Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1691

Last change on this file since 1691 was 1623, checked in by heinze, 9 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 55.3 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 1623 2015-07-20 06:30:01Z maronga $
30	#
31	# 1621 2015-07-17 11:39:33Z heinze
32	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
33	#
34	# 1575 2015-03-27 09:56:27Z raasch
35	# mpp2-queues added to lccrayh
36	#
37	# 1547 2015-01-29 15:09:12Z witha
38	# adjustments for ForWind computing cluster (lcflow)
39	#
40	# 1545 2015-01-29 06:52:23Z heinze
41	# local host name for blizzard further specified
42	#
43	# 1480 2014-10-17 14:41:49Z raasch
44	# adjustments for 2nd stage of HLRNIII
45	#
46	# 1468 2014-09-24 14:06:57Z maronga
47	# Typo removed (addres->address)
48	# Adjustments for lcxe6
49	#
50	# 1452 2014-08-22 09:41:06Z heinze
51	# local hosts for blizzard added
52	#
53	# 1450 2014-08-21 07:31:51Z heinze
54	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
55	#
56	# 1442 2014-07-28 07:09:10Z raasch
57	# HLRN-III (lccrayb/lccrayh) queues adjusted
58	#
59	# 1378 2014-04-28 06:04:58Z raasch
60	# -et option added for lctit
61	#
62	# 1350 2014-04-04 13:01:30Z maronga
63	# location of qsub updated for lcxe6
64	#
65	# 1289 2014-03-04 07:12:34Z raasch
66	# German comments translated to English
67	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
68	#
69	# 1279 2014-01-28 12:10:14Z raasch
70	# node calculation modified due to changes in mrun (tasks_per_node must not be
71	# an integral divisor of numprocs any more)
72	#
73	# 1274 2014-01-09 13:14:54Z heinze
74	# adjustments for lccrayh
75	#
76	# 1266 2013-12-11 12:07:34Z heinze
77	# further adjustments for lccrayb (use msub instead of qsub)
78	#
79	# 1264 2013-12-09 12:46:09Z fricke
80	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
81	#
82	# 1262 2013-12-09 10:57:20Z fricke
83	# further adjustments for lccrayb
84	#
85	# 1260 2013-12-04 12:48:04Z raasch
86	# jaboticaba admitted
87	#
88	# 1255 2013-11-07 14:43:35Z raasch
89	# further adjustments for lccrayb
90	#
91	# 1224 2013-09-16 07:27:23Z raasch
92	# first adjustments for lccrayb
93	#
94	# 1202 2013-07-10 16:22:07Z witha
95	# adjustments for Forwind cluster (lcflow)
96	#
97	# 1199 2013-07-05 14:52:22Z raasch
98	# adjustments for CSC Helsinki (lccrayf)
99	#
100	# use of cluster/express queue enabled (ibmh)
101	# vinessa added (imuk)
102	#
103	# 1103 2013-02-20 02:15:53Z raasch
104	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
105	# instead typeset -L/R),
106	# further adjustments for lckyuh
107	#
108	# 2013-02-10 01:47:43Z raasch
109	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
110	# and for Forwind cluster (lcflow)
111	#
112	# 1094 2013-02-03 01:52:12Z raasch
113	# new option -P for explicit setting of ssh/scp port,
114	# decalpha parts (yonsei) removed
115	#
116	# 2013-02-02 07:06:13Z raasch
117	# adjustments for Kyushu-University computing center (lckyut - tatara)
118	# old changelog messages removed
119	#
120	# 1046 2012-11-09 14:38:45Z maronga
121	# code put under GPL (PALM 3.9)
122	#
123	# 08/07/94 - Siggi - first version finished
124	# 29/06/94 - Siggi - script development started
125	#--------------------------------------------------------------------------------#
126	# subjob - script for automatic generation and submission of batch-job files
127	# for various batch queuing systems
128	#--------------------------------------------------------------------------------#
129
130
131	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
132	delete_dayfile=false
133	email_notification=none
134	group_number=none
135	locat=normal
136	no_default_queue=none
137	no_submit=false
138	job_catalog="~/job_queue"
139	job_name=none
140	local_user=$LOGNAME
141	node_usage=shared
142	numprocs=0
143	punkte="..........................................................."
