Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1857

Last change on this file since 1857 was 1842, checked in by raasch, 9 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 55.4 KB

Line
1	#!/bin/bash
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2014 Leibniz Universitaet Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1842 2016-04-07 19:24:16Z maronga $
30	#
31	# 1841 2016-04-07 19:14:06Z raasch
32	# script now running under bash
33	#
34	# 1701 2015-11-02 07:43:04Z maronga
35	# Bugfix: added missing init_cmds for lccrayh/lccrayb
36	#
37	# 1621 2015-07-17 11:39:33Z heinze
38	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
39	#
40	# 1575 2015-03-27 09:56:27Z raasch
41	# mpp2-queues added to lccrayh
42	#
43	# 1547 2015-01-29 15:09:12Z witha
44	# adjustments for ForWind computing cluster (lcflow)
45	#
46	# 1545 2015-01-29 06:52:23Z heinze
47	# local host name for blizzard further specified
48	#
49	# 1480 2014-10-17 14:41:49Z raasch
50	# adjustments for 2nd stage of HLRNIII
51	#
52	# 1468 2014-09-24 14:06:57Z maronga
53	# Typo removed (addres->address)
54	# Adjustments for lcxe6
55	#
56	# 1452 2014-08-22 09:41:06Z heinze
57	# local hosts for blizzard added
58	#
59	# 1450 2014-08-21 07:31:51Z heinze
60	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
61	#
62	# 1442 2014-07-28 07:09:10Z raasch
63	# HLRN-III (lccrayb/lccrayh) queues adjusted
64	#
65	# 1378 2014-04-28 06:04:58Z raasch
66	# -et option added for lctit
67	#
68	# 1350 2014-04-04 13:01:30Z maronga
69	# location of qsub updated for lcxe6
70	#
71	# 1289 2014-03-04 07:12:34Z raasch
72	# German comments translated to English
73	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
74	#
75	# 1279 2014-01-28 12:10:14Z raasch
76	# node calculation modified due to changes in mrun (tasks_per_node must not be
77	# an integral divisor of numprocs any more)
78	#
79	# 1274 2014-01-09 13:14:54Z heinze
80	# adjustments for lccrayh
81	#
82	# 1266 2013-12-11 12:07:34Z heinze
83	# further adjustments for lccrayb (use msub instead of qsub)
84	#
85	# 1264 2013-12-09 12:46:09Z fricke
86	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
87	#
88	# 1262 2013-12-09 10:57:20Z fricke
89	# further adjustments for lccrayb
90	#
91	# 1260 2013-12-04 12:48:04Z raasch
92	# jaboticaba admitted
93	#
94	# 1255 2013-11-07 14:43:35Z raasch
95	# further adjustments for lccrayb
96	#
97	# 1224 2013-09-16 07:27:23Z raasch
98	# first adjustments for lccrayb
99	#
100	# 1202 2013-07-10 16:22:07Z witha
101	# adjustments for Forwind cluster (lcflow)
102	#
103	# 1199 2013-07-05 14:52:22Z raasch
104	# adjustments for CSC Helsinki (lccrayf)
105	#
106	# use of cluster/express queue enabled (ibmh)
107	# vinessa added (imuk)
108	#
109	# 1103 2013-02-20 02:15:53Z raasch
110	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
111	# instead typeset -L/R),
112	# further adjustments for lckyuh
113	#
114	# 2013-02-10 01:47:43Z raasch
115	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
116	# and for Forwind cluster (lcflow)
117	#
118	# 1094 2013-02-03 01:52:12Z raasch
119	# new option -P for explicit setting of ssh/scp port,
120	# decalpha parts (yonsei) removed
121	#
122	# 2013-02-02 07:06:13Z raasch
123	# adjustments for Kyushu-University computing center (lckyut - tatara)
124	# old changelog messages removed
125	#
126	# 1046 2012-11-09 14:38:45Z maronga
127	# code put under GPL (PALM 3.9)
128	#
129	# 08/07/94 - Siggi - first version finished
130	# 29/06/94 - Siggi - script development started
131	#--------------------------------------------------------------------------------#
132	# subjob - script for automatic generation and submission of batch-job files
133	# for various batch queuing systems
134	#--------------------------------------------------------------------------------#
135
136
137	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
138	delete_dayfile=false
139	email_notification=none
140	group_number=none
141	locat=normal
142	no_default_queue=none
143	no_submit=false
144	job_catalog="~/job_queue"
145	job_name=none
146	local_user=$LOGNAME
147	node_usage=shared
148	numprocs=0
149	punkte="..........................................................."
