Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1620

Last change on this file since 1620 was 1620, checked in by heinze, 9 years ago
adjustments for Mistral at DKRZ Hamburg (lcbullhh)
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	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1620 2015-07-17 11:37:41Z heinze $
30	#
31	# 1575 2015-03-27 09:56:27Z raasch
32	# mpp2-queues added to lccrayh
33	#
34	# 1547 2015-01-29 15:09:12Z witha
35	# adjustments for ForWind computing cluster (lcflow)
36	#
37	# 1545 2015-01-29 06:52:23Z heinze
38	# local host name for blizzard further specified
39	#
40	# 1480 2014-10-17 14:41:49Z raasch
41	# adjustments for 2nd stage of HLRNIII
42	#
43	# 1468 2014-09-24 14:06:57Z maronga
44	# Typo removed (addres->address)
45	# Adjustments for lcxe6
46	#
47	# 1452 2014-08-22 09:41:06Z heinze
48	# local hosts for blizzard added
49	#
50	# 1450 2014-08-21 07:31:51Z heinze
51	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
52	#
53	# 1442 2014-07-28 07:09:10Z raasch
54	# HLRN-III (lccrayb/lccrayh) queues adjusted
55	#
56	# 1378 2014-04-28 06:04:58Z raasch
57	# -et option added for lctit
58	#
59	# 1350 2014-04-04 13:01:30Z maronga
60	# location of qsub updated for lcxe6
61	#
62	# 1289 2014-03-04 07:12:34Z raasch
63	# German comments translated to English
64	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
65	#
66	# 1279 2014-01-28 12:10:14Z raasch
67	# node calculation modified due to changes in mrun (tasks_per_node must not be
68	# an integral divisor of numprocs any more)
69	#
70	# 1274 2014-01-09 13:14:54Z heinze
71	# adjustments for lccrayh
72	#
73	# 1266 2013-12-11 12:07:34Z heinze
74	# further adjustments for lccrayb (use msub instead of qsub)
75	#
76	# 1264 2013-12-09 12:46:09Z fricke
77	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
78	#
79	# 1262 2013-12-09 10:57:20Z fricke
80	# further adjustments for lccrayb
81	#
82	# 1260 2013-12-04 12:48:04Z raasch
83	# jaboticaba admitted
84	#
85	# 1255 2013-11-07 14:43:35Z raasch
86	# further adjustments for lccrayb
87	#
88	# 1224 2013-09-16 07:27:23Z raasch
89	# first adjustments for lccrayb
90	#
91	# 1202 2013-07-10 16:22:07Z witha
92	# adjustments for Forwind cluster (lcflow)
93	#
94	# 1199 2013-07-05 14:52:22Z raasch
95	# adjustments for CSC Helsinki (lccrayf)
96	#
97	# use of cluster/express queue enabled (ibmh)
98	# vinessa added (imuk)
99	#
100	# 1103 2013-02-20 02:15:53Z raasch
101	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
102	# instead typeset -L/R),
103	# further adjustments for lckyuh
104	#
105	# 2013-02-10 01:47:43Z raasch
106	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
107	# and for Forwind cluster (lcflow)
108	#
109	# 1094 2013-02-03 01:52:12Z raasch
110	# new option -P for explicit setting of ssh/scp port,
111	# decalpha parts (yonsei) removed
112	#
113	# 2013-02-02 07:06:13Z raasch
114	# adjustments for Kyushu-University computing center (lckyut - tatara)
115	# old changelog messages removed
116	#
117	# 1046 2012-11-09 14:38:45Z maronga
118	# code put under GPL (PALM 3.9)
119	#
120	# 08/07/94 - Siggi - first version finished
121	# 29/06/94 - Siggi - script development started
122	#--------------------------------------------------------------------------------#
123	# subjob - script for automatic generation and submission of batch-job files
124	# for various batch queuing systems
125	#--------------------------------------------------------------------------------#
126
127
128	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
129	delete_dayfile=false
130	email_notification=none
131	group_number=none
132	locat=normal
133	no_default_queue=none
134	no_submit=false
135	job_catalog="~/job_queue"
136	job_name=none
137	local_user=$LOGNAME
138	node_usage=shared
139	numprocs=0
140	punkte="..........................................................."
