Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2297

Last change on this file since 2297 was 2295, checked in by raasch, 7 years ago
script adjustments for using lcgeohu, cpp_opts removed
Property svn:keywords set to `Id Rev`
File size: 55.7 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 2295 2017-06-27 14:25:52Z scharf $
30	# adjustments for using lcgeohu (cirrus @ HUB)
31	#
32	# 2266 2017-06-09 09:27:21Z raasch
33	# nech related parts removed
34	#
35	# 2257 2017-06-07 14:07:05Z witha
36	# adjustments for lceddy, removed lcflow-specific code
37	#
38	# 2188 2017-03-21 06:42:42Z raasch
39	#
40	# 2187 2017-03-21 06:41:25Z raasch
41	# adjustment of compute node names for lckyuh
42	#
43	# 2184 2017-03-21 04:31:22Z raasch
44	# bugfix: localhost renamed local_host
45	#
46	# 2148 2017-02-09 16:56:42Z scharf
47	# added kuma and gharbi to the list of known hosts
48	#
49	# 2134 2017-02-02 07:33:46Z raasch
50	# option -E added to msub commands on HLRN-III machines to allow output of more
51	# job informations in the job protocol files
52	#
53	# 1944 2016-06-15 06:29:00Z raasch
54	# adjustments for using HLRN ssh-keys
55	#
56	# 1940 2016-06-14 05:15:20Z raasch
57	# adjustments for lckiaps
58	#
59	# 1866 2016-04-15 06:50:59Z raasch
60	# adjusted for lcocean
61	#
62	# 1841 2016-04-07 19:14:06Z raasch
63	# script now running under bash
64	#
65	# 1701 2015-11-02 07:43:04Z maronga
66	# Bugfix: added missing init_cmds for lccrayh/lccrayb
67	#
68	# 1621 2015-07-17 11:39:33Z heinze
69	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
70	#
71	# 1575 2015-03-27 09:56:27Z raasch
72	# mpp2-queues added to lccrayh
73	#
74	# 1547 2015-01-29 15:09:12Z witha
75	# adjustments for ForWind computing cluster (lcflow)
76	#
77	# 1545 2015-01-29 06:52:23Z heinze
78	# local host name for blizzard further specified
79	#
80	# 1480 2014-10-17 14:41:49Z raasch
81	# adjustments for 2nd stage of HLRNIII
82	#
83	# 1468 2014-09-24 14:06:57Z maronga
84	# Typo removed (addres->address)
85	# Adjustments for lcxe6
86	#
87	# 1452 2014-08-22 09:41:06Z heinze
88	# local hosts for blizzard added
89	#
90	# 1450 2014-08-21 07:31:51Z heinze
91	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
92	#
93	# 1442 2014-07-28 07:09:10Z raasch
94	# HLRN-III (lccrayb/lccrayh) queues adjusted
95	#
96	# 1378 2014-04-28 06:04:58Z raasch
97	# -et option added for lctit
98	#
99	# 1350 2014-04-04 13:01:30Z maronga
100	# location of qsub updated for lcxe6
101	#
102	# 1289 2014-03-04 07:12:34Z raasch
103	# German comments translated to English
104	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
105	#
106	# 1279 2014-01-28 12:10:14Z raasch
107	# node calculation modified due to changes in mrun (tasks_per_node must not be
108	# an integral divisor of numprocs any more)
109	#
110	# 1274 2014-01-09 13:14:54Z heinze
111	# adjustments for lccrayh
112	#
113	# 1266 2013-12-11 12:07:34Z heinze
114	# further adjustments for lccrayb (use msub instead of qsub)
115	#
116	# 1264 2013-12-09 12:46:09Z fricke
117	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
118	#
119	# 1262 2013-12-09 10:57:20Z fricke
120	# further adjustments for lccrayb
121	#
122	# 1260 2013-12-04 12:48:04Z raasch
123	# jaboticaba admitted
124	#
125	# 1255 2013-11-07 14:43:35Z raasch
126	# further adjustments for lccrayb
127	#
128	# 1224 2013-09-16 07:27:23Z raasch
129	# first adjustments for lccrayb
130	#
131	# 1202 2013-07-10 16:22:07Z witha
132	# adjustments for Forwind cluster (lcflow)
133	#
134	# 1199 2013-07-05 14:52:22Z raasch
135	# adjustments for CSC Helsinki (lccrayf)
136	#
137	# use of cluster/express queue enabled (ibmh)
138	# vinessa added (imuk)
139	#
140	# 1103 2013-02-20 02:15:53Z raasch
141	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
142	# instead typeset -L/R),
143	# further adjustments for lckyuh
144	#
145	# 2013-02-10 01:47:43Z raasch
146	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
147	# and for Forwind cluster (lcflow)
148	#
149	# 1094 2013-02-03 01:52:12Z raasch
150	# new option -P for explicit setting of ssh/scp port,
151	# decalpha parts (yonsei) removed
152	#
153	# 2013-02-02 07:06:13Z raasch
154	# adjustments for Kyushu-University computing center (lckyut - tatara)
155	# old changelog messages removed
156	#
157	# 1046 2012-11-09 14:38:45Z maronga
158	# code put under GPL (PALM 3.9)
159	#
160	# 08/07/94 - Siggi - first version finished
161	# 29/06/94 - Siggi - script development started
162	#--------------------------------------------------------------------------------#
163	# subjob - script for automatic generation and submission of batch-job files
164	# for various batch queuing systems
165	#--------------------------------------------------------------------------------#
166
167
168	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
169	delete_dayfile=false
170	email_notification=none
171	group_number=none
172	locat=normal
173	no_default_queue=none
174	no_submit=false
175	job_catalog="~/job_queue"
176	job_name=none
177	local_user=$LOGNAME
178	node_usage=shared
179	numprocs=0
180	punkte="..........................................................."
