Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2187

Last change on this file since 2187 was 2187, checked in by raasch, 7 years ago
adjustment of compute node names for lckyuh
Property svn:keywords set to `Id Rev`
File size: 58.5 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	# adjustment of compute node names for lckyuh
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 2187 2017-03-21 06:41:25Z raasch $
30	#
31	# 2184 2017-03-21 04:31:22Z raasch
32	# bugfix: localhost renamed local_host
33	#
34	# 2148 2017-02-09 16:56:42Z scharf
35	# added kuma and gharbi to the list of known hosts
36	#
37	# 2134 2017-02-02 07:33:46Z raasch
38	# option -E added to msub commands on HLRN-III machines to allow output of more
39	# job informations in the job protocol files
40	#
41	# 1944 2016-06-15 06:29:00Z raasch
42	# adjustments for using HLRN ssh-keys
43	#
44	# 1940 2016-06-14 05:15:20Z raasch
45	# adjustments for lckiaps
46	#
47	# 1866 2016-04-15 06:50:59Z raasch
48	# adjusted for lcocean
49	#
50	# 1841 2016-04-07 19:14:06Z raasch
51	# script now running under bash
52	#
53	# 1701 2015-11-02 07:43:04Z maronga
54	# Bugfix: added missing init_cmds for lccrayh/lccrayb
55	#
56	# 1621 2015-07-17 11:39:33Z heinze
57	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
58	#
59	# 1575 2015-03-27 09:56:27Z raasch
60	# mpp2-queues added to lccrayh
61	#
62	# 1547 2015-01-29 15:09:12Z witha
63	# adjustments for ForWind computing cluster (lcflow)
64	#
65	# 1545 2015-01-29 06:52:23Z heinze
66	# local host name for blizzard further specified
67	#
68	# 1480 2014-10-17 14:41:49Z raasch
69	# adjustments for 2nd stage of HLRNIII
70	#
71	# 1468 2014-09-24 14:06:57Z maronga
72	# Typo removed (addres->address)
73	# Adjustments for lcxe6
74	#
75	# 1452 2014-08-22 09:41:06Z heinze
76	# local hosts for blizzard added
77	#
78	# 1450 2014-08-21 07:31:51Z heinze
79	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
80	#
81	# 1442 2014-07-28 07:09:10Z raasch
82	# HLRN-III (lccrayb/lccrayh) queues adjusted
83	#
84	# 1378 2014-04-28 06:04:58Z raasch
85	# -et option added for lctit
86	#
87	# 1350 2014-04-04 13:01:30Z maronga
88	# location of qsub updated for lcxe6
89	#
90	# 1289 2014-03-04 07:12:34Z raasch
91	# German comments translated to English
92	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
93	#
94	# 1279 2014-01-28 12:10:14Z raasch
95	# node calculation modified due to changes in mrun (tasks_per_node must not be
96	# an integral divisor of numprocs any more)
97	#
98	# 1274 2014-01-09 13:14:54Z heinze
99	# adjustments for lccrayh
100	#
101	# 1266 2013-12-11 12:07:34Z heinze
102	# further adjustments for lccrayb (use msub instead of qsub)
103	#
104	# 1264 2013-12-09 12:46:09Z fricke
105	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
106	#
107	# 1262 2013-12-09 10:57:20Z fricke
108	# further adjustments for lccrayb
109	#
110	# 1260 2013-12-04 12:48:04Z raasch
111	# jaboticaba admitted
112	#
113	# 1255 2013-11-07 14:43:35Z raasch
114	# further adjustments for lccrayb
115	#
116	# 1224 2013-09-16 07:27:23Z raasch
117	# first adjustments for lccrayb
118	#
119	# 1202 2013-07-10 16:22:07Z witha
120	# adjustments for Forwind cluster (lcflow)
121	#
122	# 1199 2013-07-05 14:52:22Z raasch
123	# adjustments for CSC Helsinki (lccrayf)
124	#
125	# use of cluster/express queue enabled (ibmh)
126	# vinessa added (imuk)
127	#
128	# 1103 2013-02-20 02:15:53Z raasch
129	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
130	# instead typeset -L/R),
131	# further adjustments for lckyuh
132	#
133	# 2013-02-10 01:47:43Z raasch
134	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
135	# and for Forwind cluster (lcflow)
136	#
137	# 1094 2013-02-03 01:52:12Z raasch
138	# new option -P for explicit setting of ssh/scp port,
139	# decalpha parts (yonsei) removed
140	#
141	# 2013-02-02 07:06:13Z raasch
142	# adjustments for Kyushu-University computing center (lckyut - tatara)
143	# old changelog messages removed
144	#
145	# 1046 2012-11-09 14:38:45Z maronga
146	# code put under GPL (PALM 3.9)
147	#
148	# 08/07/94 - Siggi - first version finished
149	# 29/06/94 - Siggi - script development started
150	#--------------------------------------------------------------------------------#
151	# subjob - script for automatic generation and submission of batch-job files
152	# for various batch queuing systems
153	#--------------------------------------------------------------------------------#
154
155
156	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
157	delete_dayfile=false
158	email_notification=none
159	group_number=none
160	locat=normal
161	no_default_queue=none
162	no_submit=false
163	job_catalog="~/job_queue"
164	job_name=none
165	local_user=$LOGNAME
166	node_usage=shared
167	numprocs=0
168	punkte="..........................................................."
