Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2258

Last change on this file since 2258 was 2257, checked in by witha, 7 years ago
Bugfix in PALM-WTM, modifications for lceddy
Property svn:keywords set to `Id Rev`
File size: 56.2 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 2257 2017-06-07 14:07:05Z suehring $
30	# adjustments for lceddy, removed lcflow-specific code
31	#
32	# 2188 2017-03-21 06:42:42Z raasch
33	#
34	# 2187 2017-03-21 06:41:25Z raasch
35	# adjustment of compute node names for lckyuh
36	#
37	# 2184 2017-03-21 04:31:22Z raasch
38	# bugfix: localhost renamed local_host
39	#
40	# 2148 2017-02-09 16:56:42Z scharf
41	# added kuma and gharbi to the list of known hosts
42	#
43	# 2134 2017-02-02 07:33:46Z raasch
44	# option -E added to msub commands on HLRN-III machines to allow output of more
45	# job informations in the job protocol files
46	#
47	# 1944 2016-06-15 06:29:00Z raasch
48	# adjustments for using HLRN ssh-keys
49	#
50	# 1940 2016-06-14 05:15:20Z raasch
51	# adjustments for lckiaps
52	#
53	# 1866 2016-04-15 06:50:59Z raasch
54	# adjusted for lcocean
55	#
56	# 1841 2016-04-07 19:14:06Z raasch
57	# script now running under bash
58	#
59	# 1701 2015-11-02 07:43:04Z maronga
60	# Bugfix: added missing init_cmds for lccrayh/lccrayb
61	#
62	# 1621 2015-07-17 11:39:33Z heinze
63	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
64	#
65	# 1575 2015-03-27 09:56:27Z raasch
66	# mpp2-queues added to lccrayh
67	#
68	# 1547 2015-01-29 15:09:12Z witha
69	# adjustments for ForWind computing cluster (lcflow)
70	#
71	# 1545 2015-01-29 06:52:23Z heinze
72	# local host name for blizzard further specified
73	#
74	# 1480 2014-10-17 14:41:49Z raasch
75	# adjustments for 2nd stage of HLRNIII
76	#
77	# 1468 2014-09-24 14:06:57Z maronga
78	# Typo removed (addres->address)
79	# Adjustments for lcxe6
80	#
81	# 1452 2014-08-22 09:41:06Z heinze
82	# local hosts for blizzard added
83	#
84	# 1450 2014-08-21 07:31:51Z heinze
85	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
86	#
87	# 1442 2014-07-28 07:09:10Z raasch
88	# HLRN-III (lccrayb/lccrayh) queues adjusted
89	#
90	# 1378 2014-04-28 06:04:58Z raasch
91	# -et option added for lctit
92	#
93	# 1350 2014-04-04 13:01:30Z maronga
94	# location of qsub updated for lcxe6
95	#
96	# 1289 2014-03-04 07:12:34Z raasch
97	# German comments translated to English
98	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
99	#
100	# 1279 2014-01-28 12:10:14Z raasch
101	# node calculation modified due to changes in mrun (tasks_per_node must not be
102	# an integral divisor of numprocs any more)
103	#
104	# 1274 2014-01-09 13:14:54Z heinze
105	# adjustments for lccrayh
106	#
107	# 1266 2013-12-11 12:07:34Z heinze
108	# further adjustments for lccrayb (use msub instead of qsub)
109	#
110	# 1264 2013-12-09 12:46:09Z fricke
111	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
112	#
113	# 1262 2013-12-09 10:57:20Z fricke
114	# further adjustments for lccrayb
115	#
116	# 1260 2013-12-04 12:48:04Z raasch
117	# jaboticaba admitted
118	#
119	# 1255 2013-11-07 14:43:35Z raasch
120	# further adjustments for lccrayb
121	#
122	# 1224 2013-09-16 07:27:23Z raasch
123	# first adjustments for lccrayb
124	#
125	# 1202 2013-07-10 16:22:07Z witha
126	# adjustments for Forwind cluster (lcflow)
127	#
128	# 1199 2013-07-05 14:52:22Z raasch
129	# adjustments for CSC Helsinki (lccrayf)
130	#
131	# use of cluster/express queue enabled (ibmh)
132	# vinessa added (imuk)
133	#
134	# 1103 2013-02-20 02:15:53Z raasch
135	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
136	# instead typeset -L/R),
137	# further adjustments for lckyuh
138	#
139	# 2013-02-10 01:47:43Z raasch
140	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
141	# and for Forwind cluster (lcflow)
142	#
143	# 1094 2013-02-03 01:52:12Z raasch
144	# new option -P for explicit setting of ssh/scp port,
145	# decalpha parts (yonsei) removed
146	#
147	# 2013-02-02 07:06:13Z raasch
148	# adjustments for Kyushu-University computing center (lckyut - tatara)
149	# old changelog messages removed
150	#
151	# 1046 2012-11-09 14:38:45Z maronga
152	# code put under GPL (PALM 3.9)
153	#
154	# 08/07/94 - Siggi - first version finished
155	# 29/06/94 - Siggi - script development started
156	#--------------------------------------------------------------------------------#
157	# subjob - script for automatic generation and submission of batch-job files
158	# for various batch queuing systems
159	#--------------------------------------------------------------------------------#
160
161
162	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
163	delete_dayfile=false
164	email_notification=none
165	group_number=none
166	locat=normal
167	no_default_queue=none
168	no_submit=false
169	job_catalog="~/job_queue"
170	job_name=none
171	local_user=$LOGNAME
172	node_usage=shared
173	numprocs=0
174	punkte="..........................................................."
