Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2149

Last change on this file since 2149 was 2148, checked in by scharf, 8 years ago
added header documentation for last commit
Property svn:keywords set to `Id Rev`
File size: 58.3 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	# added kuma and gharbi to the list of known hosts
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 2148 2017-02-09 16:56:42Z scharf $
30	#
31	# 2134 2017-02-02 07:33:46Z raasch
32	# option -E added to msub commands on HLRN-III machines to allow output of more
33	# job informations in the job protocol files
34	#
35	# 1944 2016-06-15 06:29:00Z raasch
36	# adjustments for using HLRN ssh-keys
37	#
38	# 1940 2016-06-14 05:15:20Z raasch
39	# adjustments for lckiaps
40	#
41	# 1866 2016-04-15 06:50:59Z raasch
42	# adjusted for lcocean
43	#
44	# 1841 2016-04-07 19:14:06Z raasch
45	# script now running under bash
46	#
47	# 1701 2015-11-02 07:43:04Z maronga
48	# Bugfix: added missing init_cmds for lccrayh/lccrayb
49	#
50	# 1621 2015-07-17 11:39:33Z heinze
51	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
52	#
53	# 1575 2015-03-27 09:56:27Z raasch
54	# mpp2-queues added to lccrayh
55	#
56	# 1547 2015-01-29 15:09:12Z witha
57	# adjustments for ForWind computing cluster (lcflow)
58	#
59	# 1545 2015-01-29 06:52:23Z heinze
60	# local host name for blizzard further specified
61	#
62	# 1480 2014-10-17 14:41:49Z raasch
63	# adjustments for 2nd stage of HLRNIII
64	#
65	# 1468 2014-09-24 14:06:57Z maronga
66	# Typo removed (addres->address)
67	# Adjustments for lcxe6
68	#
69	# 1452 2014-08-22 09:41:06Z heinze
70	# local hosts for blizzard added
71	#
72	# 1450 2014-08-21 07:31:51Z heinze
73	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
74	#
75	# 1442 2014-07-28 07:09:10Z raasch
76	# HLRN-III (lccrayb/lccrayh) queues adjusted
77	#
78	# 1378 2014-04-28 06:04:58Z raasch
79	# -et option added for lctit
80	#
81	# 1350 2014-04-04 13:01:30Z maronga
82	# location of qsub updated for lcxe6
83	#
84	# 1289 2014-03-04 07:12:34Z raasch
85	# German comments translated to English
86	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
87	#
88	# 1279 2014-01-28 12:10:14Z raasch
89	# node calculation modified due to changes in mrun (tasks_per_node must not be
90	# an integral divisor of numprocs any more)
91	#
92	# 1274 2014-01-09 13:14:54Z heinze
93	# adjustments for lccrayh
94	#
95	# 1266 2013-12-11 12:07:34Z heinze
96	# further adjustments for lccrayb (use msub instead of qsub)
97	#
98	# 1264 2013-12-09 12:46:09Z fricke
99	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
100	#
101	# 1262 2013-12-09 10:57:20Z fricke
102	# further adjustments for lccrayb
103	#
104	# 1260 2013-12-04 12:48:04Z raasch
105	# jaboticaba admitted
106	#
107	# 1255 2013-11-07 14:43:35Z raasch
108	# further adjustments for lccrayb
109	#
110	# 1224 2013-09-16 07:27:23Z raasch
111	# first adjustments for lccrayb
112	#
113	# 1202 2013-07-10 16:22:07Z witha
114	# adjustments for Forwind cluster (lcflow)
115	#
116	# 1199 2013-07-05 14:52:22Z raasch
117	# adjustments for CSC Helsinki (lccrayf)
118	#
119	# use of cluster/express queue enabled (ibmh)
120	# vinessa added (imuk)
121	#
122	# 1103 2013-02-20 02:15:53Z raasch
123	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
124	# instead typeset -L/R),
125	# further adjustments for lckyuh
126	#
127	# 2013-02-10 01:47:43Z raasch
128	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
129	# and for Forwind cluster (lcflow)
130	#
131	# 1094 2013-02-03 01:52:12Z raasch
132	# new option -P for explicit setting of ssh/scp port,
133	# decalpha parts (yonsei) removed
134	#
135	# 2013-02-02 07:06:13Z raasch
136	# adjustments for Kyushu-University computing center (lckyut - tatara)
137	# old changelog messages removed
138	#
139	# 1046 2012-11-09 14:38:45Z maronga
140	# code put under GPL (PALM 3.9)
141	#
142	# 08/07/94 - Siggi - first version finished
143	# 29/06/94 - Siggi - script development started
144	#--------------------------------------------------------------------------------#
145	# subjob - script for automatic generation and submission of batch-job files
146	# for various batch queuing systems
147	#--------------------------------------------------------------------------------#
148
149
150	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
151	delete_dayfile=false
152	email_notification=none
153	group_number=none
154	locat=normal
155	no_default_queue=none
156	no_submit=false
157	job_catalog="~/job_queue"
158	job_name=none
159	local_user=$LOGNAME
160	node_usage=shared
161	numprocs=0
162	punkte="..........................................................."
