Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2249

Last change on this file since 2249 was 2188, checked in by raasch, 8 years ago
last commit documented
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	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 2188 2017-03-21 06:42:42Z sward $
30	#
31	# 2187 2017-03-21 06:41:25Z raasch
32	# adjustment of compute node names for lckyuh
33	#
34	# 2184 2017-03-21 04:31:22Z raasch
35	# bugfix: localhost renamed local_host
36	#
37	# 2148 2017-02-09 16:56:42Z scharf
38	# added kuma and gharbi to the list of known hosts
39	#
40	# 2134 2017-02-02 07:33:46Z raasch
41	# option -E added to msub commands on HLRN-III machines to allow output of more
42	# job informations in the job protocol files
43	#
44	# 1944 2016-06-15 06:29:00Z raasch
45	# adjustments for using HLRN ssh-keys
46	#
47	# 1940 2016-06-14 05:15:20Z raasch
48	# adjustments for lckiaps
49	#
50	# 1866 2016-04-15 06:50:59Z raasch
51	# adjusted for lcocean
52	#
53	# 1841 2016-04-07 19:14:06Z raasch
54	# script now running under bash
55	#
56	# 1701 2015-11-02 07:43:04Z maronga
57	# Bugfix: added missing init_cmds for lccrayh/lccrayb
58	#
59	# 1621 2015-07-17 11:39:33Z heinze
60	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
61	#
62	# 1575 2015-03-27 09:56:27Z raasch
63	# mpp2-queues added to lccrayh
64	#
65	# 1547 2015-01-29 15:09:12Z witha
66	# adjustments for ForWind computing cluster (lcflow)
67	#
68	# 1545 2015-01-29 06:52:23Z heinze
69	# local host name for blizzard further specified
70	#
71	# 1480 2014-10-17 14:41:49Z raasch
72	# adjustments for 2nd stage of HLRNIII
73	#
74	# 1468 2014-09-24 14:06:57Z maronga
75	# Typo removed (addres->address)
76	# Adjustments for lcxe6
77	#
78	# 1452 2014-08-22 09:41:06Z heinze
79	# local hosts for blizzard added
80	#
81	# 1450 2014-08-21 07:31:51Z heinze
82	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
83	#
84	# 1442 2014-07-28 07:09:10Z raasch
85	# HLRN-III (lccrayb/lccrayh) queues adjusted
86	#
87	# 1378 2014-04-28 06:04:58Z raasch
88	# -et option added for lctit
89	#
90	# 1350 2014-04-04 13:01:30Z maronga
91	# location of qsub updated for lcxe6
92	#
93	# 1289 2014-03-04 07:12:34Z raasch
94	# German comments translated to English
95	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
96	#
97	# 1279 2014-01-28 12:10:14Z raasch
98	# node calculation modified due to changes in mrun (tasks_per_node must not be
99	# an integral divisor of numprocs any more)
100	#
101	# 1274 2014-01-09 13:14:54Z heinze
102	# adjustments for lccrayh
103	#
104	# 1266 2013-12-11 12:07:34Z heinze
105	# further adjustments for lccrayb (use msub instead of qsub)
106	#
107	# 1264 2013-12-09 12:46:09Z fricke
108	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
109	#
110	# 1262 2013-12-09 10:57:20Z fricke
111	# further adjustments for lccrayb
112	#
113	# 1260 2013-12-04 12:48:04Z raasch
114	# jaboticaba admitted
115	#
116	# 1255 2013-11-07 14:43:35Z raasch
117	# further adjustments for lccrayb
118	#
119	# 1224 2013-09-16 07:27:23Z raasch
120	# first adjustments for lccrayb
121	#
122	# 1202 2013-07-10 16:22:07Z witha
123	# adjustments for Forwind cluster (lcflow)
124	#
125	# 1199 2013-07-05 14:52:22Z raasch
126	# adjustments for CSC Helsinki (lccrayf)
127	#
128	# use of cluster/express queue enabled (ibmh)
129	# vinessa added (imuk)
130	#
131	# 1103 2013-02-20 02:15:53Z raasch
132	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
133	# instead typeset -L/R),
134	# further adjustments for lckyuh
135	#
136	# 2013-02-10 01:47:43Z raasch
137	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
138	# and for Forwind cluster (lcflow)
139	#
140	# 1094 2013-02-03 01:52:12Z raasch
141	# new option -P for explicit setting of ssh/scp port,
142	# decalpha parts (yonsei) removed
143	#
144	# 2013-02-02 07:06:13Z raasch
145	# adjustments for Kyushu-University computing center (lckyut - tatara)
146	# old changelog messages removed
147	#
148	# 1046 2012-11-09 14:38:45Z maronga
149	# code put under GPL (PALM 3.9)
150	#
151	# 08/07/94 - Siggi - first version finished
152	# 29/06/94 - Siggi - script development started
153	#--------------------------------------------------------------------------------#
154	# subjob - script for automatic generation and submission of batch-job files
155	# for various batch queuing systems
156	#--------------------------------------------------------------------------------#
157
158
159	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
160	delete_dayfile=false
161	email_notification=none
162	group_number=none
163	locat=normal
164	no_default_queue=none
165	no_submit=false
166	job_catalog="~/job_queue"
167	job_name=none
168	local_user=$LOGNAME
169	node_usage=shared
170	numprocs=0
171	punkte="..........................................................."