181	submcom=qsub
182	queue=default
183	remote_host=none
184	remote_user=""
185	verify=true
186
187	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
188	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
189
190
191
192	# ERROR HANDLING
193	# IN CASE OF EXIT:
194	trap 'if [[ $locat != normal ]]
195	then
196	case $locat in
197	(option) printf "\n --> available optios can be displayed"
198	printf " by typing:"
199	printf "\n \"subjob ?\" \n";;
200	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
201	(*) printf "\n +++ unknown error"
202	printf "\n please inform S. Raasch!\n"
203	esac
204	[[ -f $job_to_send ]] && rm $job_to_send
205	printf "\n\n+++ SUBJOB killed \n\n"
206	fi' exit
207
208
209	# IN CASE OF TERMINAL-BREAK:
210	trap '[[ -f $job_to_send ]] && rm $job_to_send
211	printf "\n\n+++ SUBJOB killed \n\n"
212	exit
213	' 2
214
215
216	# DETERMINE NAME OF LOCAL HOST
217	local_host=$(hostname)
218
219	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
220	# IS ADMITTED AT ALL)
221	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
222	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
223	case $local_host in
224	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
225	(atmos) local_address=172.20.25.35; local_host=lcide;;
226	(austru) local_address=130.75.105.128; local_host=lcmuk;;
227	(autan) local_address=130.75.105.57; local_host=lcmuk;;
228	(bora) local_address=130.75.105.103; local_host=lcmuk;;
229	(a0\|b0) local_address=133.5.4.33; local_host=lckyuh;;
230	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
231	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
232	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
233	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
234	(breva) local_address=130.75.105.98; local_host=lcmuk;;
235	(buran) local_address=130.75.105.58; local_host=lcmuk;;
236	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
237	(climate*) local_address=165.132.26.68; local_host=lcyon;;
238	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
239	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
240	(hpcl*) local_address=eddy.hpc.uni-oldenburg.de; local_host=lceddy;;
241	(cfd*) local_address=eddy.hpc.uni-oldenburg.de; local_host=lceddy;;
242	(node*) local_address=165.132.26.61 local_host=lck;;
243	# (node*) local_address=210.219.61.8 local_host=lckordi;;
244	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
245	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
246	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
247	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
248	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
249	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
250	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
251	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
252	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
253	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
254	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
255	(sno) local_address=130.75.105.113; local_host=lcmuk;;
256	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
257	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
258	(login*) local_address=118.128.66.201; local_host=lckiaps;;
259	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
260	(meller) local_address=134.106.74.155; local_host=lcfor;;
261	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
262	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
263	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
264	(nobel*) local_address=150.183.5.101; local_host=ibms;;
265	(ocean) local_address="ocean"; local_host=lcocean;;
266	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
267	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
268	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
269	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
270	(pingui) local_address=134.106.74.118; local_host=lcfor;;
271	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
272	(rte*) local_address=133.5.185.60; local_host=lcrte;;
273	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
274	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
275	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
276	(solano) local_address=130.75.105.110; local_host=lcmuk;;
277	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
278	(tc*) local_address="ocean"; local_host=lcocean;;
279	(t2a*) local_address=10.1.6.165; local_host=lctit;;
280	(urban*) local_address=147.46.30.151 local_host=lcsb;;
281	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
282	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
283	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
284	(*.cluster) local_address=192.168.1.254; local_host=lcgeohu;;
285	(*) printf "\n +++ \"$local_host\" unknown";
286	printf "\n please contact the PALM group at IMUK";
287	locat=parameter; exit;;
288	esac
289
290
291
292	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
293	remote_host=$local_host
294
295
296
297
298	# READ THE SHELLSCRIPT-OPTIONS
299	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
300	do
301	case $option in
302	(c) job_catalog=$OPTARG;;
303	(d) delete_dayfile=true;;
304	(D) no_submit=true;;
305	(e) email_notification=$OPTARG;;
306	(g) group_number=$OPTARG;;
307	(h) remote_host=$OPTARG;;
308	(m) memory=$OPTARG;;
309	(n) job_name=$OPTARG;;
310	(N) node_usage=$OPTARG;;
311	(O) threads_per_task=$OPTARG;;
312	(P) scp_port=$OPTARG;;
313	(q) no_default_queue=$OPTARG;;
314	(t) cputime=$OPTARG;;
315	(T) tasks_per_node=$OPTARG;;
316	(u) remote_user=$OPTARG;;
317	(v) verify=false;;
318	(X) numprocs=$OPTARG;;
319	(\?) printf "\n +++ Option $OPTARG unknown \n";
320	locat=option; exit;;
321	esac
322	done
323
324
325	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
326	(( to_shift = $OPTIND - 1 ))
327	shift $to_shift; file_to_send=$1
328
329
330	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
331	if [ "$1" = "?" ]
332	then
333	(printf "\n *** subjob can be called as follows:\n"
334	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
335	printf "\n Description of available options:\n"
336	printf "\n Option Description Default-Value"
337	printf "\n -c job-input- and output-catalog ~/job_queue"
338	printf "\n -d no job-protocol will be created ---"
339	printf "\n -D only the job-file will be created ---"
340	printf "\n -h execution host, available hosts: $remote_host"
341	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
342	printf "\n lckiaps, lctit"
343	printf "\n -m memory demand per process in MByte ---"
344	printf "\n -n jobname <jobdatei>"
345	printf "\n -O threads per task (for OpenMP usage) 1"
346	printf "\n -P ssh/scp port default port"
347	printf "\n -q job-queue to be used default"
348	printf "\n -t allowed cpu-time in seconds ---"
349	printf "\n -T tasks per node (on parallel hosts) ---"
350	printf "\n -u username on execution host from .netrc"
351	printf "\n -v no prompt for confirmation ---"
352	printf "\n -X # of processors (on parallel hosts) 1"
353	printf "\n "
354	printf "\n The only possible positional parameter is <jobfile>:"
355	printf "\n The complete NQS-job must be provided here."
