Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1621

Last change on this file since 1621 was 1621, checked in by heinze, 9 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 55.2 KB

Line
1	#!/bin/ksh
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 1621 2015-07-17 11:39:33Z heinze $
30	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
31	#
32	# 1575 2015-03-27 09:56:27Z raasch
33	# mpp2-queues added to lccrayh
34	#
35	# 1547 2015-01-29 15:09:12Z witha
36	# adjustments for ForWind computing cluster (lcflow)
37	#
38	# 1545 2015-01-29 06:52:23Z heinze
39	# local host name for blizzard further specified
40	#
41	# 1480 2014-10-17 14:41:49Z raasch
42	# adjustments for 2nd stage of HLRNIII
43	#
44	# 1468 2014-09-24 14:06:57Z maronga
45	# Typo removed (addres->address)
46	# Adjustments for lcxe6
47	#
48	# 1452 2014-08-22 09:41:06Z heinze
49	# local hosts for blizzard added
50	#
51	# 1450 2014-08-21 07:31:51Z heinze
52	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
53	#
54	# 1442 2014-07-28 07:09:10Z raasch
55	# HLRN-III (lccrayb/lccrayh) queues adjusted
56	#
57	# 1378 2014-04-28 06:04:58Z raasch
58	# -et option added for lctit
59	#
60	# 1350 2014-04-04 13:01:30Z maronga
61	# location of qsub updated for lcxe6
62	#
63	# 1289 2014-03-04 07:12:34Z raasch
64	# German comments translated to English
65	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
66	#
67	# 1279 2014-01-28 12:10:14Z raasch
68	# node calculation modified due to changes in mrun (tasks_per_node must not be
69	# an integral divisor of numprocs any more)
70	#
71	# 1274 2014-01-09 13:14:54Z heinze
72	# adjustments for lccrayh
73	#
74	# 1266 2013-12-11 12:07:34Z heinze
75	# further adjustments for lccrayb (use msub instead of qsub)
76	#
77	# 1264 2013-12-09 12:46:09Z fricke
78	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
79	#
80	# 1262 2013-12-09 10:57:20Z fricke
81	# further adjustments for lccrayb
82	#
83	# 1260 2013-12-04 12:48:04Z raasch
84	# jaboticaba admitted
85	#
86	# 1255 2013-11-07 14:43:35Z raasch
87	# further adjustments for lccrayb
88	#
89	# 1224 2013-09-16 07:27:23Z raasch
90	# first adjustments for lccrayb
91	#
92	# 1202 2013-07-10 16:22:07Z witha
93	# adjustments for Forwind cluster (lcflow)
94	#
95	# 1199 2013-07-05 14:52:22Z raasch
96	# adjustments for CSC Helsinki (lccrayf)
97	#
98	# use of cluster/express queue enabled (ibmh)
99	# vinessa added (imuk)
100	#
101	# 1103 2013-02-20 02:15:53Z raasch
102	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
103	# instead typeset -L/R),
104	# further adjustments for lckyuh
105	#
106	# 2013-02-10 01:47:43Z raasch
107	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
108	# and for Forwind cluster (lcflow)
109	#
110	# 1094 2013-02-03 01:52:12Z raasch
111	# new option -P for explicit setting of ssh/scp port,
112	# decalpha parts (yonsei) removed
113	#
114	# 2013-02-02 07:06:13Z raasch
115	# adjustments for Kyushu-University computing center (lckyut - tatara)
116	# old changelog messages removed
117	#
118	# 1046 2012-11-09 14:38:45Z maronga
119	# code put under GPL (PALM 3.9)
120	#
121	# 08/07/94 - Siggi - first version finished
122	# 29/06/94 - Siggi - script development started
123	#--------------------------------------------------------------------------------#
124	# subjob - script for automatic generation and submission of batch-job files
125	# for various batch queuing systems
126	#--------------------------------------------------------------------------------#
127
128
129	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
130	delete_dayfile=false
131	email_notification=none
132	group_number=none
133	locat=normal
134	no_default_queue=none
135	no_submit=false
136	job_catalog="~/job_queue"
137	job_name=none
138	local_user=$LOGNAME
139	node_usage=shared
140	numprocs=0
141	punkte="..........................................................."