172	submcom=qsub
173	queue=default
174	remote_host=none
175	remote_user=""
176	verify=true
177
178	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
179	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
180
181
182
183	# ERROR HANDLING
184	# IN CASE OF EXIT:
185	trap 'if [[ $locat != normal ]]
186	then
187	case $locat in
188	(option) printf "\n --> available optios can be displayed"
189	printf " by typing:"
190	printf "\n \"subjob ?\" \n";;
191	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
192	(*) printf "\n +++ unknown error"
193	printf "\n please inform S. Raasch!\n"
194	esac
195	[[ -f $job_to_send ]] && rm $job_to_send
196	printf "\n\n+++ SUBJOB killed \n\n"
197	fi' exit
198
199
200	# IN CASE OF TERMINAL-BREAK:
201	trap '[[ -f $job_to_send ]] && rm $job_to_send
202	printf "\n\n+++ SUBJOB killed \n\n"
203	exit
204	' 2
205
206
207	# DETERMINE NAME OF LOCAL HOST
208	local_host=$(hostname)
209
210	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
211	# IS ADMITTED AT ALL)
212	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
213	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
214	case $local_host in
215	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
216	(atmos) local_address=172.20.25.35; local_host=lcide;;
217	(austru) local_address=130.75.105.128; local_host=lcmuk;;
218	(autan) local_address=130.75.105.57; local_host=lcmuk;;
219	(bora) local_address=130.75.105.103; local_host=lcmuk;;
220	(a0\|b0) local_address=133.5.4.33; local_host=lckyuh;;
221	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
222	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
223	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
224	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
225	(breva) local_address=130.75.105.98; local_host=lcmuk;;
226	(buran) local_address=130.75.105.58; local_host=lcmuk;;
227	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
228	(climate*) local_address=165.132.26.68; local_host=lcyon;;
229	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
230	(cs*) local_address=136.172.44.131; local_host=nech;;
231	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
232	(flow01) local_address=10.141.255.71; local_host=lcflow;;
233	(flow02) local_address=10.141.255.72; local_host=lcflow;;
234	(node*) local_address=165.132.26.61 local_host=lck;;
235	# (node*) local_address=210.219.61.8 local_host=lckordi;;
236	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
237	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
238	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
239	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
240	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
241	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
242	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
243	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
244	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
245	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
246	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
247	(sno) local_address=130.75.105.113; local_host=lcmuk;;
248	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
249	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
250	(login*) local_address=118.128.66.201; local_host=lckiaps;;
251	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
252	(meller) local_address=134.106.74.155; local_host=lcfor;;
253	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
254	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
255	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
256	(nobel*) local_address=150.183.5.101; local_host=ibms;;
257	(ocean) local_address="ocean"; local_host=lcocean;;
258	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
259	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
260	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
261	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
262	(pingui) local_address=134.106.74.118; local_host=lcfor;;
263	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
264	(rte*) local_address=133.5.185.60; local_host=lcrte;;
265	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
266	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
267	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
268	(solano) local_address=130.75.105.110; local_host=lcmuk;;
269	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
270	(tc*) local_address="ocean"; local_host=lcocean;;
271	(t2a*) local_address=10.1.6.165; local_host=lctit;;
272	(urban*) local_address=147.46.30.151 local_host=lcsb;;
273	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
274	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
275	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
276	(*) printf "\n +++ \"$local_host\" unknown";
277	printf "\n please contact the PALM group at IMUK";
278	locat=parameter; exit;;
279	esac
280
281
282
283	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
284	remote_host=$local_host
285
286
287
288
289	# READ THE SHELLSCRIPT-OPTIONS
290	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
291	do
292	case $option in
293	(c) job_catalog=$OPTARG;;
294	(d) delete_dayfile=true;;
295	(D) no_submit=true;;
296	(e) email_notification=$OPTARG;;
297	(g) group_number=$OPTARG;;
298	(h) remote_host=$OPTARG;;
299	(m) memory=$OPTARG;;
300	(n) job_name=$OPTARG;;
301	(N) node_usage=$OPTARG;;
302	(O) threads_per_task=$OPTARG;;
303	(P) scp_port=$OPTARG;;
304	(q) no_default_queue=$OPTARG;;
305	(t) cputime=$OPTARG;;
306	(T) tasks_per_node=$OPTARG;;
307	(u) remote_user=$OPTARG;;
308	(v) verify=false;;
309	(X) numprocs=$OPTARG;;
310	(\?) printf "\n +++ Option $OPTARG unknown \n";
311	locat=option; exit;;
312	esac
313	done
314
315
316	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
317	(( to_shift = $OPTIND - 1 ))
318	shift $to_shift; file_to_send=$1
319
320
321	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
322	if [ "$1" = "?" ]
323	then
324	(printf "\n *** subjob can be called as follows:\n"
325	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
326	printf "\n Description of available options:\n"
327	printf "\n Option Description Default-Value"
328	printf "\n -c job-input- and output-catalog ~/job_queue"
329	printf "\n -d no job-protocol will be created ---"
330	printf "\n -D only the job-file will be created ---"
331	printf "\n -h execution host, available hosts: $remote_host"
332	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
333	printf "\n lckiaps, lctit, nech"
334	printf "\n -m memory demand per process in MByte ---"
335	printf "\n -n jobname <jobdatei>"
336	printf "\n -O threads per task (for OpenMP usage) 1"
337	printf "\n -P ssh/scp port default port"
338	printf "\n -q job-queue to be used default"
339	printf "\n -t allowed cpu-time in seconds ---"
340	printf "\n -T tasks per node (on parallel hosts) ---"
341	printf "\n -u username on execution host from .netrc"
342	printf "\n -v no prompt for confirmation ---"
343	printf "\n -X # of processors (on parallel hosts) 1"
344	printf "\n "
345	printf "\n The only possible positional parameter is <jobfile>:"
346	printf "\n The complete NQS-job must be provided here."