356	printf "\n <jobfile>=? creates this outline\n\n") \| more
357	exit
358	fi
359
360
361
362	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
363	if [[ "$file_to_send" = "" ]]
364	then
365	printf "\n +++ job-file missing"
366	locat=parameter; exit
367	else
368	if [[ -f $file_to_send ]]
369	then
370	true
371	else
372	printf "\n +++ job-file: "
373	printf "\n $file_to_send"
374	printf "\n does not exist"
375	locat=parameter; exit
376	fi
377	fi
378
379
380
381	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
382	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
383	if [[ $job_name = none ]]
384	then
385	job_name=$file_to_send
386	fi
387	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
388	then
389	printf "\n +++ job-file name: "
390	printf "\n $job_name"
391	printf "\n must not contain \"/\"-characters"
392	locat=parameter; exit
393	fi
394
395
396
397
398	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
399	# OR IF NO HOST HAS BEEN GIVEN
400	if [[ $remote_host = none ]]
401	then
402	printf "\n +++ host missing"
403	locat=option; exit
404	else
405	case $remote_host in
406	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
407	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
408	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
409	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
410	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
411	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
412	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
413	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
414	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
415	(lceddy) remote_address=eddy.hpc.uni-oldenburg.de; submcom=sbatch;;
416	(lcgeohu) remote_address=cirrus.geo.hu-berlin.de; submcom=sbatch;;
417	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
418	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
419	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
420	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
421	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
422	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
423	(lcocean) remote_address="ocean"; submcom=qsub;;
424	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
425	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
426	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
427	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
428	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
429	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
430	locat=parameter; exit;;
431	esac
432	fi
433
434
435	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
436	if [[ $no_default_queue != none ]]
437	then
438	error=false
439	ndq=$no_default_queue
440	case $remote_host in
441	(ibm) case $ndq in
442	(p690_express\|p690_standard\|p690_long) error=false;;
443	(*) error=true;;
444	esac;;
445	(ibmh) case $ndq in
446	(cluster\|express) error=false;;
447	(*) error=true;;
448	esac;;
449	(ibmkisti) case $ndq in
450	(class.32plus\|class.1-2\|class.2-32) error=false;;
451	(*) error=true;;
452	esac;;
453	(ibmku) case $ndq in
454	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
455	(*) error=true;;
456	esac;;
457	(ibms) case $ndq in
458	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
459	(*) error=true;;
460	esac;;
461	(lcbullhh) case $ndq in
462	(compute\|compute2\|shared) error=false;;
463	(*) error=true;;
464	esac;;
465	(lccrayb) case $ndq in
466	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
467	(*) error=true;;
468	esac;;
469	(lccrayh) case $ndq in
470	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
471	(*) error=true;;
472	esac;;
473	(lccrayf) case $ndq in
474	(usup\|test*\|small\|large) error=false;;
475	(*) error=true;;
476	esac;;
477	(lceddy) case $ndq in
478	(eddy.p\|cfdh.p\|cfdl.p\|carl.p\|mpcs.p\|mpcl.p\|mpcb.p\|all_nodes.p) error=false;;
479	(*) error=true;;
480	esac;;
481	(lckiaps) case $ndq in
482	(express\|normal\|normal20\|quickq) error=false;;
483	(*) error=true;;
484	esac;;
485	(lckyoto) case $ndq in
486	(eh\|ph) error=false;;
487	(*) error=true;;
488	esac;;
489	(lckyuh) case $ndq in
490	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
491	(*) error=true;;
492	esac;;
493	(lckyut) case $ndq in
494	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
495	(*) error=true;;
496	esac;;
497	(lctit) case $ndq in
498	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
499	(*) error=true;;
500	esac;;
501	(t3eb) case $ndq in
502	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
503	(*) error=true;;
504	esac;;
505	(t3eh) case $ndq in
506	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
507	(*) error=true;;
508	esac;;
509	(t3ej2\|t3ej5) case $ndq in
510	(low\|normal\|high) error=false;;
511	(*) error=true;;
512	esac;;
513	(t3es) case $ndq in
514	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
515	(*) error=true;;
516	esac;;
517	esac
518	if [[ $error = true ]]
519	then
520	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
521	locat=parameter; exit
522	else
523	queue=$no_default_queue
524	fi
525	fi
526
527
528
529	# CHECK THE CPU-TIME
530	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
531	done=false
532	while [[ $done = false ]]
533	do
534	if (( $cputime <= 0 ))
535	then
536	printf "\n +++ wrong cpu-time or cpu-time missing"
537	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
538	printf "\n >>> "
539	read cputime 1>/dev/null 2>&1
540	else
541	done=true
542	fi
543	done
544	(( stunden = cputime / 3600 ))
545	(( resttime = cputime - stunden * 3600 ))
546	(( minuten = resttime / 60 ))
547	(( sekunden = resttime - minuten * 60 ))
548	timestring=${stunden}:${minuten}:${sekunden}
549
550
551
552	# CHECK THE MEMORY DEMAND
553	done=false
554	while [[ $done = false ]]
555	do
556	if (( memory <= 0 ))