169	submcom=qsub
170	queue=default
171	remote_host=none
172	remote_user=""
173	verify=true
174
175	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
176	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
177
178
179
180	# ERROR HANDLING
181	# IN CASE OF EXIT:
182	trap 'if [[ $locat != normal ]]
183	then
184	case $locat in
185	(option) printf "\n --> available optios can be displayed"
186	printf " by typing:"
187	printf "\n \"subjob ?\" \n";;
188	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
189	(*) printf "\n +++ unknown error"
190	printf "\n please inform S. Raasch!\n"
191	esac
192	[[ -f $job_to_send ]] && rm $job_to_send
193	printf "\n\n+++ SUBJOB killed \n\n"
194	fi' exit
195
196
197	# IN CASE OF TERMINAL-BREAK:
198	trap '[[ -f $job_to_send ]] && rm $job_to_send
199	printf "\n\n+++ SUBJOB killed \n\n"
200	exit
201	' 2
202
203
204	# DETERMINE NAME OF LOCAL HOST
205	local_host=$(hostname)
206
207	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
208	# IS ADMITTED AT ALL)
209	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
210	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
211	case $local_host in
212	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
213	(atmos) local_address=172.20.25.35; local_host=lcide;;
214	(austru) local_address=130.75.105.128; local_host=lcmuk;;
215	(autan) local_address=130.75.105.57; local_host=lcmuk;;
216	(bora) local_address=130.75.105.103; local_host=lcmuk;;
217	(a0\|b0) local_address=133.5.4.33; local_host=lckyuh;;
218	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
219	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
220	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
221	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
222	(breva) local_address=130.75.105.98; local_host=lcmuk;;
223	(buran) local_address=130.75.105.58; local_host=lcmuk;;
224	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
225	(climate*) local_address=165.132.26.68; local_host=lcyon;;
226	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
227	(cs*) local_address=136.172.44.131; local_host=nech;;
228	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
229	(flow01) local_address=10.141.255.71; local_host=lcflow;;
230	(flow02) local_address=10.141.255.72; local_host=lcflow;;
231	(node*) local_address=165.132.26.61 local_host=lck;;
232	# (node*) local_address=210.219.61.8 local_host=lckordi;;
233	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
234	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
235	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
236	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
237	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
238	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
239	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
240	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
241	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
242	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
243	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
244	(sno) local_address=130.75.105.113; local_host=lcmuk;;
245	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
246	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
247	(login*) local_address=118.128.66.201; local_host=lckiaps;;
248	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
249	(meller) local_address=134.106.74.155; local_host=lcfor;;
250	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
251	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
252	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
253	(nobel*) local_address=150.183.5.101; local_host=ibms;;
254	(ocean) local_address="ocean"; local_host=lcocean;;
255	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
256	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
257	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
258	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
259	(pingui) local_address=134.106.74.118; local_host=lcfor;;
260	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
261	(rte*) local_address=133.5.185.60; local_host=lcrte;;
262	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
263	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
264	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
265	(solano) local_address=130.75.105.110; local_host=lcmuk;;
266	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
267	(tc*) local_address="ocean"; local_host=lcocean;;
268	(t2a*) local_address=10.1.6.165; local_host=lctit;;
269	(urban*) local_address=147.46.30.151 local_host=lcsb;;
270	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
271	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
272	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
273	(*) printf "\n +++ \"$local_host\" unknown";
274	printf "\n please contact the PALM group at IMUK";
275	locat=parameter; exit;;
276	esac
277
278
279
280	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
281	remote_host=$local_host
282
283
284
285
286	# READ THE SHELLSCRIPT-OPTIONS
287	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
288	do
289	case $option in
290	(c) job_catalog=$OPTARG;;
291	(d) delete_dayfile=true;;
292	(D) no_submit=true;;
293	(e) email_notification=$OPTARG;;
294	(g) group_number=$OPTARG;;
295	(h) remote_host=$OPTARG;;
296	(m) memory=$OPTARG;;
297	(n) job_name=$OPTARG;;
298	(N) node_usage=$OPTARG;;
299	(O) threads_per_task=$OPTARG;;
300	(P) scp_port=$OPTARG;;
301	(q) no_default_queue=$OPTARG;;
302	(t) cputime=$OPTARG;;
303	(T) tasks_per_node=$OPTARG;;
304	(u) remote_user=$OPTARG;;
305	(v) verify=false;;
306	(X) numprocs=$OPTARG;;
307	(\?) printf "\n +++ Option $OPTARG unknown \n";
308	locat=option; exit;;
309	esac
310	done
311
312
313	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
314	(( to_shift = $OPTIND - 1 ))
315	shift $to_shift; file_to_send=$1
316
317
318	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
319	if [ "$1" = "?" ]
320	then
321	(printf "\n *** subjob can be called as follows:\n"
322	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
323	printf "\n Description of available options:\n"
324	printf "\n Option Description Default-Value"
325	printf "\n -c job-input- and output-catalog ~/job_queue"
326	printf "\n -d no job-protocol will be created ---"
327	printf "\n -D only the job-file will be created ---"
328	printf "\n -h execution host, available hosts: $remote_host"
329	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
330	printf "\n lckiaps, lctit, nech"
331	printf "\n -m memory demand per process in MByte ---"
332	printf "\n -n jobname <jobdatei>"
333	printf "\n -O threads per task (for OpenMP usage) 1"
334	printf "\n -P ssh/scp port default port"
335	printf "\n -q job-queue to be used default"
336	printf "\n -t allowed cpu-time in seconds ---"
337	printf "\n -T tasks per node (on parallel hosts) ---"
338	printf "\n -u username on execution host from .netrc"
339	printf "\n -v no prompt for confirmation ---"
340	printf "\n -X # of processors (on parallel hosts) 1"
341	printf "\n "
342	printf "\n The only possible positional parameter is <jobfile>:"
343	printf "\n The complete NQS-job must be provided here."
