|p690_standard\|p690_long) error=false;;
423	(*) error=true;;
424	esac;;
425	(ibmh) case $ndq in
426	(cluster\|express) error=false;;
427	(*) error=true;;
428	esac;;
429	(ibmkisti) case $ndq in
430	(class.32plus\|class.1-2\|class.2-32) error=false;;
431	(*) error=true;;
432	esac;;
433	(ibmku) case $ndq in
434	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
435	(*) error=true;;
436	esac;;
437	(ibms) case $ndq in
438	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
439	(*) error=true;;
440	esac;;
441	(lcbullhh) case $ndq in
442	(compute\|shared) error=false;;
443	(*) error=true;;
444	esac;;
445	(lccrayb) case $ndq in
446	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
447	(*) error=true;;
448	esac;;
449	(lccrayh) case $ndq in
450	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
451	(*) error=true;;
452	esac;;
453	(lccrayf) case $ndq in
454	(usup\|test*\|small\|large) error=false;;
455	(*) error=true;;
456	esac;;
457	(lcflow) case $ndq in
458	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q\|cfd_ivy_shrt.q) error=false;;
459	(*) error=true;;
460	esac;;
461	(lckiaps) case $ndq in
462	(express\|normal\|normal20\|quickq) error=false;;
463	(*) error=true;;
464	esac;;
465	(lckyoto) case $ndq in
466	(eh\|ph) error=false;;
467	(*) error=true;;
468	esac;;
469	(lckyuh) case $ndq in
470	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
471	(*) error=true;;
472	esac;;
473	(lckyut) case $ndq in
474	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
475	(*) error=true;;
476	esac;;
477	(lctit) case $ndq in
478	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
479	(*) error=true;;
480	esac;;
481	(t3eb) case $ndq in
482	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
483	(*) error=true;;
484	esac;;
485	(t3eh) case $ndq in
486	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
487	(*) error=true;;
488	esac;;
489	(t3ej2\|t3ej5) case $ndq in
490	(low\|normal\|high) error=false;;
491	(*) error=true;;
492	esac;;
493	(t3es) case $ndq in
494	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
495	(*) error=true;;
496	esac;;
497	esac
498	if [[ $error = true ]]
499	then
500	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
501	locat=parameter; exit
502	else
503	queue=$no_default_queue
504	fi
505	fi
506
507
508
509	# CHECK THE CPU-TIME
510	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
511	done=false
512	while [[ $done = false ]]
513	do
514	if (( $cputime <= 0 ))
515	then
516	printf "\n +++ wrong cpu-time or cpu-time missing"
517	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
518	printf "\n >>> "
519	read cputime 1>/dev/null 2>&1
520	else
521	done=true
522	fi
523	done
524	if [[ $remote_host = nech ]]
525	then
526	if (( tasks_per_node != 0 ))
527	then
528	(( cputime = cputime * tasks_per_node ))
529	elif [[ $numprocs != 0 ]]
530	then
531	(( cputime = cputime * numprocs ))
532	fi
533	fi
534	(( stunden = cputime / 3600 ))
535	(( resttime = cputime - stunden * 3600 ))
536	(( minuten = resttime / 60 ))
537	(( sekunden = resttime - minuten * 60 ))
538	timestring=${stunden}:${minuten}:${sekunden}
539
540
541
542	# CHECK THE MEMORY DEMAND
543	done=false
544	while [[ $done = false ]]
545	do
546	if (( memory <= 0 ))
547	then
548	printf "\n +++ wrong memory demand or memory demand missing"
549	printf "\n >>> Please type memory in MByte per process as INTEGER:"
550	printf "\n >>> "
551	read memory 1>/dev/null 2>&1
552	else
553	done=true
554	fi
555	done
556
557	if [[ $remote_host = nech ]]
558	then
559	if (( tasks_per_node != 0 ))
560	then
561	(( Memory = memory * tasks_per_node / 1000 ))
562	elif [[ $numprocs != 0 ]]
563	then
564	(( Memory = memory * numprocs / 1000 ))
565	else
566	(( Memory = memory / 1000 ))
567	fi
568	elif [[ $remote_host = lctit ]]
569	then
570	(( Memory = memory * tasks_per_node / 1000 ))
571	fi
572
573
574	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
575	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
576	then
577	(( memory = memory * threads_per_task ))
578	fi
579
580
581	# CALCULATE NUMBER OF REQUIRED NODES
582	if (( tasks_per_node != 0 ))
583	then
584	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
585	fi
586
587
588	# CALCULATE NUMBER OF PROCESSES PER NODE
589	(( processes_per_node = tasks_per_node * threads_per_task ))
590
591
592	# CALCULATE NUMBER OF MPI TASKS
593	(( mpi_tasks = numprocs / threads_per_task ))