150	submcom=qsub
151	queue=default
152	remote_host=none
153	remote_user=""
154	verify=true
155
156	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
157	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
158
159
160
161	# ERROR HANDLING
162	# IN CASE OF EXIT:
163	trap 'if [[ $locat != normal ]]
164	then
165	case $locat in
166	(option) printf "\n --> available optios can be displayed"
167	printf " by typing:"
168	printf "\n \"subjob ?\" \n";;
169	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
170	(*) printf "\n +++ unknown error"
171	printf "\n please inform S. Raasch!\n"
172	esac
173	[[ -f $job_to_send ]] && rm $job_to_send
174	printf "\n\n+++ SUBJOB killed \n\n"
175	fi' exit
176
177
178	# IN CASE OF TERMINAL-BREAK:
179	trap '[[ -f $job_to_send ]] && rm $job_to_send
180	printf "\n\n+++ SUBJOB killed \n\n"
181	exit
182	' 2
183
184
185	# DETERMINE NAME OF LOCAL HOST
186	local_host=$(hostname)
187
188	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
189	# IS ADMITTED AT ALL)
190	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
191	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
192	case $local_host in
193	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
194	(atmos) local_address=172.20.25.35; local_host=lcide;;
195	(austru) local_address=130.75.105.128; local_host=lcmuk;;
196	(autan) local_address=130.75.105.57; local_host=lcmuk;;
197	(bora) local_address=130.75.105.103; local_host=lcmuk;;
198	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
199	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
200	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
201	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
202	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
203	(breva) local_address=130.75.105.98; local_host=lcmuk;;
204	(buran) local_address=130.75.105.58; local_host=lcmuk;;
205	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
206	(climate*) local_address=165.132.26.68; local_host=lcyon;;
207	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
208	(cs*) local_address=136.172.44.131; local_host=nech;;
209	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
210	(flow01) local_address=10.141.255.71; local_host=lcflow;;
211	(flow02) local_address=10.141.255.72; local_host=lcflow;;
212	(node*) local_address=165.132.26.61 local_host=lck;;
213	# (node*) local_address=210.219.61.8 local_host=lckordi;;
214	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
215	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
216	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
217	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
218	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
219	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
220	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
221	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
222	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
223	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
224	(sno) local_address=130.75.105.113; local_host=lcmuk;;
225	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
226	(login*) local_address=118.128.66.223; local_host=lckiaps;;
227	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
228	(meller) local_address=134.106.74.155; local_host=lcfor;;
229	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
230	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
231	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
232	(nobel*) local_address=150.183.5.101; local_host=ibms;;
233	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
234	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
235	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
236	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
237	(pingui) local_address=134.106.74.118; local_host=lcfor;;
238	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
239	(rte*) local_address=133.5.185.60; local_host=lcrte;;
240	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
241	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
242	(solano) local_address=130.75.105.110; local_host=lcmuk;;
243	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
244	(t2a*) local_address=10.1.6.165; local_host=lctit;;
245	(urban*) local_address=147.46.30.151 local_host=lcsb;;
246	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
247	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
248	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
249	(*) printf "\n +++ \"$local_host\" unknown";
250	printf "\n please contact the PALM group at IMUK";
251	locat=parameter; exit;;
252	esac
253
254
255
256	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
257	remote_host=$local_host
258
259
260
261
262	# READ THE SHELLSCRIPT-OPTIONS
263	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
264	do
265	case $option in
266	(c) job_catalog=$OPTARG;;
267	(d) delete_dayfile=true;;
268	(D) no_submit=true;;
269	(e) email_notification=$OPTARG;;
270	(g) group_number=$OPTARG;;
271	(h) remote_host=$OPTARG;;
272	(m) memory=$OPTARG;;
273	(n) job_name=$OPTARG;;
274	(N) node_usage=$OPTARG;;
275	(O) threads_per_task=$OPTARG;;
276	(P) scp_port=$OPTARG;;
277	(q) no_default_queue=$OPTARG;;
278	(t) cputime=$OPTARG;;
279	(T) tasks_per_node=$OPTARG;;
280	(u) remote_user=$OPTARG;;
281	(v) verify=false;;
282	(X) numprocs=$OPTARG;;
283	(\?) printf "\n +++ Option $OPTARG unknown \n";
284	locat=option; exit;;
285	esac
286	done
287
288
289	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
290	(( to_shift = $OPTIND - 1 ))
291	shift $to_shift; file_to_send=$1
292
293
294	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
295	if [ "$1" = "?" ]
296	then
297	(printf "\n *** subjob can be called as follows:\n"
298	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
299	printf "\n Description of available options:\n"
300	printf "\n Option Description Default-Value"
301	printf "\n -c job-input- and output-catalog ~/job_queue"
302	printf "\n -d no job-protocol will be created ---"
303	printf "\n -D only the job-file will be created ---"
304	printf "\n -h execution host, available hosts: $remote_host"
305	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
306	printf "\n lckiaps, lctit, nech"
307	printf "\n -m memory demand per process in MByte ---"
308	printf "\n -n jobname <jobdatei>"
309	printf "\n -O threads per task (for OpenMP usage) 1"
310	printf "\n -P ssh/scp port default port"
311	printf "\n -q job-queue to be used default"
312	printf "\n -t allowed cpu-time in seconds ---"
313	printf "\n -T tasks per node (on parallel hosts) ---"
314	printf "\n -u username on execution host from .netrc"
315	printf "\n -v no prompt for confirmation ---"
316	printf "\n -X # of processors (on parallel hosts) 1"
317	printf "\n "
318	printf "\n The only possible positional parameter is <jobfile>:"
319	printf "\n The complete NQS-job must be provided here."