344	printf "\n <jobfile>=? creates this outline\n\n") \| more
345	exit
346	fi
347
348
349
350	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
351	if [[ "$file_to_send" = "" ]]
352	then
353	printf "\n +++ job-file missing"
354	locat=parameter; exit
355	else
356	if [[ -f $file_to_send ]]
357	then
358	true
359	else
360	printf "\n +++ job-file: "
361	printf "\n $file_to_send"
362	printf "\n does not exist"
363	locat=parameter; exit
364	fi
365	fi
366
367
368
369	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
370	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
371	if [[ $job_name = none ]]
372	then
373	job_name=$file_to_send
374	fi
375	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
376	then
377	printf "\n +++ job-file name: "
378	printf "\n $job_name"
379	printf "\n must not contain \"/\"-characters"
380	locat=parameter; exit
381	fi
382
383
384
385
386	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
387	# OR IF NO HOST HAS BEEN GIVEN
388	if [[ $remote_host = none ]]
389	then
390	printf "\n +++ host missing"
391	locat=option; exit
392	else
393	case $remote_host in
394	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
395	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
396	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
397	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
398	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
399	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
400	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
401	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
402	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
403	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
404	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
405	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
406	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
407	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
408	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
409	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
410	(lcocean) remote_address="ocean"; submcom=qsub;;
411	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
412	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
413	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
414	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
415	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
416	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
417	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
418	locat=parameter; exit;;
419	esac
420	fi
421
422
423	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
424	if [[ $no_default_queue != none ]]
425	then
426	error=false
427	ndq=$no_default_queue
428	case $remote_host in
429	(ibm) case $ndq in
430	(p690_express\|p690_standard\|p690_long) error=false;;
431	(*) error=true;;
432	esac;;
433	(ibmh) case $ndq in
434	(cluster\|express) error=false;;
435	(*) error=true;;
436	esac;;
437	(ibmkisti) case $ndq in
438	(class.32plus\|class.1-2\|class.2-32) error=false;;
439	(*) error=true;;
440	esac;;
441	(ibmku) case $ndq in
442	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
443	(*) error=true;;
444	esac;;
445	(ibms) case $ndq in
446	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
447	(*) error=true;;
448	esac;;
449	(lcbullhh) case $ndq in
450	(compute\|compute2\|shared) error=false;;
451	(*) error=true;;
452	esac;;
453	(lccrayb) case $ndq in
454	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
455	(*) error=true;;
456	esac;;
457	(lccrayh) case $ndq in
458	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
459	(*) error=true;;
460	esac;;
461	(lccrayf) case $ndq in
462	(usup\|test*\|small\|large) error=false;;
463	(*) error=true;;
464	esac;;
465	(lcflow) case $ndq in
466	(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;;
467	(*) error=true;;
468	esac;;
469	(lckiaps) case $ndq in
470	(express\|normal\|normal20\|quickq) error=false;;
471	(*) error=true;;
472	esac;;
473	(lckyoto) case $ndq in
474	(eh\|ph) error=false;;
475	(*) error=true;;
476	esac;;
477	(lckyuh) case $ndq in
478	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
479	(*) error=true;;
480	esac;;
481	(lckyut) case $ndq in
482	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
483	(*) error=true;;
484	esac;;
485	(lctit) case $ndq in
486	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
487	(*) error=true;;
488	esac;;
489	(t3eb) case $ndq in
490	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
491	(*) error=true;;
492	esac;;
493	(t3eh) case $ndq in
494	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
495	(*) error=true;;
496	esac;;
497	(t3ej2\|t3ej5) case $ndq in
498	(low\|normal\|high) error=false;;
499	(*) error=true;;
500	esac;;
501	(t3es) case $ndq in
502	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
503	(*) error=true;;
504	esac;;
505	esac
506	if [[ $error = true ]]
507	then
508	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
509	locat=parameter; exit
510	else
511	queue=$no_default_queue
512	fi
513	fi
514
515
516
517	# CHECK THE CPU-TIME
518	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
519	done=false
520	while [[ $done = false ]]
521	do
522	if (( $cputime <= 0 ))
523	then
524	printf "\n +++ wrong cpu-time or cpu-time missing"
525	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
526	printf "\n >>> "
527	read cputime 1>/dev/null 2>&1
528	else
529	done=true
530	fi
531	done
532	if [[ $remote_host = nech ]]
533	then
534	if (( tasks_per_node != 0 ))
535	then
536	(( cputime = cputime * tasks_per_node ))
537	elif [[ $numprocs != 0 ]]
538	then
539	(( cputime = cputime * numprocs ))
540	fi
541	fi
542	(( stunden = cputime / 3600 ))
543	(( resttime = cputime - stunden * 3600 ))
544	(( minuten = resttime / 60 ))
545	(( sekunden = resttime - minuten * 60 ))
546	timestring=${stunden}:${minuten}:${sekunden}
547
548
549
550	# CHECK THE MEMORY DEMAND
551	done=false
552	while [[ $done = false ]]
553	do
554	if (( memory <= 0 ))
555	then
556	printf "\n +++ wrong memory demand or memory demand missing"
557	printf "\n >>> Please type memory in MByte per process as INTEGER:"