163	submcom=qsub
164	queue=default
165	remote_host=none
166	remote_user=""
167	verify=true
168
169	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
170	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
171
172
173
174	# ERROR HANDLING
175	# IN CASE OF EXIT:
176	trap 'if [[ $locat != normal ]]
177	then
178	case $locat in
179	(option) printf "\n --> available optios can be displayed"
180	printf " by typing:"
181	printf "\n \"subjob ?\" \n";;
182	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
183	(*) printf "\n +++ unknown error"
184	printf "\n please inform S. Raasch!\n"
185	esac
186	[[ -f $job_to_send ]] && rm $job_to_send
187	printf "\n\n+++ SUBJOB killed \n\n"
188	fi' exit
189
190
191	# IN CASE OF TERMINAL-BREAK:
192	trap '[[ -f $job_to_send ]] && rm $job_to_send
193	printf "\n\n+++ SUBJOB killed \n\n"
194	exit
195	' 2
196
197
198	# DETERMINE NAME OF LOCAL HOST
199	local_host=$(hostname)
200
201	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
202	# IS ADMITTED AT ALL)
203	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
204	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
205	case $local_host in
206	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
207	(atmos) local_address=172.20.25.35; local_host=lcide;;
208	(austru) local_address=130.75.105.128; local_host=lcmuk;;
209	(autan) local_address=130.75.105.57; local_host=lcmuk;;
210	(bora) local_address=130.75.105.103; local_host=lcmuk;;
211	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
212	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
213	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
214	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
215	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
216	(breva) local_address=130.75.105.98; local_host=lcmuk;;
217	(buran) local_address=130.75.105.58; local_host=lcmuk;;
218	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
219	(climate*) local_address=165.132.26.68; local_host=lcyon;;
220	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
221	(cs*) local_address=136.172.44.131; local_host=nech;;
222	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
223	(flow01) local_address=10.141.255.71; local_host=lcflow;;
224	(flow02) local_address=10.141.255.72; local_host=lcflow;;
225	(node*) local_address=165.132.26.61 local_host=lck;;
226	# (node*) local_address=210.219.61.8 local_host=lckordi;;
227	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
228	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
229	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
230	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
231	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
232	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
233	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
234	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
235	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
236	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
237	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
238	(sno) local_address=130.75.105.113; local_host=lcmuk;;
239	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
240	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
241	(login*) local_address=118.128.66.201; local_host=lckiaps;;
242	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
243	(meller) local_address=134.106.74.155; local_host=lcfor;;
244	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
245	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
246	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
247	(nobel*) local_address=150.183.5.101; local_host=ibms;;
248	(ocean) local_address="ocean"; local_host=lcocean;;
249	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
250	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
251	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
252	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
253	(pingui) local_address=134.106.74.118; local_host=lcfor;;
254	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
255	(rte*) local_address=133.5.185.60; local_host=lcrte;;
256	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
257	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
258	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
259	(solano) local_address=130.75.105.110; local_host=lcmuk;;
260	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
261	(tc*) local_address="ocean"; local_host=lcocean;;
262	(t2a*) local_address=10.1.6.165; local_host=lctit;;
263	(urban*) local_address=147.46.30.151 local_host=lcsb;;
264	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
265	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
266	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
267	(*) printf "\n +++ \"$local_host\" unknown";
268	printf "\n please contact the PALM group at IMUK";
269	locat=parameter; exit;;
270	esac
271
272
273
274	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
275	remote_host=$local_host
276
277
278
279
280	# READ THE SHELLSCRIPT-OPTIONS
281	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
282	do
283	case $option in
284	(c) job_catalog=$OPTARG;;
285	(d) delete_dayfile=true;;
286	(D) no_submit=true;;
287	(e) email_notification=$OPTARG;;
288	(g) group_number=$OPTARG;;
289	(h) remote_host=$OPTARG;;
290	(m) memory=$OPTARG;;
291	(n) job_name=$OPTARG;;
292	(N) node_usage=$OPTARG;;
293	(O) threads_per_task=$OPTARG;;
294	(P) scp_port=$OPTARG;;
295	(q) no_default_queue=$OPTARG;;
296	(t) cputime=$OPTARG;;
297	(T) tasks_per_node=$OPTARG;;
298	(u) remote_user=$OPTARG;;
299	(v) verify=false;;
300	(X) numprocs=$OPTARG;;
301	(\?) printf "\n +++ Option $OPTARG unknown \n";
302	locat=option; exit;;
303	esac
304	done
305
306
307	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
308	(( to_shift = $OPTIND - 1 ))
309	shift $to_shift; file_to_send=$1
310
311
312	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
313	if [ "$1" = "?" ]
314	then
315	(printf "\n *** subjob can be called as follows:\n"
316	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
317	printf "\n Description of available options:\n"
318	printf "\n Option Description Default-Value"
319	printf "\n -c job-input- and output-catalog ~/job_queue"
320	printf "\n -d no job-protocol will be created ---"
321	printf "\n -D only the job-file will be created ---"
322	printf "\n -h execution host, available hosts: $remote_host"
323	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
324	printf "\n lckiaps, lctit, nech"
325	printf "\n -m memory demand per process in MByte ---"
326	printf "\n -n jobname <jobdatei>"
327	printf "\n -O threads per task (for OpenMP usage) 1"
328	printf "\n -P ssh/scp port default port"
329	printf "\n -q job-queue to be used default"
330	printf "\n -t allowed cpu-time in seconds ---"
331	printf "\n -T tasks per node (on parallel hosts) ---"
332	printf "\n -u username on execution host from .netrc"
333	printf "\n -v no prompt for confirmation ---"
334	printf "\n -X # of processors (on parallel hosts) 1"
335	printf "\n "
336	printf "\n The only possible positional parameter is <jobfile>:"
337	printf "\n The complete NQS-job must be provided here."