142	submcom=qsub
143	queue=default
144	remote_host=none
145	remote_user=""
146	verify=true
147
148	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
149	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
150
151
152
153	# ERROR HANDLING
154	# IN CASE OF EXIT:
155	trap 'if [[ $locat != normal ]]
156	then
157	case $locat in
158	(option) printf "\n --> available optios can be displayed"
159	printf " by typing:"
160	printf "\n \"subjob ?\" \n";;
161	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
162	(*) printf "\n +++ unknown error"
163	printf "\n please inform S. Raasch!\n"
164	esac
165	[[ -f $job_to_send ]] && rm $job_to_send
166	printf "\n\n+++ SUBJOB killed \n\n"
167	fi' exit
168
169
170	# IN CASE OF TERMINAL-BREAK:
171	trap '[[ -f $job_to_send ]] && rm $job_to_send
172	printf "\n\n+++ SUBJOB killed \n\n"
173	exit
174	' 2
175
176
177	# DETERMINE NAME OF LOCAL HOST
178	local_host=$(hostname)
179
180	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
181	# IS ADMITTED AT ALL)
182	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
183	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
184	case $local_host in
185	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
186	(atmos) local_address=172.20.25.35; local_host=lcide;;
187	(austru) local_address=130.75.105.128; local_host=lcmuk;;
188	(autan) local_address=130.75.105.57; local_host=lcmuk;;
189	(bora) local_address=130.75.105.103; local_host=lcmuk;;
190	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
191	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
192	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
193	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
194	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
195	(breva) local_address=130.75.105.98; local_host=lcmuk;;
196	(buran) local_address=130.75.105.58; local_host=lcmuk;;
197	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
198	(climate*) local_address=165.132.26.68; local_host=lcyon;;
199	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
200	(cs*) local_address=136.172.44.131; local_host=nech;;
201	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
202	(flow01) local_address=10.141.255.71; local_host=lcflow;;
203	(flow02) local_address=10.141.255.72; local_host=lcflow;;
204	(node*) local_address=165.132.26.61 local_host=lck;;
205	# (node*) local_address=210.219.61.8 local_host=lckordi;;
206	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
207	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
208	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
209	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
210	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
211	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
212	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
213	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
214	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
215	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
216	(sno) local_address=130.75.105.113; local_host=lcmuk;;
217	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
218	(login*) local_address=118.128.66.223; local_host=lckiaps;;
219	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
220	(meller) local_address=134.106.74.155; local_host=lcfor;;
221	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
222	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
223	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
224	(nobel*) local_address=150.183.5.101; local_host=ibms;;
225	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
226	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
227	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
228	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
229	(pingui) local_address=134.106.74.118; local_host=lcfor;;
230	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
231	(rte*) local_address=133.5.185.60; local_host=lcrte;;
232	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
233	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
234	(solano) local_address=130.75.105.110; local_host=lcmuk;;
235	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
236	(t2a*) local_address=10.1.6.165; local_host=lctit;;
237	(urban*) local_address=147.46.30.151 local_host=lcsb;;
238	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
239	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
240	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
241	(*) printf "\n +++ \"$local_host\" unknown";
242	printf "\n please contact the PALM group at IMUK";
243	locat=parameter; exit;;
244	esac
245
246
247
248	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
249	remote_host=$local_host
250
251
252
253
254	# READ THE SHELLSCRIPT-OPTIONS
255	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
256	do
257	case $option in
258	(c) job_catalog=$OPTARG;;
259	(d) delete_dayfile=true;;
260	(D) no_submit=true;;
261	(e) email_notification=$OPTARG;;
262	(g) group_number=$OPTARG;;
263	(h) remote_host=$OPTARG;;
264	(m) memory=$OPTARG;;
265	(n) job_name=$OPTARG;;
266	(N) node_usage=$OPTARG;;
267	(O) threads_per_task=$OPTARG;;
268	(P) scp_port=$OPTARG;;
269	(q) no_default_queue=$OPTARG;;
270	(t) cputime=$OPTARG;;
271	(T) tasks_per_node=$OPTARG;;
272	(u) remote_user=$OPTARG;;
273	(v) verify=false;;
274	(X) numprocs=$OPTARG;;
275	(\?) printf "\n +++ Option $OPTARG unknown \n";
276	locat=option; exit;;
277	esac
278	done
279
280
281	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
282	(( to_shift = $OPTIND - 1 ))
283	shift $to_shift; file_to_send=$1
284
285
286	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
287	if [ "$1" = "?" ]
288	then
289	(printf "\n *** subjob can be called as follows:\n"
290	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
291	printf "\n Description of available options:\n"
292	printf "\n Option Description Default-Value"
293	printf "\n -c job-input- and output-catalog ~/job_queue"
294	printf "\n -d no job-protocol will be created ---"
295	printf "\n -D only the job-file will be created ---"
296	printf "\n -h execution host, available hosts: $remote_host"
297	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
298	printf "\n lckiaps, lctit, nech"
299	printf "\n -m memory demand per process in MByte ---"
300	printf "\n -n jobname <jobdatei>"
301	printf "\n -O threads per task (for OpenMP usage) 1"
302	printf "\n -P ssh/scp port default port"
303	printf "\n -q job-queue to be used default"
304	printf "\n -t allowed cpu-time in seconds ---"
305	printf "\n -T tasks per node (on parallel hosts) ---"
306	printf "\n -u username on execution host from .netrc"
307	printf "\n -v no prompt for confirmation ---"
308	printf "\n -X # of processors (on parallel hosts) 1"
309	printf "\n "
310	printf "\n The only possible positional parameter is <jobfile>:"
311	printf "\n The complete NQS-job must be provided here."