557	then
558	printf "\n +++ wrong memory demand or memory demand missing"
559	printf "\n >>> Please type memory in MByte per process as INTEGER:"
560	printf "\n >>> "
561	read memory 1>/dev/null 2>&1
562	else
563	done=true
564	fi
565	done
566
567	if [[ $remote_host = lctit ]]
568	then
569	(( Memory = memory * tasks_per_node / 1000 ))
570	fi
571
572
573	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
574	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
575	then
576	(( memory = memory * threads_per_task ))
577	fi
578
579
580	# CALCULATE NUMBER OF REQUIRED NODES
581	if (( tasks_per_node != 0 ))
582	then
583	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
584	fi
585
586
587	# CALCULATE NUMBER OF PROCESSES PER NODE
588	(( processes_per_node = tasks_per_node * threads_per_task ))
589
590
591	# CALCULATE NUMBER OF MPI TASKS
592	(( mpi_tasks = numprocs / threads_per_task ))
593
594
595	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
596	if [[ "$scp_port" != "" ]]
597	then
598	PORTOPT="-P $scp_port"
599	SSH_PORTOPT="-p $scp_port"
600	fi
601
602
603	# HEADER-OUTPUT
604	if [[ $verify = true ]]
605	then
606	printf "\n\n"
607	printf "#--------------------------------------------------------------# \n"
608	spalte1=SUBJOB;spalte2=$(date)
609	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
610	printf "\| \| \n"
611	printf "\| values of parameters/options: \| \n"
612	spalte1=$(echo local_host$punkte \| cut -c-20)
613	spalte2=$punkte$local_host
614	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
615	spalte1=$(echo remote_host$punkte \| cut -c-20)
616	spalte2=$punkte$remote_host
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	spalte1=$(echo queue$punkte \| cut -c-20)
619	spalte2=$punkte$queue
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	spalte1=$(echo memory$punkte \| cut -c-20)
622	spalte2="$punkte$memory mb"
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	spalte1=$(echo cputime$punkte \| cut -c-20)
625	spalte2="$punkte$cputime sec"
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	spalte1=$(echo job_name$punkte \| cut -c-20)
628	spalte2="$punkte$job_name"
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	printf "#--------------------------------------------------------------# \n\n"
631
632
633	# QUERY CHECK
634	antwort="dummy"
635	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
636	do
637	read antwort?" >>> continue (y/n) ? "
638	done
639	if [[ $antwort = n \|\| $antwort = N ]]
640	then
641	locat=verify; exit
642	fi
643	printf "\n"
644	fi
645
646	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
647	identifier=$RANDOM
648	job_on_remhost=${job_name}_${identifier}_$local_host
649	job_to_send=job_to_send_$identifier
650	if [[ $delete_dayfile = false ]]
651	then
652	remote_dayfile=${local_host}_${job_name}_result_$identifier
653	local_dayfile=${remote_host}_${job_name}
654	else
655	remote_dayfile=/dev/null
656	fi
657
658
659	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
660	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
661	then
662
663	# GENERAL LOADLEVELER SETTINGS
664	execute_in_shell="#!/bin/ksh"
665	use_shell="# @ shell = /bin/ksh"
666	consumable_memory="ConsumableMemory($memory mb)"
667	class="# @ class = $queue"
668	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
669	network_to_use="# @ network.mpi = sn_all,shared,us"
670	data_limit="# @ data_limit = 1.76gb"
671	image_size="# @ image_size = 50"
672	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
673
674	if [[ $email_notification = none ]]
675	then
676	notify_user=""
677	else
678	notify_user="# @ notify_user = $email_notification"
679	if [[ $delete_dayfile = true ]]
680	then
681	notification='# @ notification = never'
682	fi
683	fi
684
685	if [[ $remote_host = ibmh ]]
686	then
687	data_limit=""
688	network_to_use=""
689	class="# @ class = $queue"
690	environment=""
691	rset="# @ rset = RSET_MCM_AFFINITY"
692	task_affinity="# @ task_affinity = core(1)"
693	elif [[ $remote_host = ibmkisti ]]
694	then
695	network_to_use="# @ network.MPI = sn_all,shared,US"
696	wall_clock_limit="# @ wall_clock_limit = $timestring"
697	if [[ $threads_per_task = 1 ]]
698	then
699	rset="# @ rset = RSET_MCM_AFFINITY"
700	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
701	fi
702	environment=""
703	use_shell=""
704	data_limit=""
705	image_size=""
706	elif [[ $remote_host = ibmku ]]
707	then
708	execute_in_shell="#!/usr/bin/ksh"
709	use_shell="# @ shell = /usr/bin/ksh"
710	consumable_memory=""
711	environment=""
712	network_to_use="# @ network.mpi = sn_all,shared,us"
713	data_limit=""
714	image_size=""
715	elif [[ $remote_host = ibms ]]
716	then
717	network_to_use="# @ network.mpi = csss,shared,us"
718	fi
719
720	cat > $job_to_send << %%END%%
721	$execute_in_shell
722	$use_shell
723
724	# @ job_type = parallel
725	# @ job_name = $job_name
726	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
727	# @ output = $remote_dayfile
728	# @ error = $remote_dayfile
729	$wall_clock_limit
730	$image_size
731	$class
732	$environment
733	$network_to_use
734	$data_limit
735	$rset
736	$mcm_affinity_options
737	$task_affinity
738	$notification
739	$notify_user
740
741	%%END%%
742
743	if (( nodes > 0 ))
744	then
745
746	if [[ $remote_host != ibmkisti ]]
747	then
748
749	cat >> $job_to_send << %%END%%
750	# @ node = $nodes
751	# @ tasks_per_node = $processes_per_node
752	# @ node_usage = $node_usage
753	# @ queue
754
755	%%END%%
756
757	else
758
759	cat >> $job_to_send << %%END%%
760	# @ total_tasks = $mpi_tasks
761	# @ blocking = unlimited
762	# @ queue
763
764	%%END%%
765
766	fi
767
768	else
769
770	cat >> $job_to_send << %%END%%
771	# @ blocking = unlimited
772	# @ total_tasks = $numprocs
773	# @ node_usage = $node_usage
774	# @ queue
775
776	%%END%%