347	printf "\n <jobfile>=? creates this outline\n\n") \| more
348	exit
349	fi
350
351
352
353	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
354	if [[ "$file_to_send" = "" ]]
355	then
356	printf "\n +++ job-file missing"
357	locat=parameter; exit
358	else
359	if [[ -f $file_to_send ]]
360	then
361	true
362	else
363	printf "\n +++ job-file: "
364	printf "\n $file_to_send"
365	printf "\n does not exist"
366	locat=parameter; exit
367	fi
368	fi
369
370
371
372	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
373	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
374	if [[ $job_name = none ]]
375	then
376	job_name=$file_to_send
377	fi
378	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
379	then
380	printf "\n +++ job-file name: "
381	printf "\n $job_name"
382	printf "\n must not contain \"/\"-characters"
383	locat=parameter; exit
384	fi
385
386
387
388
389	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
390	# OR IF NO HOST HAS BEEN GIVEN
391	if [[ $remote_host = none ]]
392	then
393	printf "\n +++ host missing"
394	locat=option; exit
395	else
396	case $remote_host in
397	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
398	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
399	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
400	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
401	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
402	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
403	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
404	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
405	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
406	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
407	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
408	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
409	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
410	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
411	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
412	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
413	(lcocean) remote_address="ocean"; submcom=qsub;;
414	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
415	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
416	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
417	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
418	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
419	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
420	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
421	locat=parameter; exit;;
422	esac
423	fi
424
425
426	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
427	if [[ $no_default_queue != none ]]
428	then
429	error=false
430	ndq=$no_default_queue
431	case $remote_host in
432	(ibm) case $ndq in
433	(p690_express\|p690_standard\|p690_long) error=false;;
434	(*) error=true;;
435	esac;;
436	(ibmh) case $ndq in
437	(cluster\|express) error=false;;
438	(*) error=true;;
439	esac;;
440	(ibmkisti) case $ndq in
441	(class.32plus\|class.1-2\|class.2-32) error=false;;
442	(*) error=true;;
443	esac;;
444	(ibmku) case $ndq in
445	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
446	(*) error=true;;
447	esac;;
448	(ibms) case $ndq in
449	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
450	(*) error=true;;
451	esac;;
452	(lcbullhh) case $ndq in
453	(compute\|compute2\|shared) error=false;;
454	(*) error=true;;
455	esac;;
456	(lccrayb) case $ndq in
457	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
458	(*) error=true;;
459	esac;;
460	(lccrayh) case $ndq in
461	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
462	(*) error=true;;
463	esac;;
464	(lccrayf) case $ndq in
465	(usup\|test*\|small\|large) error=false;;
466	(*) error=true;;
467	esac;;
468	(lcflow) case $ndq in
469	(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;;
470	(*) error=true;;
471	esac;;
472	(lckiaps) case $ndq in
473	(express\|normal\|normal20\|quickq) error=false;;
474	(*) error=true;;
475	esac;;
476	(lckyoto) case $ndq in
477	(eh\|ph) error=false;;
478	(*) error=true;;
479	esac;;
480	(lckyuh) case $ndq in
481	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
482	(*) error=true;;
483	esac;;
484	(lckyut) case $ndq in
485	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
486	(*) error=true;;
487	esac;;
488	(lctit) case $ndq in
489	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
490	(*) error=true;;
491	esac;;
492	(t3eb) case $ndq in
493	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
494	(*) error=true;;
495	esac;;
496	(t3eh) case $ndq in
497	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
498	(*) error=true;;
499	esac;;
500	(t3ej2\|t3ej5) case $ndq in
501	(low\|normal\|high) error=false;;
502	(*) error=true;;
503	esac;;
504	(t3es) case $ndq in
505	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
506	(*) error=true;;
507	esac;;
508	esac
509	if [[ $error = true ]]
510	then
511	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
512	locat=parameter; exit
513	else
514	queue=$no_default_queue
515	fi
516	fi
517
518
519
520	# CHECK THE CPU-TIME
521	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
522	done=false
523	while [[ $done = false ]]
524	do
525	if (( $cputime <= 0 ))
526	then
527	printf "\n +++ wrong cpu-time or cpu-time missing"
528	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
529	printf "\n >>> "
530	read cputime 1>/dev/null 2>&1
531	else
532	done=true
533	fi
534	done
535	if [[ $remote_host = nech ]]
536	then
537	if (( tasks_per_node != 0 ))
538	then
539	(( cputime = cputime * tasks_per_node ))
540	elif [[ $numprocs != 0 ]]
541	then
542	(( cputime = cputime * numprocs ))
543	fi
544	fi
545	(( stunden = cputime / 3600 ))
546	(( resttime = cputime - stunden * 3600 ))
547	(( minuten = resttime / 60 ))
548	(( sekunden = resttime - minuten * 60 ))
549	timestring=${stunden}:${minuten}:${sekunden}
550
551
552
553	# CHECK THE MEMORY DEMAND
554	done=false
555	while [[ $done = false ]]
556	do
557	if (( memory <= 0 ))
558	then
559	printf "\n +++ wrong memory demand or memory demand missing"