312	printf "\n <jobfile>=? creates this outline\n\n") \| more
313	exit
314	fi
315
316
317
318	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
319	if [[ "$file_to_send" = "" ]]
320	then
321	printf "\n +++ job-file missing"
322	locat=parameter; exit
323	else
324	if [[ -f $file_to_send ]]
325	then
326	true
327	else
328	printf "\n +++ job-file: "
329	printf "\n $file_to_send"
330	printf "\n does not exist"
331	locat=parameter; exit
332	fi
333	fi
334
335
336
337	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
338	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
339	if [[ $job_name = none ]]
340	then
341	job_name=$file_to_send
342	fi
343	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
344	then
345	printf "\n +++ job-file name: "
346	printf "\n $job_name"
347	printf "\n must not contain \"/\"-characters"
348	locat=parameter; exit
349	fi
350
351
352
353
354	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
355	# OR IF NO HOST HAS BEEN GIVEN
356	if [[ $remote_host = none ]]
357	then
358	printf "\n +++ host missing"
359	locat=option; exit
360	else
361	case $remote_host in
362	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
363	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
364	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
365	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
366	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
367	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
368	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
369	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
370	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
371	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
372	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
373	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
374	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
375	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
376	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
377	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
378	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
379	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
380	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
381	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
382	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
383	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
384	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
385	locat=parameter; exit;;
386	esac
387	fi
388
389
390	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
391	if [[ $no_default_queue != none ]]
392	then
393	error=false
394	ndq=$no_default_queue
395	case $remote_host in
396	(ibm) case $ndq in
397	(p690_express\|p690_standard\|p690_long) error=false;;
398	(*) error=true;;
399	esac;;
400	(ibmh) case $ndq in
401	(cluster\|express) error=false;;
402	(*) error=true;;
403	esac;;
404	(ibmkisti) case $ndq in
405	(class.32plus\|class.1-2\|class.2-32) error=false;;
406	(*) error=true;;
407	esac;;
408	(ibmku) case $ndq in
409	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
410	(*) error=true;;
411	esac;;
412	(ibms) case $ndq in
413	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
414	(*) error=true;;
415	esac;;
416	(lcbullhh) case $ndq in
417	(compute\|shared) error=false;;
418	(*) error=true;;
419	esac;;
420	(lccrayb) case $ndq in
421	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
422	(*) error=true;;
423	esac;;
424	(lccrayh) case $ndq in
425	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
426	(*) error=true;;
427	esac;;
428	(lccrayf) case $ndq in
429	(usup\|test*\|small\|large) error=false;;
430	(*) error=true;;
431	esac;;
432	(lcflow) case $ndq in
433	(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;;
434	(*) error=true;;
435	esac;;
436	(lckiaps) case $ndq in
437	(express\|normal) error=false;;
438	(*) error=true;;
439	esac;;
440	(lckyoto) case $ndq in
441	(eh\|ph) error=false;;
442	(*) error=true;;
443	esac;;
444	(lckyuh) case $ndq in
445	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
446	(*) error=true;;
447	esac;;
448	(lckyut) case $ndq in
449	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
450	(*) error=true;;
451	esac;;
452	(lctit) case $ndq in
453	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
454	(*) error=true;;
455	esac;;
456	(t3eb) case $ndq in
457	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
458	(*) error=true;;
459	esac;;
460	(t3eh) case $ndq in
461	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
462	(*) error=true;;
463	esac;;
464	(t3ej2\|t3ej5) case $ndq in
465	(low\|normal\|high) error=false;;
466	(*) error=true;;
467	esac;;
468	(t3es) case $ndq in
469	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
470	(*) error=true;;
471	esac;;
472	esac
473	if [[ $error = true ]]
474	then
475	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
476	locat=parameter; exit
477	else
478	queue=$no_default_queue
479	fi
480	fi
481
482
483
484	# CHECK THE CPU-TIME
485	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
486	done=false
487	while [[ $done = false ]]
488	do
489	if (( $cputime <= 0 ))
490	then
491	printf "\n +++ wrong cpu-time or cpu-time missing"
492	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
493	printf "\n >>> "
494	read cputime 1>/dev/null 2>&1
495	else
496	done=true
497	fi
498	done
499	if [[ $remote_host = nech ]]
500	then
501	if (( tasks_per_node != 0 ))
502	then
503	(( cputime = cputime * tasks_per_node ))
504	elif [[ $numprocs != 0 ]]
505	then
506	(( cputime = cputime * numprocs ))
507	fi
508	fi
509	(( stunden = cputime / 3600 ))
510	(( resttime = cputime - stunden * 3600 ))
511	(( minuten = resttime / 60 ))
512	(( sekunden = resttime - minuten * 60 ))
513	timestring=${stunden}:${minuten}:${sekunden}
514
515
516
517	# CHECK THE MEMORY DEMAND