175	submcom=qsub
176	queue=default
177	remote_host=none
178	remote_user=""
179	verify=true
180
181	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
182	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
183
184
185
186	# ERROR HANDLING
187	# IN CASE OF EXIT:
188	trap 'if [[ $locat != normal ]]
189	then
190	case $locat in
191	(option) printf "\n --> available optios can be displayed"
192	printf " by typing:"
193	printf "\n \"subjob ?\" \n";;
194	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
195	(*) printf "\n +++ unknown error"
196	printf "\n please inform S. Raasch!\n"
197	esac
198	[[ -f $job_to_send ]] && rm $job_to_send
199	printf "\n\n+++ SUBJOB killed \n\n"
200	fi' exit
201
202
203	# IN CASE OF TERMINAL-BREAK:
204	trap '[[ -f $job_to_send ]] && rm $job_to_send
205	printf "\n\n+++ SUBJOB killed \n\n"
206	exit
207	' 2
208
209
210	# DETERMINE NAME OF LOCAL HOST
211	local_host=$(hostname)
212
213	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
214	# IS ADMITTED AT ALL)
215	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
216	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
217	case $local_host in
218	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
219	(atmos) local_address=172.20.25.35; local_host=lcide;;
220	(austru) local_address=130.75.105.128; local_host=lcmuk;;
221	(autan) local_address=130.75.105.57; local_host=lcmuk;;
222	(bora) local_address=130.75.105.103; local_host=lcmuk;;
223	(a0\|b0) local_address=133.5.4.33; local_host=lckyuh;;
224	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
225	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
226	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
227	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
228	(breva) local_address=130.75.105.98; local_host=lcmuk;;
229	(buran) local_address=130.75.105.58; local_host=lcmuk;;
230	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
231	(climate*) local_address=165.132.26.68; local_host=lcyon;;
232	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
233	(cs*) local_address=136.172.44.131; local_host=nech;;
234	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
235	(hpcl*) local_address=eddy.hpc.uni-oldenburg.de; local_host=lceddy;;
236	(cfd*) local_address=eddy.hpc.uni-oldenburg.de; local_host=lceddy;;
237	(node*) local_address=165.132.26.61 local_host=lck;;
238	# (node*) local_address=210.219.61.8 local_host=lckordi;;
239	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
240	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
241	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
242	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
243	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
244	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
245	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
246	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
247	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
248	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
249	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
250	(sno) local_address=130.75.105.113; local_host=lcmuk;;
251	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
252	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
253	(login*) local_address=118.128.66.201; local_host=lckiaps;;
254	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
255	(meller) local_address=134.106.74.155; local_host=lcfor;;
256	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
257	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
258	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
259	(nobel*) local_address=150.183.5.101; local_host=ibms;;
260	(ocean) local_address="ocean"; local_host=lcocean;;
261	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
262	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
263	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
264	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
265	(pingui) local_address=134.106.74.118; local_host=lcfor;;
266	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
267	(rte*) local_address=133.5.185.60; local_host=lcrte;;
268	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
269	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
270	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
271	(solano) local_address=130.75.105.110; local_host=lcmuk;;
272	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
273	(tc*) local_address="ocean"; local_host=lcocean;;
274	(t2a*) local_address=10.1.6.165; local_host=lctit;;
275	(urban*) local_address=147.46.30.151 local_host=lcsb;;
276	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
277	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
278	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
279	(*) printf "\n +++ \"$local_host\" unknown";
280	printf "\n please contact the PALM group at IMUK";
281	locat=parameter; exit;;
282	esac
283
284
285
286	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
287	remote_host=$local_host
288
289
290
291
292	# READ THE SHELLSCRIPT-OPTIONS
293	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
294	do
295	case $option in
296	(c) job_catalog=$OPTARG;;
297	(d) delete_dayfile=true;;
298	(D) no_submit=true;;
299	(e) email_notification=$OPTARG;;
300	(g) group_number=$OPTARG;;
301	(h) remote_host=$OPTARG;;
302	(m) memory=$OPTARG;;
303	(n) job_name=$OPTARG;;
304	(N) node_usage=$OPTARG;;
305	(O) threads_per_task=$OPTARG;;
306	(P) scp_port=$OPTARG;;
307	(q) no_default_queue=$OPTARG;;
308	(t) cputime=$OPTARG;;
309	(T) tasks_per_node=$OPTARG;;
310	(u) remote_user=$OPTARG;;
311	(v) verify=false;;
312	(X) numprocs=$OPTARG;;
313	(\?) printf "\n +++ Option $OPTARG unknown \n";
314	locat=option; exit;;
315	esac
316	done
317
318
319	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
320	(( to_shift = $OPTIND - 1 ))
321	shift $to_shift; file_to_send=$1
322
323
324	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
325	if [ "$1" = "?" ]
326	then
327	(printf "\n *** subjob can be called as follows:\n"
328	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
329	printf "\n Description of available options:\n"
330	printf "\n Option Description Default-Value"
331	printf "\n -c job-input- and output-catalog ~/job_queue"
332	printf "\n -d no job-protocol will be created ---"
333	printf "\n -D only the job-file will be created ---"
334	printf "\n -h execution host, available hosts: $remote_host"
335	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
336	printf "\n lckiaps, lctit, nech"
337	printf "\n -m memory demand per process in MByte ---"
338	printf "\n -n jobname <jobdatei>"
339	printf "\n -O threads per task (for OpenMP usage) 1"
340	printf "\n -P ssh/scp port default port"
341	printf "\n -q job-queue to be used default"
342	printf "\n -t allowed cpu-time in seconds ---"
343	printf "\n -T tasks per node (on parallel hosts) ---"
344	printf "\n -u username on execution host from .netrc"
345	printf "\n -v no prompt for confirmation ---"
346	printf "\n -X # of processors (on parallel hosts) 1"
347	printf "\n "
348	printf "\n The only possible positional parameter is <jobfile>:"
349	printf "\n The complete NQS-job must be provided here."