560	printf "\n >>> Please type memory in MByte per process as INTEGER:"
561	printf "\n >>> "
562	read memory 1>/dev/null 2>&1
563	else
564	done=true
565	fi
566	done
567
568	if [[ $remote_host = nech ]]
569	then
570	if (( tasks_per_node != 0 ))
571	then
572	(( Memory = memory * tasks_per_node / 1000 ))
573	elif [[ $numprocs != 0 ]]
574	then
575	(( Memory = memory * numprocs / 1000 ))
576	else
577	(( Memory = memory / 1000 ))
578	fi
579	elif [[ $remote_host = lctit ]]
580	then
581	(( Memory = memory * tasks_per_node / 1000 ))
582	fi
583
584
585	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
586	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
587	then
588	(( memory = memory * threads_per_task ))
589	fi
590
591
592	# CALCULATE NUMBER OF REQUIRED NODES
593	if (( tasks_per_node != 0 ))
594	then
595	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
596	fi
597
598
599	# CALCULATE NUMBER OF PROCESSES PER NODE
600	(( processes_per_node = tasks_per_node * threads_per_task ))
601
602
603	# CALCULATE NUMBER OF MPI TASKS
604	(( mpi_tasks = numprocs / threads_per_task ))
605
606
607	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
608	if [[ "$scp_port" != "" ]]
609	then
610	PORTOPT="-P $scp_port"
611	SSH_PORTOPT="-p $scp_port"
612	fi
613
614
615	# HEADER-OUTPUT
616	if [[ $verify = true ]]
617	then
618	printf "\n\n"
619	printf "#--------------------------------------------------------------# \n"
620	spalte1=SUBJOB;spalte2=$(date)
621	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
622	printf "\| \| \n"
623	printf "\| values of parameters/options: \| \n"
624	spalte1=$(echo local_host$punkte \| cut -c-20)
625	spalte2=$punkte$local_host
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	spalte1=$(echo remote_host$punkte \| cut -c-20)
628	spalte2=$punkte$remote_host
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	spalte1=$(echo queue$punkte \| cut -c-20)
631	spalte2=$punkte$queue
632	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
633	spalte1=$(echo memory$punkte \| cut -c-20)
634	spalte2="$punkte$memory mb"
635	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
636	spalte1=$(echo cputime$punkte \| cut -c-20)
637	spalte2="$punkte$cputime sec"
638	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
639	spalte1=$(echo job_name$punkte \| cut -c-20)
640	spalte2="$punkte$job_name"
641	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
642	printf "#--------------------------------------------------------------# \n\n"
643
644
645	# QUERY CHECK
646	antwort="dummy"
647	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
648	do
649	read antwort?" >>> continue (y/n) ? "
650	done
651	if [[ $antwort = n \|\| $antwort = N ]]
652	then
653	locat=verify; exit
654	fi
655	printf "\n"
656	fi
657
658	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
659	identifier=$RANDOM
660	job_on_remhost=${job_name}_${identifier}_$local_host
661	job_to_send=job_to_send_$identifier
662	if [[ $delete_dayfile = false ]]
663	then
664	remote_dayfile=${local_host}_${job_name}_result_$identifier
665	local_dayfile=${remote_host}_${job_name}
666	else
667	remote_dayfile=/dev/null
668	fi
669
670
671	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
672	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
673	then
674
675	# GENERAL LOADLEVELER SETTINGS
676	execute_in_shell="#!/bin/ksh"
677	use_shell="# @ shell = /bin/ksh"
678	consumable_memory="ConsumableMemory($memory mb)"
679	class="# @ class = $queue"
680	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
681	network_to_use="# @ network.mpi = sn_all,shared,us"
682	data_limit="# @ data_limit = 1.76gb"
683	image_size="# @ image_size = 50"
684	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
685
686	if [[ $email_notification = none ]]
687	then
688	notify_user=""
689	else
690	notify_user="# @ notify_user = $email_notification"
691	if [[ $delete_dayfile = true ]]
692	then
693	notification='# @ notification = never'
694	fi
695	fi
696
697	if [[ $remote_host = ibmh ]]
698	then
699	data_limit=""
700	network_to_use=""
701	class="# @ class = $queue"
702	environment=""
703	rset="# @ rset = RSET_MCM_AFFINITY"
704	task_affinity="# @ task_affinity = core(1)"
705	elif [[ $remote_host = ibmkisti ]]
706	then
707	network_to_use="# @ network.MPI = sn_all,shared,US"
708	wall_clock_limit="# @ wall_clock_limit = $timestring"
709	if [[ $threads_per_task = 1 ]]
710	then
711	rset="# @ rset = RSET_MCM_AFFINITY"
712	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
713	fi
714	environment=""
715	use_shell=""
716	data_limit=""
717	image_size=""
718	elif [[ $remote_host = ibmku ]]
719	then
720	execute_in_shell="#!/usr/bin/ksh"
721	use_shell="# @ shell = /usr/bin/ksh"
722	consumable_memory=""
723	environment=""
724	network_to_use="# @ network.mpi = sn_all,shared,us"
725	data_limit=""
726	image_size=""
727	elif [[ $remote_host = ibms ]]
728	then
729	network_to_use="# @ network.mpi = csss,shared,us"
730	fi
731
732	cat > $job_to_send << %%END%%
733	$execute_in_shell
734	$use_shell
735
736	# @ job_type = parallel
737	# @ job_name = $job_name
738	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
739	# @ output = $remote_dayfile
740	# @ error = $remote_dayfile
741	$wall_clock_limit
742	$image_size
743	$class
744	$environment
745	$network_to_use
746	$data_limit
747	$rset
748	$mcm_affinity_options
749	$task_affinity
750	$notification
751	$notify_user
752
753	%%END%%
754
755	if (( nodes > 0 ))
756	then
757
758	if [[ $remote_host != ibmkisti ]]
759	then
760
761	cat >> $job_to_send << %%END%%
762	# @ node = $nodes
763	# @ tasks_per_node = $processes_per_node
764	# @ node_usage = $node_usage
765	# @ queue
766
767	%%END%%
768
769	else
770
771	cat >> $job_to_send << %%END%%
772	# @ total_tasks = $mpi_tasks
773	# @ blocking = unlimited
774	# @ queue
775
776	%%END%%
777
778	fi
779
780	else
781
782	cat >> $job_to_send << %%END%%
783	# @ blocking = unlimited
784	# @ total_tasks = $numprocs
785	# @ node_usage = $node_usage
786	# @ queue
787
788	%%END%%
789
790	fi
791
792	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