338	printf "\n <jobfile>=? creates this outline\n\n") \| more
339	exit
340	fi
341
342
343
344	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
345	if [[ "$file_to_send" = "" ]]
346	then
347	printf "\n +++ job-file missing"
348	locat=parameter; exit
349	else
350	if [[ -f $file_to_send ]]
351	then
352	true
353	else
354	printf "\n +++ job-file: "
355	printf "\n $file_to_send"
356	printf "\n does not exist"
357	locat=parameter; exit
358	fi
359	fi
360
361
362
363	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
364	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
365	if [[ $job_name = none ]]
366	then
367	job_name=$file_to_send
368	fi
369	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
370	then
371	printf "\n +++ job-file name: "
372	printf "\n $job_name"
373	printf "\n must not contain \"/\"-characters"
374	locat=parameter; exit
375	fi
376
377
378
379
380	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
381	# OR IF NO HOST HAS BEEN GIVEN
382	if [[ $remote_host = none ]]
383	then
384	printf "\n +++ host missing"
385	locat=option; exit
386	else
387	case $remote_host in
388	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
389	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
390	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
391	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
392	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
393	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
394	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
395	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
396	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
397	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
398	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
399	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
400	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
401	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
402	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
403	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
404	(lcocean) remote_address="ocean"; submcom=qsub;;
405	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
406	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
407	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
408	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
409	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
410	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
411	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
412	locat=parameter; exit;;
413	esac
414	fi
415
416
417	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
418	if [[ $no_default_queue != none ]]
419	then
420	error=false
421	ndq=$no_default_queue
422	case $remote_host in
423	(ibm) case $ndq in
424	(p690_express\|p690_standard\|p690_long) error=false;;
425	(*) error=true;;
426	esac;;
427	(ibmh) case $ndq in
428	(cluster\|express) error=false;;
429	(*) error=true;;
430	esac;;
431	(ibmkisti) case $ndq in
432	(class.32plus\|class.1-2\|class.2-32) error=false;;
433	(*) error=true;;
434	esac;;
435	(ibmku) case $ndq in
436	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
437	(*) error=true;;
438	esac;;
439	(ibms) case $ndq in
440	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
441	(*) error=true;;
442	esac;;
443	(lcbullhh) case $ndq in
444	(compute\|compute2\|shared) error=false;;
445	(*) error=true;;
446	esac;;
447	(lccrayb) case $ndq in
448	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
449	(*) error=true;;
450	esac;;
451	(lccrayh) case $ndq in
452	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
453	(*) error=true;;
454	esac;;
455	(lccrayf) case $ndq in
456	(usup\|test*\|small\|large) error=false;;
457	(*) error=true;;
458	esac;;
459	(lcflow) case $ndq in
460	(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;;
461	(*) error=true;;
462	esac;;
463	(lckiaps) case $ndq in
464	(express\|normal\|normal20\|quickq) error=false;;
465	(*) error=true;;
466	esac;;
467	(lckyoto) case $ndq in
468	(eh\|ph) error=false;;
469	(*) error=true;;
470	esac;;
471	(lckyuh) case $ndq in
472	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
473	(*) error=true;;
474	esac;;
475	(lckyut) case $ndq in
476	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
477	(*) error=true;;
478	esac;;
479	(lctit) case $ndq in
480	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
481	(*) error=true;;
482	esac;;
483	(t3eb) case $ndq in
484	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
485	(*) error=true;;
486	esac;;
487	(t3eh) case $ndq in
488	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
489	(*) error=true;;
490	esac;;
491	(t3ej2\|t3ej5) case $ndq in
492	(low\|normal\|high) error=false;;
493	(*) error=true;;
494	esac;;
495	(t3es) case $ndq in
496	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
497	(*) error=true;;
498	esac;;
499	esac
500	if [[ $error = true ]]
501	then
502	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
503	locat=parameter; exit
504	else
505	queue=$no_default_queue
506	fi
507	fi
508
509
510
511	# CHECK THE CPU-TIME
512	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
513	done=false
514	while [[ $done = false ]]
515	do
516	if (( $cputime <= 0 ))
517	then
518	printf "\n +++ wrong cpu-time or cpu-time missing"
519	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
520	printf "\n >>> "
521	read cputime 1>/dev/null 2>&1
522	else
523	done=true
524	fi
525	done
526	if [[ $remote_host = nech ]]
527	then
528	if (( tasks_per_node != 0 ))
529	then
530	(( cputime = cputime * tasks_per_node ))
531	elif [[ $numprocs != 0 ]]
532	then
533	(( cputime = cputime * numprocs ))
534	fi
535	fi
536	(( stunden = cputime / 3600 ))
537	(( resttime = cputime - stunden * 3600 ))
538	(( minuten = resttime / 60 ))
539	(( sekunden = resttime - minuten * 60 ))
540	timestring=${stunden}:${minuten}:${sekunden}
541
542
543
544	# CHECK THE MEMORY DEMAND
545	done=false
546	while [[ $done = false ]]
547	do
548	if (( memory <= 0 ))
549	then
550	printf "\n +++ wrong memory demand or memory demand missing"