320	printf "\n <jobfile>=? creates this outline\n\n") \| more
321	exit
322	fi
323
324
325
326	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
327	if [[ "$file_to_send" = "" ]]
328	then
329	printf "\n +++ job-file missing"
330	locat=parameter; exit
331	else
332	if [[ -f $file_to_send ]]
333	then
334	true
335	else
336	printf "\n +++ job-file: "
337	printf "\n $file_to_send"
338	printf "\n does not exist"
339	locat=parameter; exit
340	fi
341	fi
342
343
344
345	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
346	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
347	if [[ $job_name = none ]]
348	then
349	job_name=$file_to_send
350	fi
351	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
352	then
353	printf "\n +++ job-file name: "
354	printf "\n $job_name"
355	printf "\n must not contain \"/\"-characters"
356	locat=parameter; exit
357	fi
358
359
360
361
362	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
363	# OR IF NO HOST HAS BEEN GIVEN
364	if [[ $remote_host = none ]]
365	then
366	printf "\n +++ host missing"
367	locat=option; exit
368	else
369	case $remote_host in
370	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
371	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
372	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
373	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
374	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
375	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
376	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
377	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
378	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
379	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
380	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
381	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
382	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
383	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
384	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
385	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
386	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
387	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
388	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
389	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
390	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
391	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
392	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
393	locat=parameter; exit;;
394	esac
395	fi
396
397
398	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
399	if [[ $no_default_queue != none ]]
400	then
401	error=false
402	ndq=$no_default_queue
403	case $remote_host in
404	(ibm) case $ndq in
405	(p690_express\|p690_standard\|p690_long) error=false;;
406	(*) error=true;;
407	esac;;
408	(ibmh) case $ndq in
409	(cluster\|express) error=false;;
410	(*) error=true;;
411	esac;;
412	(ibmkisti) case $ndq in
413	(class.32plus\|class.1-2\|class.2-32) error=false;;
414	(*) error=true;;
415	esac;;
416	(ibmku) case $ndq in
417	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
418	(*) error=true;;
419	esac;;
420	(ibms) case $ndq in
421	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
422	(*) error=true;;
423	esac;;
424	(lcbullhh) case $ndq in
425	(compute\|shared) error=false;;
426	(*) error=true;;
427	esac;;
428	(lccrayb) case $ndq in
429	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
430	(*) error=true;;
431	esac;;
432	(lccrayh) case $ndq in
433	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
434	(*) error=true;;
435	esac;;
436	(lccrayf) case $ndq in
437	(usup\|test*\|small\|large) error=false;;
438	(*) error=true;;
439	esac;;
440	(lcflow) case $ndq in
441	(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;;
442	(*) error=true;;
443	esac;;
444	(lckiaps) case $ndq in
445	(express\|normal) error=false;;
446	(*) error=true;;
447	esac;;
448	(lckyoto) case $ndq in
449	(eh\|ph) error=false;;
450	(*) error=true;;
451	esac;;
452	(lckyuh) case $ndq in
453	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
454	(*) error=true;;
455	esac;;
456	(lckyut) case $ndq in
457	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
458	(*) error=true;;
459	esac;;
460	(lctit) case $ndq in
461	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
462	(*) error=true;;
463	esac;;
464	(t3eb) case $ndq in
465	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
466	(*) error=true;;
467	esac;;
468	(t3eh) case $ndq in
469	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
470	(*) error=true;;
471	esac;;
472	(t3ej2\|t3ej5) case $ndq in
473	(low\|normal\|high) error=false;;
474	(*) error=true;;
475	esac;;
476	(t3es) case $ndq in
477	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
478	(*) error=true;;
479	esac;;
480	esac
481	if [[ $error = true ]]
482	then
483	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
484	locat=parameter; exit
485	else
486	queue=$no_default_queue
487	fi
488	fi
489
490
491
492	# CHECK THE CPU-TIME
493	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
494	done=false
495	while [[ $done = false ]]
496	do
497	if (( $cputime <= 0 ))
498	then
499	printf "\n +++ wrong cpu-time or cpu-time missing"
500	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
501	printf "\n >>> "
502	read cputime 1>/dev/null 2>&1
503	else
504	done=true
505	fi
506	done
507	if [[ $remote_host = nech ]]
508	then
509	if (( tasks_per_node != 0 ))
510	then
511	(( cputime = cputime * tasks_per_node ))
512	elif [[ $numprocs != 0 ]]
513	then
514	(( cputime = cputime * numprocs ))
515	fi
516	fi
517	(( stunden = cputime / 3600 ))
518	(( resttime = cputime - stunden * 3600 ))
519	(( minuten = resttime / 60 ))
520	(( sekunden = resttime - minuten * 60 ))
521	timestring=${stunden}:${minuten}:${sekunden}
522
523
524
525	# CHECK THE MEMORY DEMAND
526	done=false