558	printf "\n >>> "
559	read memory 1>/dev/null 2>&1
560	else
561	done=true
562	fi
563	done
564
565	if [[ $remote_host = nech ]]
566	then
567	if (( tasks_per_node != 0 ))
568	then
569	(( Memory = memory * tasks_per_node / 1000 ))
570	elif [[ $numprocs != 0 ]]
571	then
572	(( Memory = memory * numprocs / 1000 ))
573	else
574	(( Memory = memory / 1000 ))
575	fi
576	elif [[ $remote_host = lctit ]]
577	then
578	(( Memory = memory * tasks_per_node / 1000 ))
579	fi
580
581
582	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
583	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
584	then
585	(( memory = memory * threads_per_task ))
586	fi
587
588
589	# CALCULATE NUMBER OF REQUIRED NODES
590	if (( tasks_per_node != 0 ))
591	then
592	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
593	fi
594
595
596	# CALCULATE NUMBER OF PROCESSES PER NODE
597	(( processes_per_node = tasks_per_node * threads_per_task ))
598
599
600	# CALCULATE NUMBER OF MPI TASKS
601	(( mpi_tasks = numprocs / threads_per_task ))
602
603
604	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
605	if [[ "$scp_port" != "" ]]
606	then
607	PORTOPT="-P $scp_port"
608	SSH_PORTOPT="-p $scp_port"
609	fi
610
611
612	# HEADER-OUTPUT
613	if [[ $verify = true ]]
614	then
615	printf "\n\n"
616	printf "#--------------------------------------------------------------# \n"
617	spalte1=SUBJOB;spalte2=$(date)
618	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
619	printf "\| \| \n"
620	printf "\| values of parameters/options: \| \n"
621	spalte1=$(echo local_host$punkte \| cut -c-20)
622	spalte2=$punkte$local_host
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	spalte1=$(echo remote_host$punkte \| cut -c-20)
625	spalte2=$punkte$remote_host
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	spalte1=$(echo queue$punkte \| cut -c-20)
628	spalte2=$punkte$queue
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	spalte1=$(echo memory$punkte \| cut -c-20)
631	spalte2="$punkte$memory mb"
632	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
633	spalte1=$(echo cputime$punkte \| cut -c-20)
634	spalte2="$punkte$cputime sec"
635	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
636	spalte1=$(echo job_name$punkte \| cut -c-20)
637	spalte2="$punkte$job_name"
638	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
639	printf "#--------------------------------------------------------------# \n\n"
640
641
642	# QUERY CHECK
643	antwort="dummy"
644	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
645	do
646	read antwort?" >>> continue (y/n) ? "
647	done
648	if [[ $antwort = n \|\| $antwort = N ]]
649	then
650	locat=verify; exit
651	fi
652	printf "\n"
653	fi
654
655	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
656	identifier=$RANDOM
657	job_on_remhost=${job_name}_${identifier}_$local_host
658	job_to_send=job_to_send_$identifier
659	if [[ $delete_dayfile = false ]]
660	then
661	remote_dayfile=${local_host}_${job_name}_result_$identifier
662	local_dayfile=${remote_host}_${job_name}
663	else
664	remote_dayfile=/dev/null
665	fi
666
667
668	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
669	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
670	then
671
672	# GENERAL LOADLEVELER SETTINGS
673	execute_in_shell="#!/bin/ksh"
674	use_shell="# @ shell = /bin/ksh"
675	consumable_memory="ConsumableMemory($memory mb)"
676	class="# @ class = $queue"
677	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
678	network_to_use="# @ network.mpi = sn_all,shared,us"
679	data_limit="# @ data_limit = 1.76gb"
680	image_size="# @ image_size = 50"
681	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
682
683	if [[ $email_notification = none ]]
684	then
685	notify_user=""
686	else
687	notify_user="# @ notify_user = $email_notification"
688	if [[ $delete_dayfile = true ]]
689	then
690	notification='# @ notification = never'
691	fi
692	fi
693
694	if [[ $remote_host = ibmh ]]
695	then
696	data_limit=""
697	network_to_use=""
698	class="# @ class = $queue"
699	environment=""
700	rset="# @ rset = RSET_MCM_AFFINITY"
701	task_affinity="# @ task_affinity = core(1)"
702	elif [[ $remote_host = ibmkisti ]]
703	then
704	network_to_use="# @ network.MPI = sn_all,shared,US"
705	wall_clock_limit="# @ wall_clock_limit = $timestring"
706	if [[ $threads_per_task = 1 ]]
707	then
708	rset="# @ rset = RSET_MCM_AFFINITY"
709	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
710	fi
711	environment=""
712	use_shell=""
713	data_limit=""
714	image_size=""
715	elif [[ $remote_host = ibmku ]]
716	then
717	execute_in_shell="#!/usr/bin/ksh"
718	use_shell="# @ shell = /usr/bin/ksh"
719	consumable_memory=""
720	environment=""
721	network_to_use="# @ network.mpi = sn_all,shared,us"
722	data_limit=""
723	image_size=""
724	elif [[ $remote_host = ibms ]]
725	then
726	network_to_use="# @ network.mpi = csss,shared,us"
727	fi
728
729	cat > $job_to_send << %%END%%
730	$execute_in_shell
731	$use_shell
732
733	# @ job_type = parallel
734	# @ job_name = $job_name
735	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
736	# @ output = $remote_dayfile
737	# @ error = $remote_dayfile
738	$wall_clock_limit
739	$image_size
740	$class
741	$environment
742	$network_to_use
743	$data_limit
744	$rset
745	$mcm_affinity_options
746	$task_affinity
747	$notification
748	$notify_user
749
750	%%END%%
751
752	if (( nodes > 0 ))
753	then
754
755	if [[ $remote_host != ibmkisti ]]
756	then
757
758	cat >> $job_to_send << %%END%%
759	# @ node = $nodes
760	# @ tasks_per_node = $processes_per_node
761	# @ node_usage = $node_usage
762	# @ queue
763
764	%%END%%
765
766	else
767
768	cat >> $job_to_send << %%END%%
769	# @ total_tasks = $mpi_tasks
770	# @ blocking = unlimited
771	# @ queue
772
773	%%END%%
774
775	fi
776
777	else
778
779	cat >> $job_to_send << %%END%%
780	# @ blocking = unlimited
781	# @ total_tasks = $numprocs
782	# @ node_usage = $node_usage
783	# @ queue
784
785	%%END%%
786
787	fi
788
789	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
790	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