350	printf "\n <jobfile>=? creates this outline\n\n") \| more
351	exit
352	fi
353
354
355
356	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
357	if [[ "$file_to_send" = "" ]]
358	then
359	printf "\n +++ job-file missing"
360	locat=parameter; exit
361	else
362	if [[ -f $file_to_send ]]
363	then
364	true
365	else
366	printf "\n +++ job-file: "
367	printf "\n $file_to_send"
368	printf "\n does not exist"
369	locat=parameter; exit
370	fi
371	fi
372
373
374
375	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
376	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
377	if [[ $job_name = none ]]
378	then
379	job_name=$file_to_send
380	fi
381	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
382	then
383	printf "\n +++ job-file name: "
384	printf "\n $job_name"
385	printf "\n must not contain \"/\"-characters"
386	locat=parameter; exit
387	fi
388
389
390
391
392	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
393	# OR IF NO HOST HAS BEEN GIVEN
394	if [[ $remote_host = none ]]
395	then
396	printf "\n +++ host missing"
397	locat=option; exit
398	else
399	case $remote_host in
400	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
401	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
402	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
403	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
404	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
405	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
406	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
407	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
408	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
409	(lceddy) remote_address=eddy.hpc.uni-oldenburg.de; submcom=sbatch;;
410	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
411	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
412	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
413	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
414	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
415	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
416	(lcocean) remote_address="ocean"; submcom=qsub;;
417	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
418	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
419	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
420	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
421	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
422	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
423	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
424	locat=parameter; exit;;
425	esac
426	fi
427
428
429	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
430	if [[ $no_default_queue != none ]]
431	then
432	error=false
433	ndq=$no_default_queue
434	case $remote_host in
435	(ibm) case $ndq in
436	(p690_express\|p690_standard\|p690_long) error=false;;
437	(*) error=true;;
438	esac;;
439	(ibmh) case $ndq in
440	(cluster\|express) error=false;;
441	(*) error=true;;
442	esac;;
443	(ibmkisti) case $ndq in
444	(class.32plus\|class.1-2\|class.2-32) error=false;;
445	(*) error=true;;
446	esac;;
447	(ibmku) case $ndq in
448	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
449	(*) error=true;;
450	esac;;
451	(ibms) case $ndq in
452	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
453	(*) error=true;;
454	esac;;
455	(lcbullhh) case $ndq in
456	(compute\|compute2\|shared) error=false;;
457	(*) error=true;;
458	esac;;
459	(lccrayb) case $ndq in
460	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
461	(*) error=true;;
462	esac;;
463	(lccrayh) case $ndq in
464	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
465	(*) error=true;;
466	esac;;
467	(lccrayf) case $ndq in
468	(usup\|test*\|small\|large) error=false;;
469	(*) error=true;;
470	esac;;
471	(lceddy) case $ndq in
472	(eddy.p\|cfdh.p\|cfdl.p\|carl.p\|mpcs.p\|mpcl.p\|mpcb.p\|all_nodes.p) error=false;;
473	(*) error=true;;
474	esac;;
475	(lckiaps) case $ndq in
476	(express\|normal\|normal20\|quickq) error=false;;
477	(*) error=true;;
478	esac;;
479	(lckyoto) case $ndq in
480	(eh\|ph) error=false;;
481	(*) error=true;;
482	esac;;
483	(lckyuh) case $ndq in
484	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
485	(*) error=true;;
486	esac;;
487	(lckyut) case $ndq in
488	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
489	(*) error=true;;
490	esac;;
491	(lctit) case $ndq in
492	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
493	(*) error=true;;
494	esac;;
495	(t3eb) case $ndq in
496	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
497	(*) error=true;;
498	esac;;
499	(t3eh) case $ndq in
500	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
501	(*) error=true;;
502	esac;;
503	(t3ej2\|t3ej5) case $ndq in
504	(low\|normal\|high) error=false;;
505	(*) error=true;;
506	esac;;
507	(t3es) case $ndq in
508	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
509	(*) error=true;;
510	esac;;
511	esac
512	if [[ $error = true ]]
513	then
514	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
515	locat=parameter; exit
516	else
517	queue=$no_default_queue
518	fi
519	fi
520
521
522
523	# CHECK THE CPU-TIME
524	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
525	done=false
526	while [[ $done = false ]]
527	do
528	if (( $cputime <= 0 ))
529	then
530	printf "\n +++ wrong cpu-time or cpu-time missing"
531	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
532	printf "\n >>> "
533	read cputime 1>/dev/null 2>&1
534	else
535	done=true
536	fi
537	done
538	if [[ $remote_host = nech ]]
539	then
540	if (( tasks_per_node != 0 ))
541	then
542	(( cputime = cputime * tasks_per_node ))
543	elif [[ $numprocs != 0 ]]
544	then
545	(( cputime = cputime * numprocs ))
546	fi
547	fi
548	(( stunden = cputime / 3600 ))
549	(( resttime = cputime - stunden * 3600 ))
550	(( minuten = resttime / 60 ))
551	(( sekunden = resttime - minuten * 60 ))
552	timestring=${stunden}:${minuten}:${sekunden}