141	submcom=qsub
142	queue=default
143	remote_host=none
144	remote_user=""
145	verify=true
146
147	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
148	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
149
150
151
152	# ERROR HANDLING
153	# IN CASE OF EXIT:
154	trap 'if [[ $locat != normal ]]
155	then
156	case $locat in
157	(option) printf "\n --> available optios can be displayed"
158	printf " by typing:"
159	printf "\n \"subjob ?\" \n";;
160	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
161	(*) printf "\n +++ unknown error"
162	printf "\n please inform S. Raasch!\n"
163	esac
164	[[ -f $job_to_send ]] && rm $job_to_send
165	printf "\n\n+++ SUBJOB killed \n\n"
166	fi' exit
167
168
169	# IN CASE OF TERMINAL-BREAK:
170	trap '[[ -f $job_to_send ]] && rm $job_to_send
171	printf "\n\n+++ SUBJOB killed \n\n"
172	exit
173	' 2
174
175
176	# DETERMINE NAME OF LOCAL HOST
177	local_host=$(hostname)
178
179	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
180	# IS ADMITTED AT ALL)
181	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
182	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
183	case $local_host in
184	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
185	(atmos) local_address=172.20.25.35; local_host=lcide;;
186	(austru) local_address=130.75.105.128; local_host=lcmuk;;
187	(autan) local_address=130.75.105.57; local_host=lcmuk;;
188	(bora) local_address=130.75.105.103; local_host=lcmuk;;
189	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
190	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
191	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
192	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
193	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
194	(breva) local_address=130.75.105.98; local_host=lcmuk;;
195	(buran) local_address=130.75.105.58; local_host=lcmuk;;
196	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
197	(climate*) local_address=165.132.26.68; local_host=lcyon;;
198	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
199	(cs*) local_address=136.172.44.131; local_host=nech;;
200	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
201	(flow01) local_address=10.141.255.71; local_host=lcflow;;
202	(flow02) local_address=10.141.255.72; local_host=lcflow;;
203	(node*) local_address=165.132.26.61 local_host=lck;;
204	# (node*) local_address=210.219.61.8 local_host=lckordi;;
205	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
206	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
207	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
208	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
209	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
210	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
211	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
212	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
213	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
214	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
215	(sno) local_address=130.75.105.113; local_host=lcmuk;;
216	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
217	(login*) local_address=118.128.66.223; local_host=lckiaps;;
218	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
219	(meller) local_address=134.106.74.155; local_host=lcfor;;
220	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
221	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
222	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
223	(nobel*) local_address=150.183.5.101; local_host=ibms;;
224	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
225	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
226	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
227	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
228	(pingui) local_address=134.106.74.118; local_host=lcfor;;
229	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
230	(rte*) local_address=133.5.185.60; local_host=lcrte;;
231	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
232	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
233	(solano) local_address=130.75.105.110; local_host=lcmuk;;
234	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
235	(t2a*) local_address=10.1.6.165; local_host=lctit;;
236	(urban*) local_address=147.46.30.151 local_host=lcsb;;
237	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
238	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
239	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
240	(*) printf "\n +++ \"$local_host\" unknown";
241	printf "\n please contact the PALM group at IMUK";
242	locat=parameter; exit;;
243	esac
244
245
246
247	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
248	remote_host=$local_host
249
250
251
252
253	# READ THE SHELLSCRIPT-OPTIONS
254	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
255	do
256	case $option in
257	(c) job_catalog=$OPTARG;;
258	(d) delete_dayfile=true;;
259	(D) no_submit=true;;
260	(e) email_notification=$OPTARG;;
261	(g) group_number=$OPTARG;;
262	(h) remote_host=$OPTARG;;
263	(m) memory=$OPTARG;;
264	(n) job_name=$OPTARG;;
265	(N) node_usage=$OPTARG;;
266	(O) threads_per_task=$OPTARG;;
267	(P) scp_port=$OPTARG;;
268	(q) no_default_queue=$OPTARG;;
269	(t) cputime=$OPTARG;;
270	(T) tasks_per_node=$OPTARG;;
271	(u) remote_user=$OPTARG;;
272	(v) verify=false;;
273	(X) numprocs=$OPTARG;;
274	(\?) printf "\n +++ Option $OPTARG unknown \n";
275	locat=option; exit;;
276	esac
277	done
278
279
280	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
281	(( to_shift = $OPTIND - 1 ))
282	shift $to_shift; file_to_send=$1
283
284
285	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
286	if [ "$1" = "?" ]
287	then
288	(printf "\n *** subjob can be called as follows:\n"
289	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
290	printf "\n Description of available options:\n"
291	printf "\n Option Description Default-Value"
292	printf "\n -c job-input- and output-catalog ~/job_queue"
293	printf "\n -d no job-protocol will be created ---"
294	printf "\n -D only the job-file will be created ---"
295	printf "\n -h execution host, available hosts: $remote_host"
296	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
297	printf "\n lckiaps, lctit, nech"
298	printf "\n -m memory demand per process in MByte ---"
299	printf "\n -n jobname <jobdatei>"
300	printf "\n -O threads per task (for OpenMP usage) 1"
301	printf "\n -P ssh/scp port default port"
302	printf "\n -q job-queue to be used default"
303	printf "\n -t allowed cpu-time in seconds ---"
304	printf "\n -T tasks per node (on parallel hosts) ---"
305	printf "\n -u username on execution host from .netrc"
306	printf "\n -v no prompt for confirmation ---"
307	printf "\n -X # of processors (on parallel hosts) 1"
308	printf "\n "
309	printf "\n The only possible positional parameter is <jobfile>:"
310	printf "\n The complete NQS-job must be provided here."