527	while [[ $done = false ]]
528	do
529	if (( memory <= 0 ))
530	then
531	printf "\n +++ wrong memory demand or memory demand missing"
532	printf "\n >>> Please type memory in MByte per process as INTEGER:"
533	printf "\n >>> "
534	read memory 1>/dev/null 2>&1
535	else
536	done=true
537	fi
538	done
539
540	if [[ $remote_host = nech ]]
541	then
542	if (( tasks_per_node != 0 ))
543	then
544	(( Memory = memory * tasks_per_node / 1000 ))
545	elif [[ $numprocs != 0 ]]
546	then
547	(( Memory = memory * numprocs / 1000 ))
548	else
549	(( Memory = memory / 1000 ))
550	fi
551	elif [[ $remote_host = lctit ]]
552	then
553	(( Memory = memory * tasks_per_node / 1000 ))
554	fi
555
556
557	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
558	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
559	then
560	(( memory = memory * threads_per_task ))
561	fi
562
563
564	# CALCULATE NUMBER OF REQUIRED NODES
565	if (( tasks_per_node != 0 ))
566	then
567	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
568	fi
569
570
571	# CALCULATE NUMBER OF PROCESSES PER NODE
572	(( processes_per_node = tasks_per_node * threads_per_task ))
573
574
575	# CALCULATE NUMBER OF MPI TASKS
576	(( mpi_tasks = numprocs / threads_per_task ))
577
578
579	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
580	if [[ "$scp_port" != "" ]]
581	then
582	PORTOPT="-P $scp_port"
583	SSH_PORTOPT="-p $scp_port"
584	fi
585
586
587	# HEADER-OUTPUT
588	if [[ $verify = true ]]
589	then
590	printf "\n\n"
591	printf "#--------------------------------------------------------------# \n"
592	spalte1=SUBJOB;spalte2=$(date)
593	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
594	printf "\| \| \n"
595	printf "\| values of parameters/options: \| \n"
596	spalte1=$(echo local_host$punkte \| cut -c-20)
597	spalte2=$punkte$local_host
598	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
599	spalte1=$(echo remote_host$punkte \| cut -c-20)
600	spalte2=$punkte$remote_host
601	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
602	spalte1=$(echo queue$punkte \| cut -c-20)
603	spalte2=$punkte$queue
604	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
605	spalte1=$(echo memory$punkte \| cut -c-20)
606	spalte2="$punkte$memory mb"
607	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
608	spalte1=$(echo cputime$punkte \| cut -c-20)
609	spalte2="$punkte$cputime sec"
610	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
611	spalte1=$(echo job_name$punkte \| cut -c-20)
612	spalte2="$punkte$job_name"
613	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
614	printf "#--------------------------------------------------------------# \n\n"
615
616
617	# QUERY CHECK
618	antwort="dummy"
619	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
620	do
621	read antwort?" >>> continue (y/n) ? "
622	done
623	if [[ $antwort = n \|\| $antwort = N ]]
624	then
625	locat=verify; exit
626	fi
627	printf "\n"
628	fi
629
630	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
631	identifier=$RANDOM
632	job_on_remhost=${job_name}_${identifier}_$local_host
633	job_to_send=job_to_send_$identifier
634	if [[ $delete_dayfile = false ]]
635	then
636	remote_dayfile=${local_host}_${job_name}_result_$identifier
637	local_dayfile=${remote_host}_${job_name}
638	else
639	remote_dayfile=/dev/null
640	fi
641
642
643	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
644	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
645	then
646
647	# GENERAL LOADLEVELER SETTINGS
648	execute_in_shell="#!/bin/ksh"
649	use_shell="# @ shell = /bin/ksh"
650	consumable_memory="ConsumableMemory($memory mb)"
651	class="# @ class = $queue"
652	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
653	network_to_use="# @ network.mpi = sn_all,shared,us"
654	data_limit="# @ data_limit = 1.76gb"
655	image_size="# @ image_size = 50"
656	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
657
658	if [[ $email_notification = none ]]
659	then
660	notify_user=""
661	else
662	notify_user="# @ notify_user = $email_notification"
663	if [[ $delete_dayfile = true ]]
664	then
665	notification='# @ notification = never'
666	fi
667	fi
668
669	if [[ $remote_host = ibmh ]]
670	then
671	data_limit=""
672	network_to_use=""
673	class="# @ class = $queue"
674	environment=""
675	rset="# @ rset = RSET_MCM_AFFINITY"
676	task_affinity="# @ task_affinity = core(1)"
677	elif [[ $remote_host = ibmkisti ]]
678	then
679	network_to_use="# @ network.MPI = sn_all,shared,US"
680	wall_clock_limit="# @ wall_clock_limit = $timestring"
681	if [[ $threads_per_task = 1 ]]
682	then
683	rset="# @ rset = RSET_MCM_AFFINITY"
684	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
685	fi
686	environment=""
687	use_shell=""
688	data_limit=""
689	image_size=""
690	elif [[ $remote_host = ibmku ]]
691	then
692	execute_in_shell="#!/usr/bin/ksh"
693	use_shell="# @ shell = /usr/bin/ksh"
694	consumable_memory=""
695	environment=""
696	network_to_use="# @ network.mpi = sn_all,shared,us"
697	data_limit=""
698	image_size=""
699	elif [[ $remote_host = ibms ]]
700	then
701	network_to_use="# @ network.mpi = csss,shared,us"
702	fi
703
704	cat > $job_to_send << %%END%%
705	$execute_in_shell
706	$use_shell
707
708	# @ job_type = parallel
709	# @ job_name = $job_name
710	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
711	# @ output = $remote_dayfile
712	# @ error = $remote_dayfile
713	$wall_clock_limit
714	$image_size
715	$class
716	$environment
717	$network_to_use
718	$data_limit
719	$rset
720	$mcm_affinity_options
721	$task_affinity
722	$notification
723	$notify_user
724
725	%%END%%
726
727	if (( nodes > 0 ))
728	then
729
730	if [[ $remote_host != ibmkisti ]]
731	then
732
733	cat >> $job_to_send << %%END%%
734	# @ node = $nodes
735	# @ tasks_per_node = $processes_per_node
736	# @ node_usage = $node_usage
737	# @ queue
738
739	%%END%%
740
741	else
742