553
554
555
556	# CHECK THE MEMORY DEMAND
557	done=false
558	while [[ $done = false ]]
559	do
560	if (( memory <= 0 ))
561	then
562	printf "\n +++ wrong memory demand or memory demand missing"
563	printf "\n >>> Please type memory in MByte per process as INTEGER:"
564	printf "\n >>> "
565	read memory 1>/dev/null 2>&1
566	else
567	done=true
568	fi
569	done
570
571	if [[ $remote_host = nech ]]
572	then
573	if (( tasks_per_node != 0 ))
574	then
575	(( Memory = memory * tasks_per_node / 1000 ))
576	elif [[ $numprocs != 0 ]]
577	then
578	(( Memory = memory * numprocs / 1000 ))
579	else
580	(( Memory = memory / 1000 ))
581	fi
582	elif [[ $remote_host = lctit ]]
583	then
584	(( Memory = memory * tasks_per_node / 1000 ))
585	fi
586
587
588	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
589	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
590	then
591	(( memory = memory * threads_per_task ))
592	fi
593
594
595	# CALCULATE NUMBER OF REQUIRED NODES
596	if (( tasks_per_node != 0 ))
597	then
598	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
599	fi
600
601
602	# CALCULATE NUMBER OF PROCESSES PER NODE
603	(( processes_per_node = tasks_per_node * threads_per_task ))
604
605
606	# CALCULATE NUMBER OF MPI TASKS
607	(( mpi_tasks = numprocs / threads_per_task ))
608
609
610	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
611	if [[ "$scp_port" != "" ]]
612	then
613	PORTOPT="-P $scp_port"
614	SSH_PORTOPT="-p $scp_port"
615	fi
616
617
618	# HEADER-OUTPUT
619	if [[ $verify = true ]]
620	then
621	printf "\n\n"
622	printf "#--------------------------------------------------------------# \n"
623	spalte1=SUBJOB;spalte2=$(date)
624	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
625	printf "\| \| \n"
626	printf "\| values of parameters/options: \| \n"
627	spalte1=$(echo local_host$punkte \| cut -c-20)
628	spalte2=$punkte$local_host
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	spalte1=$(echo remote_host$punkte \| cut -c-20)
631	spalte2=$punkte$remote_host
632	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
633	spalte1=$(echo queue$punkte \| cut -c-20)
634	spalte2=$punkte$queue
635	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
636	spalte1=$(echo memory$punkte \| cut -c-20)
637	spalte2="$punkte$memory mb"
638	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
639	spalte1=$(echo cputime$punkte \| cut -c-20)
640	spalte2="$punkte$cputime sec"
641	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
642	spalte1=$(echo job_name$punkte \| cut -c-20)
643	spalte2="$punkte$job_name"
644	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
645	printf "#--------------------------------------------------------------# \n\n"
646
647
648	# QUERY CHECK
649	antwort="dummy"
650	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
651	do
652	read antwort?" >>> continue (y/n) ? "
653	done
654	if [[ $antwort = n \|\| $antwort = N ]]
655	then
656	locat=verify; exit
657	fi
658	printf "\n"
659	fi
660
661	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
662	identifier=$RANDOM
663	job_on_remhost=${job_name}_${identifier}_$local_host
664	job_to_send=job_to_send_$identifier
665	if [[ $delete_dayfile = false ]]
666	then
667	remote_dayfile=${local_host}_${job_name}_result_$identifier
668	local_dayfile=${remote_host}_${job_name}
669	else
670	remote_dayfile=/dev/null
671	fi
672
673
674	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
675	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
676	then
677
678	# GENERAL LOADLEVELER SETTINGS
679	execute_in_shell="#!/bin/ksh"
680	use_shell="# @ shell = /bin/ksh"
681	consumable_memory="ConsumableMemory($memory mb)"
682	class="# @ class = $queue"
683	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
684	network_to_use="# @ network.mpi = sn_all,shared,us"
685	data_limit="# @ data_limit = 1.76gb"
686	image_size="# @ image_size = 50"
687	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
688
689	if [[ $email_notification = none ]]
690	then
691	notify_user=""
692	else
693	notify_user="# @ notify_user = $email_notification"
694	if [[ $delete_dayfile = true ]]
695	then
696	notification='# @ notification = never'
697	fi
698	fi
699
700	if [[ $remote_host = ibmh ]]
701	then
702	data_limit=""
703	network_to_use=""
704	class="# @ class = $queue"
705	environment=""
706	rset="# @ rset = RSET_MCM_AFFINITY"
707	task_affinity="# @ task_affinity = core(1)"
708	elif [[ $remote_host = ibmkisti ]]
709	then
710	network_to_use="# @ network.MPI = sn_all,shared,US"
711	wall_clock_limit="# @ wall_clock_limit = $timestring"
712	if [[ $threads_per_task = 1 ]]
713	then
714	rset="# @ rset = RSET_MCM_AFFINITY"
715	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
716	fi
717	environment=""
718	use_shell=""
719	data_limit=""
720	image_size=""
721	elif [[ $remote_host = ibmku ]]
722	then
723	execute_in_shell="#!/usr/bin/ksh"
724	use_shell="# @ shell = /usr/bin/ksh"
725	consumable_memory=""
726	environment=""
727	network_to_use="# @ network.mpi = sn_all,shared,us"
728	data_limit=""
729	image_size=""
730	elif [[ $remote_host = ibms ]]
731	then
732	network_to_use="# @ network.mpi = csss,shared,us"
733	fi
734
735	cat > $job_to_send << %%END%%
736	$execute_in_shell
737	$use_shell
738
739	# @ job_type = parallel
740	# @ job_name = $job_name
741	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
742	# @ output = $remote_dayfile
743	# @ error = $remote_dayfile
744	$wall_clock_limit
745	$image_size
746	$class
747	$environment
748	$network_to_use
749	$data_limit
750	$rset
751	$mcm_affinity_options
752	$task_affinity
753	$notification
754	$notify_user
755
756	%%END%%
757
758	if (( nodes > 0 ))
759	then
760
761	if [[ $remote_host != ibmkisti ]]
762	then
763
764	cat >> $job_to_send << %%END%%
765	# @ node = $nodes
766	# @ tasks_per_node = $processes_per_node
767	# @ node_usage = $node_usage
768	# @ queue
