Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1826

Last change on this file since 1826 was 1779, checked in by raasch, 9 years ago
pmc array management changed from linked list to sequential loop; further small changes and cosmetics for the pmc
Property svn:keywords set to `Id Rev`
File size: 55.4 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 1779 2016-03-03 08:01:28Z maronga $
30	#
31	# 1701 2015-11-02 07:43:04Z maronga
32	# Bugfix: added missing init_cmds for lccrayh/lccrayb
33	#
34	# 1621 2015-07-17 11:39:33Z heinze
35	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
36	#
37	# 1575 2015-03-27 09:56:27Z raasch
38	# mpp2-queues added to lccrayh
39	#
40	# 1547 2015-01-29 15:09:12Z witha
41	# adjustments for ForWind computing cluster (lcflow)
42	#
43	# 1545 2015-01-29 06:52:23Z heinze
44	# local host name for blizzard further specified
45	#
46	# 1480 2014-10-17 14:41:49Z raasch
47	# adjustments for 2nd stage of HLRNIII
48	#
49	# 1468 2014-09-24 14:06:57Z maronga
50	# Typo removed (addres->address)
51	# Adjustments for lcxe6
52	#
53	# 1452 2014-08-22 09:41:06Z heinze
54	# local hosts for blizzard added
55	#
56	# 1450 2014-08-21 07:31:51Z heinze
57	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
58	#
59	# 1442 2014-07-28 07:09:10Z raasch
60	# HLRN-III (lccrayb/lccrayh) queues adjusted
61	#
62	# 1378 2014-04-28 06:04:58Z raasch
63	# -et option added for lctit
64	#
65	# 1350 2014-04-04 13:01:30Z maronga
66	# location of qsub updated for lcxe6
67	#
68	# 1289 2014-03-04 07:12:34Z raasch
69	# German comments translated to English
70	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
71	#
72	# 1279 2014-01-28 12:10:14Z raasch
73	# node calculation modified due to changes in mrun (tasks_per_node must not be
74	# an integral divisor of numprocs any more)
75	#
76	# 1274 2014-01-09 13:14:54Z heinze
77	# adjustments for lccrayh
78	#
79	# 1266 2013-12-11 12:07:34Z heinze
80	# further adjustments for lccrayb (use msub instead of qsub)
81	#
82	# 1264 2013-12-09 12:46:09Z fricke
83	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
84	#
85	# 1262 2013-12-09 10:57:20Z fricke
86	# further adjustments for lccrayb
87	#
88	# 1260 2013-12-04 12:48:04Z raasch
89	# jaboticaba admitted
90	#
91	# 1255 2013-11-07 14:43:35Z raasch
92	# further adjustments for lccrayb
93	#
94	# 1224 2013-09-16 07:27:23Z raasch
95	# first adjustments for lccrayb
96	#
97	# 1202 2013-07-10 16:22:07Z witha
98	# adjustments for Forwind cluster (lcflow)
99	#
100	# 1199 2013-07-05 14:52:22Z raasch
101	# adjustments for CSC Helsinki (lccrayf)
102	#
103	# use of cluster/express queue enabled (ibmh)
104	# vinessa added (imuk)
105	#
106	# 1103 2013-02-20 02:15:53Z raasch
107	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
108	# instead typeset -L/R),
109	# further adjustments for lckyuh
110	#
111	# 2013-02-10 01:47:43Z raasch
112	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
113	# and for Forwind cluster (lcflow)
114	#
115	# 1094 2013-02-03 01:52:12Z raasch
116	# new option -P for explicit setting of ssh/scp port,
117	# decalpha parts (yonsei) removed
118	#
119	# 2013-02-02 07:06:13Z raasch
120	# adjustments for Kyushu-University computing center (lckyut - tatara)
121	# old changelog messages removed
122	#
123	# 1046 2012-11-09 14:38:45Z maronga
124	# code put under GPL (PALM 3.9)
125	#
126	# 08/07/94 - Siggi - first version finished
127	# 29/06/94 - Siggi - script development started
128	#--------------------------------------------------------------------------------#
129	# subjob - script for automatic generation and submission of batch-job files
130	# for various batch queuing systems
131	#--------------------------------------------------------------------------------#
132
133
134	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
135	delete_dayfile=false
136	email_notification=none
137	group_number=none
138	locat=normal
139	no_default_queue=none
140	no_submit=false
141	job_catalog="~/job_queue"
142	job_name=none
143	local_user=$LOGNAME
144	node_usage=shared
145	numprocs=0
146	punkte="..........................................................."