743	cat >> $job_to_send << %%END%%
744	# @ total_tasks = $mpi_tasks
745	# @ blocking = unlimited
746	# @ queue
747
748	%%END%%
749
750	fi
751
752	else
753
754	cat >> $job_to_send << %%END%%
755	# @ blocking = unlimited
756	# @ total_tasks = $numprocs
757	# @ node_usage = $node_usage
758	# @ queue
759
760	%%END%%
761
762	fi
763
764	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
765	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
766	then
767	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
768	fi
769
770	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
771	then
772
773	cat > $job_to_send << %%END%%
774	#!/bin/ksh
775
776	# @ job_type = serial
777	# @ node_usage = $node_usage
778	# @ job_name = palm
779	# @ wall_clock_limit = ${timestring},$timestring
780	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
781	# @ output = $remote_dayfile
782	# @ error = $remote_dayfile
783	$class
784	$notification
785
786	# @ queue
787
788	%%END%%
789
790	elif [[ $remote_host = lcbullhh ]]
791	then
792	if [[ $numprocs != 0 ]]
793	then
794	cat > $job_to_send << %%END%%
795	#!/bin/bash -l
796	#SBATCH -J $job_name
797	#SBATCH -t $timestring
798	#SBATCH -N $nodes
799	#SBATCH --ntasks-per-node=$processes_per_node
800	#SBATCH -p $queue
801	#SBATCH -o $remote_dayfile
802	#SBATCH -e $remote_dayfile
803	#SBATCH -A $project_account
804
805	$init_cmds
806	$module_calls
807
808	%%END%%
809
810	else
811	cat > $job_to_send << %%END%%
812	#!/bin/bash -l
813	#SBATCH -J $job_name
814	#SBATCH -t $timestring
815	#SBATCH -l ncpus=1
816	#SBATCH -l pmem=${memory}mb
817	#SBATCH -m abe
818	#SBATCH -o $remote_dayfile
819	#SBATCH -e $remote_dayfile
820	#SBATCH -A $project_account
821
822	$init_cmds
823	$module_calls
824
825	%%END%%
826
827	fi
828
829	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
830	then
831
832	if [[ "$feature" != "" ]]
833	then
834	featuredir="#PBS -l feature=$feature"
835	fi
836
837	if [[ $numprocs != 0 ]]
838	then
839	cat > $job_to_send << %%END%%
840	#!/bin/bash
841	#PBS -N $job_name
842	#PBS -l walltime=$timestring
843	#PBS -l nodes=$nodes:ppn=$processes_per_node
844	#PBS -o $remote_dayfile
845	#PBS -j oe
846	#PBS -q $queue
847	$featuredir
848
849	$init_cmds
850	$module_calls
851
852	%%END%%
853
854	else
855
856	continue
857
858	fi
859
860	elif [[ $remote_host = lccrayf ]]
861	then
862
863	if [[ $numprocs != 0 ]]
864	then
865	cat > $job_to_send << %%END%%
866	#!/bin/bash -l
867	#SBATCH -J $job_name
868	#SBATCH -t $timestring
869	#SBATCH -N $nodes
870	#SBATCH --ntasks-per-node=$processes_per_node
871	#SBATCH -p $queue
872	#SBATCH -o $remote_dayfile
873	#SBATCH -e $remote_dayfile
874
875	$init_cmds
876	$module_calls
877
878	%%END%%
879
880	else
881	cat > $job_to_send << %%END%%
882	#!/bin/bash -l
883	#SBATCH -J $job_name
884	#SBATCH -t $timestring
885	#SBATCH -l ncpus=1
886	#SBATCH -l pmem=${memory}mb
887	#SBATCH -m abe
888	#SBATCH -o $remote_dayfile
889	#SBATCH -e $remote_dayfile
890
891	$init_cmds
892	$module_calls
893
894	%%END%%
895
896	fi
897
898	elif [[ $remote_host = lcflow ]]
899	then
900	if [[ $numprocs != 0 ]]
901	then
902	pe_set="#$ -pe impi $numprocs"
903	else
904	pe_set="#$ -pe impi 1"
905	fi
906	if [[ $queue = default ]]
907	then
908	queue_set=""
909	else
910	queue_set="#$ -q $queue"
911	fi
912	[[ "$disc_space" = "" ]] && disc_space=50
913
914	cat > $job_to_send << %%END%%
915	#!/bin/bash
916	#$ -S /bin/bash
917	#$ -N $job_name
918	#$ -cwd
919	#$ -l h_rt=$timestring
920	#$ -l h_vmem=${memory}M
921	#$ -o $remote_dayfile
922	#$ -j y
923	$pe_set
924	#$ -R y
925	#$ -l h_fsize=${disc_space}G
926	$queue_set
927
928	%%END%%
929
930	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
931	then
932
933	if [[ $numprocs != 0 ]]
934	then
935	cat > $job_to_send << %%END%%
936	#!/bin/ksh
937	#PBS -N $job_name
938	#PBS -l walltime=$timestring
939	#PBS -l ncpus=$numprocs
940	#PBS -l pmem=${memory}mb
941	#PBS -o $remote_dayfile
942	#PBS -l nodes=$nodes:ppn=${processes_per_node}
943	#PBS -j oe
944
945	mpd &
946
947	%%END%%
948
949	else
950	cat > $job_to_send << %%END%%
951	#!/bin/ksh
952	#PBS -N $job_name
953	#PBS -l walltime=$timestring
954	#PBS -l ncpus=1
955	#PBS -l pmem=${memory}mb
956	#PBS -o $remote_dayfile
957	#PBS -j oe
958
959	%%END%%
960
961	fi
962
963	elif [[ $remote_host = lckiaps ]]
964	then
965
966	if [[ $numprocs != 0 ]]
967	then
968	cat > $job_to_send << %%END%%
969	#!/bin/ksh
970	#PBS -N $job_name
971	#PBS -l walltime=$timestring
972	#PBS -l select=1:ncpus=$numprocs
973	#PBS -l pmem=${memory}mb
974	#PBS -q $queue
975	#PBS -o $remote_dayfile
976	#PBS -j oe
977	#PBS -V
978
979	%%END%%
980
981	else
982	cat > $job_to_send << %%END%%
983	#!/bin/ksh
984	#PBS -N $job_name
985	#PBS -l walltime=$timestring
986	#PBS -l ncpus=1
987	#PBS -l pmem=${memory}mb
988	#PBS -o $remote_dayfile
989	#PBS -j oe
990
991	%%END%%
992
993	fi
994
995	elif [[ $remote_host = lcyon ]]
996	then
997
998	if [[ $numprocs != 0 ]]
999	then
1000	cat > $job_to_send << %%END%%
1001	#!/bin/ksh
1002	#PBS -N $job_name
1003	#PBS -l walltime=$timestring
1004	#PBS -l ncpus=$numprocs
1005	#PBS -l pmem=${memory}mb
1006	#PBS -o $remote_dayfile
1007	#PBS -j oe
1008
1009	%%END%%
1010
1011	else
1012	cat > $job_to_send << %%END%%
1013	#!/bin/ksh
1014	#PBS -N $job_name
1015	#PBS -l walltime=$timestring
1016	#PBS -l ncpus=1
1017	#PBS -l pmem=${memory}mb
1018	#PBS -o $remote_dayfile
1019	#PBS -j oe
1020
1021	%%END%%
1022
1023	fi
1024
1025	elif [[ $remote_host = lcxe6 ]]
1026	then
1027
1028	if [[ $numprocs != 0 ]]
1029	then
1030	cat > $job_to_send << %%END%%
1031	#!/bin/ksh
1032	#PBS -S /bin/ksh
1033	#PBS -N $job_name
1034	#PBS -A $project_account
1035	#PBS -j oe
1036	#PBS -l walltime=$timestring
1037	#PBS -l mppwidth=${numprocs}
1038	#PBS -l mppnppn=${processes_per_node}
1039	#PBS -m abe
1040	#PBS -o $remote_dayfile
1041	$email_directive
1042