311	printf "\n <jobfile>=? creates this outline\n\n") \| more
312	exit
313	fi
314
315
316
317	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
318	if [[ "$file_to_send" = "" ]]
319	then
320	printf "\n +++ job-file missing"
321	locat=parameter; exit
322	else
323	if [[ -f $file_to_send ]]
324	then
325	true
326	else
327	printf "\n +++ job-file: "
328	printf "\n $file_to_send"
329	printf "\n does not exist"
330	locat=parameter; exit
331	fi
332	fi
333
334
335
336	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
337	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
338	if [[ $job_name = none ]]
339	then
340	job_name=$file_to_send
341	fi
342	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
343	then
344	printf "\n +++ job-file name: "
345	printf "\n $job_name"
346	printf "\n must not contain \"/\"-characters"
347	locat=parameter; exit
348	fi
349
350
351
352
353	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
354	# OR IF NO HOST HAS BEEN GIVEN
355	if [[ $remote_host = none ]]
356	then
357	printf "\n +++ host missing"
358	locat=option; exit
359	else
360	case $remote_host in
361	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
362	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
363	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
364	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
365	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
366	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
367	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
368	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
369	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
370	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
371	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
372	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
373	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
374	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
375	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
376	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
377	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
378	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
379	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
380	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
381	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
382	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
383	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
384	locat=parameter; exit;;
385	esac
386	fi
387
388
389	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
390	if [[ $no_default_queue != none ]]
391	then
392	error=false
393	ndq=$no_default_queue
394	case $remote_host in
395	(ibm) case $ndq in
396	(p690_express\|p690_standard\|p690_long) error=false;;
397	(*) error=true;;
398	esac;;
399	(ibmh) case $ndq in
400	(cluster\|express) error=false;;
401	(*) error=true;;
402	esac;;
403	(ibmkisti) case $ndq in
404	(class.32plus\|class.1-2\|class.2-32) error=false;;
405	(*) error=true;;
406	esac;;
407	(ibmku) case $ndq in
408	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
409	(*) error=true;;
410	esac;;
411	(ibms) case $ndq in
412	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
413	(*) error=true;;
414	esac;;
415	(lcbullhh) case $ndq in
416	(compute\|shared) error=false;;
417	(*) error=true;;
418	esac;;
419	(lccrayb) case $ndq in
420	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
421	(*) error=true;;
422	esac;;
423	(lccrayh) case $ndq in
424	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
425	(*) error=true;;
426	esac;;
427	(lccrayf) case $ndq in
428	(usup\|test*\|small\|large) error=false;;
429	(*) error=true;;
430	esac;;
431	(lcflow) case $ndq in
432	(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;;
433	(*) error=true;;
434	esac;;
435	(lckiaps) case $ndq in
436	(express\|normal) error=false;;
437	(*) error=true;;
438	esac;;
439	(lckyoto) case $ndq in
440	(eh\|ph) error=false;;
441	(*) error=true;;
442	esac;;
443	(lckyuh) case $ndq in
444	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
445	(*) error=true;;
446	esac;;
447	(lckyut) case $ndq in
448	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
449	(*) error=true;;
450	esac;;
451	(lctit) case $ndq in
452	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
453	(*) error=true;;
454	esac;;
455	(t3eb) case $ndq in
456	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
457	(*) error=true;;
458	esac;;
459	(t3eh) case $ndq in
460	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
461	(*) error=true;;
462	esac;;
463	(t3ej2\|t3ej5) case $ndq in
464	(low\|normal\|high) error=false;;
465	(*) error=true;;
466	esac;;
467	(t3es) case $ndq in
468	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
469	(*) error=true;;
470	esac;;
471	esac
472	if [[ $error = true ]]
473	then
474	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
475	locat=parameter; exit
476	else
477	queue=$no_default_queue
478	fi
479	fi
480
481
482
483	# CHECK THE CPU-TIME
484	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
485	done=false
486	while [[ $done = false ]]
487	do
488	if (( $cputime <= 0 ))
489	then
490	printf "\n +++ wrong cpu-time or cpu-time missing"
491	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
492	printf "\n >>> "
493	read cputime 1>/dev/null 2>&1
494	else
495	done=true
496	fi
497	done
498	if [[ $remote_host = nech ]]
499	then
500	if (( tasks_per_node != 0 ))
501	then
502	(( cputime = cputime * tasks_per_node ))
503	elif [[ $numprocs != 0 ]]
504	then
505	(( cputime = cputime * numprocs ))
506	fi
507	fi
508	(( stunden = cputime / 3600 ))
509	(( resttime = cputime - stunden * 3600 ))
510	(( minuten = resttime / 60 ))
511	(( sekunden = resttime - minuten * 60 ))
512	timestring=${stunden}:${minuten}:${sekunden}
513
514
515