594
595
596	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
597	if [[ "$scp_port" != "" ]]
598	then
599	PORTOPT="-P $scp_port"
600	SSH_PORTOPT="-p $scp_port"
601	fi
602
603
604	# HEADER-OUTPUT
605	if [[ $verify = true ]]
606	then
607	printf "\n\n"
608	printf "#--------------------------------------------------------------# \n"
609	spalte1=SUBJOB;spalte2=$(date)
610	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
611	printf "\| \| \n"
612	printf "\| values of parameters/options: \| \n"
613	spalte1=$(echo local_host$punkte \| cut -c-20)
614	spalte2=$punkte$local_host
615	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
616	spalte1=$(echo remote_host$punkte \| cut -c-20)
617	spalte2=$punkte$remote_host
618	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
619	spalte1=$(echo queue$punkte \| cut -c-20)
620	spalte2=$punkte$queue
621	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
622	spalte1=$(echo memory$punkte \| cut -c-20)
623	spalte2="$punkte$memory mb"
624	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
625	spalte1=$(echo cputime$punkte \| cut -c-20)
626	spalte2="$punkte$cputime sec"
627	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
628	spalte1=$(echo job_name$punkte \| cut -c-20)
629	spalte2="$punkte$job_name"
630	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
631	printf "#--------------------------------------------------------------# \n\n"
632
633
634	# QUERY CHECK
635	antwort="dummy"
636	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
637	do
638	read antwort?" >>> continue (y/n) ? "
639	done
640	if [[ $antwort = n \|\| $antwort = N ]]
641	then
642	locat=verify; exit
643	fi
644	printf "\n"
645	fi
646
647	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
648	identifier=$RANDOM
649	job_on_remhost=${job_name}_${identifier}_$local_host
650	job_to_send=job_to_send_$identifier
651	if [[ $delete_dayfile = false ]]
652	then
653	remote_dayfile=${local_host}_${job_name}_result_$identifier
654	local_dayfile=${remote_host}_${job_name}
655	else
656	remote_dayfile=/dev/null
657	fi
658
659
660	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
661	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
662	then
663
664	# GENERAL LOADLEVELER SETTINGS
665	execute_in_shell="#!/bin/ksh"
666	use_shell="# @ shell = /bin/ksh"
667	consumable_memory="ConsumableMemory($memory mb)"
668	class="# @ class = $queue"
669	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
670	network_to_use="# @ network.mpi = sn_all,shared,us"
671	data_limit="# @ data_limit = 1.76gb"
672	image_size="# @ image_size = 50"
673	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
674
675	if [[ $email_notification = none ]]
676	then
677	notify_user=""
678	else
679	notify_user="# @ notify_user = $email_notification"
680	if [[ $delete_dayfile = true ]]
681	then
682	notification='# @ notification = never'
683	fi
684	fi
685
686	if [[ $remote_host = ibmh ]]
687	then
688	data_limit=""
689	network_to_use=""
690	class="# @ class = $queue"
691	environment=""
692	rset="# @ rset = RSET_MCM_AFFINITY"
693	task_affinity="# @ task_affinity = core(1)"
694	elif [[ $remote_host = ibmkisti ]]
695	then
696	network_to_use="# @ network.MPI = sn_all,shared,US"
697	wall_clock_limit="# @ wall_clock_limit = $timestring"
698	if [[ $threads_per_task = 1 ]]
699	then
700	rset="# @ rset = RSET_MCM_AFFINITY"
701	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
702	fi
703	environment=""
704	use_shell=""
705	data_limit=""
706	image_size=""
707	elif [[ $remote_host = ibmku ]]
708	then
709	execute_in_shell="#!/usr/bin/ksh"
710	use_shell="# @ shell = /usr/bin/ksh"
711	consumable_memory=""
712	environment=""
713	network_to_use="# @ network.mpi = sn_all,shared,us"
714	data_limit=""
715	image_size=""
716	elif [[ $remote_host = ibms ]]
717	then
718	network_to_use="# @ network.mpi = csss,shared,us"
719	fi
720
721	cat > $job_to_send << %%END%%
722	$execute_in_shell
723	$use_shell
724
725	# @ job_type = parallel
726	# @ job_name = $job_name
727	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
728	# @ output = $remote_dayfile
729	# @ error = $remote_dayfile
730	$wall_clock_limit
731	$image_size
732	$class
733	$environment
734	$network_to_use
735	$data_limit
736	$rset
737	$mcm_affinity_options
738	$task_affinity
739	$notification
740	$notify_user
741
742	%%END%%
743