144	submcom=qsub
145	queue=default
146	remote_host=none
147	remote_user=""
148	verify=true
149
150	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
151	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
152
153
154
155	# ERROR HANDLING
156	# IN CASE OF EXIT:
157	trap 'if [[ $locat != normal ]]
158	then
159	case $locat in
160	(option) printf "\n --> available optios can be displayed"
161	printf " by typing:"
162	printf "\n \"subjob ?\" \n";;
163	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
164	(*) printf "\n +++ unknown error"
165	printf "\n please inform S. Raasch!\n"
166	esac
167	[[ -f $job_to_send ]] && rm $job_to_send
168	printf "\n\n+++ SUBJOB killed \n\n"
169	fi' exit
170
171
172	# IN CASE OF TERMINAL-BREAK:
173	trap '[[ -f $job_to_send ]] && rm $job_to_send
174	printf "\n\n+++ SUBJOB killed \n\n"
175	exit
176	' 2
177
178
179	# DETERMINE NAME OF LOCAL HOST
180	local_host=$(hostname)
181
182	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
183	# IS ADMITTED AT ALL)
184	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
185	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
186	case $local_host in
187	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
188	(atmos) local_address=172.20.25.35; local_host=lcide;;
189	(austru) local_address=130.75.105.128; local_host=lcmuk;;
190	(autan) local_address=130.75.105.57; local_host=lcmuk;;
191	(bora) local_address=130.75.105.103; local_host=lcmuk;;
192	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
193	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
194	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
195	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
196	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
197	(breva) local_address=130.75.105.98; local_host=lcmuk;;
198	(buran) local_address=130.75.105.58; local_host=lcmuk;;
199	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
200	(climate*) local_address=165.132.26.68; local_host=lcyon;;
201	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
202	(cs*) local_address=136.172.44.131; local_host=nech;;
203	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
204	(flow01) local_address=10.141.255.71; local_host=lcflow;;
205	(flow02) local_address=10.141.255.72; local_host=lcflow;;
206	(node*) local_address=165.132.26.61 local_host=lck;;
207	# (node*) local_address=210.219.61.8 local_host=lckordi;;
208	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
209	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
210	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
211	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
212	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
213	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
214	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
215	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
216	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
217	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
218	(sno) local_address=130.75.105.113; local_host=lcmuk;;
219	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
220	(login*) local_address=118.128.66.223; local_host=lckiaps;;
221	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
222	(meller) local_address=134.106.74.155; local_host=lcfor;;
223	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
224	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
225	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
226	(nobel*) local_address=150.183.5.101; local_host=ibms;;
227	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
228	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
229	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
230	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
231	(pingui) local_address=134.106.74.118; local_host=lcfor;;
232	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
233	(rte*) local_address=133.5.185.60; local_host=lcrte;;
234	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
235	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
236	(solano) local_address=130.75.105.110; local_host=lcmuk;;
237	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
238	(t2a*) local_address=10.1.6.165; local_host=lctit;;
239	(urban*) local_address=147.46.30.151 local_host=lcsb;;
240	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