1043	$init_cmds
1044	$module_calls
1045
1046	%%END%%
1047
1048	else
1049	cat > $job_to_send << %%END%%
1050	#!/bin/ksh
1051	#PBS -S /bin/ksh
1052	#PBS -N $job_name
1053	#PBS -A $project_account
1054	#PBS -j oe
1055	#PBS -l walltime=$timestring
1056	#PBS -l ncpus=1
1057	#PBS -l pmem=${memory}mb
1058	#PBS -m abe
1059	$email_directive
1060	#PBS -o $remote_dayfile
1061
1062	$init_cmds
1063	$module_calls
1064
1065	%%END%%
1066
1067	fi
1068
1069	elif [[ $remote_host = lckyoto ]]
1070	then
1071
1072	cat > $job_to_send << %%END%%
1073	#!/bin/ksh
1074	# @\$-o $remote_dayfile
1075	# @\$-eo -oi
1076	# @\$-lP 16
1077	# @\$-lp 1
1078	# @\$-lm 28gb -llm unlimited -ls unlimited
1079	# @\$-q $queue
1080	# @\$-Pvn abs_pack
1081	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1082	#. /thin/local/etc/setprofile/intel-11.0.sh
1083	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1084	. ~/.myprofile
1085	#. /home2/t/t51254/palm/current_version/myprofile
1086	#. /thin/apps/pgi/mpi.sh
1087	#
1088	env
1089	#
1090	set -x
1091
1092	%%END%%
1093
1094	elif [[ $remote_host = lcxt5m ]]
1095	then
1096
1097	if [[ $numprocs != 0 ]]
1098	then
1099	cat > $job_to_send << %%END%%
1100	#!/bin/ksh
1101	#PBS -S /bin/ksh
1102	#PBS -N $job_name
1103	#PBS -j oe
1104	#PBS -l walltime=$timestring
1105	#PBS -l mppwidth=${numprocs}
1106	#PBS -l mppnppn=${processes_per_node}
1107	#PBS -m abe
1108	#PBS -o $remote_dayfile
1109
1110	$init_cmds
1111	$module_calls
1112
1113	%%END%%
1114
1115	else
1116	cat > $job_to_send << %%END%%
1117	#!/bin/ksh
1118	#PBS -S /bin/ksh
1119	#PBS -N $job_name
1120	#PBS -j oe
1121	#PBS -l walltime=$timestring
1122	#PBS -l ncpus=1
1123	#PBS -l pmem=${memory}mb
1124	#PBS -m abe
1125	#PBS -o $remote_dayfile
1126
1127	$init_cmds
1128	$module_calls
1129
1130	%%END%%
1131
1132	fi
1133
1134	elif [[ $remote_host = lckyuh ]]
1135	then
1136	cat > $job_to_send << %%END%%
1137	#!/bin/bash
1138	#PJM -L "rscgrp=$queue"
1139	#PJM -L "node=$nodes"
1140	#PJM --mpi "proc=$numprocs"
1141	#PJM -L "elapse=$timestring"
1142	#PJM -o $remote_dayfile
1143	#PJM -j
1144	#PJM -X
1145	#PJM --no-stging
1146
1147	export LANG=en_US.UTF-8
1148	%%END%%
1149
1150	elif [[ $remote_host = lckyut ]]
1151	then
1152	cat > $job_to_send << %%END%%
1153	#!/bin/bash
1154	#PJM -L "rscgrp=$queue"
1155	#PJM -L "vnode=$numprocs"
1156	#PJM -L "vnode-core=1"
1157	#PJM -L "elapse=$timestring"
1158	#PJM --mpi proc=$numprocs
1159	#PJM -o $remote_dayfile
1160	#PJM -j
1161	#PJM -X
1162	#PJM --no-stging
1163
1164	export LANG=en_US.UTF-8
1165	%%END%%
1166
1167	elif [[ $remote_host = nech ]]
1168	then
1169
1170	if (( nodes > 1 ))
1171	then
1172	cat > $job_to_send << %%END%%
1173	#!/bin/ksh
1174	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1175	#PBS -l ${qsubmem}=${Memory}gb
1176	#PBS -b $nodes
1177	#PBS -o $remote_dayfile
1178	#PBS -N palm
1179	#PBS -j o
1180	#PBS -T mpisx
1181
1182	%%END%%
1183
1184	elif [[ $numprocs != 0 ]]
1185	then
1186	cat > $job_to_send << %%END%%
1187	#!/bin/ksh
1188	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1189	#PBS -l ${qsubmem}=${Memory}gb
1190	#PBS -o $remote_dayfile
1191	#PBS -N palm
1192	#PBS -j o
1193
1194	%%END%%
1195
1196	else
1197	cat > $job_to_send << %%END%%
1198	#!/bin/ksh
1199	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1200	#PBS -o $remote_dayfile
1201	#PBS -j o
1202
1203	%%END%%
1204
1205	fi
1206
1207	elif [[ $remote_host = lctit ]]
1208	then
1209	cat > $job_to_send << %%END%%
1210	#!/bin/ksh
1211	$init_cmds
1212	$module_calls
1213
1214	%%END%%
1215
1216	# SET OPTIONS FOR SUBMIT-COMMAND
1217	if [[ $tasks_per_node != $processes_per_node ]]
1218	then
1219	submcom="$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 $queue "
1220	else
1221	submcom="$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 "
1222	fi
1223
1224	else
1225
1226	cat > $job_to_send << %%END%%
1227	# @\$-q ${queue}
1228	# @\$-l${qsubtime} $timestring
1229	# @\$-l${qsubmem} ${memory}mb
1230	# @\$-o $remote_dayfile
1231	# @\$-eo
1232
1233	%%END%%
1234
1235	fi
1236
1237
1238	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1239	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1240	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1241	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1242	then
1243	echo "set +vx" >> $job_to_send
1244	echo "trap '" >> $job_to_send
1245	echo "set +vx" >> $job_to_send
1246	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 ]]
1247	then
1248	if [[ $remote_host = ibmh ]]
1249	then
1250	return_queue=c1
1251	elif [[ $remote_host = ibmkisti ]]
1252	then
1253	return_queue=class.1-2
1254	elif [[ $remote_host = ibmku ]]
1255	then
1256	return_queue=sdbg2
1257	elif [[ $remote_host = ibms ]]
1258	then
1259	return_queue=p_normal
1260	elif [[ $remote_host = lcbullhh ]]
1261	then
1262	return_queue=shared
1263	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1264	then
1265	return_queue=dataq
1266	elif [[ $remote_host = lcxe6 ]]
1267	then
1268	return_queue=debug
1269	elif [[ $remote_host = lckiaps ]]
1270	then
1271	return_queue=express
1272	elif [[ $remote_host = lckyuh ]]
1273	then
1274	return_queue=cx-single
1275	elif [[ $remote_host = lckyut ]]
1276	then
1277	return_queue=cx-single
1278	else
1279	return_queue=unknown
1280	fi
1281
1282	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1283	then
1284
1285	if [[ $remote_host = ibmku ]]
1286	then
1287	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1288	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1289	else
1290	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1291	fi
1292	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1293	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1294	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1295	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1296	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1297	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1298	if [[ $host != "ibmh" ]]