147	submcom=qsub
148	queue=default
149	remote_host=none
150	remote_user=""
151	verify=true
152
153	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
154	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
155
156
157
158	# ERROR HANDLING
159	# IN CASE OF EXIT:
160	trap 'if [[ $locat != normal ]]
161	then
162	case $locat in
163	(option) printf "\n --> available optios can be displayed"
164	printf " by typing:"
165	printf "\n \"subjob ?\" \n";;
166	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
167	(*) printf "\n +++ unknown error"
168	printf "\n please inform S. Raasch!\n"
169	esac
170	[[ -f $job_to_send ]] && rm $job_to_send
171	printf "\n\n+++ SUBJOB killed \n\n"
172	fi' exit
173
174
175	# IN CASE OF TERMINAL-BREAK:
176	trap '[[ -f $job_to_send ]] && rm $job_to_send
177	printf "\n\n+++ SUBJOB killed \n\n"
178	exit
179	' 2
180
181
182	# DETERMINE NAME OF LOCAL HOST
183	local_host=$(hostname)
184
185	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
186	# IS ADMITTED AT ALL)
187	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
188	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
189	case $local_host in
190	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
191	(atmos) local_address=172.20.25.35; local_host=lcide;;
192	(austru) local_address=130.75.105.128; local_host=lcmuk;;
193	(autan) local_address=130.75.105.57; local_host=lcmuk;;
194	(bora) local_address=130.75.105.103; local_host=lcmuk;;
195	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
196	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
197	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
198	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
199	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
200	(breva) local_address=130.75.105.98; local_host=lcmuk;;
201	(buran) local_address=130.75.105.58; local_host=lcmuk;;
202	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
203	(climate*) local_address=165.132.26.68; local_host=lcyon;;
204	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
205	(cs*) local_address=136.172.44.131; local_host=nech;;
206	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
207	(flow01) local_address=10.141.255.71; local_host=lcflow;;
208	(flow02) local_address=10.141.255.72; local_host=lcflow;;
209	(node*) local_address=165.132.26.61 local_host=lck;;
210	# (node*) local_address=210.219.61.8 local_host=lckordi;;
211	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
212	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
213	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
214	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
215	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
216	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
217	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
218	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
219	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
220	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
221	(sno) local_address=130.75.105.113; local_host=lcmuk;;
222	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
223	(login*) local_address=118.128.66.223; local_host=lckiaps;;
224	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
225	(meller) local_address=134.106.74.155; local_host=lcfor;;
226	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
227	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
228	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
229	(nobel*) local_address=150.183.5.101; local_host=ibms;;
230	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
231	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
232	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
233	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
234	(pingui) local_address=134.106.74.118; local_host=lcfor;;
235	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
236	(rte*) local_address=133.5.185.60; local_host=lcrte;;
237	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
238	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
239	(solano) local_address=130.75.105.110; local_host=lcmuk;;
240	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
241	(t2a*) local_address=10.1.6.165; local_host=lctit;;
242	(urban*) local_address=147.46.30.151 local_host=lcsb;;
243	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
244	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
245	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
246	(*) printf "\n +++ \"$local_host\" unknown";
247	printf "\n please contact the PALM group at IMUK";
248	locat=parameter; exit;;
249	esac
250
251
252
253	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
254	remote_host=$local_host
255
256
257
258
259	# READ THE SHELLSCRIPT-OPTIONS
260	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
261	do
262	case $option in
263	(c) job_catalog=$OPTARG;;
264	(d) delete_dayfile=true;;
265	(D) no_submit=true;;
266	(e) email_notification=$OPTARG;;
267	(g) group_number=$OPTARG;;
268	(h) remote_host=$OPTARG;;
269	(m) memory=$OPTARG;;
270	(n) job_name=$OPTARG;;
271	(N) node_usage=$OPTARG;;
272	(O) threads_per_task=$OPTARG;;
273	(P) scp_port=$OPTARG;;
274	(q) no_default_queue=$OPTARG;;
275	(t) cputime=$OPTARG;;
276	(T) tasks_per_node=$OPTARG;;
277	(u) remote_user=$OPTARG;;
278	(v) verify=false;;
279	(X) numprocs=$OPTARG;;
280	(\?) printf "\n +++ Option $OPTARG unknown \n";
281	locat=option; exit;;
282	esac
283	done
284
285
286	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
287	(( to_shift = $OPTIND - 1 ))
288	shift $to_shift; file_to_send=$1
289
290
291	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
292	if [ "$1" = "?" ]
293	then
294	(printf "\n *** subjob can be called as follows:\n"
295	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
296	printf "\n Description of available options:\n"
297	printf "\n Option Description Default-Value"
298	printf "\n -c job-input- and output-catalog ~/job_queue"
299	printf "\n -d no job-protocol will be created ---"
300	printf "\n -D only the job-file will be created ---"
301	printf "\n -h execution host, available hosts: $remote_host"
302	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
303	printf "\n lckiaps, lctit, nech"
304	printf "\n -m memory demand per process in MByte ---"
305	printf "\n -n jobname <jobdatei>"
306	printf "\n -O threads per task (for OpenMP usage) 1"
307	printf "\n -P ssh/scp port default port"
308	printf "\n -q job-queue to be used default"
309	printf "\n -t allowed cpu-time in seconds ---"
310	printf "\n -T tasks per node (on parallel hosts) ---"
311	printf "\n -u username on execution host from .netrc"
312	printf "\n -v no prompt for confirmation ---"
313	printf "\n -X # of processors (on parallel hosts) 1"
314	printf "\n "
315	printf "\n The only possible positional parameter is <jobfile>:"
316	printf "\n The complete NQS-job must be provided here."