518	done=false
519	while [[ $done = false ]]
520	do
521	if (( memory <= 0 ))
522	then
523	printf "\n +++ wrong memory demand or memory demand missing"
524	printf "\n >>> Please type memory in MByte per process as INTEGER:"
525	printf "\n >>> "
526	read memory 1>/dev/null 2>&1
527	else
528	done=true
529	fi
530	done
531
532	if [[ $remote_host = nech ]]
533	then
534	if (( tasks_per_node != 0 ))
535	then
536	(( Memory = memory * tasks_per_node / 1000 ))
537	elif [[ $numprocs != 0 ]]
538	then
539	(( Memory = memory * numprocs / 1000 ))
540	else
541	(( Memory = memory / 1000 ))
542	fi
543	elif [[ $remote_host = lctit ]]
544	then
545	(( Memory = memory * tasks_per_node / 1000 ))
546	fi
547
548
549	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
550	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
551	then
552	(( memory = memory * threads_per_task ))
553	fi
554
555
556	# CALCULATE NUMBER OF REQUIRED NODES
557	if (( tasks_per_node != 0 ))
558	then
559	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
560	fi
561
562
563	# CALCULATE NUMBER OF PROCESSES PER NODE
564	(( processes_per_node = tasks_per_node * threads_per_task ))
565
566
567	# CALCULATE NUMBER OF MPI TASKS
568	(( mpi_tasks = numprocs / threads_per_task ))
569
570
571	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
572	if [[ "$scp_port" != "" ]]
573	then
574	PORTOPT="-P $scp_port"
575	SSH_PORTOPT="-p $scp_port"
576	fi
577
578
579	# HEADER-OUTPUT
580	if [[ $verify = true ]]
581	then
582	printf "\n\n"
583	printf "#--------------------------------------------------------------# \n"
584	spalte1=SUBJOB;spalte2=$(date)
585	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
586	printf "\| \| \n"
587	printf "\| values of parameters/options: \| \n"
588	spalte1=$(echo local_host$punkte \| cut -c-20)
589	spalte2=$punkte$local_host
590	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
591	spalte1=$(echo remote_host$punkte \| cut -c-20)
592	spalte2=$punkte$remote_host
593	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
594	spalte1=$(echo queue$punkte \| cut -c-20)
595	spalte2=$punkte$queue
596	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
597	spalte1=$(echo memory$punkte \| cut -c-20)
598	spalte2="$punkte$memory mb"
599	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
600	spalte1=$(echo cputime$punkte \| cut -c-20)
601	spalte2="$punkte$cputime sec"
602	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
603	spalte1=$(echo job_name$punkte \| cut -c-20)
604	spalte2="$punkte$job_name"
605	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
606	printf "#--------------------------------------------------------------# \n\n"
607
608
609	# QUERY CHECK
610	antwort="dummy"
611	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
612	do
613	read antwort?" >>> continue (y/n) ? "
614	done
615	if [[ $antwort = n \|\| $antwort = N ]]
616	then
617	locat=verify; exit
618	fi
619	printf "\n"
620	fi
621
622	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
623	identifier=$RANDOM
624	job_on_remhost=${job_name}_${identifier}_$local_host
625	job_to_send=job_to_send_$identifier
626	if [[ $delete_dayfile = false ]]
627	then
628	remote_dayfile=${local_host}_${job_name}_result_$identifier
629	local_dayfile=${remote_host}_${job_name}
630	else
631	remote_dayfile=/dev/null
632	fi
633
634
635	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
636	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
637	then
638
639	# GENERAL LOADLEVELER SETTINGS
640	execute_in_shell="#!/bin/ksh"
641	use_shell="# @ shell = /bin/ksh"
642	consumable_memory="ConsumableMemory($memory mb)"
643	class="# @ class = $queue"
644	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
645	network_to_use="# @ network.mpi = sn_all,shared,us"
646	data_limit="# @ data_limit = 1.76gb"
647	image_size="# @ image_size = 50"
648	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
649
650	if [[ $email_notification = none ]]
651	then
652	notify_user=""
653	else
654	notify_user="# @ notify_user = $email_notification"
655	if [[ $delete_dayfile = true ]]
656	then
657	notification='# @ notification = never'
658	fi
659	fi
660
661	if [[ $remote_host = ibmh ]]
662	then
663	data_limit=""
664	network_to_use=""
665	class="# @ class = $queue"
666	environment=""
667	rset="# @ rset = RSET_MCM_AFFINITY"
668	task_affinity="# @ task_affinity = core(1)"
669	elif [[ $remote_host = ibmkisti ]]
670	then
671	network_to_use="# @ network.MPI = sn_all,shared,US"
672	wall_clock_limit="# @ wall_clock_limit = $timestring"
673	if [[ $threads_per_task = 1 ]]
674	then
675	rset="# @ rset = RSET_MCM_AFFINITY"
676	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
677	fi
678	environment=""
679	use_shell=""
680	data_limit=""
681	image_size=""
682	elif [[ $remote_host = ibmku ]]
683	then
684	execute_in_shell="#!/usr/bin/ksh"
685	use_shell="# @ shell = /usr/bin/ksh"
686	consumable_memory=""
687	environment=""
688	network_to_use="# @ network.mpi = sn_all,shared,us"
689	data_limit=""
690	image_size=""
691	elif [[ $remote_host = ibms ]]
692	then
693	network_to_use="# @ network.mpi = csss,shared,us"
694	fi
695
696	cat > $job_to_send << %%END%%
697	$execute_in_shell
698	$use_shell
699
700	# @ job_type = parallel
701	# @ job_name = $job_name
702	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
703	# @ output = $remote_dayfile
704	# @ error = $remote_dayfile
705	$wall_clock_limit
706	$image_size
707	$class
708	$environment
709	$network_to_use
710	$data_limit
711	$rset
712	$mcm_affinity_options
713	$task_affinity
714	$notification
715	$notify_user
716
717	%%END%%
718
719	if (( nodes > 0 ))
720	then
721
722	if [[ $remote_host != ibmkisti ]]
723	then
724
725	cat >> $job_to_send << %%END%%
726	# @ node = $nodes
727	# @ tasks_per_node = $processes_per_node