793	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
794	then
795	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
796	fi
797
798	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
799	then
800
801	cat > $job_to_send << %%END%%
802	#!/bin/ksh
803
804	# @ job_type = serial
805	# @ node_usage = $node_usage
806	# @ job_name = palm
807	# @ wall_clock_limit = ${timestring},$timestring
808	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
809	# @ output = $remote_dayfile
810	# @ error = $remote_dayfile
811	$class
812	$notification
813
814	# @ queue
815
816	%%END%%
817
818	elif [[ $remote_host = lcbullhh ]]
819	then
820	if [[ $numprocs != 0 ]]
821	then
822	cat > $job_to_send << %%END%%
823	#!/bin/bash -l
824	#SBATCH -J $job_name
825	#SBATCH -t $timestring
826	#SBATCH -N $nodes
827	#SBATCH --ntasks-per-node=$processes_per_node
828	#SBATCH -p $queue
829	#SBATCH -o $remote_dayfile
830	#SBATCH -e $remote_dayfile
831	#SBATCH -A $project_account
832
833	$init_cmds
834	$module_calls
835
836	%%END%%
837
838	else
839	cat > $job_to_send << %%END%%
840	#!/bin/bash -l
841	#SBATCH -J $job_name
842	#SBATCH -t $timestring
843	#SBATCH -l ncpus=1
844	#SBATCH -l pmem=${memory}mb
845	#SBATCH -m abe
846	#SBATCH -o $remote_dayfile
847	#SBATCH -e $remote_dayfile
848	#SBATCH -A $project_account
849
850	$init_cmds
851	$module_calls
852
853	%%END%%
854
855	fi
856
857	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
858	then
859
860	if [[ "$feature" != "" ]]
861	then
862	featuredir="#PBS -l feature=$feature"
863	fi
864
865	if [[ $numprocs != 0 ]]
866	then
867	cat > $job_to_send << %%END%%
868	#!/bin/bash
869	#PBS -N $job_name
870	#PBS -l walltime=$timestring
871	#PBS -l nodes=$nodes:ppn=$processes_per_node
872	#PBS -o $remote_dayfile
873	#PBS -j oe
874	#PBS -q $queue
875	$featuredir
876
877	$init_cmds
878	$module_calls
879
880	%%END%%
881
882	else
883
884	continue
885
886	fi
887
888	elif [[ $remote_host = lccrayf ]]
889	then
890
891	if [[ $numprocs != 0 ]]
892	then
893	cat > $job_to_send << %%END%%
894	#!/bin/bash -l
895	#SBATCH -J $job_name
896	#SBATCH -t $timestring
897	#SBATCH -N $nodes
898	#SBATCH --ntasks-per-node=$processes_per_node
899	#SBATCH -p $queue
900	#SBATCH -o $remote_dayfile
901	#SBATCH -e $remote_dayfile
902
903	$init_cmds
904	$module_calls
905
906	%%END%%
907
908	else
909	cat > $job_to_send << %%END%%
910	#!/bin/bash -l
911	#SBATCH -J $job_name
912	#SBATCH -t $timestring
913	#SBATCH -l ncpus=1
914	#SBATCH -l pmem=${memory}mb
915	#SBATCH -m abe
916	#SBATCH -o $remote_dayfile
917	#SBATCH -e $remote_dayfile
918
919	$init_cmds
920	$module_calls
921
922	%%END%%
923
924	fi
925
926	elif [[ $remote_host = lcflow ]]
927	then
928	if [[ $numprocs != 0 ]]
929	then
930	pe_set="#$ -pe impi $numprocs"
931	else
932	pe_set="#$ -pe impi 1"
933	fi
934	if [[ $queue = default ]]
935	then
936	queue_set=""
937	else
938	queue_set="#$ -q $queue"
939	fi
940	[[ "$disc_space" = "" ]] && disc_space=50
941
942	cat > $job_to_send << %%END%%
943	#!/bin/bash
944	#$ -S /bin/bash
945	#$ -N $job_name
946	#$ -cwd
947	#$ -l h_rt=$timestring
948	#$ -l h_vmem=${memory}M
949	#$ -o $remote_dayfile
950	#$ -j y
951	$pe_set
952	#$ -R y
953	#$ -l h_fsize=${disc_space}G
954	$queue_set
955
956	%%END%%
957
958	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
959	then
960
961	if [[ $numprocs != 0 ]]
962	then
963	cat > $job_to_send << %%END%%
964	#!/bin/ksh
965	#PBS -N $job_name
966	#PBS -l walltime=$timestring
967	#PBS -l ncpus=$numprocs
968	#PBS -l pmem=${memory}mb
969	#PBS -o $remote_dayfile
970	#PBS -l nodes=$nodes:ppn=${processes_per_node}
971	#PBS -j oe
972
973	mpd &
974
975	%%END%%
976
977	else
978	cat > $job_to_send << %%END%%
979	#!/bin/ksh
980	#PBS -N $job_name
981	#PBS -l walltime=$timestring
982	#PBS -l ncpus=1
983	#PBS -l pmem=${memory}mb
984	#PBS -o $remote_dayfile
985	#PBS -j oe
986
987	%%END%%
988
989	fi
990
991	elif [[ $remote_host = lckiaps ]]
992	then
993
994	if [[ $numprocs != 0 ]]
995	then
996	cat > $job_to_send << %%END%%
997	#!/bin/bash
998	#PBS -N $job_name
999	#PBS -l walltime=$timestring
1000	#PBS -l select=1:ncpus=$numprocs
1001	#PBS -l pmem=${memory}mb
1002	#PBS -q $queue
1003	#PBS -o $remote_dayfile
1004	#PBS -j oe
1005	#PBS -V
1006
1007	%%END%%
1008
1009	else
1010	cat > $job_to_send << %%END%%
1011	#!/bin/bash
1012	#PBS -N $job_name
1013	#PBS -l walltime=$timestring
1014	#PBS -l ncpus=1
1015	#PBS -l pmem=${memory}mb
1016	#PBS -o $remote_dayfile
1017	#PBS -j oe
1018
1019	%%END%%
1020
1021	fi
1022
1023	elif [[ $remote_host = lcyon ]]
1024	then
1025
1026	if [[ $numprocs != 0 ]]
1027	then
1028	cat > $job_to_send << %%END%%
1029	#!/bin/ksh
1030	#PBS -N $job_name
1031	#PBS -l walltime=$timestring
1032	#PBS -l ncpus=$numprocs
1033	#PBS -l pmem=${memory}mb
1034	#PBS -o $remote_dayfile
1035	#PBS -j oe
1036
1037	%%END%%
1038
1039	else
1040	cat > $job_to_send << %%END%%
1041	#!/bin/ksh
1042	#PBS -N $job_name
1043	#PBS -l walltime=$timestring
1044	#PBS -l ncpus=1
1045	#PBS -l pmem=${memory}mb
1046	#PBS -o $remote_dayfile
1047	#PBS -j oe
1048
1049	%%END%%
1050
1051	fi
1052
1053	elif [[ $remote_host = lcxe6 ]]
1054	then
1055
1056	if [[ $numprocs != 0 ]]
1057	then
1058	cat > $job_to_send << %%END%%
1059	#!/bin/ksh
1060	#PBS -S /bin/ksh
1061	#PBS -N $job_name
1062	#PBS -A $project_account
1063	#PBS -j oe
1064	#PBS -l walltime=$timestring
1065	#PBS -l mppwidth=${numprocs}
1066	#PBS -l mppnppn=${processes_per_node}
1067	#PBS -m abe
1068	#PBS -o $remote_dayfile
1069	$email_directive
1070
1071	$init_cmds
1072	$module_calls
1073
1074	%%END%%
1075
1076	else
1077	cat > $job_to_send << %%END%%
1078	#!/bin/ksh
1079	#PBS -S /bin/ksh
1080	#PBS -N $job_name
1081	#PBS -A $project_account
1082	#PBS -j oe
1083	#PBS -l walltime=$timestring
1084	#PBS -l ncpus=1
1085	#PBS -l pmem=${memory}mb
1086	#PBS -m abe
1087	$email_directive
1088	#PBS -o $remote_dayfile
1089
1090	$init_cmds
1091	$module_calls
1092
1093	%%END%%
1094
1095	fi
1096
1097	elif [[ $remote_host = lckyoto ]]
1098	then
1099
1100	cat > $job_to_send << %%END%%
1101	#!/bin/ksh
1102	# @\$-o $remote_dayfile
1103	# @\$-eo -oi
1104	# @\$-lP 16