744	if (( nodes > 0 ))
745	then
746
747	if [[ $remote_host != ibmkisti ]]
748	then
749
750	cat >> $job_to_send << %%END%%
751	# @ node = $nodes
752	# @ tasks_per_node = $processes_per_node
753	# @ node_usage = $node_usage
754	# @ queue
755
756	%%END%%
757
758	else
759
760	cat >> $job_to_send << %%END%%
761	# @ total_tasks = $mpi_tasks
762	# @ blocking = unlimited
763	# @ queue
764
765	%%END%%
766
767	fi
768
769	else
770
771	cat >> $job_to_send << %%END%%
772	# @ blocking = unlimited
773	# @ total_tasks = $numprocs
774	# @ node_usage = $node_usage
775	# @ queue
776
777	%%END%%
778
779	fi
780
781	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
782	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
783	then
784	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
785	fi
786
787	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
788	then
789
790	cat > $job_to_send << %%END%%
791	#!/bin/ksh
792
793	# @ job_type = serial
794	# @ node_usage = $node_usage
795	# @ job_name = palm
796	# @ wall_clock_limit = ${timestring},$timestring
797	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
798	# @ output = $remote_dayfile
799	# @ error = $remote_dayfile
800	$class
801	$notification
802
803	# @ queue
804
805	%%END%%
806
807	elif [[ $remote_host = lcbullhh ]]
808	then
809	if [[ $numprocs != 0 ]]
810	then
811	cat > $job_to_send << %%END%%
812	#!/bin/bash -l
813	#SBATCH -J $job_name
814	#SBATCH -t $timestring
815	#SBATCH -N $nodes
816	#SBATCH --ntasks-per-node=$processes_per_node
817	#SBATCH -p $queue
818	#SBATCH -o $remote_dayfile
819	#SBATCH -e $remote_dayfile
820	#SBATCH -A $project_account
821
822	$init_cmds
823	$module_calls
824
825	%%END%%
826
827	else
828	cat > $job_to_send << %%END%%
829	#!/bin/bash -l
830	#SBATCH -J $job_name
831	#SBATCH -t $timestring
832	#SBATCH -l ncpus=1
833	#SBATCH -l pmem=${memory}mb
834	#SBATCH -m abe
835	#SBATCH -o $remote_dayfile
836	#SBATCH -e $remote_dayfile
837	#SBATCH -A $project_account
838
839	$init_cmds
840	$module_calls
841
842	%%END%%
843
844	fi
845
846	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
847	then
848
849	if [[ "$feature" != "" ]]
850	then
851	featuredir="#PBS -l feature=$feature"
852	fi
853
854	if [[ $numprocs != 0 ]]
855	then
856	cat > $job_to_send << %%END%%
857	#!/bin/bash
858	#PBS -N $job_name
859	#PBS -l walltime=$timestring
860	#PBS -l nodes=$nodes:ppn=$processes_per_node
861	#PBS -o $remote_dayfile
862	#PBS -j oe
863	#PBS -q $queue
864	$featuredir
865
866	$init_cmds
867	$module_calls
868
869	%%END%%
870
871	else
872
873	continue
874
875	fi
876
877	elif [[ $remote_host = lccrayf ]]
878	then
879
880	if [[ $numprocs != 0 ]]
881	then
882	cat > $job_to_send << %%END%%
883	#!/bin/bash -l
884	#SBATCH -J $job_name
885	#SBATCH -t $timestring
886	#SBATCH -N $nodes
887	#SBATCH --ntasks-per-node=$processes_per_node
888	#SBATCH -p $queue
889	#SBATCH -o $remote_dayfile
890	#SBATCH -e $remote_dayfile
891
892	$init_cmds
893	$module_calls
894
895	%%END%%
896
897	else
898	cat > $job_to_send << %%END%%
899	#!/bin/bash -l
900	#SBATCH -J $job_name
901	#SBATCH -t $timestring
902	#SBATCH -l ncpus=1
903	#SBATCH -l pmem=${memory}mb
904	#SBATCH -m abe
905	#SBATCH -o $remote_dayfile
906	#SBATCH -e $remote_dayfile
907
908	$init_cmds
909	$module_calls
910
911	%%END%%
912
913	fi
914
915	elif [[ $remote_host = lcflow ]]
916	then
917	if [[ $numprocs != 0 ]]
918	then
919	pe_set="#$ -pe impi $numprocs"
920	else
921	pe_set="#$ -pe impi 1"
922	fi
923	if [[ $queue = default ]]
924	then
925	queue_set=""
926	else
927	queue_set="#$ -q $queue"
928	fi
929	[[ "$disc_space" = "" ]] && disc_space=50
930
931	cat > $job_to_send << %%END%%
932	#!/bin/bash
933	#$ -S /bin/bash
934	#$ -N $job_name
935	#$ -cwd
936	#$ -l h_rt=$timestring
937	#$ -l h_vmem=${memory}M
938	#$ -o $remote_dayfile
939	#$ -j y
940	$pe_set
941	#$ -R y
942	#$ -l h_fsize=${disc_space}G
943	$queue_set
944
945	%%END%%
946
947	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
948	then
949
950	if [[ $numprocs != 0 ]]
951	then
952	cat > $job_to_send << %%END%%
953	#!/bin/ksh
954	#PBS -N $job_name
955	#PBS -l walltime=$timestring
956	#PBS -l ncpus=$numprocs
957	#PBS -l pmem=${memory}mb
958	#PBS -o $remote_dayfile
959	#PBS -l nodes=$nodes:ppn=${processes_per_node}
960	#PBS -j oe
961
962	mpd &
963
964	%%END%%
965
966	else
967	cat > $job_to_send << %%END%%
968	#!/bin/ksh
969	#PBS -N $job_name