317	printf "\n <jobfile>=? creates this outline\n\n") \| more
318	exit
319	fi
320
321
322
323	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
324	if [[ "$file_to_send" = "" ]]
325	then
326	printf "\n +++ job-file missing"
327	locat=parameter; exit
328	else
329	if [[ -f $file_to_send ]]
330	then
331	true
332	else
333	printf "\n +++ job-file: "
334	printf "\n $file_to_send"
335	printf "\n does not exist"
336	locat=parameter; exit
337	fi
338	fi
339
340
341
342	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
343	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
344	if [[ $job_name = none ]]
345	then
346	job_name=$file_to_send
347	fi
348	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
349	then
350	printf "\n +++ job-file name: "
351	printf "\n $job_name"
352	printf "\n must not contain \"/\"-characters"
353	locat=parameter; exit
354	fi
355
356
357
358
359	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
360	# OR IF NO HOST HAS BEEN GIVEN
361	if [[ $remote_host = none ]]
362	then
363	printf "\n +++ host missing"
364	locat=option; exit
365	else
366	case $remote_host in
367	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
368	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
369	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
370	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
371	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
372	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
373	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
374	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
375	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
376	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
377	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
378	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
379	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
380	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
381	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
382	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
383	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
384	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
385	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
386	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
387	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
388	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
389	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
390	locat=parameter; exit;;
391	esac
392	fi
393
394
395	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
396	if [[ $no_default_queue != none ]]
397	then
398	error=false
399	ndq=$no_default_queue
400	case $remote_host in
401	(ibm) case $ndq in
402	(p690_express\|p690_standard\|p690_long) error=false;;
403	(*) error=true;;
404	esac;;
405	(ibmh) case $ndq in
406	(cluster\|express) error=false;;
407	(*) error=true;;
408	esac;;
409	(ibmkisti) case $ndq in
410	(class.32plus\|class.1-2\|class.2-32) error=false;;
411	(*) error=true;;
412	esac;;
413	(ibmku) case $ndq in
414	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
415	(*) error=true;;
416	esac;;
417	(ibms) case $ndq in
418	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
419	(*) error=true;;
420	esac;;
421	(lcbullhh) case $ndq in
422	(compute\|shared) error=false;;
423	(*) error=true;;
424	esac;;
425	(lccrayb) case $ndq in
426	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
427	(*) error=true;;
428	esac;;
429	(lccrayh) case $ndq in
430	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
431	(*) error=true;;
432	esac;;
433	(lccrayf) case $ndq in
434	(usup\|test*\|small\|large) error=false;;
435	(*) error=true;;
436	esac;;
437	(lcflow) case $ndq in
438	(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;;
439	(*) error=true;;
440	esac;;
441	(lckiaps) case $ndq in
442	(express\|normal) error=false;;
443	(*) error=true;;
444	esac;;
445	(lckyoto) case $ndq in
446	(eh\|ph) error=false;;
447	(*) error=true;;
448	esac;;
449	(lckyuh) case $ndq in
450	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
451	(*) error=true;;
452	esac;;
453	(lckyut) case $ndq in
454	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
455	(*) error=true;;
456	esac;;
457	(lctit) case $ndq in
458	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
459	(*) error=true;;
460	esac;;
461	(t3eb) case $ndq in
462	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
463	(*) error=true;;
464	esac;;
465	(t3eh) case $ndq in
466	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
467	(*) error=true;;
468	esac;;
469	(t3ej2\|t3ej5) case $ndq in
470	(low\|normal\|high) error=false;;
471	(*) error=true;;
472	esac;;
473	(t3es) case $ndq in
474	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
475	(*) error=true;;
476	esac;;
477	esac
478	if [[ $error = true ]]
479	then
480	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
481	locat=parameter; exit
482	else
483	queue=$no_default_queue
484	fi
485	fi
486
487
488
489	# CHECK THE CPU-TIME
490	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
491	done=false
492	while [[ $done = false ]]
493	do
494	if (( $cputime <= 0 ))
495	then
496	printf "\n +++ wrong cpu-time or cpu-time missing"
497	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
498	printf "\n >>> "
499	read cputime 1>/dev/null 2>&1
500	else
501	done=true
502	fi
503	done
504	if [[ $remote_host = nech ]]
505	then
506	if (( tasks_per_node != 0 ))
507	then
508	(( cputime = cputime * tasks_per_node ))
509	elif [[ $numprocs != 0 ]]
510	then
511	(( cputime = cputime * numprocs ))
512	fi
513	fi
514	(( stunden = cputime / 3600 ))
515	(( resttime = cputime - stunden * 3600 ))
516	(( minuten = resttime / 60 ))
517	(( sekunden = resttime - minuten * 60 ))
518	timestring=${stunden}:${minuten}:${sekunden}
519
520
521
522	# CHECK THE MEMORY DEMAND
523	done=false