241	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
242	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
243	(*) printf "\n +++ \"$local_host\" unknown";
244	printf "\n please contact the PALM group at IMUK";
245	locat=parameter; exit;;
246	esac
247
248
249
250	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
251	remote_host=$local_host
252
253
254
255
256	# READ THE SHELLSCRIPT-OPTIONS
257	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
258	do
259	case $option in
260	(c) job_catalog=$OPTARG;;
261	(d) delete_dayfile=true;;
262	(D) no_submit=true;;
263	(e) email_notification=$OPTARG;;
264	(g) group_number=$OPTARG;;
265	(h) remote_host=$OPTARG;;
266	(m) memory=$OPTARG;;
267	(n) job_name=$OPTARG;;
268	(N) node_usage=$OPTARG;;
269	(O) threads_per_task=$OPTARG;;
270	(P) scp_port=$OPTARG;;
271	(q) no_default_queue=$OPTARG;;
272	(t) cputime=$OPTARG;;
273	(T) tasks_per_node=$OPTARG;;
274	(u) remote_user=$OPTARG;;
275	(v) verify=false;;
276	(X) numprocs=$OPTARG;;
277	(\?) printf "\n +++ Option $OPTARG unknown \n";
278	locat=option; exit;;
279	esac
280	done
281
282
283	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
284	(( to_shift = $OPTIND - 1 ))
285	shift $to_shift; file_to_send=$1
286
287
288	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
289	if [ "$1" = "?" ]
290	then
291	(printf "\n *** subjob can be called as follows:\n"
292	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
293	printf "\n Description of available options:\n"
294	printf "\n Option Description Default-Value"
295	printf "\n -c job-input- and output-catalog ~/job_queue"
296	printf "\n -d no job-protocol will be created ---"
297	printf "\n -D only the job-file will be created ---"
298	printf "\n -h execution host, available hosts: $remote_host"
299	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
300	printf "\n lckiaps, lctit, nech"
301	printf "\n -m memory demand per process in MByte ---"
302	printf "\n -n jobname <jobdatei>"
303	printf "\n -O threads per task (for OpenMP usage) 1"
304	printf "\n -P ssh/scp port default port"
305	printf "\n -q job-queue to be used default"
306	printf "\n -t allowed cpu-time in seconds ---"
307	printf "\n -T tasks per node (on parallel hosts) ---"
308	printf "\n -u username on execution host from .netrc"
309	printf "\n -v no prompt for confirmation ---"
310	printf "\n -X # of processors (on parallel hosts) 1"
311	printf "\n "
312	printf "\n The only possible positional parameter is <jobfile>:"
313	printf "\n The complete NQS-job must be provided here."
314	printf "\n <jobfile>=? creates this outline\n\n") \| more
315	exit
316	fi
317
318
319
320	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
321	if [[ "$file_to_send" = "" ]]
322	then
323	printf "\n +++ job-file missing"
324	locat=parameter; exit
325	else
326	if [[ -f $file_to_send ]]
327	then
328	true
329	else
330	printf "\n +++ job-file: "
331	printf "\n $file_to_send"
332	printf "\n does not exist"
333	locat=parameter; exit
334	fi
335	fi
336
337
338
339	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
340	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
341	if [[ $job_name = none ]]
342	then
343	job_name=$file_to_send
344	fi
345	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
346	then
347	printf "\n +++ job-file name: "
348	printf "\n $job_name"
349	printf "\n must not contain \"/\"-characters"
350	locat=parameter; exit
351	fi
352
353
354
355
356	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
357	# OR IF NO HOST HAS BEEN GIVEN
358	if [[ $remote_host = none ]]
359	then
360	printf "\n +++ host missing"
361	locat=option; exit
362	else
363	case $remote_host in
364	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
365	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
366	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
367	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
368	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
369	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
370	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
371	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
372	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
373	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
374	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
375	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
376	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