516	# CHECK THE MEMORY DEMAND
517	done=false
518	while [[ $done = false ]]
519	do
520	if (( memory <= 0 ))
521	then
522	printf "\n +++ wrong memory demand or memory demand missing"
523	printf "\n >>> Please type memory in MByte per process as INTEGER:"
524	printf "\n >>> "
525	read memory 1>/dev/null 2>&1
526	else
527	done=true
528	fi
529	done
530
531	if [[ $remote_host = nech ]]
532	then
533	if (( tasks_per_node != 0 ))
534	then
535	(( Memory = memory * tasks_per_node / 1000 ))
536	elif [[ $numprocs != 0 ]]
537	then
538	(( Memory = memory * numprocs / 1000 ))
539	else
540	(( Memory = memory / 1000 ))
541	fi
542	elif [[ $remote_host = lctit ]]
543	then
544	(( Memory = memory * tasks_per_node / 1000 ))
545	fi
546
547
548	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
549	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
550	then
551	(( memory = memory * threads_per_task ))
552	fi
553
554
555	# CALCULATE NUMBER OF REQUIRED NODES
556	if (( tasks_per_node != 0 ))
557	then
558	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
559	fi
560
561
562	# CALCULATE NUMBER OF PROCESSES PER NODE
563	(( processes_per_node = tasks_per_node * threads_per_task ))
564
565
566	# CALCULATE NUMBER OF MPI TASKS
567	(( mpi_tasks = numprocs / threads_per_task ))
568
569
570	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
571	if [[ "$scp_port" != "" ]]
572	then
573	PORTOPT="-P $scp_port"
574	SSH_PORTOPT="-p $scp_port"
575	fi
576
577
578	# HEADER-OUTPUT
579	if [[ $verify = true ]]
580	then
581	printf "\n\n"
582	printf "#--------------------------------------------------------------# \n"
583	spalte1=SUBJOB;spalte2=$(date)
584	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
585	printf "\| \| \n"
586	printf "\| values of parameters/options: \| \n"
587	spalte1=$(echo local_host$punkte \| cut -c-20)
588	spalte2=$punkte$local_host
589	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
590	spalte1=$(echo remote_host$punkte \| cut -c-20)
591	spalte2=$punkte$remote_host
592	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
593	spalte1=$(echo queue$punkte \| cut -c-20)
594	spalte2=$punkte$queue
595	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
596	spalte1=$(echo memory$punkte \| cut -c-20)
597	spalte2="$punkte$memory mb"
598	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
599	spalte1=$(echo cputime$punkte \| cut -c-20)
600	spalte2="$punkte$cputime sec"
601	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
602	spalte1=$(echo job_name$punkte \| cut -c-20)
603	spalte2="$punkte$job_name"
604	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
605	printf "#--------------------------------------------------------------# \n\n"
606
607
608	# QUERY CHECK
609	antwort="dummy"
610	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
611	do
612	read antwort?" >>> continue (y/n) ? "
613	done
614	if [[ $antwort = n \|\| $antwort = N ]]
615	then
616	locat=verify; exit
617	fi
618	printf "\n"
619	fi
620
621	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
622	identifier=$RANDOM
623	job_on_remhost=${job_name}_${identifier}_$local_host
624	job_to_send=job_to_send_$identifier
625	if [[ $delete_dayfile = false ]]
626	then
627	remote_dayfile=${local_host}_${job_name}_result_$identifier
628	local_dayfile=${remote_host}_${job_name}
629	else
630	remote_dayfile=/dev/null
631	fi
632
633
634	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
635	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
636	then
637
638	# GENERAL LOADLEVELER SETTINGS
639	execute_in_shell="#!/bin/ksh"
640	use_shell="# @ shell = /bin/ksh"
641	consumable_memory="ConsumableMemory($memory mb)"
642	class="# @ class = $queue"
643	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
644	network_to_use="# @ network.mpi = sn_all,shared,us"
645	data_limit="# @ data_limit = 1.76gb"
646	image_size="# @ image_size = 50"
647	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
648
649	if [[ $email_notification = none ]]
650	then
651	notify_user=""
652	else
653	notify_user="# @ notify_user = $email_notification"
654	if [[ $delete_dayfile = true ]]
655	then
656	notification='# @ notification = never'
657	fi
658	fi
659
660	if [[ $remote_host = ibmh ]]
661	then
662	data_limit=""
663	network_to_use=""
664	class="# @ class = $queue"
665	environment=""
666	rset="# @ rset = RSET_MCM_AFFINITY"
667	task_affinity="# @ task_affinity = core(1)"
668	elif [[ $remote_host = ibmkisti ]]
669	then
670	network_to_use="# @ network.MPI = sn_all,shared,US"
671	wall_clock_limit="# @ wall_clock_limit = $timestring"
672	if [[ $threads_per_task = 1 ]]
673	then
674	rset="# @ rset = RSET_MCM_AFFINITY"
675	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
676	fi
677	environment=""
678	use_shell=""
679	data_limit=""
680	image_size=""
681	elif [[ $remote_host = ibmku ]]
682	then
683	execute_in_shell="#!/usr/bin/ksh"
684	use_shell="# @ shell = /usr/bin/ksh"
685	consumable_memory=""
686	environment=""
687	network_to_use="# @ network.mpi = sn_all,shared,us"
688	data_limit=""
689	image_size=""
690	elif [[ $remote_host = ibms ]]
691	then
692	network_to_use="# @ network.mpi = csss,shared,us"
693	fi
694
695	cat > $job_to_send << %%END%%
696	$execute_in_shell
697	$use_shell
698
699	# @ job_type = parallel
700	# @ job_name = $job_name
701	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
702	# @ output = $remote_dayfile
703	# @ error = $remote_dayfile
704	$wall_clock_limit
705	$image_size
706	$class
707	$environment
708	$network_to_use
709	$data_limit
710	$rset
711	$mcm_affinity_options
712	$task_affinity
713	$notification
714	$notify_user
715
716	%%END%%
717
718	if (( nodes > 0 ))
719	then
720
721	if [[ $remote_host != ibmkisti ]]
722	then
723
724	cat >> $job_to_send << %%END%%
725	# @ node = $nodes
726	# @ tasks_per_node = $processes_per_node