524	while [[ $done = false ]]
525	do
526	if (( memory <= 0 ))
527	then
528	printf "\n +++ wrong memory demand or memory demand missing"
529	printf "\n >>> Please type memory in MByte per process as INTEGER:"
530	printf "\n >>> "
531	read memory 1>/dev/null 2>&1
532	else
533	done=true
534	fi
535	done
536
537	if [[ $remote_host = nech ]]
538	then
539	if (( tasks_per_node != 0 ))
540	then
541	(( Memory = memory * tasks_per_node / 1000 ))
542	elif [[ $numprocs != 0 ]]
543	then
544	(( Memory = memory * numprocs / 1000 ))
545	else
546	(( Memory = memory / 1000 ))
547	fi
548	elif [[ $remote_host = lctit ]]
549	then
550	(( Memory = memory * tasks_per_node / 1000 ))
551	fi
552
553
554	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
555	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
556	then
557	(( memory = memory * threads_per_task ))
558	fi
559
560
561	# CALCULATE NUMBER OF REQUIRED NODES
562	if (( tasks_per_node != 0 ))
563	then
564	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
565	fi
566
567
568	# CALCULATE NUMBER OF PROCESSES PER NODE
569	(( processes_per_node = tasks_per_node * threads_per_task ))
570
571
572	# CALCULATE NUMBER OF MPI TASKS
573	(( mpi_tasks = numprocs / threads_per_task ))
574
575
576	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
577	if [[ "$scp_port" != "" ]]
578	then
579	PORTOPT="-P $scp_port"
580	SSH_PORTOPT="-p $scp_port"
581	fi
582
583
584	# HEADER-OUTPUT
585	if [[ $verify = true ]]
586	then
587	printf "\n\n"
588	printf "#--------------------------------------------------------------# \n"
589	spalte1=SUBJOB;spalte2=$(date)
590	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
591	printf "\| \| \n"
592	printf "\| values of parameters/options: \| \n"
593	spalte1=$(echo local_host$punkte \| cut -c-20)
594	spalte2=$punkte$local_host
595	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
596	spalte1=$(echo remote_host$punkte \| cut -c-20)
597	spalte2=$punkte$remote_host
598	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
599	spalte1=$(echo queue$punkte \| cut -c-20)
600	spalte2=$punkte$queue
601	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
602	spalte1=$(echo memory$punkte \| cut -c-20)
603	spalte2="$punkte$memory mb"
604	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
605	spalte1=$(echo cputime$punkte \| cut -c-20)
606	spalte2="$punkte$cputime sec"
607	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
608	spalte1=$(echo job_name$punkte \| cut -c-20)
609	spalte2="$punkte$job_name"
610	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
611	printf "#--------------------------------------------------------------# \n\n"
612
613
614	# QUERY CHECK
615	antwort="dummy"
616	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
617	do
618	read antwort?" >>> continue (y/n) ? "
619	done
620	if [[ $antwort = n \|\| $antwort = N ]]
621	then
622	locat=verify; exit
623	fi
624	printf "\n"
625	fi
626
627	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
628	identifier=$RANDOM
629	job_on_remhost=${job_name}_${identifier}_$local_host
630	job_to_send=job_to_send_$identifier
631	if [[ $delete_dayfile = false ]]
632	then
633	remote_dayfile=${local_host}_${job_name}_result_$identifier
634	local_dayfile=${remote_host}_${job_name}
635	else
636	remote_dayfile=/dev/null
637	fi
638
639
640	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
641	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
642	then
643
644	# GENERAL LOADLEVELER SETTINGS
645	execute_in_shell="#!/bin/ksh"
646	use_shell="# @ shell = /bin/ksh"
647	consumable_memory="ConsumableMemory($memory mb)"
648	class="# @ class = $queue"
649	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
650	network_to_use="# @ network.mpi = sn_all,shared,us"
651	data_limit="# @ data_limit = 1.76gb"
652	image_size="# @ image_size = 50"
653	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
654
655	if [[ $email_notification = none ]]
656	then
657	notify_user=""
658	else
659	notify_user="# @ notify_user = $email_notification"
660	if [[ $delete_dayfile = true ]]
661	then
662	notification='# @ notification = never'
663	fi
664	fi
665
666	if [[ $remote_host = ibmh ]]
667	then
668	data_limit=""
669	network_to_use=""
670	class="# @ class = $queue"
671	environment=""
672	rset="# @ rset = RSET_MCM_AFFINITY"
673	task_affinity="# @ task_affinity = core(1)"
674	elif [[ $remote_host = ibmkisti ]]
675	then
676	network_to_use="# @ network.MPI = sn_all,shared,US"
677	wall_clock_limit="# @ wall_clock_limit = $timestring"
678	if [[ $threads_per_task = 1 ]]
679	then
680	rset="# @ rset = RSET_MCM_AFFINITY"
681	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
682	fi
683	environment=""
684	use_shell=""
685	data_limit=""
686	image_size=""
687	elif [[ $remote_host = ibmku ]]
688	then
689	execute_in_shell="#!/usr/bin/ksh"
690	use_shell="# @ shell = /usr/bin/ksh"
691	consumable_memory=""
692	environment=""
693	network_to_use="# @ network.mpi = sn_all,shared,us"
694	data_limit=""
695	image_size=""
696	elif [[ $remote_host = ibms ]]
697	then
698	network_to_use="# @ network.mpi = csss,shared,us"
699	fi
700
701	cat > $job_to_send << %%END%%
702	$execute_in_shell
703	$use_shell
704
705	# @ job_type = parallel
706	# @ job_name = $job_name
707	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
708	# @ output = $remote_dayfile
709	# @ error = $remote_dayfile
710	$wall_clock_limit
711	$image_size
712	$class
713	$environment
714	$network_to_use
715	$data_limit
716	$rset
717	$mcm_affinity_options
718	$task_affinity
719	$notification
720	$notify_user
721
722	%%END%%
723
724	if (( nodes > 0 ))
725	then
726
727	if [[ $remote_host != ibmkisti ]]
728	then
729
730	cat >> $job_to_send << %%END%%
731	# @ node = $nodes
732	# @ tasks_per_node = $processes_per_node
733	# @ node_usage = $node_usage
734	# @ queue
735
736	%%END%%
737
738	else
739