777
778	fi
779
780	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
781	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
782	then
783	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
784	fi
785
786	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
787	then
788
789	cat > $job_to_send << %%END%%
790	#!/bin/ksh
791
792	# @ job_type = serial
793	# @ node_usage = $node_usage
794	# @ job_name = palm
795	# @ wall_clock_limit = ${timestring},$timestring
796	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
797	# @ output = $remote_dayfile
798	# @ error = $remote_dayfile
799	$class
800	$notification
801
802	# @ queue
803
804	%%END%%
805
806	elif [[ $remote_host = lcbullhh ]]
807	then
808	if [[ $numprocs != 0 ]]
809	then
810	cat > $job_to_send << %%END%%
811	#!/bin/bash -l
812	#SBATCH -J $job_name
813	#SBATCH -t $timestring
814	#SBATCH -N $nodes
815	#SBATCH --ntasks-per-node=$processes_per_node
816	#SBATCH -p $queue
817	#SBATCH -o $remote_dayfile
818	#SBATCH -e $remote_dayfile
819	#SBATCH -A $project_account
820
821	$init_cmds
822	$module_calls
823
824	%%END%%
825
826	else
827	cat > $job_to_send << %%END%%
828	#!/bin/bash -l
829	#SBATCH -J $job_name
830	#SBATCH -t $timestring
831	#SBATCH -l ncpus=1
832	#SBATCH -l pmem=${memory}mb
833	#SBATCH -m abe
834	#SBATCH -o $remote_dayfile
835	#SBATCH -e $remote_dayfile
836	#SBATCH -A $project_account
837
838	$init_cmds
839	$module_calls
840
841	%%END%%
842
843	fi
844
845	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
846	then
847
848	if [[ "$feature" != "" ]]
849	then
850	featuredir="#PBS -l feature=$feature"
851	fi
852
853	if [[ $numprocs != 0 ]]
854	then
855	cat > $job_to_send << %%END%%
856	#!/bin/bash
857	#PBS -N $job_name
858	#PBS -l walltime=$timestring
859	#PBS -l nodes=$nodes:ppn=$processes_per_node
860	#PBS -o $remote_dayfile
861	#PBS -j oe
862	#PBS -q $queue
863	$featuredir
864
865	$init_cmds
866	$module_calls
867
868	%%END%%
869
870	else
871
872	continue
873
874	fi
875
876	elif [[ $remote_host = lccrayf ]]
877	then
878
879	if [[ $numprocs != 0 ]]
880	then
881	cat > $job_to_send << %%END%%
882	#!/bin/bash -l
883	#SBATCH -J $job_name
884	#SBATCH -t $timestring
885	#SBATCH -N $nodes
886	#SBATCH --ntasks-per-node=$processes_per_node
887	#SBATCH -p $queue
888	#SBATCH -o $remote_dayfile
889	#SBATCH -e $remote_dayfile
890
891	$init_cmds
892	$module_calls
893
894	%%END%%
895
896	else
897	cat > $job_to_send << %%END%%
898	#!/bin/bash -l
899	#SBATCH -J $job_name
900	#SBATCH -t $timestring
901	#SBATCH -l ncpus=1
902	#SBATCH -l pmem=${memory}mb
903	#SBATCH -m abe
904	#SBATCH -o $remote_dayfile
905	#SBATCH -e $remote_dayfile
906
907	$init_cmds
908	$module_calls
909
910	%%END%%
911
912	fi
913
914
915
916	elif [[ $remote_host = lceddy ]]
917	then
918	cat > $job_to_send << %%END%%
919	#!/bin/bash
920	#SBATCH -J $job_name
921	#SBATCH -t $timestring
922	#SBATCH -n $numprocs
923	#SBATCH -N $nodes
924	#SBATCH --cpus-per-task 1
925	#SBATCH -p $queue
926	#SBATCH -o $remote_dayfile
927	#SBATCH -e $remote_dayfile
928	#SBATCH --mem-per-cpu $memory
929	#SBATCH --exclusive
930
931	module load PALMDependencies/.gcc
932
933	$init_cmds
934	$module_calls
935
936	%%END%%
937
938	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
939	then
940
941	if [[ $numprocs != 0 ]]
942	then
943	cat > $job_to_send << %%END%%
944	#!/bin/ksh
945	#PBS -N $job_name
946	#PBS -l walltime=$timestring
947	#PBS -l ncpus=$numprocs
948	#PBS -l pmem=${memory}mb
949	#PBS -o $remote_dayfile
950	#PBS -l nodes=$nodes:ppn=${processes_per_node}
951	#PBS -j oe
952
953	mpd &
954
955	%%END%%
956
957	else
958	cat > $job_to_send << %%END%%
959	#!/bin/ksh
960	#PBS -N $job_name
961	#PBS -l walltime=$timestring
962	#PBS -l ncpus=1
963	#PBS -l pmem=${memory}mb
964	#PBS -o $remote_dayfile
965	#PBS -j oe
966
967	%%END%%
968
969	fi
970
971	elif [[ $remote_host = lcgeohu ]]
972	then
973
974	if [[ $email_notification = none ]]
975	then
976	notify_user=""
977	else
978	notify_user="#SBATCH --mail-type=ALL"
979	fi
980
981	if [[ $numprocs != 0 ]]
982	then
983	cat > $job_to_send << %%END%%
984	#!/bin/bash
985	#SBATCH --job-name=$job_name
986	#SBATCH --ntasks=$processes_per_node
987	#SBATCH --time=$timestring
988	#SBATCH --output=$remote_dayfile
989	#SBATCH --error=$remote_dayfile
990	#SBATCH --qos=short
991	$notify_user
992
993	$init_cmds
994	$module_calls
995
996	%%END%%
997
998	else
999	cat > $job_to_send << %%END%%
1000	#!/bin/bash
1001	#SBATCH --job-name=$job_name
1002	#SBATCH --ntasks=$processes_per_node
1003	#SBATCH --output=$remote_dayfile
1004	#SBATCH --error=$remote_dayfile
1005	#SBATCH -l ncpus=1
1006
1007	$init_cmds
1008	$module_calls
1009
1010	%%END%%
1011
1012	fi
1013
1014	elif [[ $remote_host = lckiaps ]]
1015	then
1016
1017	if [[ $numprocs != 0 ]]
1018	then
1019	cat > $job_to_send << %%END%%
1020	#!/bin/bash
1021	#PBS -N $job_name
1022	#PBS -l walltime=$timestring
1023	#PBS -l select=1:ncpus=$numprocs
1024	#PBS -l pmem=${memory}mb
1025	#PBS -q $queue
1026	#PBS -o $remote_dayfile
1027	#PBS -j oe
1028	#PBS -V
1029
1030	%%END%%
1031
1032	else
1033	cat > $job_to_send << %%END%%
1034	#!/bin/bash
1035	#PBS -N $job_name
1036	#PBS -l walltime=$timestring
1037	#PBS -l ncpus=1
1038	#PBS -l pmem=${memory}mb
1039	#PBS -o $remote_dayfile
1040	#PBS -j oe
1041
1042	%%END%%
1043
1044	fi
1045
1046	elif [[ $remote_host = lcyon ]]
1047	then
1048
1049	if [[ $numprocs != 0 ]]
1050	then
1051	cat > $job_to_send << %%END%%
1052	#!/bin/ksh
1053	#PBS -N $job_name
1054	#PBS -l walltime=$timestring
1055	#PBS -l ncpus=$numprocs
1056	#PBS -l pmem=${memory}mb
1057	#PBS -o $remote_dayfile
1058	#PBS -j oe
1059
1060	%%END%%
1061
1062	else
1063	cat > $job_to_send << %%END%%
1064	#!/bin/ksh
1065	#PBS -N $job_name
1066	#PBS -l walltime=$timestring
1067	#PBS -l ncpus=1
1068	#PBS -l pmem=${memory}mb