970	#PBS -l walltime=$timestring
971	#PBS -l ncpus=1
972	#PBS -l pmem=${memory}mb
973	#PBS -o $remote_dayfile
974	#PBS -j oe
975
976	%%END%%
977
978	fi
979
980	elif [[ $remote_host = lckiaps ]]
981	then
982
983	if [[ $numprocs != 0 ]]
984	then
985	cat > $job_to_send << %%END%%
986	#!/bin/bash
987	#PBS -N $job_name
988	#PBS -l walltime=$timestring
989	#PBS -l select=1:ncpus=$numprocs
990	#PBS -l pmem=${memory}mb
991	#PBS -q $queue
992	#PBS -o $remote_dayfile
993	#PBS -j oe
994	#PBS -V
995
996	%%END%%
997
998	else
999	cat > $job_to_send << %%END%%
1000	#!/bin/bash
1001	#PBS -N $job_name
1002	#PBS -l walltime=$timestring
1003	#PBS -l ncpus=1
1004	#PBS -l pmem=${memory}mb
1005	#PBS -o $remote_dayfile
1006	#PBS -j oe
1007
1008	%%END%%
1009
1010	fi
1011
1012	elif [[ $remote_host = lcyon ]]
1013	then
1014
1015	if [[ $numprocs != 0 ]]
1016	then
1017	cat > $job_to_send << %%END%%
1018	#!/bin/ksh
1019	#PBS -N $job_name
1020	#PBS -l walltime=$timestring
1021	#PBS -l ncpus=$numprocs
1022	#PBS -l pmem=${memory}mb
1023	#PBS -o $remote_dayfile
1024	#PBS -j oe
1025
1026	%%END%%
1027
1028	else
1029	cat > $job_to_send << %%END%%
1030	#!/bin/ksh
1031	#PBS -N $job_name
1032	#PBS -l walltime=$timestring
1033	#PBS -l ncpus=1
1034	#PBS -l pmem=${memory}mb
1035	#PBS -o $remote_dayfile
1036	#PBS -j oe
1037
1038	%%END%%
1039
1040	fi
1041
1042	elif [[ $remote_host = lcxe6 ]]
1043	then
1044
1045	if [[ $numprocs != 0 ]]
1046	then
1047	cat > $job_to_send << %%END%%
1048	#!/bin/ksh
1049	#PBS -S /bin/ksh
1050	#PBS -N $job_name
1051	#PBS -A $project_account
1052	#PBS -j oe
1053	#PBS -l walltime=$timestring
1054	#PBS -l mppwidth=${numprocs}
1055	#PBS -l mppnppn=${processes_per_node}
1056	#PBS -m abe
1057	#PBS -o $remote_dayfile
1058	$email_directive
1059
1060	$init_cmds
1061	$module_calls
1062
1063	%%END%%
1064
1065	else
1066	cat > $job_to_send << %%END%%
1067	#!/bin/ksh
1068	#PBS -S /bin/ksh
1069	#PBS -N $job_name
1070	#PBS -A $project_account
1071	#PBS -j oe
1072	#PBS -l walltime=$timestring
1073	#PBS -l ncpus=1
1074	#PBS -l pmem=${memory}mb
1075	#PBS -m abe
1076	$email_directive
1077	#PBS -o $remote_dayfile
1078
1079	$init_cmds
1080	$module_calls
1081
1082	%%END%%
1083
1084	fi
1085
1086	elif [[ $remote_host = lckyoto ]]
1087	then
1088
1089	cat > $job_to_send << %%END%%
1090	#!/bin/ksh
1091	# @\$-o $remote_dayfile
1092	# @\$-eo -oi
1093	# @\$-lP 16
1094	# @\$-lp 1
1095	# @\$-lm 28gb -llm unlimited -ls unlimited
1096	# @\$-q $queue
1097	# @\$-Pvn abs_pack
1098	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1099	#. /thin/local/etc/setprofile/intel-11.0.sh
1100	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1101	. ~/.myprofile
1102	#. /home2/t/t51254/palm/current_version/myprofile
1103	#. /thin/apps/pgi/mpi.sh
1104	#
1105	env
1106	#
1107	set -x
1108
1109	%%END%%
1110
1111	elif [[ $remote_host = lcxt5m ]]
1112	then
1113
1114	if [[ $numprocs != 0 ]]
1115	then
1116	cat > $job_to_send << %%END%%
1117	#!/bin/ksh
1118	#PBS -S /bin/ksh
1119	#PBS -N $job_name
1120	#PBS -j oe
1121	#PBS -l walltime=$timestring
1122	#PBS -l mppwidth=${numprocs}
1123	#PBS -l mppnppn=${processes_per_node}
1124	#PBS -m abe
1125	#PBS -o $remote_dayfile
1126
1127	$init_cmds
1128	$module_calls
1129
1130	%%END%%
1131
1132	else
1133	cat > $job_to_send << %%END%%
1134	#!/bin/ksh
1135	#PBS -S /bin/ksh
1136	#PBS -N $job_name
1137	#PBS -j oe
1138	#PBS -l walltime=$timestring
1139	#PBS -l ncpus=1
1140	#PBS -l pmem=${memory}mb
1141	#PBS -m abe
1142	#PBS -o $remote_dayfile
1143
1144	$init_cmds
1145	$module_calls
1146
1147	%%END%%
1148
1149	fi
1150
1151	elif [[ $remote_host = lckyuh ]]
1152	then
1153	cat > $job_to_send << %%END%%
1154	#!/bin/bash
1155	#PJM -L "rscgrp=$queue"
1156	#PJM -L "node=$nodes"
1157	#PJM --mpi "proc=$numprocs"
1158	#PJM -L "elapse=$timestring"
1159	#PJM -o $remote_dayfile
1160	#PJM -j
1161	#PJM -X
1162	#PJM --no-stging
1163
1164	export LANG=en_US.UTF-8
1165	%%END%%
1166
1167	elif [[ $remote_host = lckyut ]]
1168	then
1169	cat > $job_to_send << %%END%%
1170	#!/bin/bash
1171	#PJM -L "rscgrp=$queue"
1172	#PJM -L "vnode=$numprocs"
1173	#PJM -L "vnode-core=1"
1174	#PJM -L "elapse=$timestring"
1175	#PJM --mpi proc=$numprocs
1176	#PJM -o $remote_dayfile
1177	#PJM -j
1178	#PJM -X
1179	#PJM --no-stging
1180
1181	export LANG=en_US.UTF-8
1182	%%END%%
1183
1184	elif [[ $remote_host = lcocean ]]
1185	then