740	cat >> $job_to_send << %%END%%
741	# @ total_tasks = $mpi_tasks
742	# @ blocking = unlimited
743	# @ queue
744
745	%%END%%
746
747	fi
748
749	else
750
751	cat >> $job_to_send << %%END%%
752	# @ blocking = unlimited
753	# @ total_tasks = $numprocs
754	# @ node_usage = $node_usage
755	# @ queue
756
757	%%END%%
758
759	fi
760
761	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
762	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
763	then
764	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
765	fi
766
767	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
768	then
769
770	cat > $job_to_send << %%END%%
771	#!/bin/ksh
772
773	# @ job_type = serial
774	# @ node_usage = $node_usage
775	# @ job_name = palm
776	# @ wall_clock_limit = ${timestring},$timestring
777	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
778	# @ output = $remote_dayfile
779	# @ error = $remote_dayfile
780	$class
781	$notification
782
783	# @ queue
784
785	%%END%%
786
787	elif [[ $remote_host = lcbullhh ]]
788	then
789	if [[ $numprocs != 0 ]]
790	then
791	cat > $job_to_send << %%END%%
792	#!/bin/bash -l
793	#SBATCH -J $job_name
794	#SBATCH -t $timestring
795	#SBATCH -N $nodes
796	#SBATCH --ntasks-per-node=$processes_per_node
797	#SBATCH -p $queue
798	#SBATCH -o $remote_dayfile
799	#SBATCH -e $remote_dayfile
800	#SBATCH -A $project_account
801
802	$init_cmds
803	$module_calls
804
805	%%END%%
806
807	else
808	cat > $job_to_send << %%END%%
809	#!/bin/bash -l
810	#SBATCH -J $job_name
811	#SBATCH -t $timestring
812	#SBATCH -l ncpus=1
813	#SBATCH -l pmem=${memory}mb
814	#SBATCH -m abe
815	#SBATCH -o $remote_dayfile
816	#SBATCH -e $remote_dayfile
817	#SBATCH -A $project_account
818
819	$init_cmds
820	$module_calls
821
822	%%END%%
823
824	fi
825
826	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
827	then
828
829	if [[ "$feature" != "" ]]
830	then
831	featuredir="#PBS -l feature=$feature"
832	fi
833
834	if [[ $numprocs != 0 ]]
835	then
836	cat > $job_to_send << %%END%%
837	#!/bin/bash
838	#PBS -N $job_name
839	#PBS -l walltime=$timestring
840	#PBS -l nodes=$nodes:ppn=$processes_per_node
841	#PBS -o $remote_dayfile
842	#PBS -j oe
843	#PBS -q $queue
844	$featuredir
845
846	$init_cmds
847	$module_calls
848
849	%%END%%
850
851	else
852
853	continue
854
855	fi
856
857	elif [[ $remote_host = lccrayf ]]
858	then
859
860	if [[ $numprocs != 0 ]]
861	then
862	cat > $job_to_send << %%END%%
863	#!/bin/bash -l
864	#SBATCH -J $job_name
865	#SBATCH -t $timestring
866	#SBATCH -N $nodes
867	#SBATCH --ntasks-per-node=$processes_per_node
868	#SBATCH -p $queue
869	#SBATCH -o $remote_dayfile
870	#SBATCH -e $remote_dayfile
871
872	$init_cmds
873	$module_calls
874
875	%%END%%
876
877	else
878	cat > $job_to_send << %%END%%
879	#!/bin/bash -l
880	#SBATCH -J $job_name
881	#SBATCH -t $timestring
882	#SBATCH -l ncpus=1
883	#SBATCH -l pmem=${memory}mb
884	#SBATCH -m abe
885	#SBATCH -o $remote_dayfile
886	#SBATCH -e $remote_dayfile
887
888	$init_cmds
889	$module_calls
890
891	%%END%%
892
893	fi
894
895	elif [[ $remote_host = lcflow ]]
896	then
897	if [[ $numprocs != 0 ]]
898	then
899	pe_set="#$ -pe impi $numprocs"
900	else
901	pe_set="#$ -pe impi 1"
902	fi
903	if [[ $queue = default ]]
904	then
905	queue_set=""
906	else
907	queue_set="#$ -q $queue"
908	fi
909	[[ "$disc_space" = "" ]] && disc_space=50
910
911	cat > $job_to_send << %%END%%
912	#!/bin/bash
913	#$ -S /bin/bash
914	#$ -N $job_name
915	#$ -cwd
916	#$ -l h_rt=$timestring
917	#$ -l h_vmem=${memory}M
918	#$ -o $remote_dayfile
919	#$ -j y
920	$pe_set
921	#$ -R y
922	#$ -l h_fsize=${disc_space}G
923	$queue_set
924
925	%%END%%
926
927	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
928	then
929
930	if [[ $numprocs != 0 ]]
931	then
932	cat > $job_to_send << %%END%%
933	#!/bin/ksh
934	#PBS -N $job_name
935	#PBS -l walltime=$timestring
936	#PBS -l ncpus=$numprocs
937	#PBS -l pmem=${memory}mb
938	#PBS -o $remote_dayfile
939	#PBS -l nodes=$nodes:ppn=${processes_per_node}
940	#PBS -j oe
941
942	mpd &
943
944	%%END%%
945
946	else
947	cat > $job_to_send << %%END%%
948	#!/bin/ksh
949	#PBS -N $job_name
950	#PBS -l walltime=$timestring
951	#PBS -l ncpus=1
952	#PBS -l pmem=${memory}mb
953	#PBS -o $remote_dayfile
954	#PBS -j oe
955
956	%%END%%
957
958	fi
959
960	elif [[ $remote_host = lckiaps ]]
961	then
962
963	if [[ $numprocs != 0 ]]
964	then
965	cat > $job_to_send << %%END%%
966	#!/bin/ksh
967	#PBS -N $job_name
968	#PBS -l walltime=$timestring
969	#PBS -l select=1:ncpus=$numprocs
970	#PBS -l pmem=${memory}mb
971	#PBS -q $queue
972	#PBS -o $remote_dayfile
973	#PBS -j oe
974	#PBS -V
975
976	%%END%%
977
978	else
979	cat > $job_to_send << %%END%%
980	#!/bin/ksh
981	#PBS -N $job_name
982	#PBS -l walltime=$timestring
983	#PBS -l ncpus=1
984	#PBS -l pmem=${memory}mb
985	#PBS -o $remote_dayfile
986	#PBS -j oe
987
988	%%END%%
989
990	fi
991
992	elif [[ $remote_host = lcyon ]]
993	then
994
995	if [[ $numprocs != 0 ]]
996	then
997	cat > $job_to_send << %%END%%
998	#!/bin/ksh
999	#PBS -N $job_name
1000	#PBS -l walltime=$timestring
1001	#PBS -l ncpus=$numprocs
1002	#PBS -l pmem=${memory}mb
1003	#PBS -o $remote_dayfile
1004	#PBS -j oe
1005
1006	%%END%%
1007
1008	else
1009	cat > $job_to_send << %%END%%
1010	#!/bin/ksh
1011	#PBS -N $job_name
1012	#PBS -l walltime=$timestring
1013	#PBS -l ncpus=1
1014	#PBS -l pmem=${memory}mb
1015	#PBS -o $remote_dayfile
1016	#PBS -j oe
1017
1018	%%END%%
1019
1020	fi
1021
1022	elif [[ $remote_host = lcxe6 ]]
1023	then
1024
1025	if [[ $numprocs != 0 ]]
1026	then
1027	cat > $job_to_send << %%END%%
1028	#!/bin/ksh
1029	#PBS -S /bin/ksh
1030	#PBS -N $job_name
1031	#PBS -A $project_account
1032	#PBS -j oe
1033	#PBS -l walltime=$timestring
1034	#PBS -l mppwidth=${numprocs}
1035	#PBS -l mppnppn=${processes_per_node}
1036	#PBS -m abe
1037	#PBS -o $remote_dayfile
1038	$email_directive
1039