728	# @ node_usage = $node_usage
729	# @ queue
730
731	%%END%%
732
733	else
734
735	cat >> $job_to_send << %%END%%
736	# @ total_tasks = $mpi_tasks
737	# @ blocking = unlimited
738	# @ queue
739
740	%%END%%
741
742	fi
743
744	else
745
746	cat >> $job_to_send << %%END%%
747	# @ blocking = unlimited
748	# @ total_tasks = $numprocs
749	# @ node_usage = $node_usage
750	# @ queue
751
752	%%END%%
753
754	fi
755
756	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
757	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
758	then
759	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
760	fi
761
762	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
763	then
764
765	cat > $job_to_send << %%END%%
766	#!/bin/ksh
767
768	# @ job_type = serial
769	# @ node_usage = $node_usage
770	# @ job_name = palm
771	# @ wall_clock_limit = ${timestring},$timestring
772	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
773	# @ output = $remote_dayfile
774	# @ error = $remote_dayfile
775	$class
776	$notification
777
778	# @ queue
779
780	%%END%%
781
782	elif [[ $remote_host = lcbullhh ]]
783	then
784	if [[ $numprocs != 0 ]]
785	then
786	cat > $job_to_send << %%END%%
787	#!/bin/bash -l
788	#SBATCH -J $job_name
789	#SBATCH -t $timestring
790	#SBATCH -N $nodes
791	#SBATCH --ntasks-per-node=$processes_per_node
792	#SBATCH -p $queue
793	#SBATCH -o $remote_dayfile
794	#SBATCH -e $remote_dayfile
795	#SBATCH -A $project_account
796
797	$init_cmds
798	$module_calls
799
800	%%END%%
801
802	else
803	cat > $job_to_send << %%END%%
804	#!/bin/bash -l
805	#SBATCH -J $job_name
806	#SBATCH -t $timestring
807	#SBATCH -l ncpus=1
808	#SBATCH -l pmem=${memory}mb
809	#SBATCH -m abe
810	#SBATCH -o $remote_dayfile
811	#SBATCH -e $remote_dayfile
812	#SBATCH -A $project_account
813
814	$init_cmds
815	$module_calls
816
817	%%END%%
818
819	fi
820
821	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
822	then
823
824	if [[ "$feature" != "" ]]
825	then
826	featuredir="#PBS -l feature=$feature"
827	fi
828
829	if [[ $numprocs != 0 ]]
830	then
831	cat > $job_to_send << %%END%%
832	#!/bin/bash
833	#PBS -N $job_name
834	#PBS -l walltime=$timestring
835	#PBS -l nodes=$nodes:ppn=$processes_per_node
836	#PBS -o $remote_dayfile
837	#PBS -j oe
838	#PBS -q $queue
839	$featuredir
840
841	$module_calls
842
843	%%END%%
844
845	else
846
847	continue
848
849	fi
850
851	elif [[ $remote_host = lccrayf ]]
852	then
853
854	if [[ $numprocs != 0 ]]
855	then
856	cat > $job_to_send << %%END%%
857	#!/bin/bash -l
858	#SBATCH -J $job_name
859	#SBATCH -t $timestring
860	#SBATCH -N $nodes
861	#SBATCH --ntasks-per-node=$processes_per_node
862	#SBATCH -p $queue
863	#SBATCH -o $remote_dayfile
864	#SBATCH -e $remote_dayfile
865
866	$init_cmds
867	$module_calls
868
869	%%END%%
870
871	else
872	cat > $job_to_send << %%END%%
873	#!/bin/bash -l
874	#SBATCH -J $job_name
875	#SBATCH -t $timestring
876	#SBATCH -l ncpus=1
877	#SBATCH -l pmem=${memory}mb
878	#SBATCH -m abe
879	#SBATCH -o $remote_dayfile
880	#SBATCH -e $remote_dayfile
881
882	$init_cmds
883	$module_calls
884
885	%%END%%
886
887	fi
888
889	elif [[ $remote_host = lcflow ]]
890	then
891	if [[ $numprocs != 0 ]]
892	then
893	pe_set="#$ -pe impi $numprocs"
894	else
895	pe_set="#$ -pe impi 1"
896	fi
897	if [[ $queue = default ]]
898	then
899	queue_set=""
900	else
901	queue_set="#$ -q $queue"
902	fi
903	[[ "$disc_space" = "" ]] && disc_space=50
904
905	cat > $job_to_send << %%END%%
906	#!/bin/bash
907	#$ -S /bin/bash
908	#$ -N $job_name
909	#$ -cwd
910	#$ -l h_rt=$timestring
911	#$ -l h_vmem=${memory}M
912	#$ -o $remote_dayfile
913	#$ -j y
914	$pe_set
915	#$ -R y
916	#$ -l h_fsize=${disc_space}G
917	$queue_set
918
919	%%END%%
920
921	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
922	then
923
924	if [[ $numprocs != 0 ]]
925	then
926	cat > $job_to_send << %%END%%
927	#!/bin/ksh
928	#PBS -N $job_name
929	#PBS -l walltime=$timestring
930	#PBS -l ncpus=$numprocs
931	#PBS -l pmem=${memory}mb
932	#PBS -o $remote_dayfile
933	#PBS -l nodes=$nodes:ppn=${processes_per_node}
934	#PBS -j oe
935
936	mpd &
937
938	%%END%%
939
940	else
941	cat > $job_to_send << %%END%%
942	#!/bin/ksh
943	#PBS -N $job_name
944	#PBS -l walltime=$timestring
945	#PBS -l ncpus=1
946	#PBS -l pmem=${memory}mb
947	#PBS -o $remote_dayfile
948	#PBS -j oe
949
950	%%END%%
951
952	fi
953
954	elif [[ $remote_host = lckiaps ]]
955	then
956
957	if [[ $numprocs != 0 ]]
958	then
959	cat > $job_to_send << %%END%%
960	#!/bin/ksh
961	#PBS -N $job_name
962	#PBS -l walltime=$timestring
963	#PBS -l select=1:ncpus=$numprocs
964	#PBS -l pmem=${memory}mb
965	#PBS -q $queue
966	#PBS -o $remote_dayfile
967	#PBS -j oe
968	#PBS -V
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 = lcyon ]]
987	then
988
989	if [[ $numprocs != 0 ]]
990	then
991	cat > $job_to_send << %%END%%
992	#!/bin/ksh
993	#PBS -N $job_name
994	#PBS -l walltime=$timestring
995	#PBS -l ncpus=$numprocs
996	#PBS -l pmem=${memory}mb
997	#PBS -o $remote_dayfile
998	#PBS -j oe
999
1000	%%END%%
1001
1002	else
1003	cat > $job_to_send << %%END%%
1004	#!/bin/ksh
1005	#PBS -N $job_name
1006	#PBS -l walltime=$timestring
1007	#PBS -l ncpus=1
1008	#PBS -l pmem=${memory}mb
1009	#PBS -o $remote_dayfile
1010	#PBS -j oe
1011
1012	%%END%%
1013
1014	fi
1015
1016	elif [[ $remote_host = lcxe6 ]]
1017	then
1018
1019	if [[ $numprocs != 0 ]]
1020	then
1021	cat > $job_to_send << %%END%%
1022	#!/bin/ksh
1023	#PBS -S /bin/ksh
1024	#PBS -N $job_name
1025	#PBS -A $project_account
1026	#PBS -j oe
1027	#PBS -l walltime=$timestring
1028	#PBS -l mppwidth=${numprocs}
1029	#PBS -l mppnppn=${processes_per_node}
1030	#PBS -m abe
1031	#PBS -o $remote_dayfile
1032	$email_directive
1033
1034	$init_cmds
1035	$module_calls
1036
1037	%%END%%
1038
1039	else
1040	cat > $job_to_send << %%END%%
1041	#!/bin/ksh