1186	cat > $job_to_send << %%END%%
1187	#!/bin/bash
1188	#$ -cwd
1189	#$ -V
1190	#$ -N $job_name
1191	#$ -pe orte $numprocs
1192	#$ -o $remote_dayfile
1193	#$ -j y
1194	#$ -R y
1195	$init_cmds
1196	$module_calls
1197
1198	%%END%%
1199
1200	elif [[ $remote_host = nech ]]
1201	then
1202
1203	if (( nodes > 1 ))
1204	then
1205	cat > $job_to_send << %%END%%
1206	#!/bin/ksh
1207	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1208	#PBS -l ${qsubmem}=${Memory}gb
1209	#PBS -b $nodes
1210	#PBS -o $remote_dayfile
1211	#PBS -N palm
1212	#PBS -j o
1213	#PBS -T mpisx
1214
1215	%%END%%
1216
1217	elif [[ $numprocs != 0 ]]
1218	then
1219	cat > $job_to_send << %%END%%
1220	#!/bin/ksh
1221	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1222	#PBS -l ${qsubmem}=${Memory}gb
1223	#PBS -o $remote_dayfile
1224	#PBS -N palm
1225	#PBS -j o
1226
1227	%%END%%
1228
1229	else
1230	cat > $job_to_send << %%END%%
1231	#!/bin/ksh
1232	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1233	#PBS -o $remote_dayfile
1234	#PBS -j o
1235
1236	%%END%%
1237
1238	fi
1239
1240	elif [[ $remote_host = lctit ]]
1241	then
1242	cat > $job_to_send << %%END%%
1243	#!/bin/ksh
1244	$init_cmds
1245	$module_calls
1246
1247	%%END%%
1248
1249	# SET OPTIONS FOR SUBMIT-COMMAND
1250	if [[ $tasks_per_node != $processes_per_node ]]
1251	then
1252	submcom="$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 $queue "
1253	else
1254	submcom="$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 "
1255	fi
1256
1257	else
1258
1259	cat > $job_to_send << %%END%%
1260	# @\$-q ${queue}
1261	# @\$-l${qsubtime} $timestring
1262	# @\$-l${qsubmem} ${memory}mb
1263	# @\$-o $remote_dayfile
1264	# @\$-eo
1265
1266	%%END%%
1267
1268	fi
1269
1270
1271	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1272	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1273	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1274	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1275	then
1276	echo "set +vx" >> $job_to_send
1277	echo "trap '" >> $job_to_send
1278	echo "set +vx" >> $job_to_send
1279	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lcbullhh \|\| $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 \|\| $remote_host = lcocean ]]
1280	then
1281	if [[ $remote_host = ibmh ]]
1282	then
1283	return_queue=c1
1284	elif [[ $remote_host = ibmkisti ]]
1285	then
1286	return_queue=class.1-2
1287	elif [[ $remote_host = ibmku ]]
1288	then
1289	return_queue=sdbg2
1290	elif [[ $remote_host = ibms ]]
1291	then
1292	return_queue=p_normal
1293	elif [[ $remote_host = lcbullhh ]]
1294	then
1295	return_queue=shared
1296	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1297	then
1298	return_queue=dataq
1299	elif [[ $remote_host = lcxe6 ]]
1300	then
1301	return_queue=debug
1302	elif [[ $remote_host = lckiaps ]]
1303	then
1304	return_queue=express
1305	elif [[ $remote_host = lckyuh ]]
1306	then
1307	return_queue=cx-single
1308	elif [[ $remote_host = lckyut ]]
1309	then
1310	return_queue=cx-single
1311	else
1312	return_queue=unknown
1313	fi
1314
1315	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1316	then
1317
1318	if [[ $remote_host = ibmku ]]
1319	then
1320	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1321	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1322	else
1323	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1324	fi
1325	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1326	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1327	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1328	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1329	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1330	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1331	if [[ $host != "ibmh" ]]
1332	then
1333	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1334	fi
1335	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1336	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1337
1338	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1339	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1340
1341	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1342	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
1343	if [[ $remote_host = ibmku ]]
1344	then
1345	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1346	fi
1347	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1348
1349	elif [[ $remote_host = nech ]]