727	# @ node_usage = $node_usage
728	# @ queue
729
730	%%END%%
731
732	else
733
734	cat >> $job_to_send << %%END%%
735	# @ total_tasks = $mpi_tasks
736	# @ blocking = unlimited
737	# @ queue
738
739	%%END%%
740
741	fi
742
743	else
744
745	cat >> $job_to_send << %%END%%
746	# @ blocking = unlimited
747	# @ total_tasks = $numprocs
748	# @ node_usage = $node_usage
749	# @ queue
750
751	%%END%%
752
753	fi
754
755	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
756	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
757	then
758	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
759	fi
760
761	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
762	then
763
764	cat > $job_to_send << %%END%%
765	#!/bin/ksh
766
767	# @ job_type = serial
768	# @ node_usage = $node_usage
769	# @ job_name = palm
770	# @ wall_clock_limit = ${timestring},$timestring
771	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
772	# @ output = $remote_dayfile
773	# @ error = $remote_dayfile
774	$class
775	$notification
776
777	# @ queue
778
779	%%END%%
780
781	elif [[ $remote_host = lcbullhh ]]
782	then
783	if [[ $numprocs != 0 ]]
784	then
785	cat > $job_to_send << %%END%%
786	#!/bin/bash -l
787	#SBATCH -J $job_name
788	#SBATCH -t $timestring
789	#SBATCH -N $nodes
790	#SBATCH --ntasks-per-node=$processes_per_node
791	#SBATCH -p $queue
792	#SBATCH -o $remote_dayfile
793	#SBATCH -e $remote_dayfile
794	#SBATCH -A $project_account
795
796	$init_cmds
797	$module_calls
798
799	%%END%%
800
801	else
802	cat > $job_to_send << %%END%%
803	#!/bin/bash -l
804	#SBATCH -J $job_name
805	#SBATCH -t $timestring
806	#SBATCH -l ncpus=1
807	#SBATCH -l pmem=${memory}mb
808	#SBATCH -m abe
809	#SBATCH -o $remote_dayfile
810	#SBATCH -e $remote_dayfile
811	#SBATCH -A $project_account
812
813	$init_cmds
814	$module_calls
815
816	%%END%%
817
818	fi
819
820	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
821	then
822
823	if [[ "$feature" != "" ]]
824	then
825	featuredir="#PBS -l feature=$feature"
826	fi
827
828	if [[ $numprocs != 0 ]]
829	then
830	cat > $job_to_send << %%END%%
831	#!/bin/bash
832	#PBS -N $job_name
833	#PBS -l walltime=$timestring
834	#PBS -l nodes=$nodes:ppn=$processes_per_node
835	#PBS -o $remote_dayfile
836	#PBS -j oe
837	#PBS -q $queue
838	$featuredir
839
840	$module_calls
841
842	%%END%%
843
844	else
845
846	continue
847
848	fi
849
850	elif [[ $remote_host = lccrayf ]]
851	then
852
853	if [[ $numprocs != 0 ]]
854	then
855	cat > $job_to_send << %%END%%
856	#!/bin/bash -l
857	#SBATCH -J $job_name
858	#SBATCH -t $timestring
859	#SBATCH -N $nodes
860	#SBATCH --ntasks-per-node=$processes_per_node
861	#SBATCH -p $queue
862	#SBATCH -o $remote_dayfile
863	#SBATCH -e $remote_dayfile
864
865	$init_cmds
866	$module_calls
867
868	%%END%%
869
870	else
871	cat > $job_to_send << %%END%%
872	#!/bin/bash -l
873	#SBATCH -J $job_name
874	#SBATCH -t $timestring
875	#SBATCH -l ncpus=1
876	#SBATCH -l pmem=${memory}mb
877	#SBATCH -m abe
878	#SBATCH -o $remote_dayfile
879	#SBATCH -e $remote_dayfile
880
881	$init_cmds
882	$module_calls
883
884	%%END%%
885
886	fi
887
888	elif [[ $remote_host = lcflow ]]
889	then
890	if [[ $numprocs != 0 ]]
891	then
892	pe_set="#$ -pe impi $numprocs"
893	else
894	pe_set="#$ -pe impi 1"
895	fi
896	if [[ $queue = default ]]
897	then
898	queue_set=""
899	else
900	queue_set="#$ -q $queue"
901	fi
902	[[ "$disc_space" = "" ]] && disc_space=50
903
904	cat > $job_to_send << %%END%%
905	#!/bin/bash
906	#$ -S /bin/bash
907	#$ -N $job_name
908	#$ -cwd
909	#$ -l h_rt=$timestring
910	#$ -l h_vmem=${memory}M
911	#$ -o $remote_dayfile
912	#$ -j y
913	$pe_set
914	#$ -R y
915	#$ -l h_fsize=${disc_space}G
916	$queue_set
917
918	%%END%%
919
920	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
921	then
922
923	if [[ $numprocs != 0 ]]
924	then
925	cat > $job_to_send << %%END%%
926	#!/bin/ksh
927	#PBS -N $job_name
928	#PBS -l walltime=$timestring
929	#PBS -l ncpus=$numprocs
930	#PBS -l pmem=${memory}mb
931	#PBS -o $remote_dayfile
932	#PBS -l nodes=$nodes:ppn=${processes_per_node}
933	#PBS -j oe
934
935	mpd &
936
937	%%END%%
938
939	else
940	cat > $job_to_send << %%END%%
941	#!/bin/ksh
942	#PBS -N $job_name
943	#PBS -l walltime=$timestring
944	#PBS -l ncpus=1
945	#PBS -l pmem=${memory}mb
946	#PBS -o $remote_dayfile
947	#PBS -j oe
948
949	%%END%%
950
951	fi
952
953	elif [[ $remote_host = lckiaps ]]
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 select=1:ncpus=$numprocs
963	#PBS -l pmem=${memory}mb
964	#PBS -q $queue
965	#PBS -o $remote_dayfile
966	#PBS -j oe
967	#PBS -V
968
969	%%END%%
970
971	else
972	cat > $job_to_send << %%END%%
973	#!/bin/ksh
974	#PBS -N $job_name
975	#PBS -l walltime=$timestring
976	#PBS -l ncpus=1
977	#PBS -l pmem=${memory}mb
978	#PBS -o $remote_dayfile
979	#PBS -j oe
980
981	%%END%%
982
983	fi
984
985	elif [[ $remote_host = lcyon ]]
986	then
987
988	if [[ $numprocs != 0 ]]
989	then
990	cat > $job_to_send << %%END%%
991	#!/bin/ksh
992	#PBS -N $job_name
993	#PBS -l walltime=$timestring
994	#PBS -l ncpus=$numprocs
995	#PBS -l pmem=${memory}mb
996	#PBS -o $remote_dayfile
997	#PBS -j oe
998
999	%%END%%
1000
1001	else
1002	cat > $job_to_send << %%END%%
1003	#!/bin/ksh
1004	#PBS -N $job_name
1005	#PBS -l walltime=$timestring
1006	#PBS -l ncpus=1
1007	#PBS -l pmem=${memory}mb
1008	#PBS -o $remote_dayfile
1009	#PBS -j oe
1010
1011	%%END%%
1012
1013	fi
1014
1015	elif [[ $remote_host = lcxe6 ]]
1016	then
1017
1018	if [[ $numprocs != 0 ]]
1019	then
1020	cat > $job_to_send << %%END%%
1021	#!/bin/ksh
1022	#PBS -S /bin/ksh
1023	#PBS -N $job_name
1024	#PBS -A $project_account
1025	#PBS -j oe
1026	#PBS -l walltime=$timestring
1027	#PBS -l mppwidth=${numprocs}
1028	#PBS -l mppnppn=${processes_per_node}
1029	#PBS -m abe
1030	#PBS -o $remote_dayfile