1069	#PBS -o $remote_dayfile
1070	#PBS -j oe
1071
1072	%%END%%
1073
1074	fi
1075
1076	elif [[ $remote_host = lcxe6 ]]
1077	then
1078
1079	if [[ $numprocs != 0 ]]
1080	then
1081	cat > $job_to_send << %%END%%
1082	#!/bin/ksh
1083	#PBS -S /bin/ksh
1084	#PBS -N $job_name
1085	#PBS -A $project_account
1086	#PBS -j oe
1087	#PBS -l walltime=$timestring
1088	#PBS -l mppwidth=${numprocs}
1089	#PBS -l mppnppn=${processes_per_node}
1090	#PBS -m abe
1091	#PBS -o $remote_dayfile
1092	$email_directive
1093
1094	$init_cmds
1095	$module_calls
1096
1097	%%END%%
1098
1099	else
1100	cat > $job_to_send << %%END%%
1101	#!/bin/ksh
1102	#PBS -S /bin/ksh
1103	#PBS -N $job_name
1104	#PBS -A $project_account
1105	#PBS -j oe
1106	#PBS -l walltime=$timestring
1107	#PBS -l ncpus=1
1108	#PBS -l pmem=${memory}mb
1109	#PBS -m abe
1110	$email_directive
1111	#PBS -o $remote_dayfile
1112
1113	$init_cmds
1114	$module_calls
1115
1116	%%END%%
1117
1118	fi
1119
1120	elif [[ $remote_host = lckyoto ]]
1121	then
1122
1123	cat > $job_to_send << %%END%%
1124	#!/bin/ksh
1125	# @\$-o $remote_dayfile
1126	# @\$-eo -oi
1127	# @\$-lP 16
1128	# @\$-lp 1
1129	# @\$-lm 28gb -llm unlimited -ls unlimited
1130	# @\$-q $queue
1131	# @\$-Pvn abs_pack
1132	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1133	#. /thin/local/etc/setprofile/intel-11.0.sh
1134	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1135	. ~/.myprofile
1136	#. /home2/t/t51254/palm/current_version/myprofile
1137	#. /thin/apps/pgi/mpi.sh
1138	#
1139	env
1140	#
1141	set -x
1142
1143	%%END%%
1144
1145	elif [[ $remote_host = lcxt5m ]]
1146	then
1147
1148	if [[ $numprocs != 0 ]]
1149	then
1150	cat > $job_to_send << %%END%%
1151	#!/bin/ksh
1152	#PBS -S /bin/ksh
1153	#PBS -N $job_name
1154	#PBS -j oe
1155	#PBS -l walltime=$timestring
1156	#PBS -l mppwidth=${numprocs}
1157	#PBS -l mppnppn=${processes_per_node}
1158	#PBS -m abe
1159	#PBS -o $remote_dayfile
1160
1161	$init_cmds
1162	$module_calls
1163
1164	%%END%%
1165
1166	else
1167	cat > $job_to_send << %%END%%
1168	#!/bin/ksh
1169	#PBS -S /bin/ksh
1170	#PBS -N $job_name
1171	#PBS -j oe
1172	#PBS -l walltime=$timestring
1173	#PBS -l ncpus=1
1174	#PBS -l pmem=${memory}mb
1175	#PBS -m abe
1176	#PBS -o $remote_dayfile
1177
1178	$init_cmds
1179	$module_calls
1180
1181	%%END%%
1182
1183	fi
1184
1185	elif [[ $remote_host = lckyuh ]]
1186	then
1187	cat > $job_to_send << %%END%%
1188	#!/bin/bash
1189	#PJM -L "rscgrp=$queue"
1190	#PJM -L "node=$nodes"
1191	#PJM --mpi "proc=$numprocs"
1192	#PJM -L "elapse=$timestring"
1193	#PJM -o $remote_dayfile
1194	#PJM -j
1195	#PJM -X
1196	#PJM --no-stging
1197
1198	export LANG=en_US.UTF-8
1199	%%END%%
1200
1201	elif [[ $remote_host = lckyut ]]
1202	then
1203	cat > $job_to_send << %%END%%
1204	#!/bin/bash
1205	#PJM -L "rscgrp=$queue"
1206	#PJM -L "vnode=$numprocs"
1207	#PJM -L "vnode-core=1"
1208	#PJM -L "elapse=$timestring"
1209	#PJM --mpi proc=$numprocs
1210	#PJM -o $remote_dayfile
1211	#PJM -j
1212	#PJM -X
1213	#PJM --no-stging
1214
1215	export LANG=en_US.UTF-8
1216	%%END%%
1217
1218	elif [[ $remote_host = lcocean ]]
1219	then
1220	cat > $job_to_send << %%END%%
1221	#!/bin/bash
1222	#$ -cwd
1223	#$ -V
1224	#$ -N $job_name
1225	#$ -pe orte $numprocs
1226	#$ -o $remote_dayfile
1227	#$ -j y
1228	#$ -R y
1229	$init_cmds
1230	$module_calls
1231
1232	%%END%%
1233
1234	elif [[ $remote_host = lctit ]]
1235	then
1236	cat > $job_to_send << %%END%%
1237	#!/bin/ksh
1238	$init_cmds
1239	$module_calls
1240
1241	%%END%%
1242
1243	# SET OPTIONS FOR SUBMIT-COMMAND
1244	if [[ $tasks_per_node != $processes_per_node ]]
1245	then
1246	submcom="$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 $queue "
1247	else
1248	submcom="$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 "
1249	fi
1250
1251	else
1252
1253	cat > $job_to_send << %%END%%
1254	# @\$-q ${queue}
1255	# @\$-l${qsubtime} $timestring
1256	# @\$-l${qsubmem} ${memory}mb
1257	# @\$-o $remote_dayfile
1258	# @\$-eo
1259
1260	%%END%%
1261
1262	fi
1263
1264
1265	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1266	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1267	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1268	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1269	then
1270	echo "set +vx" >> $job_to_send
1271	echo "trap '" >> $job_to_send
1272	echo "set +vx" >> $job_to_send
1273	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 = lckiaps \|\| $remote_host = lckyu* \|\| $remote_host = lcxe6 \|\| $remote_host = lcocean \|\| $remote_host = lcgeohu ]]
1274	then
1275	if [[ $remote_host = ibmh ]]
1276	then
1277	return_queue=c1
1278	elif [[ $remote_host = ibmkisti ]]
1279	then
1280	return_queue=class.1-2
1281	elif [[ $remote_host = ibmku ]]
1282	then
1283	return_queue=sdbg2
1284	elif [[ $remote_host = ibms ]]
1285	then
1286	return_queue=p_normal
1287	elif [[ $remote_host = lcbullhh ]]
1288	then
1289	return_queue=shared
1290	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1291	then
1292	return_queue=dataq
1293	elif [[ $remote_host = lcxe6 ]]
1294	then
1295	return_queue=debug
1296	elif [[ $remote_host = lckiaps ]]
1297	then
1298	return_queue=express
1299	elif [[ $remote_host = lckyuh ]]
1300	then
1301	return_queue=cx-single
1302	elif [[ $remote_host = lckyut ]]
1303	then
1304	return_queue=cx-single
1305	else
1306	return_queue=unknown
1307	fi
1308
1309	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1310	then
1311
1312	if [[ $remote_host = ibmku ]]
1313	then
1314	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1315	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1316	else
1317	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1318	fi