1040	$init_cmds
1041	$module_calls
1042
1043	%%END%%
1044
1045	else
1046	cat > $job_to_send << %%END%%
1047	#!/bin/ksh
1048	#PBS -S /bin/ksh
1049	#PBS -N $job_name
1050	#PBS -A $project_account
1051	#PBS -j oe
1052	#PBS -l walltime=$timestring
1053	#PBS -l ncpus=1
1054	#PBS -l pmem=${memory}mb
1055	#PBS -m abe
1056	$email_directive
1057	#PBS -o $remote_dayfile
1058
1059	$init_cmds
1060	$module_calls
1061
1062	%%END%%
1063
1064	fi
1065
1066	elif [[ $remote_host = lckyoto ]]
1067	then
1068
1069	cat > $job_to_send << %%END%%
1070	#!/bin/ksh
1071	# @\$-o $remote_dayfile
1072	# @\$-eo -oi
1073	# @\$-lP 16
1074	# @\$-lp 1
1075	# @\$-lm 28gb -llm unlimited -ls unlimited
1076	# @\$-q $queue
1077	# @\$-Pvn abs_pack
1078	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1079	#. /thin/local/etc/setprofile/intel-11.0.sh
1080	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1081	. ~/.myprofile
1082	#. /home2/t/t51254/palm/current_version/myprofile
1083	#. /thin/apps/pgi/mpi.sh
1084	#
1085	env
1086	#
1087	set -x
1088
1089	%%END%%
1090
1091	elif [[ $remote_host = lcxt5m ]]
1092	then
1093
1094	if [[ $numprocs != 0 ]]
1095	then
1096	cat > $job_to_send << %%END%%
1097	#!/bin/ksh
1098	#PBS -S /bin/ksh
1099	#PBS -N $job_name
1100	#PBS -j oe
1101	#PBS -l walltime=$timestring
1102	#PBS -l mppwidth=${numprocs}
1103	#PBS -l mppnppn=${processes_per_node}
1104	#PBS -m abe
1105	#PBS -o $remote_dayfile
1106
1107	$init_cmds
1108	$module_calls
1109
1110	%%END%%
1111
1112	else
1113	cat > $job_to_send << %%END%%
1114	#!/bin/ksh
1115	#PBS -S /bin/ksh
1116	#PBS -N $job_name
1117	#PBS -j oe
1118	#PBS -l walltime=$timestring
1119	#PBS -l ncpus=1
1120	#PBS -l pmem=${memory}mb
1121	#PBS -m abe
1122	#PBS -o $remote_dayfile
1123
1124	$init_cmds
1125	$module_calls
1126
1127	%%END%%
1128
1129	fi
1130
1131	elif [[ $remote_host = lckyuh ]]
1132	then
1133	cat > $job_to_send << %%END%%
1134	#!/bin/bash
1135	#PJM -L "rscgrp=$queue"
1136	#PJM -L "node=$nodes"
1137	#PJM --mpi "proc=$numprocs"
1138	#PJM -L "elapse=$timestring"
1139	#PJM -o $remote_dayfile
1140	#PJM -j
1141	#PJM -X
1142	#PJM --no-stging
1143
1144	export LANG=en_US.UTF-8
1145	%%END%%
1146
1147	elif [[ $remote_host = lckyut ]]
1148	then
1149	cat > $job_to_send << %%END%%
1150	#!/bin/bash
1151	#PJM -L "rscgrp=$queue"
1152	#PJM -L "vnode=$numprocs"
1153	#PJM -L "vnode-core=1"
1154	#PJM -L "elapse=$timestring"
1155	#PJM --mpi proc=$numprocs
1156	#PJM -o $remote_dayfile
1157	#PJM -j
1158	#PJM -X
1159	#PJM --no-stging
1160
1161	export LANG=en_US.UTF-8
1162	%%END%%
1163
1164	elif [[ $remote_host = nech ]]
1165	then
1166
1167	if (( nodes > 1 ))
1168	then
1169	cat > $job_to_send << %%END%%
1170	#!/bin/ksh
1171	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1172	#PBS -l ${qsubmem}=${Memory}gb
1173	#PBS -b $nodes
1174	#PBS -o $remote_dayfile
1175	#PBS -N palm
1176	#PBS -j o
1177	#PBS -T mpisx
1178
1179	%%END%%
1180
1181	elif [[ $numprocs != 0 ]]
1182	then
1183	cat > $job_to_send << %%END%%
1184	#!/bin/ksh
1185	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1186	#PBS -l ${qsubmem}=${Memory}gb
1187	#PBS -o $remote_dayfile
1188	#PBS -N palm
1189	#PBS -j o
1190
1191	%%END%%
1192
1193	else
1194	cat > $job_to_send << %%END%%
1195	#!/bin/ksh
1196	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1197	#PBS -o $remote_dayfile
1198	#PBS -j o
1199
1200	%%END%%
1201
1202	fi
1203
1204	elif [[ $remote_host = lctit ]]
1205	then
1206	cat > $job_to_send << %%END%%
1207	#!/bin/ksh
1208	$init_cmds
1209	$module_calls
1210
1211	%%END%%
1212
1213	# SET OPTIONS FOR SUBMIT-COMMAND
1214	if [[ $tasks_per_node != $processes_per_node ]]
1215	then
1216	submcom="$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 $queue "
1217	else
1218	submcom="$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 "
1219	fi
1220
1221	else
1222
1223	cat > $job_to_send << %%END%%
1224	# @\$-q ${queue}
1225	# @\$-l${qsubtime} $timestring
1226	# @\$-l${qsubmem} ${memory}mb
1227	# @\$-o $remote_dayfile
1228	# @\$-eo
1229
1230	%%END%%
1231
1232	fi
1233
1234
1235	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1236	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1237	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1238	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1239	then
1240	echo "set +vx" >> $job_to_send
1241	echo "trap '" >> $job_to_send
1242	echo "set +vx" >> $job_to_send
1243	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 ]]
1244	then
1245	if [[ $remote_host = ibmh ]]
1246	then
1247	return_queue=c1
1248	elif [[ $remote_host = ibmkisti ]]
1249	then
1250	return_queue=class.1-2
1251	elif [[ $remote_host = ibmku ]]
1252	then
1253	return_queue=sdbg2
1254	elif [[ $remote_host = ibms ]]
1255	then
1256	return_queue=p_normal
1257	elif [[ $remote_host = lcbullhh ]]
1258	then
1259	return_queue=shared
1260	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1261	then
1262	return_queue=dataq
1263	elif [[ $remote_host = lcxe6 ]]
1264	then
1265	return_queue=debug
1266	elif [[ $remote_host = lckiaps ]]
1267	then
1268	return_queue=express
1269	elif [[ $remote_host = lckyuh ]]
1270	then
1271	return_queue=cx-single
1272	elif [[ $remote_host = lckyut ]]
1273	then
1274	return_queue=cx-single
1275	else
1276	return_queue=unknown
1277	fi
1278
1279	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1280	then
1281
1282	if [[ $remote_host = ibmku ]]
1283	then
1284	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1285	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1286	else
1287	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1288	fi
1289	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1290	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1291	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1292	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1293	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1294	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1295	if [[ $host != "ibmh" ]]