1042	#PBS -S /bin/ksh
1043	#PBS -N $job_name
1044	#PBS -A $project_account
1045	#PBS -j oe
1046	#PBS -l walltime=$timestring
1047	#PBS -l ncpus=1
1048	#PBS -l pmem=${memory}mb
1049	#PBS -m abe
1050	$email_directive
1051	#PBS -o $remote_dayfile
1052
1053	$init_cmds
1054	$module_calls
1055
1056	%%END%%
1057
1058	fi
1059
1060	elif [[ $remote_host = lckyoto ]]
1061	then
1062
1063	cat > $job_to_send << %%END%%
1064	#!/bin/ksh
1065	# @\$-o $remote_dayfile
1066	# @\$-eo -oi
1067	# @\$-lP 16
1068	# @\$-lp 1
1069	# @\$-lm 28gb -llm unlimited -ls unlimited
1070	# @\$-q $queue
1071	# @\$-Pvn abs_pack
1072	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1073	#. /thin/local/etc/setprofile/intel-11.0.sh
1074	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1075	. ~/.myprofile
1076	#. /home2/t/t51254/palm/current_version/myprofile
1077	#. /thin/apps/pgi/mpi.sh
1078	#
1079	env
1080	#
1081	set -x
1082
1083	%%END%%
1084
1085	elif [[ $remote_host = lcxt5m ]]
1086	then
1087
1088	if [[ $numprocs != 0 ]]
1089	then
1090	cat > $job_to_send << %%END%%
1091	#!/bin/ksh
1092	#PBS -S /bin/ksh
1093	#PBS -N $job_name
1094	#PBS -j oe
1095	#PBS -l walltime=$timestring
1096	#PBS -l mppwidth=${numprocs}
1097	#PBS -l mppnppn=${processes_per_node}
1098	#PBS -m abe
1099	#PBS -o $remote_dayfile
1100
1101	$init_cmds
1102	$module_calls
1103
1104	%%END%%
1105
1106	else
1107	cat > $job_to_send << %%END%%
1108	#!/bin/ksh
1109	#PBS -S /bin/ksh
1110	#PBS -N $job_name
1111	#PBS -j oe
1112	#PBS -l walltime=$timestring
1113	#PBS -l ncpus=1
1114	#PBS -l pmem=${memory}mb
1115	#PBS -m abe
1116	#PBS -o $remote_dayfile
1117
1118	$init_cmds
1119	$module_calls
1120
1121	%%END%%
1122
1123	fi
1124
1125	elif [[ $remote_host = lckyuh ]]
1126	then
1127	cat > $job_to_send << %%END%%
1128	#!/bin/bash
1129	#PJM -L "rscgrp=$queue"
1130	#PJM -L "node=$nodes"
1131	#PJM --mpi "proc=$numprocs"
1132	#PJM -L "elapse=$timestring"
1133	#PJM -o $remote_dayfile
1134	#PJM -j
1135	#PJM -X
1136	#PJM --no-stging
1137
1138	export LANG=en_US.UTF-8
1139	%%END%%
1140
1141	elif [[ $remote_host = lckyut ]]
1142	then
1143	cat > $job_to_send << %%END%%
1144	#!/bin/bash
1145	#PJM -L "rscgrp=$queue"
1146	#PJM -L "vnode=$numprocs"
1147	#PJM -L "vnode-core=1"
1148	#PJM -L "elapse=$timestring"
1149	#PJM --mpi proc=$numprocs
1150	#PJM -o $remote_dayfile
1151	#PJM -j
1152	#PJM -X
1153	#PJM --no-stging
1154
1155	export LANG=en_US.UTF-8
1156	%%END%%
1157
1158	elif [[ $remote_host = nech ]]
1159	then
1160
1161	if (( nodes > 1 ))
1162	then
1163	cat > $job_to_send << %%END%%
1164	#!/bin/ksh
1165	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1166	#PBS -l ${qsubmem}=${Memory}gb
1167	#PBS -b $nodes
1168	#PBS -o $remote_dayfile
1169	#PBS -N palm
1170	#PBS -j o
1171	#PBS -T mpisx
1172
1173	%%END%%
1174
1175	elif [[ $numprocs != 0 ]]
1176	then
1177	cat > $job_to_send << %%END%%
1178	#!/bin/ksh
1179	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1180	#PBS -l ${qsubmem}=${Memory}gb
1181	#PBS -o $remote_dayfile
1182	#PBS -N palm
1183	#PBS -j o
1184
1185	%%END%%
1186
1187	else
1188	cat > $job_to_send << %%END%%
1189	#!/bin/ksh
1190	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1191	#PBS -o $remote_dayfile
1192	#PBS -j o
1193
1194	%%END%%
1195
1196	fi
1197
1198	elif [[ $remote_host = lctit ]]
1199	then
1200	cat > $job_to_send << %%END%%
1201	#!/bin/ksh
1202	$init_cmds
1203	$module_calls
1204
1205	%%END%%
1206
1207	# SET OPTIONS FOR SUBMIT-COMMAND
1208	if [[ $tasks_per_node != $processes_per_node ]]
1209	then
1210	submcom="$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 $queue "
1211	else
1212	submcom="$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 "
1213	fi
1214
1215	else
1216
1217	cat > $job_to_send << %%END%%
1218	# @\$-q ${queue}
1219	# @\$-l${qsubtime} $timestring
1220	# @\$-l${qsubmem} ${memory}mb
1221	# @\$-o $remote_dayfile
1222	# @\$-eo
1223
1224	%%END%%
1225
1226	fi
1227
1228
1229	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1230	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1231	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1232	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1233	then
1234	echo "set +vx" >> $job_to_send
1235	echo "trap '" >> $job_to_send
1236	echo "set +vx" >> $job_to_send
1237	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 ]]
1238	then
1239	if [[ $remote_host = ibmh ]]
1240	then
1241	return_queue=c1
1242	elif [[ $remote_host = ibmkisti ]]
1243	then
1244	return_queue=class.1-2
1245	elif [[ $remote_host = ibmku ]]
1246	then
1247	return_queue=sdbg2
1248	elif [[ $remote_host = ibms ]]
1249	then
1250	return_queue=p_normal
1251	elif [[ $remote_host = lcbullhh ]]
1252	then
1253	return_queue=shared
1254	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1255	then
1256	return_queue=dataq
1257	elif [[ $remote_host = lcxe6 ]]
1258	then
1259	return_queue=debug
1260	elif [[ $remote_host = lckiaps ]]
1261	then
1262	return_queue=express
1263	elif [[ $remote_host = lckyuh ]]
1264	then
1265	return_queue=cx-single
1266	elif [[ $remote_host = lckyut ]]
1267	then
1268	return_queue=cx-single
1269	else
1270	return_queue=unknown
1271	fi
1272
1273	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1274	then
1275
1276	if [[ $remote_host = ibmku ]]
1277	then
1278	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1279	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1280	else
1281	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1282	fi
1283	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1284	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1285	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1286	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1287	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1288	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1289	if [[ $host != "ibmh" ]]