1031	$email_directive
1032
1033	$init_cmds
1034	$module_calls
1035
1036	%%END%%
1037
1038	else
1039	cat > $job_to_send << %%END%%
1040	#!/bin/ksh
1041	#PBS -S /bin/ksh
1042	#PBS -N $job_name
1043	#PBS -A $project_account
1044	#PBS -j oe
1045	#PBS -l walltime=$timestring
1046	#PBS -l ncpus=1
1047	#PBS -l pmem=${memory}mb
1048	#PBS -m abe
1049	$email_directive
1050	#PBS -o $remote_dayfile
1051
1052	$init_cmds
1053	$module_calls
1054
1055	%%END%%
1056
1057	fi
1058
1059	elif [[ $remote_host = lckyoto ]]
1060	then
1061
1062	cat > $job_to_send << %%END%%
1063	#!/bin/ksh
1064	# @\$-o $remote_dayfile
1065	# @\$-eo -oi
1066	# @\$-lP 16
1067	# @\$-lp 1
1068	# @\$-lm 28gb -llm unlimited -ls unlimited
1069	# @\$-q $queue
1070	# @\$-Pvn abs_pack
1071	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1072	#. /thin/local/etc/setprofile/intel-11.0.sh
1073	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1074	. ~/.myprofile
1075	#. /home2/t/t51254/palm/current_version/myprofile
1076	#. /thin/apps/pgi/mpi.sh
1077	#
1078	env
1079	#
1080	set -x
1081
1082	%%END%%
1083
1084	elif [[ $remote_host = lcxt5m ]]
1085	then
1086
1087	if [[ $numprocs != 0 ]]
1088	then
1089	cat > $job_to_send << %%END%%
1090	#!/bin/ksh
1091	#PBS -S /bin/ksh
1092	#PBS -N $job_name
1093	#PBS -j oe
1094	#PBS -l walltime=$timestring
1095	#PBS -l mppwidth=${numprocs}
1096	#PBS -l mppnppn=${processes_per_node}
1097	#PBS -m abe
1098	#PBS -o $remote_dayfile
1099
1100	$init_cmds
1101	$module_calls
1102
1103	%%END%%
1104
1105	else
1106	cat > $job_to_send << %%END%%
1107	#!/bin/ksh
1108	#PBS -S /bin/ksh
1109	#PBS -N $job_name
1110	#PBS -j oe
1111	#PBS -l walltime=$timestring
1112	#PBS -l ncpus=1
1113	#PBS -l pmem=${memory}mb
1114	#PBS -m abe
1115	#PBS -o $remote_dayfile
1116
1117	$init_cmds
1118	$module_calls
1119
1120	%%END%%
1121
1122	fi
1123
1124	elif [[ $remote_host = lckyuh ]]
1125	then
1126	cat > $job_to_send << %%END%%
1127	#!/bin/bash
1128	#PJM -L "rscgrp=$queue"
1129	#PJM -L "node=$nodes"
1130	#PJM --mpi "proc=$numprocs"
1131	#PJM -L "elapse=$timestring"
1132	#PJM -o $remote_dayfile
1133	#PJM -j
1134	#PJM -X
1135	#PJM --no-stging
1136
1137	export LANG=en_US.UTF-8
1138	%%END%%
1139
1140	elif [[ $remote_host = lckyut ]]
1141	then
1142	cat > $job_to_send << %%END%%
1143	#!/bin/bash
1144	#PJM -L "rscgrp=$queue"
1145	#PJM -L "vnode=$numprocs"
1146	#PJM -L "vnode-core=1"
1147	#PJM -L "elapse=$timestring"
1148	#PJM --mpi proc=$numprocs
1149	#PJM -o $remote_dayfile
1150	#PJM -j
1151	#PJM -X
1152	#PJM --no-stging
1153
1154	export LANG=en_US.UTF-8
1155	%%END%%
1156
1157	elif [[ $remote_host = nech ]]
1158	then
1159
1160	if (( nodes > 1 ))
1161	then
1162	cat > $job_to_send << %%END%%
1163	#!/bin/ksh
1164	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1165	#PBS -l ${qsubmem}=${Memory}gb
1166	#PBS -b $nodes
1167	#PBS -o $remote_dayfile
1168	#PBS -N palm
1169	#PBS -j o
1170	#PBS -T mpisx
1171
1172	%%END%%
1173
1174	elif [[ $numprocs != 0 ]]
1175	then
1176	cat > $job_to_send << %%END%%
1177	#!/bin/ksh
1178	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1179	#PBS -l ${qsubmem}=${Memory}gb
1180	#PBS -o $remote_dayfile
1181	#PBS -N palm
1182	#PBS -j o
1183
1184	%%END%%
1185
1186	else
1187	cat > $job_to_send << %%END%%
1188	#!/bin/ksh
1189	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1190	#PBS -o $remote_dayfile
1191	#PBS -j o
1192
1193	%%END%%
1194
1195	fi
1196
1197	elif [[ $remote_host = lctit ]]
1198	then
1199	cat > $job_to_send << %%END%%
1200	#!/bin/ksh
1201	$init_cmds
1202	$module_calls
1203
1204	%%END%%
1205
1206	# SET OPTIONS FOR SUBMIT-COMMAND
1207	if [[ $tasks_per_node != $processes_per_node ]]
1208	then
1209	submcom="$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 $queue "
1210	else
1211	submcom="$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 "
1212	fi
1213
1214	else
1215
1216	cat > $job_to_send << %%END%%
1217	# @\$-q ${queue}
1218	# @\$-l${qsubtime} $timestring
1219	# @\$-l${qsubmem} ${memory}mb
1220	# @\$-o $remote_dayfile
1221	# @\$-eo
1222
1223	%%END%%
1224
1225	fi
1226
1227
1228	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1229	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1230	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1231	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1232	then
1233	echo "set +vx" >> $job_to_send
1234	echo "trap '" >> $job_to_send
1235	echo "set +vx" >> $job_to_send
1236	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 ]]
1237	then
1238	if [[ $remote_host = ibmh ]]
1239	then
1240	return_queue=c1
1241	elif [[ $remote_host = ibmkisti ]]
1242	then
1243	return_queue=class.1-2
1244	elif [[ $remote_host = ibmku ]]
1245	then
1246	return_queue=sdbg2
1247	elif [[ $remote_host = ibms ]]
1248	then
1249	return_queue=p_normal
1250	elif [[ $remote_host = lcbullhh ]]
1251	then
1252	return_queue=shared
1253	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1254	then
1255	return_queue=dataq
1256	elif [[ $remote_host = lcxe6 ]]
1257	then
1258	return_queue=debug
1259	elif [[ $remote_host = lckiaps ]]
1260	then
1261	return_queue=express
1262	elif [[ $remote_host = lckyuh ]]
1263	then
1264	return_queue=cx-single
1265	elif [[ $remote_host = lckyut ]]
1266	then
1267	return_queue=cx-single
1268	else
1269	return_queue=unknown
1270	fi
1271
1272	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1273	then
1274
1275	if [[ $remote_host = ibmku ]]
1276	then
1277	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1278	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1279	else
1280	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1281	fi
1282	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1283	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1284	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1285	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1286	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1287	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1288	if [[ $host != "ibmh" ]]