551	printf "\n >>> Please type memory in MByte per process as INTEGER:"
552	printf "\n >>> "
553	read memory 1>/dev/null 2>&1
554	else
555	done=true
556	fi
557	done
558
559	if [[ $remote_host = nech ]]
560	then
561	if (( tasks_per_node != 0 ))
562	then
563	(( Memory = memory * tasks_per_node / 1000 ))
564	elif [[ $numprocs != 0 ]]
565	then
566	(( Memory = memory * numprocs / 1000 ))
567	else
568	(( Memory = memory / 1000 ))
569	fi
570	elif [[ $remote_host = lctit ]]
571	then
572	(( Memory = memory * tasks_per_node / 1000 ))
573	fi
574
575
576	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
577	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
578	then
579	(( memory = memory * threads_per_task ))
580	fi
581
582
583	# CALCULATE NUMBER OF REQUIRED NODES
584	if (( tasks_per_node != 0 ))
585	then
586	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
587	fi
588
589
590	# CALCULATE NUMBER OF PROCESSES PER NODE
591	(( processes_per_node = tasks_per_node * threads_per_task ))
592
593
594	# CALCULATE NUMBER OF MPI TASKS
595	(( mpi_tasks = numprocs / threads_per_task ))
596
597
598	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
599	if [[ "$scp_port" != "" ]]
600	then
601	PORTOPT="-P $scp_port"
602	SSH_PORTOPT="-p $scp_port"
603	fi
604
605
606	# HEADER-OUTPUT
607	if [[ $verify = true ]]
608	then
609	printf "\n\n"
610	printf "#--------------------------------------------------------------# \n"
611	spalte1=SUBJOB;spalte2=$(date)
612	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
613	printf "\| \| \n"
614	printf "\| values of parameters/options: \| \n"
615	spalte1=$(echo local_host$punkte \| cut -c-20)
616	spalte2=$punkte$local_host
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	spalte1=$(echo remote_host$punkte \| cut -c-20)
619	spalte2=$punkte$remote_host
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	spalte1=$(echo queue$punkte \| cut -c-20)
622	spalte2=$punkte$queue
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	spalte1=$(echo memory$punkte \| cut -c-20)
625	spalte2="$punkte$memory mb"
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	spalte1=$(echo cputime$punkte \| cut -c-20)
628	spalte2="$punkte$cputime sec"
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	spalte1=$(echo job_name$punkte \| cut -c-20)
631	spalte2="$punkte$job_name"
632	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
633	printf "#--------------------------------------------------------------# \n\n"
634
635
636	# QUERY CHECK
637	antwort="dummy"
638	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
639	do
640	read antwort?" >>> continue (y/n) ? "
641	done
642	if [[ $antwort = n \|\| $antwort = N ]]
643	then
644	locat=verify; exit
645	fi
646	printf "\n"
647	fi
648
649	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
650	identifier=$RANDOM
651	job_on_remhost=${job_name}_${identifier}_$local_host
652	job_to_send=job_to_send_$identifier
653	if [[ $delete_dayfile = false ]]
654	then
655	remote_dayfile=${local_host}_${job_name}_result_$identifier
656	local_dayfile=${remote_host}_${job_name}
657	else
658	remote_dayfile=/dev/null
659	fi
660
661
662	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
663	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
664	then
665
666	# GENERAL LOADLEVELER SETTINGS
667	execute_in_shell="#!/bin/ksh"
668	use_shell="# @ shell = /bin/ksh"
669	consumable_memory="ConsumableMemory($memory mb)"
670	class="# @ class = $queue"
671	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
672	network_to_use="# @ network.mpi = sn_all,shared,us"
673	data_limit="# @ data_limit = 1.76gb"
674	image_size="# @ image_size = 50"
675	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
676
677	if [[ $email_notification = none ]]
678	then
679	notify_user=""
680	else
681	notify_user="# @ notify_user = $email_notification"
682	if [[ $delete_dayfile = true ]]
683	then
684	notification='# @ notification = never'
685	fi
686	fi
687
688	if [[ $remote_host = ibmh ]]
689	then
690	data_limit=""
691	network_to_use=""
692	class="# @ class = $queue"
693	environment=""
694	rset="# @ rset = RSET_MCM_AFFINITY"
695	task_affinity="# @ task_affinity = core(1)"
696	elif [[ $remote_host = ibmkisti ]]
697	then
698	network_to_use="# @ network.MPI = sn_all,shared,US"
699	wall_clock_limit="# @ wall_clock_limit = $timestring"
700	if [[ $threads_per_task = 1 ]]
701	then
702	rset="# @ rset = RSET_MCM_AFFINITY"
703	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
704	fi
705	environment=""
706	use_shell=""
707	data_limit=""
708	image_size=""
709	elif [[ $remote_host = ibmku ]]
710	then
711	execute_in_shell="#!/usr/bin/ksh"
712	use_shell="# @ shell = /usr/bin/ksh"
713	consumable_memory=""
714	environment=""
715	network_to_use="# @ network.mpi = sn_all,shared,us"
716	data_limit=""
717	image_size=""
718	elif [[ $remote_host = ibms ]]
719	then
720	network_to_use="# @ network.mpi = csss,shared,us"
721	fi
722
723	cat > $job_to_send << %%END%%
724	$execute_in_shell
725	$use_shell
726
727	# @ job_type = parallel
728	# @ job_name = $job_name
729	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
730	# @ output = $remote_dayfile
731	# @ error = $remote_dayfile
732	$wall_clock_limit
733	$image_size
734	$class
735	$environment
736	$network_to_use
737	$data_limit
738	$rset
739	$mcm_affinity_options
740	$task_affinity
741	$notification
742	$notify_user
743
744	%%END%%
745
746	if (( nodes > 0 ))
747	then
748
749	if [[ $remote_host != ibmkisti ]]
750	then
751
752	cat >> $job_to_send << %%END%%
753	# @ node = $nodes
754	# @ tasks_per_node = $processes_per_node
755	# @ node_usage = $node_usage
756	# @ queue
757
758	%%END%%
759
760	else
761
762	cat >> $job_to_send << %%END%%
763	# @ total_tasks = $mpi_tasks
764	# @ blocking = unlimited
765	# @ queue
766
767	%%END%%
768
769	fi
770
771	else
772
773	cat >> $job_to_send << %%END%%
774	# @ blocking = unlimited
775	# @ total_tasks = $numprocs
776	# @ node_usage = $node_usage
777	# @ queue
778
779	%%END%%
780
781	fi
782
783	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