1105	# @\$-lp 1
1106	# @\$-lm 28gb -llm unlimited -ls unlimited
1107	# @\$-q $queue
1108	# @\$-Pvn abs_pack
1109	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1110	#. /thin/local/etc/setprofile/intel-11.0.sh
1111	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1112	. ~/.myprofile
1113	#. /home2/t/t51254/palm/current_version/myprofile
1114	#. /thin/apps/pgi/mpi.sh
1115	#
1116	env
1117	#
1118	set -x
1119
1120	%%END%%
1121
1122	elif [[ $remote_host = lcxt5m ]]
1123	then
1124
1125	if [[ $numprocs != 0 ]]
1126	then
1127	cat > $job_to_send << %%END%%
1128	#!/bin/ksh
1129	#PBS -S /bin/ksh
1130	#PBS -N $job_name
1131	#PBS -j oe
1132	#PBS -l walltime=$timestring
1133	#PBS -l mppwidth=${numprocs}
1134	#PBS -l mppnppn=${processes_per_node}
1135	#PBS -m abe
1136	#PBS -o $remote_dayfile
1137
1138	$init_cmds
1139	$module_calls
1140
1141	%%END%%
1142
1143	else
1144	cat > $job_to_send << %%END%%
1145	#!/bin/ksh
1146	#PBS -S /bin/ksh
1147	#PBS -N $job_name
1148	#PBS -j oe
1149	#PBS -l walltime=$timestring
1150	#PBS -l ncpus=1
1151	#PBS -l pmem=${memory}mb
1152	#PBS -m abe
1153	#PBS -o $remote_dayfile
1154
1155	$init_cmds
1156	$module_calls
1157
1158	%%END%%
1159
1160	fi
1161
1162	elif [[ $remote_host = lckyuh ]]
1163	then
1164	cat > $job_to_send << %%END%%
1165	#!/bin/bash
1166	#PJM -L "rscgrp=$queue"
1167	#PJM -L "node=$nodes"
1168	#PJM --mpi "proc=$numprocs"
1169	#PJM -L "elapse=$timestring"
1170	#PJM -o $remote_dayfile
1171	#PJM -j
1172	#PJM -X
1173	#PJM --no-stging
1174
1175	export LANG=en_US.UTF-8
1176	%%END%%
1177
1178	elif [[ $remote_host = lckyut ]]
1179	then
1180	cat > $job_to_send << %%END%%
1181	#!/bin/bash
1182	#PJM -L "rscgrp=$queue"
1183	#PJM -L "vnode=$numprocs"
1184	#PJM -L "vnode-core=1"
1185	#PJM -L "elapse=$timestring"
1186	#PJM --mpi proc=$numprocs
1187	#PJM -o $remote_dayfile
1188	#PJM -j
1189	#PJM -X
1190	#PJM --no-stging
1191
1192	export LANG=en_US.UTF-8
1193	%%END%%
1194
1195	elif [[ $remote_host = lcocean ]]
1196	then
1197	cat > $job_to_send << %%END%%
1198	#!/bin/bash
1199	#$ -cwd
1200	#$ -V
1201	#$ -N $job_name
1202	#$ -pe orte $numprocs
1203	#$ -o $remote_dayfile
1204	#$ -j y
1205	#$ -R y
1206	$init_cmds
1207	$module_calls
1208
1209	%%END%%
1210
1211	elif [[ $remote_host = nech ]]
1212	then
1213
1214	if (( nodes > 1 ))
1215	then
1216	cat > $job_to_send << %%END%%
1217	#!/bin/ksh
1218	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1219	#PBS -l ${qsubmem}=${Memory}gb
1220	#PBS -b $nodes
1221	#PBS -o $remote_dayfile
1222	#PBS -N palm
1223	#PBS -j o
1224	#PBS -T mpisx
1225
1226	%%END%%
1227
1228	elif [[ $numprocs != 0 ]]
1229	then
1230	cat > $job_to_send << %%END%%
1231	#!/bin/ksh
1232	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1233	#PBS -l ${qsubmem}=${Memory}gb
1234	#PBS -o $remote_dayfile
1235	#PBS -N palm
1236	#PBS -j o
1237
1238	%%END%%
1239
1240	else
1241	cat > $job_to_send << %%END%%
1242	#!/bin/ksh
1243	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1244	#PBS -o $remote_dayfile
1245	#PBS -j o
1246
1247	%%END%%
1248
1249	fi
1250
1251	elif [[ $remote_host = lctit ]]
1252	then
1253	cat > $job_to_send << %%END%%
1254	#!/bin/ksh
1255	$init_cmds
1256	$module_calls
1257
1258	%%END%%
1259
1260	# SET OPTIONS FOR SUBMIT-COMMAND
1261	if [[ $tasks_per_node != $processes_per_node ]]
1262	then
1263	submcom="$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 $queue "
1264	else
1265	submcom="$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 "
1266	fi
1267
1268	else
1269
1270	cat > $job_to_send << %%END%%
1271	# @\$-q ${queue}
1272	# @\$-l${qsubtime} $timestring
1273	# @\$-l${qsubmem} ${memory}mb
1274	# @\$-o $remote_dayfile
1275	# @\$-eo
1276
1277	%%END%%
1278
1279	fi
1280
1281
1282	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1283	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1284	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1285	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1286	then
1287	echo "set +vx" >> $job_to_send
1288	echo "trap '" >> $job_to_send
1289	echo "set +vx" >> $job_to_send
1290	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 ]]
1291	then
1292	if [[ $remote_host = ibmh ]]
1293	then
1294	return_queue=c1
1295	elif [[ $remote_host = ibmkisti ]]
1296	then
1297	return_queue=class.1-2
1298	elif [[ $remote_host = ibmku ]]
1299	then
1300	return_queue=sdbg2
1301	elif [[ $remote_host = ibms ]]
1302	then
1303	return_queue=p_normal
1304	elif [[ $remote_host = lcbullhh ]]
1305	then
1306	return_queue=shared
1307	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1308	then
1309	return_queue=dataq
1310	elif [[ $remote_host = lcxe6 ]]
1311	then
1312	return_queue=debug
1313	elif [[ $remote_host = lckiaps ]]
1314	then
1315	return_queue=express
1316	elif [[ $remote_host = lckyuh ]]
1317	then
1318	return_queue=cx-single
1319	elif [[ $remote_host = lckyut ]]
1320	then
1321	return_queue=cx-single
1322	else
1323	return_queue=unknown
1324	fi
1325
1326	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1327	then
1328
1329	if [[ $remote_host = ibmku ]]
1330	then
1331	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1333	else
1334	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1335	fi
1336	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1337	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1338	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1339	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1340	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1341	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1342	if [[ $host != "ibmh" ]]
1343	then
1344	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1345	fi
1346	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1347	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1348
1349	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1350	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1351
1352	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1353	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
1354	if [[ $remote_host = ibmku ]]
1355	then
1356	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1357	fi