791	then
792	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
793	fi
794
795	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
796	then
797
798	cat > $job_to_send << %%END%%
799	#!/bin/ksh
800
801	# @ job_type = serial
802	# @ node_usage = $node_usage
803	# @ job_name = palm
804	# @ wall_clock_limit = ${timestring},$timestring
805	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
806	# @ output = $remote_dayfile
807	# @ error = $remote_dayfile
808	$class
809	$notification
810
811	# @ queue
812
813	%%END%%
814
815	elif [[ $remote_host = lcbullhh ]]
816	then
817	if [[ $numprocs != 0 ]]
818	then
819	cat > $job_to_send << %%END%%
820	#!/bin/bash -l
821	#SBATCH -J $job_name
822	#SBATCH -t $timestring
823	#SBATCH -N $nodes
824	#SBATCH --ntasks-per-node=$processes_per_node
825	#SBATCH -p $queue
826	#SBATCH -o $remote_dayfile
827	#SBATCH -e $remote_dayfile
828	#SBATCH -A $project_account
829
830	$init_cmds
831	$module_calls
832
833	%%END%%
834
835	else
836	cat > $job_to_send << %%END%%
837	#!/bin/bash -l
838	#SBATCH -J $job_name
839	#SBATCH -t $timestring
840	#SBATCH -l ncpus=1
841	#SBATCH -l pmem=${memory}mb
842	#SBATCH -m abe
843	#SBATCH -o $remote_dayfile
844	#SBATCH -e $remote_dayfile
845	#SBATCH -A $project_account
846
847	$init_cmds
848	$module_calls
849
850	%%END%%
851
852	fi
853
854	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
855	then
856
857	if [[ "$feature" != "" ]]
858	then
859	featuredir="#PBS -l feature=$feature"
860	fi
861
862	if [[ $numprocs != 0 ]]
863	then
864	cat > $job_to_send << %%END%%
865	#!/bin/bash
866	#PBS -N $job_name
867	#PBS -l walltime=$timestring
868	#PBS -l nodes=$nodes:ppn=$processes_per_node
869	#PBS -o $remote_dayfile
870	#PBS -j oe
871	#PBS -q $queue
872	$featuredir
873
874	$init_cmds
875	$module_calls
876
877	%%END%%
878
879	else
880
881	continue
882
883	fi
884
885	elif [[ $remote_host = lccrayf ]]
886	then
887
888	if [[ $numprocs != 0 ]]
889	then
890	cat > $job_to_send << %%END%%
891	#!/bin/bash -l
892	#SBATCH -J $job_name
893	#SBATCH -t $timestring
894	#SBATCH -N $nodes
895	#SBATCH --ntasks-per-node=$processes_per_node
896	#SBATCH -p $queue
897	#SBATCH -o $remote_dayfile
898	#SBATCH -e $remote_dayfile
899
900	$init_cmds
901	$module_calls
902
903	%%END%%
904
905	else
906	cat > $job_to_send << %%END%%
907	#!/bin/bash -l
908	#SBATCH -J $job_name
909	#SBATCH -t $timestring
910	#SBATCH -l ncpus=1
911	#SBATCH -l pmem=${memory}mb
912	#SBATCH -m abe
913	#SBATCH -o $remote_dayfile
914	#SBATCH -e $remote_dayfile
915
916	$init_cmds
917	$module_calls
918
919	%%END%%
920
921	fi
922
923	elif [[ $remote_host = lcflow ]]
924	then
925	if [[ $numprocs != 0 ]]
926	then
927	pe_set="#$ -pe impi $numprocs"
928	else
929	pe_set="#$ -pe impi 1"
930	fi
931	if [[ $queue = default ]]
932	then
933	queue_set=""
934	else
935	queue_set="#$ -q $queue"
936	fi
937	[[ "$disc_space" = "" ]] && disc_space=50
938
939	cat > $job_to_send << %%END%%
940	#!/bin/bash
941	#$ -S /bin/bash
942	#$ -N $job_name
943	#$ -cwd
944	#$ -l h_rt=$timestring
945	#$ -l h_vmem=${memory}M
946	#$ -o $remote_dayfile
947	#$ -j y
948	$pe_set
949	#$ -R y
950	#$ -l h_fsize=${disc_space}G
951	$queue_set
952
953	%%END%%
954
955	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
956	then
957
958	if [[ $numprocs != 0 ]]
959	then
960	cat > $job_to_send << %%END%%
961	#!/bin/ksh
962	#PBS -N $job_name
963	#PBS -l walltime=$timestring
964	#PBS -l ncpus=$numprocs
965	#PBS -l pmem=${memory}mb
966	#PBS -o $remote_dayfile
967	#PBS -l nodes=$nodes:ppn=${processes_per_node}
968	#PBS -j oe
969
970	mpd &
971
972	%%END%%
973
974	else
975	cat > $job_to_send << %%END%%
976	#!/bin/ksh
977	#PBS -N $job_name
978	#PBS -l walltime=$timestring
979	#PBS -l ncpus=1
980	#PBS -l pmem=${memory}mb
981	#PBS -o $remote_dayfile
982	#PBS -j oe
983
984	%%END%%
985
986	fi
987
988	elif [[ $remote_host = lckiaps ]]
989	then
990
991	if [[ $numprocs != 0 ]]
992	then
993	cat > $job_to_send << %%END%%
994	#!/bin/bash
995	#PBS -N $job_name
996	#PBS -l walltime=$timestring
997	#PBS -l select=1:ncpus=$numprocs
998	#PBS -l pmem=${memory}mb
999	#PBS -q $queue
1000	#PBS -o $remote_dayfile
1001	#PBS -j oe
1002	#PBS -V
1003
1004	%%END%%
1005
1006	else
1007	cat > $job_to_send << %%END%%
1008	#!/bin/bash
1009	#PBS -N $job_name
1010	#PBS -l walltime=$timestring
1011	#PBS -l ncpus=1
1012	#PBS -l pmem=${memory}mb
1013	#PBS -o $remote_dayfile
1014	#PBS -j oe
1015
1016	%%END%%
1017
1018	fi
1019
1020	elif [[ $remote_host = lcyon ]]
1021	then
1022
1023	if [[ $numprocs != 0 ]]
1024	then
1025	cat > $job_to_send << %%END%%
1026	#!/bin/ksh
1027	#PBS -N $job_name
1028	#PBS -l walltime=$timestring
1029	#PBS -l ncpus=$numprocs
1030	#PBS -l pmem=${memory}mb
1031	#PBS -o $remote_dayfile
1032	#PBS -j oe
1033
1034	%%END%%
1035
1036	else
1037	cat > $job_to_send << %%END%%
1038	#!/bin/ksh
1039	#PBS -N $job_name
1040	#PBS -l walltime=$timestring
1041	#PBS -l ncpus=1
1042	#PBS -l pmem=${memory}mb
1043	#PBS -o $remote_dayfile
1044	#PBS -j oe
1045
1046	%%END%%
1047
1048	fi
1049
1050	elif [[ $remote_host = lcxe6 ]]
1051	then
1052
1053	if [[ $numprocs != 0 ]]
1054	then
1055	cat > $job_to_send << %%END%%
1056	#!/bin/ksh
1057	#PBS -S /bin/ksh
1058	#PBS -N $job_name
1059	#PBS -A $project_account
1060	#PBS -j oe
1061	#PBS -l walltime=$timestring
1062	#PBS -l mppwidth=${numprocs}
1063	#PBS -l mppnppn=${processes_per_node}
1064	#PBS -m abe
1065	#PBS -o $remote_dayfile
1066	$email_directive
1067
1068	$init_cmds
1069	$module_calls
1070
1071	%%END%%
1072
1073	else
1074	cat > $job_to_send << %%END%%
1075	#!/bin/ksh
1076	#PBS -S /bin/ksh
1077	#PBS -N $job_name
1078	#PBS -A $project_account
1079	#PBS -j oe
1080	#PBS -l walltime=$timestring
1081	#PBS -l ncpus=1
1082	#PBS -l pmem=${memory}mb
1083	#PBS -m abe
1084	$email_directive
1085	#PBS -o $remote_dayfile
1086
1087	$init_cmds
1088	$module_calls
1089
1090	%%END%%
1091
1092	fi
1093
1094	elif [[ $remote_host = lckyoto ]]
1095	then
1096
1097	cat > $job_to_send << %%END%%
1098	#!/bin/ksh
1099	# @\$-o $remote_dayfile
1100	# @\$-eo -oi
1101	# @\$-lP 16
1102	# @\$-lp 1