784	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
785	then
786	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
787	fi
788
789	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
790	then
791
792	cat > $job_to_send << %%END%%
793	#!/bin/ksh
794
795	# @ job_type = serial
796	# @ node_usage = $node_usage
797	# @ job_name = palm
798	# @ wall_clock_limit = ${timestring},$timestring
799	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
800	# @ output = $remote_dayfile
801	# @ error = $remote_dayfile
802	$class
803	$notification
804
805	# @ queue
806
807	%%END%%
808
809	elif [[ $remote_host = lcbullhh ]]
810	then
811	if [[ $numprocs != 0 ]]
812	then
813	cat > $job_to_send << %%END%%
814	#!/bin/bash -l
815	#SBATCH -J $job_name
816	#SBATCH -t $timestring
817	#SBATCH -N $nodes
818	#SBATCH --ntasks-per-node=$processes_per_node
819	#SBATCH -p $queue
820	#SBATCH -o $remote_dayfile
821	#SBATCH -e $remote_dayfile
822	#SBATCH -A $project_account
823
824	$init_cmds
825	$module_calls
826
827	%%END%%
828
829	else
830	cat > $job_to_send << %%END%%
831	#!/bin/bash -l
832	#SBATCH -J $job_name
833	#SBATCH -t $timestring
834	#SBATCH -l ncpus=1
835	#SBATCH -l pmem=${memory}mb
836	#SBATCH -m abe
837	#SBATCH -o $remote_dayfile
838	#SBATCH -e $remote_dayfile
839	#SBATCH -A $project_account
840
841	$init_cmds
842	$module_calls
843
844	%%END%%
845
846	fi
847
848	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
849	then
850
851	if [[ "$feature" != "" ]]
852	then
853	featuredir="#PBS -l feature=$feature"
854	fi
855
856	if [[ $numprocs != 0 ]]
857	then
858	cat > $job_to_send << %%END%%
859	#!/bin/bash
860	#PBS -N $job_name
861	#PBS -l walltime=$timestring
862	#PBS -l nodes=$nodes:ppn=$processes_per_node
863	#PBS -o $remote_dayfile
864	#PBS -j oe
865	#PBS -q $queue
866	$featuredir
867
868	$init_cmds
869	$module_calls
870
871	%%END%%
872
873	else
874
875	continue
876
877	fi
878
879	elif [[ $remote_host = lccrayf ]]
880	then
881
882	if [[ $numprocs != 0 ]]
883	then
884	cat > $job_to_send << %%END%%
885	#!/bin/bash -l
886	#SBATCH -J $job_name
887	#SBATCH -t $timestring
888	#SBATCH -N $nodes
889	#SBATCH --ntasks-per-node=$processes_per_node
890	#SBATCH -p $queue
891	#SBATCH -o $remote_dayfile
892	#SBATCH -e $remote_dayfile
893
894	$init_cmds
895	$module_calls
896
897	%%END%%
898
899	else
900	cat > $job_to_send << %%END%%
901	#!/bin/bash -l
902	#SBATCH -J $job_name
903	#SBATCH -t $timestring
904	#SBATCH -l ncpus=1
905	#SBATCH -l pmem=${memory}mb
906	#SBATCH -m abe
907	#SBATCH -o $remote_dayfile
908	#SBATCH -e $remote_dayfile
909
910	$init_cmds
911	$module_calls
912
913	%%END%%
914
915	fi
916
917	elif [[ $remote_host = lcflow ]]
918	then
919	if [[ $numprocs != 0 ]]
920	then
921	pe_set="#$ -pe impi $numprocs"
922	else
923	pe_set="#$ -pe impi 1"
924	fi
925	if [[ $queue = default ]]
926	then
927	queue_set=""
928	else
929	queue_set="#$ -q $queue"
930	fi
931	[[ "$disc_space" = "" ]] && disc_space=50
932
933	cat > $job_to_send << %%END%%
934	#!/bin/bash
935	#$ -S /bin/bash
936	#$ -N $job_name
937	#$ -cwd
938	#$ -l h_rt=$timestring
939	#$ -l h_vmem=${memory}M
940	#$ -o $remote_dayfile
941	#$ -j y
942	$pe_set
943	#$ -R y
944	#$ -l h_fsize=${disc_space}G
945	$queue_set
946
947	%%END%%
948
949	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
950	then
951
952	if [[ $numprocs != 0 ]]
953	then
954	cat > $job_to_send << %%END%%
955	#!/bin/ksh
956	#PBS -N $job_name
957	#PBS -l walltime=$timestring
958	#PBS -l ncpus=$numprocs
959	#PBS -l pmem=${memory}mb
960	#PBS -o $remote_dayfile
961	#PBS -l nodes=$nodes:ppn=${processes_per_node}
962	#PBS -j oe
963
964	mpd &
965
966	%%END%%
967
968	else
969	cat > $job_to_send << %%END%%
970	#!/bin/ksh
971	#PBS -N $job_name
972	#PBS -l walltime=$timestring
973	#PBS -l ncpus=1
974	#PBS -l pmem=${memory}mb
975	#PBS -o $remote_dayfile
976	#PBS -j oe
977
978	%%END%%
979
980	fi
981
982	elif [[ $remote_host = lckiaps ]]
983	then
984
985	if [[ $numprocs != 0 ]]
986	then
987	cat > $job_to_send << %%END%%
988	#!/bin/bash
989	#PBS -N $job_name
990	#PBS -l walltime=$timestring
991	#PBS -l select=1:ncpus=$numprocs
992	#PBS -l pmem=${memory}mb
993	#PBS -q $queue
994	#PBS -o $remote_dayfile
995	#PBS -j oe
996	#PBS -V
997
998	%%END%%
999
1000	else
1001	cat > $job_to_send << %%END%%
1002	#!/bin/bash
1003	#PBS -N $job_name
1004	#PBS -l walltime=$timestring
1005	#PBS -l ncpus=1
1006	#PBS -l pmem=${memory}mb
1007	#PBS -o $remote_dayfile
1008	#PBS -j oe
1009
1010	%%END%%
1011
1012	fi
1013
1014	elif [[ $remote_host = lcyon ]]
1015	then
1016
1017	if [[ $numprocs != 0 ]]
1018	then
1019	cat > $job_to_send << %%END%%
1020	#!/bin/ksh
1021	#PBS -N $job_name
1022	#PBS -l walltime=$timestring
1023	#PBS -l ncpus=$numprocs
1024	#PBS -l pmem=${memory}mb
1025	#PBS -o $remote_dayfile
1026	#PBS -j oe
1027
1028	%%END%%
1029
1030	else
1031	cat > $job_to_send << %%END%%
1032	#!/bin/ksh
1033	#PBS -N $job_name
1034	#PBS -l walltime=$timestring
1035	#PBS -l ncpus=1
1036	#PBS -l pmem=${memory}mb
1037	#PBS -o $remote_dayfile
1038	#PBS -j oe
1039
1040	%%END%%
1041
1042	fi
1043
1044	elif [[ $remote_host = lcxe6 ]]
1045	then
1046
1047	if [[ $numprocs != 0 ]]
1048	then
1049	cat > $job_to_send << %%END%%
1050	#!/bin/ksh
1051	#PBS -S /bin/ksh
1052	#PBS -N $job_name
1053	#PBS -A $project_account
1054	#PBS -j oe
1055	#PBS -l walltime=$timestring
1056	#PBS -l mppwidth=${numprocs}
1057	#PBS -l mppnppn=${processes_per_node}
1058	#PBS -m abe
1059	#PBS -o $remote_dayfile
1060	$email_directive
1061
1062	$init_cmds
1063	$module_calls
1064
1065	%%END%%
1066
1067	else
1068	cat > $job_to_send << %%END%%
1069	#!/bin/ksh
1070	#PBS -S /bin/ksh
1071	#PBS -N $job_name
1072	#PBS -A $project_account
1073	#PBS -j oe
1074	#PBS -l walltime=$timestring
1075	#PBS -l ncpus=1
1076	#PBS -l pmem=${memory}mb
1077	#PBS -m abe
1078	$email_directive
1079	#PBS -o $remote_dayfile
1080
1081	$init_cmds
1082	$module_calls
1083
1084	%%END%%
1085
1086	fi
1087
1088	elif [[ $remote_host = lckyoto ]]
1089	then
1090
1091	cat > $job_to_send << %%END%%
1092	#!/bin/ksh
1093	# @\$-o $remote_dayfile