1358	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1359
1360	elif [[ $remote_host = nech ]]
1361	then
1362	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1363	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1364	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1365	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1366	echo "#PBS -j o" >> $job_to_send
1367	echo " " >> $job_to_send
1368	echo "set -x" >> $job_to_send
1369	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1370	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1371	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1372	echo "%%END%%" >> $job_to_send
1373
1374	elif [[ $remote_host = lcbullhh ]]
1375	then
1376	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1377	echo "#!/bin/bash" >> $job_to_send
1378	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1379	echo "#SBATCH -t 00:20:00" >> $job_to_send
1380	echo "#SBATCH -N 1" >> $job_to_send
1381	echo "#SBATCH -n 1" >> $job_to_send
1382	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1383	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1384	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1385	echo "#SBATCH -A $project_account" >> $job_to_send
1386	echo "#SBATCH -p $return_queue" >> $job_to_send
1387	echo " " >> $job_to_send
1388	echo "set -x" >> $job_to_send
1389	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1390	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1391	echo "%%END%%" >> $job_to_send
1392
1393	elif [[ $remote_host = lckyuh ]]
1394	then
1395	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1396	echo "#!/bin/bash" >> $job_to_send
1397	echo "#PJM -L \"node=1\"" >> $job_to_send
1398	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1399	echo "#PJM --no-stging" >> $job_to_send
1400	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1401	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1402	echo "#PJM -j" >> $job_to_send
1403	echo " " >> $job_to_send
1404	echo "export LANG=en_US.UTF-8" >> $job_to_send
1405	echo "set -x" >> $job_to_send
1406	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1407	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1408	echo "%%END%%" >> $job_to_send
1409
1410	elif [[ $remote_host = lckyut ]]
1411	then
1412	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1413	echo "#!/bin/bash" >> $job_to_send
1414	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1415	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1416	echo "#PJM --no-stging" >> $job_to_send
1417	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1418	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1419	echo "#PJM -j" >> $job_to_send
1420	echo " " >> $job_to_send
1421	echo "export LANG=en_US.UTF-8" >> $job_to_send
1422	echo "set -x" >> $job_to_send
1423	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1424	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1425	echo "%%END%%" >> $job_to_send
1426
1427	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1428	then
1429	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1430	echo "#!/bin/bash" >> $job_to_send
1431	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1432	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1433	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1434	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1435	echo "#PBS -j oe" >> $job_to_send
1436	echo " " >> $job_to_send
1437	echo "set -x" >> $job_to_send
1438	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1439	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1440	echo "%%END%%" >> $job_to_send
1441
1442	elif [[ $remote_host = lcocean ]]
1443	then
1444	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1445	echo "#!/bin/bash" >> $job_to_send
1446	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1447	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1448	echo "SGEPREFIX -cwd" >> $job_to_send
1449	echo "SGEPREFIX -j y" >> $job_to_send
1450	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1451	echo " " >> $job_to_send
1452	echo "set -x" >> $job_to_send
1453	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1454	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1455	echo "" >> $job_to_send
1456	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1457	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1458	echo "rm -f scpjob.${identifier}" >> $job_to_send
1459	echo "%%END%%" >> $job_to_send
1460	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1461	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1462
1463	elif [[ $remote_host = lcflow ]]
1464	then
1465	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1466	echo "#!/bin/bash" >> $job_to_send
1467	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1468	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1469	echo "SGEPREFIX -cwd" >> $job_to_send
1470	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1471	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1472	echo "SGEPREFIX -j y" >> $job_to_send
1473	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1474	echo " " >> $job_to_send
1475	echo "set -x" >> $job_to_send
1476	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1477	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1478	echo "" >> $job_to_send
1479	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1480	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1481	echo "rm -f scpjob.${identifier}" >> $job_to_send
1482	echo "%%END%%" >> $job_to_send
1483	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1484	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1485	elif [[ $remote_host = lcxe6 ]]
1486	then
1487	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1488	echo "#!/bin/ksh" >> $job_to_send
1489	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1490	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1491	echo "#PBS -A $project_account" >> $job_to_send
1492	echo "#PBS -l mppwidth=1" >> $job_to_send
1493	echo "#PBS -l mppnppn=1" >> $job_to_send