1103	# @\$-lm 28gb -llm unlimited -ls unlimited
1104	# @\$-q $queue
1105	# @\$-Pvn abs_pack
1106	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1107	#. /thin/local/etc/setprofile/intel-11.0.sh
1108	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1109	. ~/.myprofile
1110	#. /home2/t/t51254/palm/current_version/myprofile
1111	#. /thin/apps/pgi/mpi.sh
1112	#
1113	env
1114	#
1115	set -x
1116
1117	%%END%%
1118
1119	elif [[ $remote_host = lcxt5m ]]
1120	then
1121
1122	if [[ $numprocs != 0 ]]
1123	then
1124	cat > $job_to_send << %%END%%
1125	#!/bin/ksh
1126	#PBS -S /bin/ksh
1127	#PBS -N $job_name
1128	#PBS -j oe
1129	#PBS -l walltime=$timestring
1130	#PBS -l mppwidth=${numprocs}
1131	#PBS -l mppnppn=${processes_per_node}
1132	#PBS -m abe
1133	#PBS -o $remote_dayfile
1134
1135	$init_cmds
1136	$module_calls
1137
1138	%%END%%
1139
1140	else
1141	cat > $job_to_send << %%END%%
1142	#!/bin/ksh
1143	#PBS -S /bin/ksh
1144	#PBS -N $job_name
1145	#PBS -j oe
1146	#PBS -l walltime=$timestring
1147	#PBS -l ncpus=1
1148	#PBS -l pmem=${memory}mb
1149	#PBS -m abe
1150	#PBS -o $remote_dayfile
1151
1152	$init_cmds
1153	$module_calls
1154
1155	%%END%%
1156
1157	fi
1158
1159	elif [[ $remote_host = lckyuh ]]
1160	then
1161	cat > $job_to_send << %%END%%
1162	#!/bin/bash
1163	#PJM -L "rscgrp=$queue"
1164	#PJM -L "node=$nodes"
1165	#PJM --mpi "proc=$numprocs"
1166	#PJM -L "elapse=$timestring"
1167	#PJM -o $remote_dayfile
1168	#PJM -j
1169	#PJM -X
1170	#PJM --no-stging
1171
1172	export LANG=en_US.UTF-8
1173	%%END%%
1174
1175	elif [[ $remote_host = lckyut ]]
1176	then
1177	cat > $job_to_send << %%END%%
1178	#!/bin/bash
1179	#PJM -L "rscgrp=$queue"
1180	#PJM -L "vnode=$numprocs"
1181	#PJM -L "vnode-core=1"
1182	#PJM -L "elapse=$timestring"
1183	#PJM --mpi proc=$numprocs
1184	#PJM -o $remote_dayfile
1185	#PJM -j
1186	#PJM -X
1187	#PJM --no-stging
1188
1189	export LANG=en_US.UTF-8
1190	%%END%%
1191
1192	elif [[ $remote_host = lcocean ]]
1193	then
1194	cat > $job_to_send << %%END%%
1195	#!/bin/bash
1196	#$ -cwd
1197	#$ -V
1198	#$ -N $job_name
1199	#$ -pe orte $numprocs
1200	#$ -o $remote_dayfile
1201	#$ -j y
1202	#$ -R y
1203	$init_cmds
1204	$module_calls
1205
1206	%%END%%
1207
1208	elif [[ $remote_host = nech ]]
1209	then
1210
1211	if (( nodes > 1 ))
1212	then
1213	cat > $job_to_send << %%END%%
1214	#!/bin/ksh
1215	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1216	#PBS -l ${qsubmem}=${Memory}gb
1217	#PBS -b $nodes
1218	#PBS -o $remote_dayfile
1219	#PBS -N palm
1220	#PBS -j o
1221	#PBS -T mpisx
1222
1223	%%END%%
1224
1225	elif [[ $numprocs != 0 ]]
1226	then
1227	cat > $job_to_send << %%END%%
1228	#!/bin/ksh
1229	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1230	#PBS -l ${qsubmem}=${Memory}gb
1231	#PBS -o $remote_dayfile
1232	#PBS -N palm
1233	#PBS -j o
1234
1235	%%END%%
1236
1237	else
1238	cat > $job_to_send << %%END%%
1239	#!/bin/ksh
1240	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1241	#PBS -o $remote_dayfile
1242	#PBS -j o
1243
1244	%%END%%
1245
1246	fi
1247
1248	elif [[ $remote_host = lctit ]]
1249	then
1250	cat > $job_to_send << %%END%%
1251	#!/bin/ksh
1252	$init_cmds
1253	$module_calls
1254
1255	%%END%%
1256
1257	# SET OPTIONS FOR SUBMIT-COMMAND
1258	if [[ $tasks_per_node != $processes_per_node ]]
1259	then
1260	submcom="$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 $queue "
1261	else
1262	submcom="$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 "
1263	fi
1264
1265	else
1266
1267	cat > $job_to_send << %%END%%
1268	# @\$-q ${queue}
1269	# @\$-l${qsubtime} $timestring
1270	# @\$-l${qsubmem} ${memory}mb
1271	# @\$-o $remote_dayfile
1272	# @\$-eo
1273
1274	%%END%%
1275
1276	fi
1277
1278
1279	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1280	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1281	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1282	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1283	then
1284	echo "set +vx" >> $job_to_send
1285	echo "trap '" >> $job_to_send
1286	echo "set +vx" >> $job_to_send
1287	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 ]]
1288	then
1289	if [[ $remote_host = ibmh ]]
1290	then
1291	return_queue=c1
1292	elif [[ $remote_host = ibmkisti ]]
1293	then
1294	return_queue=class.1-2
1295	elif [[ $remote_host = ibmku ]]
1296	then
1297	return_queue=sdbg2
1298	elif [[ $remote_host = ibms ]]
1299	then
1300	return_queue=p_normal
1301	elif [[ $remote_host = lcbullhh ]]
1302	then
1303	return_queue=shared
1304	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1305	then
1306	return_queue=dataq
1307	elif [[ $remote_host = lcxe6 ]]
1308	then
1309	return_queue=debug
1310	elif [[ $remote_host = lckiaps ]]
1311	then
1312	return_queue=express
1313	elif [[ $remote_host = lckyuh ]]
1314	then
1315	return_queue=cx-single
1316	elif [[ $remote_host = lckyut ]]
1317	then
1318	return_queue=cx-single
1319	else
1320	return_queue=unknown
1321	fi
1322
1323	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1324	then
1325
1326	if [[ $remote_host = ibmku ]]
1327	then
1328	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1329	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1330	else
1331	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1332	fi
1333	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1334	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1335	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1336	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1337	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1338	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1339	if [[ $host != "ibmh" ]]
1340	then
1341	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1342	fi
1343	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1344	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1345
1346	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1347	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1348
1349	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1350	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
1351	if [[ $remote_host = ibmku ]]
1352	then