769
770	%%END%%
771
772	else
773
774	cat >> $job_to_send << %%END%%
775	# @ total_tasks = $mpi_tasks
776	# @ blocking = unlimited
777	# @ queue
778
779	%%END%%
780
781	fi
782
783	else
784
785	cat >> $job_to_send << %%END%%
786	# @ blocking = unlimited
787	# @ total_tasks = $numprocs
788	# @ node_usage = $node_usage
789	# @ queue
790
791	%%END%%
792
793	fi
794
795	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
796	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
797	then
798	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
799	fi
800
801	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
802	then
803
804	cat > $job_to_send << %%END%%
805	#!/bin/ksh
806
807	# @ job_type = serial
808	# @ node_usage = $node_usage
809	# @ job_name = palm
810	# @ wall_clock_limit = ${timestring},$timestring
811	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
812	# @ output = $remote_dayfile
813	# @ error = $remote_dayfile
814	$class
815	$notification
816
817	# @ queue
818
819	%%END%%
820
821	elif [[ $remote_host = lcbullhh ]]
822	then
823	if [[ $numprocs != 0 ]]
824	then
825	cat > $job_to_send << %%END%%
826	#!/bin/bash -l
827	#SBATCH -J $job_name
828	#SBATCH -t $timestring
829	#SBATCH -N $nodes
830	#SBATCH --ntasks-per-node=$processes_per_node
831	#SBATCH -p $queue
832	#SBATCH -o $remote_dayfile
833	#SBATCH -e $remote_dayfile
834	#SBATCH -A $project_account
835
836	$init_cmds
837	$module_calls
838
839	%%END%%
840
841	else
842	cat > $job_to_send << %%END%%
843	#!/bin/bash -l
844	#SBATCH -J $job_name
845	#SBATCH -t $timestring
846	#SBATCH -l ncpus=1
847	#SBATCH -l pmem=${memory}mb
848	#SBATCH -m abe
849	#SBATCH -o $remote_dayfile
850	#SBATCH -e $remote_dayfile
851	#SBATCH -A $project_account
852
853	$init_cmds
854	$module_calls
855
856	%%END%%
857
858	fi
859
860	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
861	then
862
863	if [[ "$feature" != "" ]]
864	then
865	featuredir="#PBS -l feature=$feature"
866	fi
867
868	if [[ $numprocs != 0 ]]
869	then
870	cat > $job_to_send << %%END%%
871	#!/bin/bash
872	#PBS -N $job_name
873	#PBS -l walltime=$timestring
874	#PBS -l nodes=$nodes:ppn=$processes_per_node
875	#PBS -o $remote_dayfile
876	#PBS -j oe
877	#PBS -q $queue
878	$featuredir
879
880	$init_cmds
881	$module_calls
882
883	%%END%%
884
885	else
886
887	continue
888
889	fi
890
891	elif [[ $remote_host = lccrayf ]]
892	then
893
894	if [[ $numprocs != 0 ]]
895	then
896	cat > $job_to_send << %%END%%
897	#!/bin/bash -l
898	#SBATCH -J $job_name
899	#SBATCH -t $timestring
900	#SBATCH -N $nodes
901	#SBATCH --ntasks-per-node=$processes_per_node
902	#SBATCH -p $queue
903	#SBATCH -o $remote_dayfile
904	#SBATCH -e $remote_dayfile
905
906	$init_cmds
907	$module_calls
908
909	%%END%%
910
911	else
912	cat > $job_to_send << %%END%%
913	#!/bin/bash -l
914	#SBATCH -J $job_name
915	#SBATCH -t $timestring
916	#SBATCH -l ncpus=1
917	#SBATCH -l pmem=${memory}mb
918	#SBATCH -m abe
919	#SBATCH -o $remote_dayfile
920	#SBATCH -e $remote_dayfile
921
922	$init_cmds
923	$module_calls
924
925	%%END%%
926
927	fi
928
929
930
931	elif [[ $remote_host = lceddy ]]
932	then
933	cat > $job_to_send << %%END%%
934	#!/bin/bash
935	#SBATCH -J $job_name
936	#SBATCH -t $timestring
937	#SBATCH -n $numprocs
938	#SBATCH -N $nodes
939	#SBATCH --cpus-per-task 1
940	#SBATCH -p $queue
941	#SBATCH -o $remote_dayfile
942	#SBATCH -e $remote_dayfile
943	#SBATCH --mem-per-cpu $memory
944	#SBATCH --exclusive
945
946	module load PALMDependencies/.gcc
947
948	$init_cmds
949	$module_calls
950
951	%%END%%
952
953	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
954	then
955
956	if [[ $numprocs != 0 ]]
957	then
958	cat > $job_to_send << %%END%%
959	#!/bin/ksh
960	#PBS -N $job_name
961	#PBS -l walltime=$timestring
962	#PBS -l ncpus=$numprocs
963	#PBS -l pmem=${memory}mb
964	#PBS -o $remote_dayfile
965	#PBS -l nodes=$nodes:ppn=${processes_per_node}
966	#PBS -j oe
967
968	mpd &
969
970	%%END%%
971
972	else
973	cat > $job_to_send << %%END%%
974	#!/bin/ksh
975	#PBS -N $job_name
976	#PBS -l walltime=$timestring
977	#PBS -l ncpus=1
978	#PBS -l pmem=${memory}mb
979	#PBS -o $remote_dayfile
980	#PBS -j oe
981
982	%%END%%
983
984	fi
985
986	elif [[ $remote_host = lckiaps ]]
987	then
988
989	if [[ $numprocs != 0 ]]
990	then
991	cat > $job_to_send << %%END%%
992	#!/bin/bash
993	#PBS -N $job_name
994	#PBS -l walltime=$timestring
995	#PBS -l select=1:ncpus=$numprocs
996	#PBS -l pmem=${memory}mb
997	#PBS -q $queue
998	#PBS -o $remote_dayfile
999	#PBS -j oe
1000	#PBS -V
1001
1002	%%END%%
1003
1004	else
1005	cat > $job_to_send << %%END%%
1006	#!/bin/bash
1007	#PBS -N $job_name
1008	#PBS -l walltime=$timestring
1009	#PBS -l ncpus=1
1010	#PBS -l pmem=${memory}mb
1011	#PBS -o $remote_dayfile
1012	#PBS -j oe
1013
1014	%%END%%
1015
1016	fi
1017
1018	elif [[ $remote_host = lcyon ]]
1019	then
1020
1021	if [[ $numprocs != 0 ]]
1022	then
1023	cat > $job_to_send << %%END%%
1024	#!/bin/ksh
1025	#PBS -N $job_name
1026	#PBS -l walltime=$timestring
1027	#PBS -l ncpus=$numprocs
1028	#PBS -l pmem=${memory}mb
1029	#PBS -o $remote_dayfile
1030	#PBS -j oe
1031
1032	%%END%%
1033
1034	else
1035	cat > $job_to_send << %%END%%
1036	#!/bin/ksh
1037	#PBS -N $job_name
1038	#PBS -l walltime=$timestring
1039	#PBS -l ncpus=1
1040	#PBS -l pmem=${memory}mb
1041	#PBS -o $remote_dayfile
1042	#PBS -j oe
1043
1044	%%END%%
1045
1046	fi
1047
1048	elif [[ $remote_host = lcxe6 ]]
1049	then
1050
1051	if [[ $numprocs != 0 ]]
1052	then
1053	cat > $job_to_send << %%END%%
1054	#!/bin/ksh
1055	#PBS -S /bin/ksh
1056	#PBS -N $job_name
1057	#PBS -A $project_account
1058	#PBS -j oe
1059	#PBS -l walltime=$timestring
1060	#PBS -l mppwidth=${numprocs}
1061	#PBS -l mppnppn=${processes_per_node}
1062	#PBS -m abe
1063	#PBS -o $remote_dayfile
1064	$email_directive
1065
1066	$init_cmds
1067	$module_calls
1068
1069	%%END%%
1070
1071	else
1072	cat > $job_to_send << %%END%%