1319	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1320	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1321	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1322	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1323	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1324	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1325	if [[ $host != "ibmh" ]]
1326	then
1327	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1328	fi
1329	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1330	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1331
1332	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1333	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1334
1335	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1336	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
1337	if [[ $remote_host = ibmku ]]
1338	then
1339	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1340	fi
1341	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1342
1343	elif [[ $remote_host = lcbullhh ]]
1344	then
1345	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1346	echo "#!/bin/bash" >> $job_to_send
1347	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1348	echo "#SBATCH -t 00:20:00" >> $job_to_send
1349	echo "#SBATCH -N 1" >> $job_to_send
1350	echo "#SBATCH -n 1" >> $job_to_send
1351	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1352	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1353	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1354	echo "#SBATCH -A $project_account" >> $job_to_send
1355	echo "#SBATCH -p $return_queue" >> $job_to_send
1356	echo " " >> $job_to_send
1357	echo "set -x" >> $job_to_send
1358	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1359	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1360	echo "%%END%%" >> $job_to_send
1361
1362	elif [[ $remote_host = lckyuh ]]
1363	then
1364	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1365	echo "#!/bin/bash" >> $job_to_send
1366	echo "#PJM -L \"node=1\"" >> $job_to_send
1367	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1368	echo "#PJM --no-stging" >> $job_to_send
1369	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1370	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1371	echo "#PJM -j" >> $job_to_send
1372	echo " " >> $job_to_send
1373	echo "export LANG=en_US.UTF-8" >> $job_to_send
1374	echo "set -x" >> $job_to_send
1375	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1376	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1377	echo "%%END%%" >> $job_to_send
1378
1379	elif [[ $remote_host = lckyut ]]
1380	then
1381	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1382	echo "#!/bin/bash" >> $job_to_send
1383	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1384	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1385	echo "#PJM --no-stging" >> $job_to_send
1386	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1387	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1388	echo "#PJM -j" >> $job_to_send
1389	echo " " >> $job_to_send
1390	echo "export LANG=en_US.UTF-8" >> $job_to_send
1391	echo "set -x" >> $job_to_send
1392	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1393	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1394	echo "%%END%%" >> $job_to_send
1395
1396	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1397	then
1398	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1399	echo "#!/bin/bash" >> $job_to_send
1400	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1401	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1402	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1403	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1404	echo "#PBS -j oe" >> $job_to_send
1405	echo " " >> $job_to_send
1406	echo "set -x" >> $job_to_send
1407	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1408	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1409	echo "%%END%%" >> $job_to_send
1410
1411	elif [[ $remote_host = lcocean ]]
1412	then
1413	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1414	echo "#!/bin/bash" >> $job_to_send
1415	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1416	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1417	echo "SGEPREFIX -cwd" >> $job_to_send
1418	echo "SGEPREFIX -j y" >> $job_to_send
1419	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1420	echo " " >> $job_to_send
1421	echo "set -x" >> $job_to_send
1422	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1423	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1424	echo "" >> $job_to_send
1425	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1426	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1427	echo "rm -f scpjob.${identifier}" >> $job_to_send
1428	echo "%%END%%" >> $job_to_send
1429	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1430	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1431
1432	elif [[ $remote_host = lcxe6 ]]
1433	then
1434	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1435	echo "#!/bin/ksh" >> $job_to_send
1436	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1437	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1438	echo "#PBS -A $project_account" >> $job_to_send
1439	echo "#PBS -l mppwidth=1" >> $job_to_send
1440	echo "#PBS -l mppnppn=1" >> $job_to_send
1441	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1442	echo "#PBS -j oe" >> $job_to_send
1443	echo " " >> $job_to_send
1444	echo "set -x" >> $job_to_send
1445	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1446	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1447	echo "%%END%%" >> $job_to_send
1448
1449	elif [[ $remote_host = lcgeohu ]]
1450	then
1451	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1452	echo "#!/bin/bash" >> $job_to_send
1453	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1454	echo "#SBATCH -t 00:20:00" >> $job_to_send
1455	echo "#SBATCH -N 1" >> $job_to_send