1289	then
1290	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1291	fi
1292	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1293	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1294
1295	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1296	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1297
1298	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1299	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
1300	if [[ $remote_host = ibmku ]]
1301	then
1302	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1303	fi
1304	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1305
1306	elif [[ $remote_host = nech ]]
1307	then
1308	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1309	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1310	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1311	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1312	echo "#PBS -j o" >> $job_to_send
1313	echo " " >> $job_to_send
1314	echo "set -x" >> $job_to_send
1315	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1316	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1317	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1318	echo "%%END%%" >> $job_to_send
1319
1320	elif [[ $remote_host = lcbullhh ]]
1321	then
1322	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1323	echo "#!/bin/bash" >> $job_to_send
1324	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1325	echo "#SBATCH -t 00:20:00" >> $job_to_send
1326	echo "#SBATCH -N 1" >> $job_to_send
1327	echo "#SBATCH -n 1" >> $job_to_send
1328	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1329	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1330	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1331	echo "#SBATCH -A $project_account" >> $job_to_send
1332	echo "#SBATCH -p $return_queue" >> $job_to_send
1333	echo " " >> $job_to_send
1334	echo "set -x" >> $job_to_send
1335	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1336	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1337	echo "%%END%%" >> $job_to_send
1338
1339	elif [[ $remote_host = lckyuh ]]
1340	then
1341	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1342	echo "#!/bin/bash" >> $job_to_send
1343	echo "#PJM -L \"node=1\"" >> $job_to_send
1344	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1345	echo "#PJM --no-stging" >> $job_to_send
1346	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1347	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1348	echo "#PJM -j" >> $job_to_send
1349	echo " " >> $job_to_send
1350	echo "export LANG=en_US.UTF-8" >> $job_to_send
1351	echo "set -x" >> $job_to_send
1352	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1353	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1354	echo "%%END%%" >> $job_to_send
1355
1356	elif [[ $remote_host = lckyut ]]
1357	then
1358	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1359	echo "#!/bin/bash" >> $job_to_send
1360	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1361	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1362	echo "#PJM --no-stging" >> $job_to_send
1363	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1364	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1365	echo "#PJM -j" >> $job_to_send
1366	echo " " >> $job_to_send
1367	echo "export LANG=en_US.UTF-8" >> $job_to_send
1368	echo "set -x" >> $job_to_send
1369	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1370	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1371	echo "%%END%%" >> $job_to_send
1372
1373	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1374	then
1375	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1376	echo "#!/bin/bash" >> $job_to_send
1377	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1378	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1379	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1380	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1381	echo "#PBS -j oe" >> $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 [[ $remote_host = lcflow ]]
1389	then
1390	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1391	echo "#!/bin/bash" >> $job_to_send
1392	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1393	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1394	echo "SGEPREFIX -cwd" >> $job_to_send
1395	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1396	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1397	echo "SGEPREFIX -j y" >> $job_to_send
1398	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1399	echo " " >> $job_to_send
1400	echo "set -x" >> $job_to_send
1401	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1402	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1403	echo "" >> $job_to_send
1404	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1405	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1406	echo "rm -f scpjob.${identifier}" >> $job_to_send
1407	echo "%%END%%" >> $job_to_send
1408	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1409	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1410	elif [[ $remote_host = lcxe6 ]]
1411	then
1412	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1413	echo "#!/bin/ksh" >> $job_to_send
1414	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1415	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1416	echo "#PBS -A $project_account" >> $job_to_send
1417	echo "#PBS -l mppwidth=1" >> $job_to_send
1418	echo "#PBS -l mppnppn=1" >> $job_to_send
1419	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1420	echo "#PBS -j oe" >> $job_to_send
1421	echo " " >> $job_to_send