1073	#!/bin/ksh
1074	#PBS -S /bin/ksh
1075	#PBS -N $job_name
1076	#PBS -A $project_account
1077	#PBS -j oe
1078	#PBS -l walltime=$timestring
1079	#PBS -l ncpus=1
1080	#PBS -l pmem=${memory}mb
1081	#PBS -m abe
1082	$email_directive
1083	#PBS -o $remote_dayfile
1084
1085	$init_cmds
1086	$module_calls
1087
1088	%%END%%
1089
1090	fi
1091
1092	elif [[ $remote_host = lckyoto ]]
1093	then
1094
1095	cat > $job_to_send << %%END%%
1096	#!/bin/ksh
1097	# @\$-o $remote_dayfile
1098	# @\$-eo -oi
1099	# @\$-lP 16
1100	# @\$-lp 1
1101	# @\$-lm 28gb -llm unlimited -ls unlimited
1102	# @\$-q $queue
1103	# @\$-Pvn abs_pack
1104	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1105	#. /thin/local/etc/setprofile/intel-11.0.sh
1106	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1107	. ~/.myprofile
1108	#. /home2/t/t51254/palm/current_version/myprofile
1109	#. /thin/apps/pgi/mpi.sh
1110	#
1111	env
1112	#
1113	set -x
1114
1115	%%END%%
1116
1117	elif [[ $remote_host = lcxt5m ]]
1118	then
1119
1120	if [[ $numprocs != 0 ]]
1121	then
1122	cat > $job_to_send << %%END%%
1123	#!/bin/ksh
1124	#PBS -S /bin/ksh
1125	#PBS -N $job_name
1126	#PBS -j oe
1127	#PBS -l walltime=$timestring
1128	#PBS -l mppwidth=${numprocs}
1129	#PBS -l mppnppn=${processes_per_node}
1130	#PBS -m abe
1131	#PBS -o $remote_dayfile
1132
1133	$init_cmds
1134	$module_calls
1135
1136	%%END%%
1137
1138	else
1139	cat > $job_to_send << %%END%%
1140	#!/bin/ksh
1141	#PBS -S /bin/ksh
1142	#PBS -N $job_name
1143	#PBS -j oe
1144	#PBS -l walltime=$timestring
1145	#PBS -l ncpus=1
1146	#PBS -l pmem=${memory}mb
1147	#PBS -m abe
1148	#PBS -o $remote_dayfile
1149
1150	$init_cmds
1151	$module_calls
1152
1153	%%END%%
1154
1155	fi
1156
1157	elif [[ $remote_host = lckyuh ]]
1158	then
1159	cat > $job_to_send << %%END%%
1160	#!/bin/bash
1161	#PJM -L "rscgrp=$queue"
1162	#PJM -L "node=$nodes"
1163	#PJM --mpi "proc=$numprocs"
1164	#PJM -L "elapse=$timestring"
1165	#PJM -o $remote_dayfile
1166	#PJM -j
1167	#PJM -X
1168	#PJM --no-stging
1169
1170	export LANG=en_US.UTF-8
1171	%%END%%
1172
1173	elif [[ $remote_host = lckyut ]]
1174	then
1175	cat > $job_to_send << %%END%%
1176	#!/bin/bash
1177	#PJM -L "rscgrp=$queue"
1178	#PJM -L "vnode=$numprocs"
1179	#PJM -L "vnode-core=1"
1180	#PJM -L "elapse=$timestring"
1181	#PJM --mpi proc=$numprocs
1182	#PJM -o $remote_dayfile
1183	#PJM -j
1184	#PJM -X
1185	#PJM --no-stging
1186
1187	export LANG=en_US.UTF-8
1188	%%END%%
1189
1190	elif [[ $remote_host = lcocean ]]
1191	then
1192	cat > $job_to_send << %%END%%
1193	#!/bin/bash
1194	#$ -cwd
1195	#$ -V
1196	#$ -N $job_name
1197	#$ -pe orte $numprocs
1198	#$ -o $remote_dayfile
1199	#$ -j y
1200	#$ -R y
1201	$init_cmds
1202	$module_calls
1203
1204	%%END%%
1205
1206	elif [[ $remote_host = nech ]]
1207	then
1208
1209	if (( nodes > 1 ))
1210	then
1211	cat > $job_to_send << %%END%%
1212	#!/bin/ksh
1213	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1214	#PBS -l ${qsubmem}=${Memory}gb
1215	#PBS -b $nodes
1216	#PBS -o $remote_dayfile
1217	#PBS -N palm
1218	#PBS -j o
1219	#PBS -T mpisx
1220
1221	%%END%%
1222
1223	elif [[ $numprocs != 0 ]]
1224	then
1225	cat > $job_to_send << %%END%%
1226	#!/bin/ksh
1227	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1228	#PBS -l ${qsubmem}=${Memory}gb
1229	#PBS -o $remote_dayfile
1230	#PBS -N palm
1231	#PBS -j o
1232
1233	%%END%%
1234
1235	else
1236	cat > $job_to_send << %%END%%
1237	#!/bin/ksh
1238	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1239	#PBS -o $remote_dayfile
1240	#PBS -j o
1241
1242	%%END%%
1243
1244	fi
1245
1246	elif [[ $remote_host = lctit ]]
1247	then
1248	cat > $job_to_send << %%END%%
1249	#!/bin/ksh
1250	$init_cmds
1251	$module_calls
1252
1253	%%END%%
1254
1255	# SET OPTIONS FOR SUBMIT-COMMAND
1256	if [[ $tasks_per_node != $processes_per_node ]]
1257	then
1258	submcom="$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 $queue "
1259	else
1260	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -et 1 -q $queue "
1261	fi
1262
1263	else
1264
1265	cat > $job_to_send << %%END%%
1266	# @\$-q ${queue}
1267	# @\$-l${qsubtime} $timestring
1268	# @\$-l${qsubmem} ${memory}mb
1269	# @\$-o $remote_dayfile
1270	# @\$-eo
1271
1272	%%END%%
1273
1274	fi
1275
1276
1277	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1278	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1279	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1280	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1281	then
1282	echo "set +vx" >> $job_to_send
1283	echo "trap '" >> $job_to_send
1284	echo "set +vx" >> $job_to_send
1285	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 ]]
1286	then
1287	if [[ $remote_host = ibmh ]]
1288	then
1289	return_queue=c1
1290	elif [[ $remote_host = ibmkisti ]]
1291	then
1292	return_queue=class.1-2
1293	elif [[ $remote_host = ibmku ]]
1294	then
1295	return_queue=sdbg2
1296	elif [[ $remote_host = ibms ]]
1297	then
1298	return_queue=p_normal
1299	elif [[ $remote_host = lcbullhh ]]
1300	then
1301	return_queue=shared
1302	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1303	then
1304	return_queue=dataq
1305	elif [[ $remote_host = lcxe6 ]]
1306	then
1307	return_queue=debug
1308	elif [[ $remote_host = lckiaps ]]
1309	then
1310	return_queue=express
1311	elif [[ $remote_host = lckyuh ]]
1312	then
1313	return_queue=cx-single
1314	elif [[ $remote_host = lckyut ]]
1315	then
1316	return_queue=cx-single
1317	else
1318	return_queue=unknown
1319	fi
1320
1321	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1322	then
1323
1324	if [[ $remote_host = ibmku ]]
1325	then
1326	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1327	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1328	else
1329	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1330	fi