377	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
378	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
379	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
380	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
381	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
382	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
383	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
384	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
385	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
386	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
387	locat=parameter; exit;;
388	esac
389	fi
390
391
392	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
393	if [[ $no_default_queue != none ]]
394	then
395	error=false
396	ndq=$no_default_queue
397	case $remote_host in
398	(ibm) case $ndq in
399	(p690_express\|p690_standard\|p690_long) error=false;;
400	(*) error=true;;
401	esac;;
402	(ibmh) case $ndq in
403	(cluster\|express) error=false;;
404	(*) error=true;;
405	esac;;
406	(ibmkisti) case $ndq in
407	(class.32plus\|class.1-2\|class.2-32) error=false;;
408	(*) error=true;;
409	esac;;
410	(ibmku) case $ndq in
411	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
412	(*) error=true;;
413	esac;;
414	(ibms) case $ndq in
415	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
416	(*) error=true;;
417	esac;;
418	(lcbullhh) case $ndq in
419	(compute\|shared) error=false;;
420	(*) error=true;;
421	esac;;
422	(lccrayb) case $ndq in
423	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
424	(*) error=true;;
425	esac;;
426	(lccrayh) case $ndq in
427	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
428	(*) error=true;;
429	esac;;
430	(lccrayf) case $ndq in
431	(usup\|test*\|small\|large) error=false;;
432	(*) error=true;;
433	esac;;
434	(lcflow) case $ndq in
435	(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;;
436	(*) error=true;;
437	esac;;
438	(lckiaps) case $ndq in
439	(express\|normal) error=false;;
440	(*) error=true;;
441	esac;;
442	(lckyoto) case $ndq in
443	(eh\|ph) error=false;;
444	(*) error=true;;
445	esac;;
446	(lckyuh) case $ndq in
447	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
448	(*) error=true;;
449	esac;;
450	(lckyut) case $ndq in
451	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
452	(*) error=true;;
453	esac;;
454	(lctit) case $ndq in
455	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
456	(*) error=true;;
457	esac;;
458	(t3eb) case $ndq in
459	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
460	(*) error=true;;
461	esac;;
462	(t3eh) case $ndq in
463	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
464	(*) error=true;;
465	esac;;
466	(t3ej2\|t3ej5) case $ndq in
467	(low\|normal\|high) error=false;;
468	(*) error=true;;
469	esac;;
470	(t3es) case $ndq in
471	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
472	(*) error=true;;
473	esac;;
474	esac
475	if [[ $error = true ]]
476	then
477	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
478	locat=parameter; exit
479	else
480	queue=$no_default_queue
481	fi
482	fi
483
484
485
486	# CHECK THE CPU-TIME
487	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
488	done=false
489	while [[ $done = false ]]
490	do
491	if (( $cputime <= 0 ))
492	then
493	printf "\n +++ wrong cpu-time or cpu-time missing"
494	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
495	printf "\n >>> "
496	read cputime 1>/dev/null 2>&1
497	else
498	done=true
499	fi
500	done
501	if [[ $remote_host = nech ]]
502	then
503	if (( tasks_per_node != 0 ))
504	then
505	(( cputime = cputime * tasks_per_node ))
506	elif [[ $numprocs != 0 ]]
507	then
508	(( cputime = cputime * numprocs ))
509	fi
510	fi
511	(( stunden = cputime / 3600 ))
512	(( resttime = cputime - stunden * 3600 ))
513	(( minuten = resttime / 60 ))
514	(( sekunden = resttime - minuten * 60 ))
515	timestring=${stunden}:${minuten}:${sekunden}
516
517
518
519	# CHECK THE MEMORY DEMAND
520	done=false
521	while [[ $done = false ]]
522	do
523	if (( memory <= 0 ))
524	then
525	printf "\n +++ wrong memory demand or memory demand missing"
526	printf "\n >>> Please type memory in MByte per process as INTEGER:"
527	printf "\n >>> "
528	read memory 1>/dev/null 2>&1
529	else
530	done=true
531	fi
532	done
533
534	if [[ $remote_host = nech ]]
535	then
536	if (( tasks_per_node != 0 ))
537	then