1094	# @\$-eo -oi
1095	# @\$-lP 16
1096	# @\$-lp 1
1097	# @\$-lm 28gb -llm unlimited -ls unlimited
1098	# @\$-q $queue
1099	# @\$-Pvn abs_pack
1100	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1101	#. /thin/local/etc/setprofile/intel-11.0.sh
1102	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1103	. ~/.myprofile
1104	#. /home2/t/t51254/palm/current_version/myprofile
1105	#. /thin/apps/pgi/mpi.sh
1106	#
1107	env
1108	#
1109	set -x
1110
1111	%%END%%
1112
1113	elif [[ $remote_host = lcxt5m ]]
1114	then
1115
1116	if [[ $numprocs != 0 ]]
1117	then
1118	cat > $job_to_send << %%END%%
1119	#!/bin/ksh
1120	#PBS -S /bin/ksh
1121	#PBS -N $job_name
1122	#PBS -j oe
1123	#PBS -l walltime=$timestring
1124	#PBS -l mppwidth=${numprocs}
1125	#PBS -l mppnppn=${processes_per_node}
1126	#PBS -m abe
1127	#PBS -o $remote_dayfile
1128
1129	$init_cmds
1130	$module_calls
1131
1132	%%END%%
1133
1134	else
1135	cat > $job_to_send << %%END%%
1136	#!/bin/ksh
1137	#PBS -S /bin/ksh
1138	#PBS -N $job_name
1139	#PBS -j oe
1140	#PBS -l walltime=$timestring
1141	#PBS -l ncpus=1
1142	#PBS -l pmem=${memory}mb
1143	#PBS -m abe
1144	#PBS -o $remote_dayfile
1145
1146	$init_cmds
1147	$module_calls
1148
1149	%%END%%
1150
1151	fi
1152
1153	elif [[ $remote_host = lckyuh ]]
1154	then
1155	cat > $job_to_send << %%END%%
1156	#!/bin/bash
1157	#PJM -L "rscgrp=$queue"
1158	#PJM -L "node=$nodes"
1159	#PJM --mpi "proc=$numprocs"
1160	#PJM -L "elapse=$timestring"
1161	#PJM -o $remote_dayfile
1162	#PJM -j
1163	#PJM -X
1164	#PJM --no-stging
1165
1166	export LANG=en_US.UTF-8
1167	%%END%%
1168
1169	elif [[ $remote_host = lckyut ]]
1170	then
1171	cat > $job_to_send << %%END%%
1172	#!/bin/bash
1173	#PJM -L "rscgrp=$queue"
1174	#PJM -L "vnode=$numprocs"
1175	#PJM -L "vnode-core=1"
1176	#PJM -L "elapse=$timestring"
1177	#PJM --mpi proc=$numprocs
1178	#PJM -o $remote_dayfile
1179	#PJM -j
1180	#PJM -X
1181	#PJM --no-stging
1182
1183	export LANG=en_US.UTF-8
1184	%%END%%
1185
1186	elif [[ $remote_host = lcocean ]]
1187	then
1188	cat > $job_to_send << %%END%%
1189	#!/bin/bash
1190	#$ -cwd
1191	#$ -V
1192	#$ -N $job_name
1193	#$ -pe orte $numprocs
1194	#$ -o $remote_dayfile
1195	#$ -j y
1196	#$ -R y
1197	$init_cmds
1198	$module_calls
1199
1200	%%END%%
1201
1202	elif [[ $remote_host = nech ]]
1203	then
1204
1205	if (( nodes > 1 ))
1206	then
1207	cat > $job_to_send << %%END%%
1208	#!/bin/ksh
1209	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1210	#PBS -l ${qsubmem}=${Memory}gb
1211	#PBS -b $nodes
1212	#PBS -o $remote_dayfile
1213	#PBS -N palm
1214	#PBS -j o
1215	#PBS -T mpisx
1216
1217	%%END%%
1218
1219	elif [[ $numprocs != 0 ]]
1220	then
1221	cat > $job_to_send << %%END%%
1222	#!/bin/ksh
1223	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1224	#PBS -l ${qsubmem}=${Memory}gb
1225	#PBS -o $remote_dayfile
1226	#PBS -N palm
1227	#PBS -j o
1228
1229	%%END%%
1230
1231	else
1232	cat > $job_to_send << %%END%%
1233	#!/bin/ksh
1234	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1235	#PBS -o $remote_dayfile
1236	#PBS -j o
1237
1238	%%END%%
1239
1240	fi
1241
1242	elif [[ $remote_host = lctit ]]
1243	then
1244	cat > $job_to_send << %%END%%
1245	#!/bin/ksh
1246	$init_cmds
1247	$module_calls
1248
1249	%%END%%
1250
1251	# SET OPTIONS FOR SUBMIT-COMMAND
1252	if [[ $tasks_per_node != $processes_per_node ]]
1253	then
1254	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -et 1 -q $queue "
1255	else
1256	submcom="$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 "
1257	fi
1258
1259	else
1260
1261	cat > $job_to_send << %%END%%
1262	# @\$-q ${queue}
1263	# @\$-l${qsubtime} $timestring
1264	# @\$-l${qsubmem} ${memory}mb
1265	# @\$-o $remote_dayfile
1266	# @\$-eo
1267
1268	%%END%%
1269
1270	fi
1271
1272
1273	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1274	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1275	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1276	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1277	then
1278	echo "set +vx" >> $job_to_send
1279	echo "trap '" >> $job_to_send
1280	echo "set +vx" >> $job_to_send
1281	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 ]]
1282	then
1283	if [[ $remote_host = ibmh ]]
1284	then
1285	return_queue=c1
1286	elif [[ $remote_host = ibmkisti ]]
1287	then
1288	return_queue=class.1-2
1289	elif [[ $remote_host = ibmku ]]
1290	then
1291	return_queue=sdbg2
1292	elif [[ $remote_host = ibms ]]
1293	then
1294	return_queue=p_normal
1295	elif [[ $remote_host = lcbullhh ]]
1296	then
1297	return_queue=shared
1298	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1299	then
1300	return_queue=dataq
1301	elif [[ $remote_host = lcxe6 ]]
1302	then
1303	return_queue=debug
1304	elif [[ $remote_host = lckiaps ]]
1305	then
1306	return_queue=express
1307	elif [[ $remote_host = lckyuh ]]
1308	then
1309	return_queue=cx-single
1310	elif [[ $remote_host = lckyut ]]
1311	then
1312	return_queue=cx-single
1313	else
1314	return_queue=unknown
1315	fi
1316
1317	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1318	then
1319
1320	if [[ $remote_host = ibmku ]]
1321	then
1322	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1323	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1324	else
1325	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1326	fi
1327	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1328	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1329	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1330	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1331	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1333	if [[ $host != "ibmh" ]]
1334	then
1335	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1336	fi
1337	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1338	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1339
1340	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1341	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1342
1343	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1344	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