1353	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1354	fi
1355	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1356
1357	elif [[ $remote_host = nech ]]
1358	then
1359	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1360	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1361	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1362	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1363	echo "#PBS -j o" >> $job_to_send
1364	echo " " >> $job_to_send
1365	echo "set -x" >> $job_to_send
1366	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1367	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1368	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1369	echo "%%END%%" >> $job_to_send
1370
1371	elif [[ $remote_host = lcbullhh ]]
1372	then
1373	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1374	echo "#!/bin/bash" >> $job_to_send
1375	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1376	echo "#SBATCH -t 00:20:00" >> $job_to_send
1377	echo "#SBATCH -N 1" >> $job_to_send
1378	echo "#SBATCH -n 1" >> $job_to_send
1379	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1380	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1381	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1382	echo "#SBATCH -A $project_account" >> $job_to_send
1383	echo "#SBATCH -p $return_queue" >> $job_to_send
1384	echo " " >> $job_to_send
1385	echo "set -x" >> $job_to_send
1386	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1387	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1388	echo "%%END%%" >> $job_to_send
1389
1390	elif [[ $remote_host = lckyuh ]]
1391	then
1392	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1393	echo "#!/bin/bash" >> $job_to_send
1394	echo "#PJM -L \"node=1\"" >> $job_to_send
1395	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1396	echo "#PJM --no-stging" >> $job_to_send
1397	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1398	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1399	echo "#PJM -j" >> $job_to_send
1400	echo " " >> $job_to_send
1401	echo "export LANG=en_US.UTF-8" >> $job_to_send
1402	echo "set -x" >> $job_to_send
1403	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1404	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1405	echo "%%END%%" >> $job_to_send
1406
1407	elif [[ $remote_host = lckyut ]]
1408	then
1409	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1410	echo "#!/bin/bash" >> $job_to_send
1411	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1412	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1413	echo "#PJM --no-stging" >> $job_to_send
1414	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1415	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1416	echo "#PJM -j" >> $job_to_send
1417	echo " " >> $job_to_send
1418	echo "export LANG=en_US.UTF-8" >> $job_to_send
1419	echo "set -x" >> $job_to_send
1420	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1421	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1422	echo "%%END%%" >> $job_to_send
1423
1424	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1425	then
1426	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1427	echo "#!/bin/bash" >> $job_to_send
1428	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1429	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1430	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1431	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1432	echo "#PBS -j oe" >> $job_to_send
1433	echo " " >> $job_to_send
1434	echo "set -x" >> $job_to_send
1435	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1436	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1437	echo "%%END%%" >> $job_to_send
1438
1439	elif [[ $remote_host = lcocean ]]
1440	then
1441	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1442	echo "#!/bin/bash" >> $job_to_send
1443	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1444	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1445	echo "SGEPREFIX -cwd" >> $job_to_send
1446	echo "SGEPREFIX -j y" >> $job_to_send
1447	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1448	echo " " >> $job_to_send
1449	echo "set -x" >> $job_to_send
1450	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1451	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1452	echo "" >> $job_to_send
1453	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1454	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1455	echo "rm -f scpjob.${identifier}" >> $job_to_send
1456	echo "%%END%%" >> $job_to_send
1457	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1458	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1459
1460	elif [[ $remote_host = lcflow ]]
1461	then
1462	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1463	echo "#!/bin/bash" >> $job_to_send
1464	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1465	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1466	echo "SGEPREFIX -cwd" >> $job_to_send
1467	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1468	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1469	echo "SGEPREFIX -j y" >> $job_to_send
1470	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1471	echo " " >> $job_to_send
1472	echo "set -x" >> $job_to_send
1473	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1474	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1475	echo "" >> $job_to_send
1476	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1477	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1478	echo "rm -f scpjob.${identifier}" >> $job_to_send
1479	echo "%%END%%" >> $job_to_send
1480	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1481	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1482	elif [[ $remote_host = lcxe6 ]]
1483	then
1484	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1485	echo "#!/bin/ksh" >> $job_to_send
1486	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1487	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1488	echo "#PBS -A $project_account" >> $job_to_send
1489	echo "#PBS -l mppwidth=1" >> $job_to_send
1490	echo "#PBS -l mppnppn=1" >> $job_to_send