1296	then
1297	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1298	fi
1299	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1300	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1301
1302	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1303	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1304
1305	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1306	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
1307	if [[ $remote_host = ibmku ]]
1308	then
1309	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1310	fi
1311	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1312
1313	elif [[ $remote_host = nech ]]
1314	then
1315	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1316	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1317	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1318	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1319	echo "#PBS -j o" >> $job_to_send
1320	echo " " >> $job_to_send
1321	echo "set -x" >> $job_to_send
1322	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1323	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1324	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1325	echo "%%END%%" >> $job_to_send
1326
1327	elif [[ $remote_host = lcbullhh ]]
1328	then
1329	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1330	echo "#!/bin/bash" >> $job_to_send
1331	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1332	echo "#SBATCH -t 00:20:00" >> $job_to_send
1333	echo "#SBATCH -N 1" >> $job_to_send
1334	echo "#SBATCH -n 1" >> $job_to_send
1335	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1336	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1337	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1338	echo "#SBATCH -A $project_account" >> $job_to_send
1339	echo "#SBATCH -p $return_queue" >> $job_to_send
1340	echo " " >> $job_to_send
1341	echo "set -x" >> $job_to_send
1342	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1343	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1344	echo "%%END%%" >> $job_to_send
1345
1346	elif [[ $remote_host = lckyuh ]]
1347	then
1348	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1349	echo "#!/bin/bash" >> $job_to_send
1350	echo "#PJM -L \"node=1\"" >> $job_to_send
1351	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1352	echo "#PJM --no-stging" >> $job_to_send
1353	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1354	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1355	echo "#PJM -j" >> $job_to_send
1356	echo " " >> $job_to_send
1357	echo "export LANG=en_US.UTF-8" >> $job_to_send
1358	echo "set -x" >> $job_to_send
1359	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1360	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1361	echo "%%END%%" >> $job_to_send
1362
1363	elif [[ $remote_host = lckyut ]]
1364	then
1365	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1366	echo "#!/bin/bash" >> $job_to_send
1367	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1368	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1369	echo "#PJM --no-stging" >> $job_to_send
1370	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1371	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1372	echo "#PJM -j" >> $job_to_send
1373	echo " " >> $job_to_send
1374	echo "export LANG=en_US.UTF-8" >> $job_to_send
1375	echo "set -x" >> $job_to_send
1376	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1377	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1378	echo "%%END%%" >> $job_to_send
1379
1380	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1381	then
1382	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1383	echo "#!/bin/bash" >> $job_to_send
1384	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1385	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1386	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1387	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1388	echo "#PBS -j oe" >> $job_to_send
1389	echo " " >> $job_to_send
1390	echo "set -x" >> $job_to_send
1391	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1392	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1393	echo "%%END%%" >> $job_to_send
1394
1395	elif [[ $remote_host = lcflow ]]
1396	then
1397	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1398	echo "#!/bin/bash" >> $job_to_send
1399	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1400	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1401	echo "SGEPREFIX -cwd" >> $job_to_send
1402	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1403	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1404	echo "SGEPREFIX -j y" >> $job_to_send
1405	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1406	echo " " >> $job_to_send
1407	echo "set -x" >> $job_to_send
1408	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1409	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1410	echo "" >> $job_to_send
1411	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1412	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1413	echo "rm -f scpjob.${identifier}" >> $job_to_send
1414	echo "%%END%%" >> $job_to_send
1415	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1416	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1417	elif [[ $remote_host = lcxe6 ]]
1418	then
1419	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1420	echo "#!/bin/ksh" >> $job_to_send
1421	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1422	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1423	echo "#PBS -A $project_account" >> $job_to_send
1424	echo "#PBS -l mppwidth=1" >> $job_to_send
1425	echo "#PBS -l mppnppn=1" >> $job_to_send
1426	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1427	echo "#PBS -j oe" >> $job_to_send