1345	if [[ $remote_host = ibmku ]]
1346	then
1347	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1348	fi
1349	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1350
1351	elif [[ $remote_host = nech ]]
1352	then
1353	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1354	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1355	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1356	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1357	echo "#PBS -j o" >> $job_to_send
1358	echo " " >> $job_to_send
1359	echo "set -x" >> $job_to_send
1360	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1361	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1362	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1363	echo "%%END%%" >> $job_to_send
1364
1365	elif [[ $remote_host = lcbullhh ]]
1366	then
1367	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1368	echo "#!/bin/bash" >> $job_to_send
1369	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1370	echo "#SBATCH -t 00:20:00" >> $job_to_send
1371	echo "#SBATCH -N 1" >> $job_to_send
1372	echo "#SBATCH -n 1" >> $job_to_send
1373	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1374	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1375	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1376	echo "#SBATCH -A $project_account" >> $job_to_send
1377	echo "#SBATCH -p $return_queue" >> $job_to_send
1378	echo " " >> $job_to_send
1379	echo "set -x" >> $job_to_send
1380	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1381	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1382	echo "%%END%%" >> $job_to_send
1383
1384	elif [[ $remote_host = lckyuh ]]
1385	then
1386	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1387	echo "#!/bin/bash" >> $job_to_send
1388	echo "#PJM -L \"node=1\"" >> $job_to_send
1389	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1390	echo "#PJM --no-stging" >> $job_to_send
1391	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1392	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1393	echo "#PJM -j" >> $job_to_send
1394	echo " " >> $job_to_send
1395	echo "export LANG=en_US.UTF-8" >> $job_to_send
1396	echo "set -x" >> $job_to_send
1397	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1398	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1399	echo "%%END%%" >> $job_to_send
1400
1401	elif [[ $remote_host = lckyut ]]
1402	then
1403	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1404	echo "#!/bin/bash" >> $job_to_send
1405	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1406	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1407	echo "#PJM --no-stging" >> $job_to_send
1408	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1409	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1410	echo "#PJM -j" >> $job_to_send
1411	echo " " >> $job_to_send
1412	echo "export LANG=en_US.UTF-8" >> $job_to_send
1413	echo "set -x" >> $job_to_send
1414	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1415	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1416	echo "%%END%%" >> $job_to_send
1417
1418	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1419	then
1420	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1421	echo "#!/bin/bash" >> $job_to_send
1422	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1423	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1424	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1425	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1426	echo "#PBS -j oe" >> $job_to_send
1427	echo " " >> $job_to_send
1428	echo "set -x" >> $job_to_send
1429	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1430	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1431	echo "%%END%%" >> $job_to_send
1432
1433	elif [[ $remote_host = lcocean ]]
1434	then
1435	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1436	echo "#!/bin/bash" >> $job_to_send
1437	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1438	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1439	echo "SGEPREFIX -cwd" >> $job_to_send
1440	echo "SGEPREFIX -j y" >> $job_to_send
1441	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1442	echo " " >> $job_to_send
1443	echo "set -x" >> $job_to_send
1444	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1445	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1446	echo "" >> $job_to_send
1447	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1448	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1449	echo "rm -f scpjob.${identifier}" >> $job_to_send
1450	echo "%%END%%" >> $job_to_send
1451	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1452	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1453
1454	elif [[ $remote_host = lcflow ]]
1455	then
1456	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1457	echo "#!/bin/bash" >> $job_to_send
1458	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1459	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1460	echo "SGEPREFIX -cwd" >> $job_to_send
1461	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1462	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1463	echo "SGEPREFIX -j y" >> $job_to_send
1464	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1465	echo " " >> $job_to_send
1466	echo "set -x" >> $job_to_send
1467	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1468	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1469	echo "" >> $job_to_send
1470	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1471	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1472	echo "rm -f scpjob.${identifier}" >> $job_to_send
1473	echo "%%END%%" >> $job_to_send
1474	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1475	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1476	elif [[ $remote_host = lcxe6 ]]
1477	then
1478	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1479	echo "#!/bin/ksh" >> $job_to_send
1480	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1481	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1482	echo "#PBS -A $project_account" >> $job_to_send