1350	then
1351	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1352	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1353	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1354	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1355	echo "#PBS -j o" >> $job_to_send
1356	echo " " >> $job_to_send
1357	echo "set -x" >> $job_to_send
1358	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1359	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1360	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1361	echo "%%END%%" >> $job_to_send
1362
1363	elif [[ $remote_host = lcbullhh ]]
1364	then
1365	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1366	echo "#!/bin/bash" >> $job_to_send
1367	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1368	echo "#SBATCH -t 00:20:00" >> $job_to_send
1369	echo "#SBATCH -N 1" >> $job_to_send
1370	echo "#SBATCH -n 1" >> $job_to_send
1371	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1372	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1373	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1374	echo "#SBATCH -A $project_account" >> $job_to_send
1375	echo "#SBATCH -p $return_queue" >> $job_to_send
1376	echo " " >> $job_to_send
1377	echo "set -x" >> $job_to_send
1378	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1379	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1380	echo "%%END%%" >> $job_to_send
1381
1382	elif [[ $remote_host = lckyuh ]]
1383	then
1384	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1385	echo "#!/bin/bash" >> $job_to_send
1386	echo "#PJM -L \"node=1\"" >> $job_to_send
1387	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1388	echo "#PJM --no-stging" >> $job_to_send
1389	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1390	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1391	echo "#PJM -j" >> $job_to_send
1392	echo " " >> $job_to_send
1393	echo "export LANG=en_US.UTF-8" >> $job_to_send
1394	echo "set -x" >> $job_to_send
1395	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1396	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1397	echo "%%END%%" >> $job_to_send
1398
1399	elif [[ $remote_host = lckyut ]]
1400	then
1401	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1402	echo "#!/bin/bash" >> $job_to_send
1403	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1404	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1405	echo "#PJM --no-stging" >> $job_to_send
1406	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1407	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1408	echo "#PJM -j" >> $job_to_send
1409	echo " " >> $job_to_send
1410	echo "export LANG=en_US.UTF-8" >> $job_to_send
1411	echo "set -x" >> $job_to_send
1412	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1413	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1414	echo "%%END%%" >> $job_to_send
1415
1416	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1417	then
1418	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1419	echo "#!/bin/bash" >> $job_to_send
1420	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1421	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1422	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1423	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1424	echo "#PBS -j oe" >> $job_to_send
1425	echo " " >> $job_to_send
1426	echo "set -x" >> $job_to_send
1427	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1428	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1429	echo "%%END%%" >> $job_to_send
1430
1431	elif [[ $remote_host = lcocean ]]
1432	then
1433	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1434	echo "#!/bin/bash" >> $job_to_send
1435	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1436	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1437	echo "SGEPREFIX -cwd" >> $job_to_send
1438	echo "SGEPREFIX -j y" >> $job_to_send
1439	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1440	echo " " >> $job_to_send
1441	echo "set -x" >> $job_to_send
1442	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1443	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1444	echo "" >> $job_to_send
1445	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1446	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1447	echo "rm -f scpjob.${identifier}" >> $job_to_send
1448	echo "%%END%%" >> $job_to_send