1491	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1492	echo "#PBS -j oe" >> $job_to_send
1493	echo " " >> $job_to_send
1494	echo "set -x" >> $job_to_send
1495	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1496	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1497	echo "%%END%%" >> $job_to_send
1498	else
1499
1500	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1501	echo "# @\\\$-q $return_queue" >> $job_to_send
1502	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1503	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1504	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1505	then
1506	echo "# @\$-l mpp_p=0" >> $job_to_send
1507	fi
1508	echo '# @\$-lF 10mb' >> $job_to_send
1509	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1510	echo '# @\\\$-eo' >> $job_to_send
1511	echo " " >> $job_to_send
1512	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1513	then
1514	echo "set +vx" >> $job_to_send
1515	echo ". .profile" >> $job_to_send
1516	fi
1517	echo "set -x" >> $job_to_send
1518	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
1519	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1520	echo "%%END%%" >> $job_to_send
1521
1522	fi
1523
1524	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1525	then
1526	echo "llsubmit scpjob.$identifier" >> $job_to_send
1527	elif [[ $remote_host = lcbullhh ]]
1528	then
1529	echo "sbatch scpjob.$identifier" >> $job_to_send
1530	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1531	then
1532	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1533	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1534	then
1535	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1536	elif [[ $remote_host = t3es ]]
1537	then
1538	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1539	elif [[ $remote_host = lckiaps ]]
1540	then
1541	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1542	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1543	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1544	elif [[ $remote_host = lckyu* ]]
1545	then
1546	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1547	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1548	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1549	then
1550	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1551	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1552	else
1553	echo "$submcom scpjob.$identifier" >> $job_to_send
1554	fi
1555	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1556	then
1557	echo "rm scpjob.$identifier" >> $job_to_send
1558	fi
1559	if [[ $remote_host = nech ]]
1560	then
1561	echo "cd -" >> $job_to_send
1562	fi
1563	else
1564	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1565	# ??? funktioniert das ÃŒberhaupt noch ???
1566	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1567	fi
1568	echo "set -x" >> $job_to_send
1569	echo " ' exit" >> $job_to_send
1570	echo "set -x" >> $job_to_send
1571	fi
1572
1573
1574	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1575	cat $file_to_send >> $job_to_send
1576
1577	if [[ $remote_host = ibm ]]
1578	then
1579	echo " " >> $job_to_send
1580	echo "exit" >> $job_to_send
1581	fi
1582
1583	# REMOVE JOB-FILE
1584	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1585	then
1586	echo " " >> $job_to_send
1587	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1588	fi
1589
1590
1591	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1592	if [[ $no_submit = false ]]
1593	then
1594	if [[ $remote_host != $local_host ]]
1595	then
1596	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1597	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1598	then
1599	job_catalog_save=$job_catalog
1600	job_catalog=job_queue
1601	elif [[ $remote_host = nech ]]
1602	then
1603	job_catalog_save=$job_catalog
1604	job_catalog=/hpf/b/${remote_user}/job_queue
1605	fi
1606	if [[ $remote_host = nech ]]
1607	then
1608	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1609	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1610	else
1611	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1612	fi
1613	if [[ $? = 1 ]]
1614	then
1615	locat=scp; exit
1616	fi
1617	if [[ $remote_host = ibms ]]
1618	then
1619	job_catalog=$job_catalog_save
1620	fi
1621	[[ $verify = true ]] && printf "\n >>> finished\n"
1622	else
1623	eval job_catalog=$job_catalog
1624	cp $job_to_send ${job_catalog}/$job_on_remhost
1625	fi
1626
1627
1628
1629	# START NQS- / LOADLEVELER-JOB
1630	if [[ $remote_host != $local_host ]]
1631	then
1632	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1633
1634	if [[ $remote_host = ibmku ]]
1635	then
1636	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1637	elif [[ $remote_host = lcflow ]]
1638	then
1639	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1640	else
1641	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1642	fi
1643
1644	[[ $verify = true ]] && printf " >>> o.k.\n"
1645	else
1646	cd $job_catalog
1647	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1648	then
1649	eval $submcom $job_on_remhost
1650	elif [[ $local_host = lctit \|\| $local_host = lcxe6 \|\| $local_host = lck \|\| $local_host = lckordi \|\| $local_host = lcyon \|\| $local_host = lcsb \|\| $local_host = lckyu* ]]
1651	then
1652	chmod u+x $job_on_remhost
1653	eval $submcom $job_on_remhost
1654	elif [[ $local_host = nech ]]
1655	then
1656	if [[ $queue = default ]]
1657	then
1658	eval $submcom $job_on_remhost
1659	else
1660	eval $submcom -q $queue $job_on_remhost
1661	fi
1662	elif [[ $local_host = lcbullhh ]]
1663	then
1664	if [[ $queue = default ]]
1665	then
1666	eval $submcom $job_on_remhost
1667	fi
1668	else
1669	qsub $job_on_remhost
1670	fi
1671
1672	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1673	# AT THE END OF THE JOB
1674	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1675	then
1676	rm $job_on_remhost
1677	fi
1678	cd - > /dev/null
1679	fi
1680	fi
1681
1682	# FINAL ACTIONS
1683	if [[ $no_submit = false ]]
1684	then
1685	rm -f $job_to_send
1686	fi
1687	[[ $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 |