1494	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1495	echo "#PBS -j oe" >> $job_to_send
1496	echo " " >> $job_to_send
1497	echo "set -x" >> $job_to_send
1498	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1499	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1500	echo "%%END%%" >> $job_to_send
1501	else
1502
1503	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1504	echo "# @\\\$-q $return_queue" >> $job_to_send
1505	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1506	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1507	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1508	then
1509	echo "# @\$-l mpp_p=0" >> $job_to_send
1510	fi
1511	echo '# @\$-lF 10mb' >> $job_to_send
1512	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1513	echo '# @\\\$-eo' >> $job_to_send
1514	echo " " >> $job_to_send
1515	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1516	then
1517	echo "set +vx" >> $job_to_send
1518	echo ". .profile" >> $job_to_send
1519	fi
1520	echo "set -x" >> $job_to_send
1521	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
1522	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1523	echo "%%END%%" >> $job_to_send
1524
1525	fi
1526
1527	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1528	then
1529	echo "llsubmit scpjob.$identifier" >> $job_to_send
1530	elif [[ $remote_host = lcbullhh ]]
1531	then
1532	echo "sbatch scpjob.$identifier" >> $job_to_send
1533	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1534	then
1535	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1536	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1537	then
1538	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1539	elif [[ $remote_host = t3es ]]
1540	then
1541	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1542	elif [[ $remote_host = lckiaps ]]
1543	then
1544	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1545	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1546	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1547	elif [[ $remote_host = lckyu* ]]
1548	then
1549	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1550	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1551	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1552	then
1553	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1554	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1555	else
1556	echo "$submcom scpjob.$identifier" >> $job_to_send
1557	fi
1558	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1559	then
1560	echo "rm scpjob.$identifier" >> $job_to_send
1561	fi
1562	if [[ $remote_host = nech ]]
1563	then
1564	echo "cd -" >> $job_to_send
1565	fi
1566	else
1567	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1568	# ??? funktioniert das ÃŒberhaupt noch ???
1569	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1570	fi
1571	echo "set -x" >> $job_to_send
1572	echo " ' exit" >> $job_to_send
1573	echo "set -x" >> $job_to_send
1574	fi
1575
1576
1577	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1578	cat $file_to_send >> $job_to_send
1579
1580	if [[ $remote_host = ibm ]]
1581	then
1582	echo " " >> $job_to_send
1583	echo "exit" >> $job_to_send
1584	fi
1585
1586	# REMOVE JOB-FILE
1587	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1588	then
1589	echo " " >> $job_to_send
1590	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1591	fi
1592
1593
1594	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1595	if [[ $no_submit = false ]]
1596	then
1597	if [[ $remote_host != $local_host ]]
1598	then
1599	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1600	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1601	then
1602	job_catalog_save=$job_catalog
1603	job_catalog=job_queue
1604	elif [[ $remote_host = nech ]]
1605	then
1606	job_catalog_save=$job_catalog
1607	job_catalog=/hpf/b/${remote_user}/job_queue
1608	fi
1609	if [[ $remote_host = nech ]]
1610	then
1611	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1612	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1613	else
1614	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1615	fi
1616	if [[ $? = 1 ]]
1617	then
1618	locat=scp; exit
1619	fi
1620	if [[ $remote_host = ibms ]]
1621	then
1622	job_catalog=$job_catalog_save
1623	fi
1624	[[ $verify = true ]] && printf "\n >>> finished\n"
1625	else
1626	eval job_catalog=$job_catalog
1627	cp $job_to_send ${job_catalog}/$job_on_remhost
1628	fi
1629
1630
1631
1632	# START NQS- / LOADLEVELER-JOB
1633	if [[ $remote_host != $local_host ]]
1634	then
1635	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1636
1637	if [[ $remote_host = ibmku ]]
1638	then
1639	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1640	elif [[ $remote_host = lcflow ]]
1641	then
1642	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1643	else
1644	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1645	fi
1646
1647	[[ $verify = true ]] && printf " >>> o.k.\n"
1648	else
1649	cd $job_catalog
1650	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1651	then
1652	eval $submcom $job_on_remhost
1653	elif [[ $local_host = lctit \|\| $local_host = lcxe6 \|\| $local_host = lck \|\| $local_host = lckordi \|\| $local_host = lcyon \|\| $local_host = lcsb \|\| $local_host = lckyu* ]]
1654	then
1655	chmod u+x $job_on_remhost
1656	eval $submcom $job_on_remhost
1657	elif [[ $local_host = nech ]]
1658	then
1659	if [[ $queue = default ]]
1660	then
1661	eval $submcom $job_on_remhost
1662	else
1663	eval $submcom -q $queue $job_on_remhost
1664	fi
1665	elif [[ $local_host = lcbullhh ]]
1666	then
1667	if [[ $queue = default ]]
1668	then
1669	eval $submcom $job_on_remhost
1670	fi
1671	else
1672	qsub $job_on_remhost
1673	fi
1674
1675	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1676	# AT THE END OF THE JOB
1677	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1678	then
1679	rm $job_on_remhost
1680	fi
1681	cd - > /dev/null
1682	fi
1683	fi
1684
1685	# FINAL ACTIONS
1686	if [[ $no_submit = false ]]
1687	then
1688	rm -f $job_to_send
1689	fi
1690	[[ $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 |