1428	echo " " >> $job_to_send
1429	echo "set -x" >> $job_to_send
1430	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1431	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1432	echo "%%END%%" >> $job_to_send
1433	else
1434
1435	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1436	echo "# @\\\$-q $return_queue" >> $job_to_send
1437	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1438	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1439	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1440	then
1441	echo "# @\$-l mpp_p=0" >> $job_to_send
1442	fi
1443	echo '# @\$-lF 10mb' >> $job_to_send
1444	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1445	echo '# @\\\$-eo' >> $job_to_send
1446	echo " " >> $job_to_send
1447	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1448	then
1449	echo "set +vx" >> $job_to_send
1450	echo ". .profile" >> $job_to_send
1451	fi
1452	echo "set -x" >> $job_to_send
1453	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
1454	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1455	echo "%%END%%" >> $job_to_send
1456
1457	fi
1458
1459	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1460	then
1461	echo "llsubmit scpjob.$identifier" >> $job_to_send
1462	elif [[ $remote_host = lcbullhh ]]
1463	then
1464	echo "sbatch scpjob.$identifier" >> $job_to_send
1465	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1466	then
1467	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1468	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1469	then
1470	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1471	elif [[ $remote_host = t3es ]]
1472	then
1473	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1474	elif [[ $remote_host = lckiaps ]]
1475	then
1476	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1477	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1478	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1479	elif [[ $remote_host = lckyu* ]]
1480	then
1481	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1482	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1483	elif [[ $remote_host = lcflow ]]
1484	then
1485	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1486	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1487	else
1488	echo "$submcom scpjob.$identifier" >> $job_to_send
1489	fi
1490	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1491	then
1492	echo "rm scpjob.$identifier" >> $job_to_send
1493	fi
1494	if [[ $remote_host = nech ]]
1495	then
1496	echo "cd -" >> $job_to_send
1497	fi
1498	else
1499	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1500	# ??? funktioniert das ÃŒberhaupt noch ???
1501	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1502	fi
1503	echo "set -x" >> $job_to_send
1504	echo " ' exit" >> $job_to_send
1505	echo "set -x" >> $job_to_send
1506	fi
1507
1508
1509	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1510	cat $file_to_send >> $job_to_send
1511
1512	if [[ $remote_host = ibm ]]
1513	then
1514	echo " " >> $job_to_send
1515	echo "exit" >> $job_to_send
1516	fi
1517
1518	# REMOVE JOB-FILE
1519	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1520	then
1521	echo " " >> $job_to_send
1522	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1523	fi
1524
1525
1526	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1527	if [[ $no_submit = false ]]
1528	then
1529	if [[ $remote_host != $local_host ]]
1530	then
1531	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1532	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1533	then
1534	job_catalog_save=$job_catalog
1535	job_catalog=job_queue
1536	elif [[ $remote_host = nech ]]
1537	then
1538	job_catalog_save=$job_catalog
1539	job_catalog=/hpf/b/${remote_user}/job_queue
1540	fi
1541	if [[ $remote_host = nech ]]
1542	then
1543	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1544	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1545	else
1546	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1547	fi
1548	if [[ $? = 1 ]]
1549	then
1550	locat=scp; exit
1551	fi
1552	if [[ $remote_host = ibms ]]
1553	then
1554	job_catalog=$job_catalog_save
1555	fi
1556	[[ $verify = true ]] && printf "\n >>> finished\n"
1557	else
1558	eval job_catalog=$job_catalog
1559	cp $job_to_send ${job_catalog}/$job_on_remhost
1560	fi
1561
1562
1563
1564	# START NQS- / LOADLEVELER-JOB
1565	if [[ $remote_host != $local_host ]]
1566	then
1567	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1568
1569	if [[ $remote_host = ibmku ]]
1570	then
1571	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1572	elif [[ $remote_host = lcflow ]]
1573	then
1574	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1575	else
1576	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1577	fi
1578
1579	[[ $verify = true ]] && printf " >>> o.k.\n"
1580	else
1581	cd $job_catalog
1582	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1583	then
1584	eval $submcom $job_on_remhost
1585	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1586	then
1587	chmod u+x $job_on_remhost
1588	eval $submcom $job_on_remhost
1589	elif [[ $local_host = nech ]]
1590	then
1591	if [[ $queue = default ]]
1592	then
1593	eval $submcom $job_on_remhost
1594	else
1595	eval $submcom -q $queue $job_on_remhost
1596	fi
1597	elif [[ $local_host = lcbullhh ]]
1598	then
1599	if [[ $queue = default ]]
1600	then
1601	eval $submcom $job_on_remhost
1602	fi
1603	else
1604	qsub $job_on_remhost
1605	fi
1606
1607	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1608	# AT THE END OF THE JOB
1609	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1610	then
1611	rm $job_on_remhost
1612	fi
1613	cd - > /dev/null
1614	fi
1615	fi
1616
1617	# FINAL ACTIONS
1618	if [[ $no_submit = false ]]
1619	then
1620	rm -f $job_to_send
1621	fi
1622	[[ $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 |