1483	echo "#PBS -l mppwidth=1" >> $job_to_send
1484	echo "#PBS -l mppnppn=1" >> $job_to_send
1485	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1486	echo "#PBS -j oe" >> $job_to_send
1487	echo " " >> $job_to_send
1488	echo "set -x" >> $job_to_send
1489	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1490	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1491	echo "%%END%%" >> $job_to_send
1492	else
1493
1494	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1495	echo "# @\\\$-q $return_queue" >> $job_to_send
1496	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1497	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1498	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1499	then
1500	echo "# @\$-l mpp_p=0" >> $job_to_send
1501	fi
1502	echo '# @\$-lF 10mb' >> $job_to_send
1503	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1504	echo '# @\\\$-eo' >> $job_to_send
1505	echo " " >> $job_to_send
1506	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1507	then
1508	echo "set +vx" >> $job_to_send
1509	echo ". .profile" >> $job_to_send
1510	fi
1511	echo "set -x" >> $job_to_send
1512	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
1513	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1514	echo "%%END%%" >> $job_to_send
1515
1516	fi
1517
1518	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1519	then
1520	echo "llsubmit scpjob.$identifier" >> $job_to_send
1521	elif [[ $remote_host = lcbullhh ]]
1522	then
1523	echo "sbatch scpjob.$identifier" >> $job_to_send
1524	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1525	then
1526	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1527	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1528	then
1529	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1530	elif [[ $remote_host = t3es ]]
1531	then
1532	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1533	elif [[ $remote_host = lckiaps ]]
1534	then
1535	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1536	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1537	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1538	elif [[ $remote_host = lckyu* ]]
1539	then
1540	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1541	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1542	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1543	then
1544	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1545	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1546	else
1547	echo "$submcom scpjob.$identifier" >> $job_to_send
1548	fi
1549	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1550	then
1551	echo "rm scpjob.$identifier" >> $job_to_send
1552	fi
1553	if [[ $remote_host = nech ]]
1554	then
1555	echo "cd -" >> $job_to_send
1556	fi
1557	else
1558	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1559	# ??? funktioniert das ÃŒberhaupt noch ???
1560	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1561	fi
1562	echo "set -x" >> $job_to_send
1563	echo " ' exit" >> $job_to_send
1564	echo "set -x" >> $job_to_send
1565	fi
1566
1567
1568	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1569	cat $file_to_send >> $job_to_send
1570
1571	if [[ $remote_host = ibm ]]
1572	then
1573	echo " " >> $job_to_send
1574	echo "exit" >> $job_to_send
1575	fi
1576
1577	# REMOVE JOB-FILE
1578	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1579	then
1580	echo " " >> $job_to_send
1581	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1582	fi
1583
1584
1585	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1586	if [[ $no_submit = false ]]
1587	then
1588	if [[ $remote_host != $local_host ]]
1589	then
1590	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1591	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1592	then
1593	job_catalog_save=$job_catalog
1594	job_catalog=job_queue
1595	elif [[ $remote_host = nech ]]
1596	then
1597	job_catalog_save=$job_catalog
1598	job_catalog=/hpf/b/${remote_user}/job_queue
1599	fi
1600	if [[ $remote_host = nech ]]
1601	then
1602	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1603	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1604	else
1605	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1606	fi
1607	if [[ $? = 1 ]]
1608	then
1609	locat=scp; exit
1610	fi
1611	if [[ $remote_host = ibms ]]
1612	then
1613	job_catalog=$job_catalog_save
1614	fi
1615	[[ $verify = true ]] && printf "\n >>> finished\n"
1616	else
1617	eval job_catalog=$job_catalog
1618	cp $job_to_send ${job_catalog}/$job_on_remhost
1619	fi
1620
1621
1622
1623	# START NQS- / LOADLEVELER-JOB
1624	if [[ $remote_host != $local_host ]]
1625	then
1626	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1627
1628	if [[ $remote_host = ibmku ]]
1629	then
1630	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1631	elif [[ $remote_host = lcflow ]]
1632	then
1633	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1634	else
1635	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1636	fi
1637
1638	[[ $verify = true ]] && printf " >>> o.k.\n"
1639	else
1640	cd $job_catalog
1641	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1642	then
1643	eval $submcom $job_on_remhost
1644	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1645	then
1646	chmod u+x $job_on_remhost
1647	eval $submcom $job_on_remhost
1648	elif [[ $local_host = nech ]]
1649	then
1650	if [[ $queue = default ]]
1651	then
1652	eval $submcom $job_on_remhost
1653	else
1654	eval $submcom -q $queue $job_on_remhost
1655	fi
1656	elif [[ $local_host = lcbullhh ]]
1657	then
1658	if [[ $queue = default ]]
1659	then
1660	eval $submcom $job_on_remhost
1661	fi
1662	else
1663	qsub $job_on_remhost
1664	fi
1665
1666	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1667	# AT THE END OF THE JOB
1668	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1669	then
1670	rm $job_on_remhost
1671	fi
1672	cd - > /dev/null
1673	fi
1674	fi
1675
1676	# FINAL ACTIONS
1677	if [[ $no_submit = false ]]
1678	then
1679	rm -f $job_to_send
1680	fi
1681	[[ $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 |