1422	echo "set -x" >> $job_to_send
1423	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1424	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1425	echo "%%END%%" >> $job_to_send
1426	else
1427
1428	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1429	echo "# @\\\$-q $return_queue" >> $job_to_send
1430	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1431	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1432	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1433	then
1434	echo "# @\$-l mpp_p=0" >> $job_to_send
1435	fi
1436	echo '# @\$-lF 10mb' >> $job_to_send
1437	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1438	echo '# @\\\$-eo' >> $job_to_send
1439	echo " " >> $job_to_send
1440	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1441	then
1442	echo "set +vx" >> $job_to_send
1443	echo ". .profile" >> $job_to_send
1444	fi
1445	echo "set -x" >> $job_to_send
1446	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
1447	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1448	echo "%%END%%" >> $job_to_send
1449
1450	fi
1451
1452	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1453	then
1454	echo "llsubmit scpjob.$identifier" >> $job_to_send
1455	elif [[ $remote_host = lcbullhh ]]
1456	then
1457	echo "sbatch scpjob.$identifier" >> $job_to_send
1458	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1459	then
1460	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1461	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1462	then
1463	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1464	elif [[ $remote_host = t3es ]]
1465	then
1466	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1467	elif [[ $remote_host = lckiaps ]]
1468	then
1469	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1470	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1471	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1472	elif [[ $remote_host = lckyu* ]]
1473	then
1474	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1475	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1476	elif [[ $remote_host = lcflow ]]
1477	then
1478	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1479	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1480	else
1481	echo "$submcom scpjob.$identifier" >> $job_to_send
1482	fi
1483	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1484	then
1485	echo "rm scpjob.$identifier" >> $job_to_send
1486	fi
1487	if [[ $remote_host = nech ]]
1488	then
1489	echo "cd -" >> $job_to_send
1490	fi
1491	else
1492	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1493	# ??? funktioniert das ÃŒberhaupt noch ???
1494	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1495	fi
1496	echo "set -x" >> $job_to_send
1497	echo " ' exit" >> $job_to_send
1498	echo "set -x" >> $job_to_send
1499	fi
1500
1501
1502	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1503	cat $file_to_send >> $job_to_send
1504
1505	if [[ $remote_host = ibm ]]
1506	then
1507	echo " " >> $job_to_send
1508	echo "exit" >> $job_to_send
1509	fi
1510
1511	# REMOVE JOB-FILE
1512	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1513	then
1514	echo " " >> $job_to_send
1515	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1516	fi
1517
1518
1519	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1520	if [[ $no_submit = false ]]
1521	then
1522	if [[ $remote_host != $local_host ]]
1523	then
1524	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1525	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1526	then
1527	job_catalog_save=$job_catalog
1528	job_catalog=job_queue
1529	elif [[ $remote_host = nech ]]
1530	then
1531	job_catalog_save=$job_catalog
1532	job_catalog=/hpf/b/${remote_user}/job_queue
1533	fi
1534	if [[ $remote_host = nech ]]
1535	then
1536	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1537	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1538	else
1539	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1540	fi
1541	if [[ $? = 1 ]]
1542	then
1543	locat=scp; exit
1544	fi
1545	if [[ $remote_host = ibms ]]
1546	then
1547	job_catalog=$job_catalog_save
1548	fi
1549	[[ $verify = true ]] && printf "\n >>> finished\n"
1550	else
1551	eval job_catalog=$job_catalog
1552	cp $job_to_send ${job_catalog}/$job_on_remhost
1553	fi
1554
1555
1556
1557	# START NQS- / LOADLEVELER-JOB
1558	if [[ $remote_host != $local_host ]]
1559	then
1560	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1561
1562	if [[ $remote_host = ibmku ]]
1563	then
1564	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1565	elif [[ $remote_host = lcflow ]]
1566	then
1567	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1568	else
1569	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1570	fi
1571
1572	[[ $verify = true ]] && printf " >>> o.k.\n"
1573	else
1574	cd $job_catalog
1575	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1576	then
1577	eval $submcom $job_on_remhost
1578	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1579	then
1580	chmod u+x $job_on_remhost
1581	eval $submcom $job_on_remhost
1582	elif [[ $local_host = nech ]]
1583	then
1584	if [[ $queue = default ]]
1585	then
1586	eval $submcom $job_on_remhost
1587	else
1588	eval $submcom -q $queue $job_on_remhost
1589	fi
1590	elif [[ $local_host = lcbullhh ]]
1591	then
1592	if [[ $queue = default ]]
1593	then
1594	eval $submcom $job_on_remhost
1595	fi
1596	else
1597	qsub $job_on_remhost
1598	fi
1599
1600	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1601	# AT THE END OF THE JOB
1602	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1603	then
1604	rm $job_on_remhost
1605	fi
1606	cd - > /dev/null
1607	fi
1608	fi
1609
1610	# FINAL ACTIONS
1611	if [[ $no_submit = false ]]
1612	then
1613	rm -f $job_to_send
1614	fi
1615	[[ $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 |