1449	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1450	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1451
1452	elif [[ $remote_host = lcflow ]]
1453	then
1454	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1455	echo "#!/bin/bash" >> $job_to_send
1456	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1457	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1458	echo "SGEPREFIX -cwd" >> $job_to_send
1459	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1460	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1461	echo "SGEPREFIX -j y" >> $job_to_send
1462	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1463	echo " " >> $job_to_send
1464	echo "set -x" >> $job_to_send
1465	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1466	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1467	echo "" >> $job_to_send
1468	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1469	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1470	echo "rm -f scpjob.${identifier}" >> $job_to_send
1471	echo "%%END%%" >> $job_to_send
1472	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1473	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1474	elif [[ $remote_host = lcxe6 ]]
1475	then
1476	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1477	echo "#!/bin/ksh" >> $job_to_send
1478	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1479	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1480	echo "#PBS -A $project_account" >> $job_to_send
1481	echo "#PBS -l mppwidth=1" >> $job_to_send
1482	echo "#PBS -l mppnppn=1" >> $job_to_send
1483	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1484	echo "#PBS -j oe" >> $job_to_send
1485	echo " " >> $job_to_send
1486	echo "set -x" >> $job_to_send
1487	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1488	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1489	echo "%%END%%" >> $job_to_send
1490	else
1491
1492	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1493	echo "# @\\\$-q $return_queue" >> $job_to_send
1494	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1495	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1496	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1497	then
1498	echo "# @\$-l mpp_p=0" >> $job_to_send
1499	fi
1500	echo '# @\$-lF 10mb' >> $job_to_send
1501	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1502	echo '# @\\\$-eo' >> $job_to_send
1503	echo " " >> $job_to_send
1504	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1505	then
1506	echo "set +vx" >> $job_to_send
1507	echo ". .profile" >> $job_to_send
1508	fi
1509	echo "set -x" >> $job_to_send
1510	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
1511	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1512	echo "%%END%%" >> $job_to_send
1513
1514	fi
1515
1516	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1517	then
1518	echo "llsubmit scpjob.$identifier" >> $job_to_send
1519	elif [[ $remote_host = lcbullhh ]]
1520	then
1521	echo "sbatch scpjob.$identifier" >> $job_to_send
1522	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1523	then
1524	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1525	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1526	then
1527	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1528	elif [[ $remote_host = t3es ]]
1529	then
1530	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1531	elif [[ $remote_host = lckiaps ]]
1532	then
1533	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1534	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1535	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1536	elif [[ $remote_host = lckyu* ]]
1537	then
1538	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1539	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1540	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1541	then
1542	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1543	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1544	else
1545	echo "$submcom scpjob.$identifier" >> $job_to_send
1546	fi
1547	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1548	then
1549	echo "rm scpjob.$identifier" >> $job_to_send
1550	fi
1551	if [[ $remote_host = nech ]]
1552	then
1553	echo "cd -" >> $job_to_send
1554	fi
1555	else
1556	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1557	# ??? funktioniert das ÃŒberhaupt noch ???