1290	then
1291	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1292	fi
1293	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1294	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1295
1296	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1297	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1298
1299	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1300	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
1301	if [[ $remote_host = ibmku ]]
1302	then
1303	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1304	fi
1305	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1306
1307	elif [[ $remote_host = nech ]]
1308	then
1309	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1310	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1311	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1312	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1313	echo "#PBS -j o" >> $job_to_send
1314	echo " " >> $job_to_send
1315	echo "set -x" >> $job_to_send
1316	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1317	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1318	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1319	echo "%%END%%" >> $job_to_send
1320
1321	elif [[ $remote_host = lcbullhh ]]
1322	then
1323	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1324	echo "#!/bin/bash" >> $job_to_send
1325	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1326	echo "#SBATCH -t 00:20:00" >> $job_to_send
1327	echo "#SBATCH -N 1" >> $job_to_send
1328	echo "#SBATCH -n 1" >> $job_to_send
1329	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1330	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1331	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1332	echo "#SBATCH -A $project_account" >> $job_to_send
1333	echo "#SBATCH -p $return_queue" >> $job_to_send
1334	echo " " >> $job_to_send
1335	echo "set -x" >> $job_to_send
1336	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1337	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1338	echo "%%END%%" >> $job_to_send
1339
1340	elif [[ $remote_host = lckyuh ]]
1341	then
1342	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1343	echo "#!/bin/bash" >> $job_to_send
1344	echo "#PJM -L \"node=1\"" >> $job_to_send
1345	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1346	echo "#PJM --no-stging" >> $job_to_send
1347	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1348	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1349	echo "#PJM -j" >> $job_to_send
1350	echo " " >> $job_to_send
1351	echo "export LANG=en_US.UTF-8" >> $job_to_send
1352	echo "set -x" >> $job_to_send
1353	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1354	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1355	echo "%%END%%" >> $job_to_send
1356
1357	elif [[ $remote_host = lckyut ]]
1358	then
1359	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1360	echo "#!/bin/bash" >> $job_to_send
1361	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1362	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1363	echo "#PJM --no-stging" >> $job_to_send
1364	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1365	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1366	echo "#PJM -j" >> $job_to_send
1367	echo " " >> $job_to_send
1368	echo "export LANG=en_US.UTF-8" >> $job_to_send
1369	echo "set -x" >> $job_to_send
1370	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1371	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1372	echo "%%END%%" >> $job_to_send
1373
1374	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1375	then
1376	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1377	echo "#!/bin/bash" >> $job_to_send
1378	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1379	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1380	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1381	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1382	echo "#PBS -j oe" >> $job_to_send
1383	echo " " >> $job_to_send
1384	echo "set -x" >> $job_to_send
1385	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1386	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1387	echo "%%END%%" >> $job_to_send
1388
1389	elif [[ $remote_host = lcflow ]]
1390	then
1391	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1392	echo "#!/bin/bash" >> $job_to_send
1393	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1394	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1395	echo "SGEPREFIX -cwd" >> $job_to_send
1396	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1397	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1398	echo "SGEPREFIX -j y" >> $job_to_send
1399	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1400	echo " " >> $job_to_send
1401	echo "set -x" >> $job_to_send
1402	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1403	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1404	echo "" >> $job_to_send
1405	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1406	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1407	echo "rm -f scpjob.${identifier}" >> $job_to_send
1408	echo "%%END%%" >> $job_to_send
1409	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1410	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1411	elif [[ $remote_host = lcxe6 ]]
1412	then
1413	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1414	echo "#!/bin/ksh" >> $job_to_send
1415	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1416	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1417	echo "#PBS -A $project_account" >> $job_to_send
1418	echo "#PBS -l mppwidth=1" >> $job_to_send
1419	echo "#PBS -l mppnppn=1" >> $job_to_send
1420	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1421	echo "#PBS -j oe" >> $job_to_send
1422	echo " " >> $job_to_send