1456	echo "#SBATCH -n 1" >> $job_to_send
1457	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1458	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1459	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1460	echo " " >> $job_to_send
1461	echo "set -x" >> $job_to_send
1462	echo "sbatch scpjob.$identifier" >> $job_to_send
1463	echo "%%END%%" >> $job_to_send
1464
1465	else
1466
1467	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1468	echo "# @\\\$-q $return_queue" >> $job_to_send
1469	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1470	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1471	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1472	then
1473	echo "# @\$-l mpp_p=0" >> $job_to_send
1474	fi
1475	echo '# @\$-lF 10mb' >> $job_to_send
1476	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1477	echo '# @\\\$-eo' >> $job_to_send
1478	echo " " >> $job_to_send
1479	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1480	then
1481	echo "set +vx" >> $job_to_send
1482	echo ". .profile" >> $job_to_send
1483	fi
1484	echo "set -x" >> $job_to_send
1485	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
1486	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1487	echo "%%END%%" >> $job_to_send
1488
1489	fi
1490
1491	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1492	then
1493	echo "llsubmit scpjob.$identifier" >> $job_to_send
1494	elif [[ $remote_host = lcbullhh ]]
1495	then
1496	echo "sbatch scpjob.$identifier" >> $job_to_send
1497	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1498	then
1499	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1500	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1501	then
1502	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1503	elif [[ $remote_host = t3es ]]
1504	then
1505	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1506	elif [[ $remote_host = lckiaps ]]
1507	then
1508	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1509	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1510	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1511	elif [[ $remote_host = lckyu* ]]
1512	then
1513	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1514	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1515	elif [[ $remote_host = lcocean ]]
1516	then
1517	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1518	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1519	else
1520	echo "$submcom scpjob.$identifier" >> $job_to_send
1521	fi
1522	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1523	then
1524	echo "rm scpjob.$identifier" >> $job_to_send
1525	fi
1526	else
1527	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1528	# ??? funktioniert das ÃŒberhaupt noch ???
1529	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1530	fi
1531	echo "set -x" >> $job_to_send
1532	echo " ' exit" >> $job_to_send
1533	echo "set -x" >> $job_to_send
1534	fi
1535
1536
1537	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1538	cat $file_to_send >> $job_to_send
1539
1540	if [[ $remote_host = ibm ]]
1541	then
1542	echo " " >> $job_to_send
1543	echo "exit" >> $job_to_send
1544	fi
1545
1546	# REMOVE JOB-FILE
1547	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1548	then
1549	echo " " >> $job_to_send
1550	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1551	fi
1552
1553
1554	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1555	if [[ $no_submit = false ]]
1556	then
1557	if [[ $remote_host != $local_host ]]
1558	then
1559	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1560	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1561	then
1562	job_catalog_save=$job_catalog
1563	job_catalog=job_queue
1564	elif [[ $remote_host = lcgeohu ]]
1565	then
1566	job_catalog_save=$job_catalog
1567	job_catalog=/home/${remote_user}/job_queue
1568	fi
1569	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1570	if [[ $? = 1 ]]
1571	then
1572	locat=scp; exit
1573	fi
1574	if [[ $remote_host = ibms \|\| $remote_host = lcgeohu ]]
1575	then
1576	job_catalog=$job_catalog_save
1577	fi
1578	[[ $verify = true ]] && printf "\n >>> finished\n"
1579	else
1580	eval job_catalog=$job_catalog
1581	cp $job_to_send ${job_catalog}/$job_on_remhost
1582	fi
1583
1584
1585
1586	# START NQS- / LOADLEVELER-JOB
1587	if [[ $remote_host != $local_host ]]
1588	then
1589	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1590
1591	if [[ $remote_host = ibmku ]]
1592	then
1593	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1594	elif [[ $remote_host = lcgeohu ]]
1595	then
1596	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1597	else
1598	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1599	fi
1600
1601	[[ $verify = true ]] && printf " >>> o.k.\n"
1602	else
1603	cd $job_catalog
1604	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray \|\| $local_host = lceddy \|\| $local_host = lcgeohu ]]
1605	then
1606	eval $submcom $job_on_remhost
1607	elif [[ $local_host = lctit \|\| $local_host = lcxe6 \|\| $local_host = lck \|\| $local_host = lckordi \|\| $local_host = lcyon \|\| $local_host = lcsb \|\| $local_host = lckyu* ]]
1608	then
1609	chmod u+x $job_on_remhost
1610	eval $submcom $job_on_remhost
1611	elif [[ $local_host = lcbullhh ]]
1612	then
1613	if [[ $queue = default ]]
1614	then
1615	eval $submcom $job_on_remhost
1616	fi
1617	else
1618	qsub $job_on_remhost
1619	fi
1620
1621	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku. THIS WILL BE DONE
1622	# AT THE END OF THE JOB
1623	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcgeohu ]]
1624	then
1625	rm $job_on_remhost
1626	fi
1627	cd - > /dev/null
1628	fi
1629	fi
1630
1631	# FINAL ACTIONS
1632	if [[ $no_submit = false ]]
1633	then
1634	rm -f $job_to_send
1635	fi
1636	[[ $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 |