1558	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1559	fi
1560	echo "set -x" >> $job_to_send
1561	echo " ' exit" >> $job_to_send
1562	echo "set -x" >> $job_to_send
1563	fi
1564
1565
1566	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1567	cat $file_to_send >> $job_to_send
1568
1569	if [[ $remote_host = ibm ]]
1570	then
1571	echo " " >> $job_to_send
1572	echo "exit" >> $job_to_send
1573	fi
1574
1575	# REMOVE JOB-FILE
1576	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1577	then
1578	echo " " >> $job_to_send
1579	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1580	fi
1581
1582
1583	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1584	if [[ $no_submit = false ]]
1585	then
1586	if [[ $remote_host != $local_host ]]
1587	then
1588	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1589	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1590	then
1591	job_catalog_save=$job_catalog
1592	job_catalog=job_queue
1593	elif [[ $remote_host = nech ]]
1594	then
1595	job_catalog_save=$job_catalog
1596	job_catalog=/hpf/b/${remote_user}/job_queue
1597	fi
1598	if [[ $remote_host = nech ]]
1599	then
1600	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1601	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1602	else
1603	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1604	fi
1605	if [[ $? = 1 ]]
1606	then
1607	locat=scp; exit
1608	fi
1609	if [[ $remote_host = ibms ]]
1610	then
1611	job_catalog=$job_catalog_save
1612	fi
1613	[[ $verify = true ]] && printf "\n >>> finished\n"
1614	else
1615	eval job_catalog=$job_catalog
1616	cp $job_to_send ${job_catalog}/$job_on_remhost
1617	fi
1618
1619
1620
1621	# START NQS- / LOADLEVELER-JOB
1622	if [[ $remote_host != $local_host ]]
1623	then
1624	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1625
1626	if [[ $remote_host = ibmku ]]
1627	then
1628	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1629	elif [[ $remote_host = lcflow ]]
1630	then
1631	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1632	else
1633	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1634	fi
1635
1636	[[ $verify = true ]] && printf " >>> o.k.\n"
1637	else
1638	cd $job_catalog
1639	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1640	then
1641	eval $submcom $job_on_remhost
1642	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1643	then
1644	chmod u+x $job_on_remhost
1645	eval $submcom $job_on_remhost
1646	elif [[ $local_host = nech ]]
1647	then
1648	if [[ $queue = default ]]
1649	then
1650	eval $submcom $job_on_remhost
1651	else
1652	eval $submcom -q $queue $job_on_remhost
1653	fi
1654	elif [[ $local_host = lcbullhh ]]
1655	then
1656	if [[ $queue = default ]]
1657	then
1658	eval $submcom $job_on_remhost
1659	fi
1660	else
1661	qsub $job_on_remhost
1662	fi
1663
1664	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1665	# AT THE END OF THE JOB
1666	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1667	then
1668	rm $job_on_remhost
1669	fi
1670	cd - > /dev/null
1671	fi
1672	fi
1673
1674	# FINAL ACTIONS
1675	if [[ $no_submit = false ]]
1676	then
1677	rm -f $job_to_send
1678	fi
1679	[[ $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 |