1331	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1333	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1334	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1335	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1336	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1337	if [[ $host != "ibmh" ]]
1338	then
1339	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1340	fi
1341	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1342	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1343
1344	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1345	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1346
1347	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1348	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
1349	if [[ $remote_host = ibmku ]]
1350	then
1351	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1352	fi
1353	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1354
1355	elif [[ $remote_host = nech ]]
1356	then
1357	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1358	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1359	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1360	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1361	echo "#PBS -j o" >> $job_to_send
1362	echo " " >> $job_to_send
1363	echo "set -x" >> $job_to_send
1364	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1365	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1366	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1367	echo "%%END%%" >> $job_to_send
1368
1369	elif [[ $remote_host = lcbullhh ]]
1370	then
1371	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1372	echo "#!/bin/bash" >> $job_to_send
1373	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1374	echo "#SBATCH -t 00:20:00" >> $job_to_send
1375	echo "#SBATCH -N 1" >> $job_to_send
1376	echo "#SBATCH -n 1" >> $job_to_send
1377	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1378	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1379	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1380	echo "#SBATCH -A $project_account" >> $job_to_send
1381	echo "#SBATCH -p $return_queue" >> $job_to_send
1382	echo " " >> $job_to_send
1383	echo "set -x" >> $job_to_send
1384	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1385	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1386	echo "%%END%%" >> $job_to_send
1387
1388	elif [[ $remote_host = lckyuh ]]
1389	then
1390	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1391	echo "#!/bin/bash" >> $job_to_send
1392	echo "#PJM -L \"node=1\"" >> $job_to_send
1393	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1394	echo "#PJM --no-stging" >> $job_to_send
1395	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1396	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1397	echo "#PJM -j" >> $job_to_send
1398	echo " " >> $job_to_send
1399	echo "export LANG=en_US.UTF-8" >> $job_to_send
1400	echo "set -x" >> $job_to_send
1401	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1402	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1403	echo "%%END%%" >> $job_to_send
1404
1405	elif [[ $remote_host = lckyut ]]
1406	then
1407	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1408	echo "#!/bin/bash" >> $job_to_send
1409	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1410	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1411	echo "#PJM --no-stging" >> $job_to_send
1412	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1413	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1414	echo "#PJM -j" >> $job_to_send
1415	echo " " >> $job_to_send
1416	echo "export LANG=en_US.UTF-8" >> $job_to_send
1417	echo "set -x" >> $job_to_send
1418	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1419	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1420	echo "%%END%%" >> $job_to_send
1421
1422	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1423	then
1424	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1425	echo "#!/bin/bash" >> $job_to_send
1426	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1427	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1428	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1429	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1430	echo "#PBS -j oe" >> $job_to_send
1431	echo " " >> $job_to_send
1432	echo "set -x" >> $job_to_send
1433	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1434	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1435	echo "%%END%%" >> $job_to_send
1436
1437	elif [[ $remote_host = lcocean ]]
1438	then
1439	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1440	echo "#!/bin/bash" >> $job_to_send
1441	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1442	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1443	echo "SGEPREFIX -cwd" >> $job_to_send
1444	echo "SGEPREFIX -j y" >> $job_to_send
1445	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1446	echo " " >> $job_to_send
1447	echo "set -x" >> $job_to_send
1448	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1449	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1450	echo "" >> $job_to_send
1451	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1452	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1453	echo "rm -f scpjob.${identifier}" >> $job_to_send
1454	echo "%%END%%" >> $job_to_send
1455	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1456	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1457
1458	elif [[ $remote_host = lcxe6 ]]
1459	then
1460	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1461	echo "#!/bin/ksh" >> $job_to_send
1462	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1463	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1464	echo "#PBS -A $project_account" >> $job_to_send
1465	echo "#PBS -l mppwidth=1" >> $job_to_send
1466	echo "#PBS -l mppnppn=1" >> $job_to_send
1467	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1468	echo "#PBS -j oe" >> $job_to_send
1469	echo " " >> $job_to_send