1299	then
1300	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1301	fi
1302	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1303	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1304
1305	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1306	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1307
1308	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1309	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
1310	if [[ $remote_host = ibmku ]]
1311	then
1312	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1313	fi
1314	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1315
1316	elif [[ $remote_host = nech ]]
1317	then
1318	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1319	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1320	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1321	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1322	echo "#PBS -j o" >> $job_to_send
1323	echo " " >> $job_to_send
1324	echo "set -x" >> $job_to_send
1325	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1326	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1327	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1328	echo "%%END%%" >> $job_to_send
1329
1330	elif [[ $remote_host = lcbullhh ]]
1331	then
1332	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1333	echo "#!/bin/bash" >> $job_to_send
1334	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1335	echo "#SBATCH -t 00:20:00" >> $job_to_send
1336	echo "#SBATCH -N 1" >> $job_to_send
1337	echo "#SBATCH -n 1" >> $job_to_send
1338	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1339	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1340	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1341	echo "#SBATCH -A $project_account" >> $job_to_send
1342	echo "#SBATCH -p $return_queue" >> $job_to_send
1343	echo " " >> $job_to_send
1344	echo "set -x" >> $job_to_send
1345	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1346	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1347	echo "%%END%%" >> $job_to_send
1348
1349	elif [[ $remote_host = lckyuh ]]
1350	then
1351	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1352	echo "#!/bin/bash" >> $job_to_send
1353	echo "#PJM -L \"node=1\"" >> $job_to_send
1354	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1355	echo "#PJM --no-stging" >> $job_to_send
1356	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1357	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1358	echo "#PJM -j" >> $job_to_send
1359	echo " " >> $job_to_send
1360	echo "export LANG=en_US.UTF-8" >> $job_to_send
1361	echo "set -x" >> $job_to_send
1362	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1363	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1364	echo "%%END%%" >> $job_to_send
1365
1366	elif [[ $remote_host = lckyut ]]
1367	then
1368	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1369	echo "#!/bin/bash" >> $job_to_send
1370	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1371	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1372	echo "#PJM --no-stging" >> $job_to_send
1373	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1374	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1375	echo "#PJM -j" >> $job_to_send
1376	echo " " >> $job_to_send
1377	echo "export LANG=en_US.UTF-8" >> $job_to_send
1378	echo "set -x" >> $job_to_send
1379	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1380	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1381	echo "%%END%%" >> $job_to_send
1382
1383	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1384	then
1385	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1386	echo "#!/bin/bash" >> $job_to_send
1387	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1388	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1389	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1390	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1391	echo "#PBS -j oe" >> $job_to_send
1392	echo " " >> $job_to_send
1393	echo "set -x" >> $job_to_send
1394	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1395	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1396	echo "%%END%%" >> $job_to_send
1397
1398	elif [[ $remote_host = lcflow ]]
1399	then
1400	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1401	echo "#!/bin/bash" >> $job_to_send
1402	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1403	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1404	echo "SGEPREFIX -cwd" >> $job_to_send
1405	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1406	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1407	echo "SGEPREFIX -j y" >> $job_to_send
1408	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1409	echo " " >> $job_to_send
1410	echo "set -x" >> $job_to_send
1411	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1412	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1413	echo "" >> $job_to_send
1414	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1415	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1416	echo "rm -f scpjob.${identifier}" >> $job_to_send
1417	echo "%%END%%" >> $job_to_send
1418	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1419	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1420	elif [[ $remote_host = lcxe6 ]]
1421	then
1422	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1423	echo "#!/bin/ksh" >> $job_to_send
1424	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1425	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1426	echo "#PBS -A $project_account" >> $job_to_send
1427	echo "#PBS -l mppwidth=1" >> $job_to_send
1428	echo "#PBS -l mppnppn=1" >> $job_to_send
1429	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1430	echo "#PBS -j oe" >> $job_to_send