1423	echo "set -x" >> $job_to_send
1424	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1425	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1426	echo "%%END%%" >> $job_to_send
1427	else
1428
1429	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1430	echo "# @\\\$-q $return_queue" >> $job_to_send
1431	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1432	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1433	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1434	then
1435	echo "# @\$-l mpp_p=0" >> $job_to_send
1436	fi
1437	echo '# @\$-lF 10mb' >> $job_to_send
1438	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1439	echo '# @\\\$-eo' >> $job_to_send
1440	echo " " >> $job_to_send
1441	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1442	then
1443	echo "set +vx" >> $job_to_send
1444	echo ". .profile" >> $job_to_send
1445	fi
1446	echo "set -x" >> $job_to_send
1447	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
1448	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1449	echo "%%END%%" >> $job_to_send
1450
1451	fi
1452
1453	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1454	then
1455	echo "llsubmit scpjob.$identifier" >> $job_to_send
1456	elif [[ $remote_host = lcbullhh ]]
1457	then
1458	echo "sbatch scpjob.$identifier" >> $job_to_send
1459	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1460	then
1461	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1462	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1463	then
1464	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1465	elif [[ $remote_host = t3es ]]
1466	then
1467	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1468	elif [[ $remote_host = lckiaps ]]
1469	then
1470	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1471	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1472	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1473	elif [[ $remote_host = lckyu* ]]
1474	then
1475	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1476	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1477	elif [[ $remote_host = lcflow ]]
1478	then
1479	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1480	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1481	else
1482	echo "$submcom scpjob.$identifier" >> $job_to_send
1483	fi
1484	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1485	then
1486	echo "rm scpjob.$identifier" >> $job_to_send
1487	fi
1488	if [[ $remote_host = nech ]]
1489	then
1490	echo "cd -" >> $job_to_send
1491	fi
1492	else
1493	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1494	# ??? funktioniert das ÃŒberhaupt noch ???
1495	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1496	fi
1497	echo "set -x" >> $job_to_send
1498	echo " ' exit" >> $job_to_send
1499	echo "set -x" >> $job_to_send
1500	fi
1501
1502
1503	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1504	cat $file_to_send >> $job_to_send
1505
1506	if [[ $remote_host = ibm ]]
1507	then
1508	echo " " >> $job_to_send
1509	echo "exit" >> $job_to_send
1510	fi
1511
1512	# REMOVE JOB-FILE
1513	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1514	then
1515	echo " " >> $job_to_send
1516	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1517	fi
1518
1519
1520	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1521	if [[ $no_submit = false ]]
1522	then
1523	if [[ $remote_host != $local_host ]]
1524	then
1525	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1526	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1527	then
1528	job_catalog_save=$job_catalog
1529	job_catalog=job_queue
1530	elif [[ $remote_host = nech ]]
1531	then
1532	job_catalog_save=$job_catalog
1533	job_catalog=/hpf/b/${remote_user}/job_queue
1534	fi
1535	if [[ $remote_host = nech ]]
1536	then
1537	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1538	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1539	else
1540	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1541	fi
1542	if [[ $? = 1 ]]
1543	then
1544	locat=scp; exit
1545	fi
1546	if [[ $remote_host = ibms ]]
1547	then
1548	job_catalog=$job_catalog_save
1549	fi
1550	[[ $verify = true ]] && printf "\n >>> finished\n"
1551	else
1552	eval job_catalog=$job_catalog
1553	cp $job_to_send ${job_catalog}/$job_on_remhost
1554	fi
1555
1556
1557
1558	# START NQS- / LOADLEVELER-JOB
1559	if [[ $remote_host != $local_host ]]
1560	then
1561	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1562
1563	if [[ $remote_host = ibmku ]]
1564	then
1565	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1566	elif [[ $remote_host = lcflow ]]
1567	then
1568	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1569	else
1570	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1571	fi
1572
1573	[[ $verify = true ]] && printf " >>> o.k.\n"
1574	else
1575	cd $job_catalog
1576	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1577	then
1578	eval $submcom $job_on_remhost
1579	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1580	then
1581	chmod u+x $job_on_remhost
1582	eval $submcom $job_on_remhost
1583	elif [[ $local_host = nech ]]
1584	then
1585	if [[ $queue = default ]]
1586	then
1587	eval $submcom $job_on_remhost
1588	else
1589	eval $submcom -q $queue $job_on_remhost
1590	fi
1591	elif [[ $local_host = lcbullhh ]]
1592	then
1593	if [[ $queue = default ]]
1594	then
1595	eval $submcom $job_on_remhost
1596	fi
1597	else
1598	qsub $job_on_remhost
1599	fi
1600
1601	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1602	# AT THE END OF THE JOB
1603	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1604	then
1605	rm $job_on_remhost
1606	fi
1607	cd - > /dev/null
1608	fi
1609	fi
1610
1611	# FINAL ACTIONS
1612	if [[ $no_submit = false ]]
1613	then
1614	rm -f $job_to_send
1615	fi
1616	[[ $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 |