538	(( Memory = memory * tasks_per_node / 1000 ))
539	elif [[ $numprocs != 0 ]]
540	then
541	(( Memory = memory * numprocs / 1000 ))
542	else
543	(( Memory = memory / 1000 ))
544	fi
545	elif [[ $remote_host = lctit ]]
546	then
547	(( Memory = memory * tasks_per_node / 1000 ))
548	fi
549
550
551	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
552	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
553	then
554	(( memory = memory * threads_per_task ))
555	fi
556
557
558	# CALCULATE NUMBER OF REQUIRED NODES
559	if (( tasks_per_node != 0 ))
560	then
561	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
562	fi
563
564
565	# CALCULATE NUMBER OF PROCESSES PER NODE
566	(( processes_per_node = tasks_per_node * threads_per_task ))
567
568
569	# CALCULATE NUMBER OF MPI TASKS
570	(( mpi_tasks = numprocs / threads_per_task ))
571
572
573	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
574	if [[ "$scp_port" != "" ]]
575	then
576	PORTOPT="-P $scp_port"
577	SSH_PORTOPT="-p $scp_port"
578	fi
579
580
581	# HEADER-OUTPUT
582	if [[ $verify = true ]]
583	then
584	printf "\n\n"
585	printf "#--------------------------------------------------------------# \n"
586	spalte1=SUBJOB;spalte2=$(date)
587	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
588	printf "\| \| \n"
589	printf "\| values of parameters/options: \| \n"
590	spalte1=$(echo local_host$punkte \| cut -c-20)
591	spalte2=$punkte$local_host
592	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
593	spalte1=$(echo remote_host$punkte \| cut -c-20)
594	spalte2=$punkte$remote_host
595	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
596	spalte1=$(echo queue$punkte \| cut -c-20)
597	spalte2=$punkte$queue
598	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
599	spalte1=$(echo memory$punkte \| cut -c-20)
600	spalte2="$punkte$memory mb"
601	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
602	spalte1=$(echo cputime$punkte \| cut -c-20)
603	spalte2="$punkte$cputime sec"
604	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
605	spalte1=$(echo job_name$punkte \| cut -c-20)
606	spalte2="$punkte$job_name"
607	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
608	printf "#--------------------------------------------------------------# \n\n"
609
610
611	# QUERY CHECK
612	antwort="dummy"
613	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
614	do
615	read antwort?" >>> continue (y/n) ? "
616	done
617	if [[ $antwort = n \|\| $antwort = N ]]
618	then
619	locat=verify; exit
620	fi
621	printf "\n"
622	fi
623
624	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
625	identifier=$RANDOM
626	job_on_remhost=${job_name}_${identifier}_$local_host
627	job_to_send=job_to_send_$identifier
628	if [[ $delete_dayfile = false ]]
629	then
630	remote_dayfile=${local_host}_${job_name}_result_$identifier
631	local_dayfile=${remote_host}_${job_name}
632	else
633	remote_dayfile=/dev/null
634	fi
635
636
637	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
638	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
639	then
640
641	# GENERAL LOADLEVELER SETTINGS
642	execute_in_shell="#!/bin/ksh"
643	use_shell="# @ shell = /bin/ksh"
644	consumable_memory="ConsumableMemory($memory mb)"
645	class="# @ class = $queue"
646	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
647	network_to_use="# @ network.mpi = sn_all,shared,us"
648	data_limit="# @ data_limit = 1.76gb"
649	image_size="# @ image_size = 50"
650	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
651
652	if [[ $email_notification = none ]]
653	then
654	notify_user=""
655	else
656	notify_user="# @ notify_user = $email_notification"
657	if [[ $delete_dayfile = true ]]
658	then
659	notification='# @ notification = never'
660	fi
661	fi
662
663	if [[ $remote_host = ibmh ]]
664	then
665	data_limit=""
666	network_to_use=""
667	class="# @ class = $queue"
668	environment=""
669	rset="# @ rset = RSET_MCM_AFFINITY"
670	task_affinity="# @ task_affinity = core(1)"
671	elif [[ $remote_host = ibmkisti ]]
672	then
673	network_to_use="# @ network.MPI = sn_all,shared,US"
674	wall_clock_limit="# @ wall_clock_limit = $timestring"
675	if [[ $threads_per_task = 1 ]]
676	then
677	rset="# @ rset = RSET_MCM_AFFINITY"
678	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
679	fi
680	environment=""
681	use_shell=""
682	data_limit=""
683	image_size=""
684	elif [[ $remote_host = ibmku ]]