1431	echo " " >> $job_to_send
1432	echo "set -x" >> $job_to_send
1433	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1434	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1435	echo "%%END%%" >> $job_to_send
1436	else
1437
1438	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1439	echo "# @\\\$-q $return_queue" >> $job_to_send
1440	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1441	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1442	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1443	then
1444	echo "# @\$-l mpp_p=0" >> $job_to_send
1445	fi
1446	echo '# @\$-lF 10mb' >> $job_to_send
1447	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1448	echo '# @\\\$-eo' >> $job_to_send
1449	echo " " >> $job_to_send
1450	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1451	then
1452	echo "set +vx" >> $job_to_send
1453	echo ". .profile" >> $job_to_send
1454	fi
1455	echo "set -x" >> $job_to_send
1456	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
1457	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1458	echo "%%END%%" >> $job_to_send
1459
1460	fi
1461
1462	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1463	then
1464	echo "llsubmit scpjob.$identifier" >> $job_to_send
1465	elif [[ $remote_host = lcbullhh ]]
1466	then
1467	echo "sbatch scpjob.$identifier" >> $job_to_send
1468	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1469	then
1470	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1471	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1472	then
1473	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1474	elif [[ $remote_host = t3es ]]
1475	then
1476	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1477	elif [[ $remote_host = lckiaps ]]
1478	then
1479	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1480	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1481	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1482	elif [[ $remote_host = lckyu* ]]
1483	then
1484	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1485	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1486	elif [[ $remote_host = lcflow ]]
1487	then
1488	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1489	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1490	else
1491	echo "$submcom scpjob.$identifier" >> $job_to_send
1492	fi
1493	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1494	then
1495	echo "rm scpjob.$identifier" >> $job_to_send
1496	fi
1497	if [[ $remote_host = nech ]]
1498	then
1499	echo "cd -" >> $job_to_send
1500	fi
1501	else
1502	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1503	# ??? funktioniert das ÃŒberhaupt noch ???
1504	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1505	fi
1506	echo "set -x" >> $job_to_send
1507	echo " ' exit" >> $job_to_send
1508	echo "set -x" >> $job_to_send
1509	fi
1510
1511
1512	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1513	cat $file_to_send >> $job_to_send
1514
1515	if [[ $remote_host = ibm ]]
1516	then
1517	echo " " >> $job_to_send
1518	echo "exit" >> $job_to_send
1519	fi
1520
1521	# REMOVE JOB-FILE
1522	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1523	then
1524	echo " " >> $job_to_send
1525	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1526	fi
1527
1528
1529	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1530	if [[ $no_submit = false ]]
1531	then
1532	if [[ $remote_host != $local_host ]]
1533	then
1534	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1535	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1536	then
1537	job_catalog_save=$job_catalog
1538	job_catalog=job_queue
1539	elif [[ $remote_host = nech ]]
1540	then
1541	job_catalog_save=$job_catalog
1542	job_catalog=/hpf/b/${remote_user}/job_queue
1543	fi
1544	if [[ $remote_host = nech ]]
1545	then
1546	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1547	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1548	else
1549	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1550	fi
1551	if [[ $? = 1 ]]
1552	then
1553	locat=scp; exit
1554	fi
1555	if [[ $remote_host = ibms ]]
1556	then
1557	job_catalog=$job_catalog_save
1558	fi
1559	[[ $verify = true ]] && printf "\n >>> finished\n"
1560	else
1561	eval job_catalog=$job_catalog
1562	cp $job_to_send ${job_catalog}/$job_on_remhost
1563	fi
1564
1565
1566
1567	# START NQS- / LOADLEVELER-JOB
1568	if [[ $remote_host != $local_host ]]
1569	then
1570	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1571
1572	if [[ $remote_host = ibmku ]]
1573	then
1574	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1575	elif [[ $remote_host = lcflow ]]
1576	then
1577	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1578	else
1579	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1580	fi
1581
1582	[[ $verify = true ]] && printf " >>> o.k.\n"
1583	else
1584	cd $job_catalog
1585	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1586	then
1587	eval $submcom $job_on_remhost
1588	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1589	then
1590	chmod u+x $job_on_remhost
1591	eval $submcom $job_on_remhost
1592	elif [[ $local_host = nech ]]
1593	then
1594	if [[ $queue = default ]]
1595	then
1596	eval $submcom $job_on_remhost
1597	else
1598	eval $submcom -q $queue $job_on_remhost
1599	fi
1600	elif [[ $local_host = lcbullhh ]]
1601	then
1602	if [[ $queue = default ]]
1603	then
1604	eval $submcom $job_on_remhost
1605	fi
1606	else
1607	qsub $job_on_remhost
1608	fi
1609
1610	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1611	# AT THE END OF THE JOB
1612	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1613	then
1614	rm $job_on_remhost
1615	fi
1616	cd - > /dev/null
1617	fi
1618	fi
1619
1620	# FINAL ACTIONS
1621	if [[ $no_submit = false ]]
1622	then
1623	rm -f $job_to_send
1624	fi
1625	[[ $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 |