1470	echo "set -x" >> $job_to_send
1471	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1472	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1473	echo "%%END%%" >> $job_to_send
1474	else
1475
1476	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1477	echo "# @\\\$-q $return_queue" >> $job_to_send
1478	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1479	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1480	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1481	then
1482	echo "# @\$-l mpp_p=0" >> $job_to_send
1483	fi
1484	echo '# @\$-lF 10mb' >> $job_to_send
1485	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1486	echo '# @\\\$-eo' >> $job_to_send
1487	echo " " >> $job_to_send
1488	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1489	then
1490	echo "set +vx" >> $job_to_send
1491	echo ". .profile" >> $job_to_send
1492	fi
1493	echo "set -x" >> $job_to_send
1494	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
1495	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1496	echo "%%END%%" >> $job_to_send
1497
1498	fi
1499
1500	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1501	then
1502	echo "llsubmit scpjob.$identifier" >> $job_to_send
1503	elif [[ $remote_host = lcbullhh ]]
1504	then
1505	echo "sbatch scpjob.$identifier" >> $job_to_send
1506	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1507	then
1508	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1509	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1510	then
1511	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1512	elif [[ $remote_host = t3es ]]
1513	then
1514	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1515	elif [[ $remote_host = lckiaps ]]
1516	then
1517	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1518	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1519	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1520	elif [[ $remote_host = lckyu* ]]
1521	then
1522	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1523	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1524	elif [[ $remote_host = lcocean ]]
1525	then
1526	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1527	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1528	else
1529	echo "$submcom scpjob.$identifier" >> $job_to_send
1530	fi
1531	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1532	then
1533	echo "rm scpjob.$identifier" >> $job_to_send
1534	fi
1535	if [[ $remote_host = nech ]]
1536	then
1537	echo "cd -" >> $job_to_send
1538	fi
1539	else
1540	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1541	# ??? funktioniert das ÃŒberhaupt noch ???
1542	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1543	fi
1544	echo "set -x" >> $job_to_send
1545	echo " ' exit" >> $job_to_send
1546	echo "set -x" >> $job_to_send
1547	fi
1548
1549
1550	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1551	cat $file_to_send >> $job_to_send
1552
1553	if [[ $remote_host = ibm ]]
1554	then
1555	echo " " >> $job_to_send
1556	echo "exit" >> $job_to_send
1557	fi
1558
1559	# REMOVE JOB-FILE
1560	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1561	then
1562	echo " " >> $job_to_send
1563	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1564	fi
1565
1566
1567	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1568	if [[ $no_submit = false ]]
1569	then
1570	if [[ $remote_host != $local_host ]]
1571	then
1572	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1573	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1574	then
1575	job_catalog_save=$job_catalog
1576	job_catalog=job_queue
1577	elif [[ $remote_host = nech ]]
1578	then
1579	job_catalog_save=$job_catalog
1580	job_catalog=/hpf/b/${remote_user}/job_queue
1581	fi
1582	if [[ $remote_host = nech ]]
1583	then
1584	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1585	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1586	else
1587	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1588	fi
1589	if [[ $? = 1 ]]
1590	then
1591	locat=scp; exit
1592	fi
1593	if [[ $remote_host = ibms ]]
1594	then
1595	job_catalog=$job_catalog_save
1596	fi
1597	[[ $verify = true ]] && printf "\n >>> finished\n"
1598	else
1599	eval job_catalog=$job_catalog
1600	cp $job_to_send ${job_catalog}/$job_on_remhost
1601	fi
1602
1603
1604
1605	# START NQS- / LOADLEVELER-JOB
1606	if [[ $remote_host != $local_host ]]
1607	then
1608	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1609
1610	if [[ $remote_host = ibmku ]]
1611	then
1612	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1613	else
1614	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1615	fi
1616
1617	[[ $verify = true ]] && printf " >>> o.k.\n"
1618	else
1619	cd $job_catalog
1620	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray \|\| $local_host = lceddy ]]
1621	then
1622	eval $submcom $job_on_remhost
1623	elif [[ $local_host = lctit \|\| $local_host = lcxe6 \|\| $local_host = lck \|\| $local_host = lckordi \|\| $local_host = lcyon \|\| $local_host = lcsb \|\| $local_host = lckyu* ]]
1624	then
1625	chmod u+x $job_on_remhost
1626	eval $submcom $job_on_remhost
1627	elif [[ $local_host = nech ]]
1628	then
1629	if [[ $queue = default ]]
1630	then
1631	eval $submcom $job_on_remhost
1632	else
1633	eval $submcom -q $queue $job_on_remhost
1634	fi
1635	elif [[ $local_host = lcbullhh ]]
1636	then
1637	if [[ $queue = default ]]
1638	then
1639	eval $submcom $job_on_remhost
1640	fi
1641	else
1642	qsub $job_on_remhost
1643	fi
1644
1645	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku. THIS WILL BE DONE
1646	# AT THE END OF THE JOB
1647	if [[ $local_host != lctit && $local_host != ibmku ]]
1648	then
1649	rm $job_on_remhost
1650	fi
1651	cd - > /dev/null
1652	fi
1653	fi
1654
1655	# FINAL ACTIONS
1656	if [[ $no_submit = false ]]
1657	then
1658	rm -f $job_to_send
1659	fi
1660	[[ $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 |