685	then
686	execute_in_shell="#!/usr/bin/ksh"
687	use_shell="# @ shell = /usr/bin/ksh"
688	consumable_memory=""
689	environment=""
690	network_to_use="# @ network.mpi = sn_all,shared,us"
691	data_limit=""
692	image_size=""
693	elif [[ $remote_host = ibms ]]
694	then
695	network_to_use="# @ network.mpi = csss,shared,us"
696	fi
697
698	cat > $job_to_send << %%END%%
699	$execute_in_shell
700	$use_shell
701
702	# @ job_type = parallel
703	# @ job_name = $job_name
704	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
705	# @ output = $remote_dayfile
706	# @ error = $remote_dayfile
707	$wall_clock_limit
708	$image_size
709	$class
710	$environment
711	$network_to_use
712	$data_limit
713	$rset
714	$mcm_affinity_options
715	$task_affinity
716	$notification
717	$notify_user
718
719	%%END%%
720
721	if (( nodes > 0 ))
722	then
723
724	if [[ $remote_host != ibmkisti ]]
725	then
726
727	cat >> $job_to_send << %%END%%
728	# @ node = $nodes
729	# @ tasks_per_node = $processes_per_node
730	# @ node_usage = $node_usage
731	# @ queue
732
733	%%END%%
734
735	else
736
737	cat >> $job_to_send << %%END%%
738	# @ total_tasks = $mpi_tasks
739	# @ blocking = unlimited
740	# @ queue
741
742	%%END%%
743
744	fi
745
746	else
747
748	cat >> $job_to_send << %%END%%
749	# @ blocking = unlimited
750	# @ total_tasks = $numprocs
751	# @ node_usage = $node_usage
752	# @ queue
753
754	%%END%%
755
756	fi
757
758	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
759	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
760	then
761	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
762	fi
763
764	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
765	then
766
767	cat > $job_to_send << %%END%%
768	#!/bin/ksh
769
770	# @ job_type = serial
771	# @ node_usage = $node_usage
772	# @ job_name = palm
773	# @ wall_clock_limit = ${timestring},$timestring
774	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
775	# @ output = $remote_dayfile
776	# @ error = $remote_dayfile
777	$class
778	$notification
779
780	# @ queue
781
782	%%END%%
783
784	elif [[ $remote_host = lcbullhh ]]
785	then
786	if [[ $numprocs != 0 ]]
787	then
788	cat > $job_to_send << %%END%%
789	#!/bin/bash -l
790	#SBATCH -J $job_name
791	#SBATCH -t $timestring
792	#SBATCH -N $nodes
793	#SBATCH --ntasks-per-node=$processes_per_node
794	#SBATCH -p $queue
795	#SBATCH -o $remote_dayfile
796	#SBATCH -e $remote_dayfile
797	#SBATCH -A $project_account
798
799	$init_cmds
800	$module_calls
801
802	%%END%%
803
804	else
805	cat > $job_to_send << %%END%%
806	#!/bin/bash -l
807	#SBATCH -J $job_name
808	#SBATCH -t $timestring
809	#SBATCH -l ncpus=1
810	#SBATCH -l pmem=${memory}mb
811	#SBATCH -m abe
812	#SBATCH -o $remote_dayfile
813	#SBATCH -e $remote_dayfile
814	#SBATCH -A $project_account
815
816	$init_cmds
817	$module_calls
818
819	%%END%%
820
821	fi
822
823	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
824	then
825
826	if [[ "$feature" != "" ]]
827	then
828	featuredir="#PBS -l feature=$feature"
829	fi
830
831	if [[ $numprocs != 0 ]]
832	then
833	cat > $job_to_send << %%END%%
834	#!/bin/bash
835	#PBS -N $job_name
836	#PBS -l walltime=$timestring
837	#PBS -l nodes=$nodes:ppn=$processes_per_node
838	#PBS -o $remote_dayfile
839	#PBS -j oe
840	#PBS -q $queue
841	$featuredir
842
843	$module_calls
844
845	%%END%%
846
847	else
848
849	continue
850
851	fi
852
853	elif [[ $remote_host = lccrayf ]]
854	then
855
856	if [[ $numprocs != 0 ]]
857	then
858	cat > $job_to_send << %%END%%
859	#!/bin/bash -l
860	#SBATCH -J $job_name
861	#SBATCH -t $timestring
862	#SBATCH -N $nodes
863	#SBATCH --ntasks-per-node=$processes_per_node
864	#SBATCH -p $queue
865	#SBATCH -o $remote_dayfile
866	#SBATCH -e $remote_dayfile
867
868	$init_cmds
869	$module_calls
870
871	%%END%%
872
873	else
874	cat > $job_to_send << %%END%%
875	#!/bin/bash -l
876	#SBATCH -J $job_name
877	#SBATCH -t $timestring
878	#SBATCH -l ncpus=1
879	#SBATCH -l pmem=${memory}mb
880	#SBATCH -m abe
881	#SBATCH -o $remote_dayfile
882	#SBATCH -e $remote_dayfile
883
884	$init_cmds
885	$module_calls
886
887	%%END%%
888
889	fi
890
891	elif [[ $remote_host = lcflow ]]
892	then
893	if [[ $numprocs != 0 ]]
894	then
895	pe_set="#$ -pe impi $numprocs"
896	else
897	pe_set="#$ -pe impi 1"
898	fi
899	if [[ $queue = default ]]
900	then
901	queue_set=""
902	else
903	queue_set="#$ -q $queue"
904	fi
905	[[ "$disc_space" = "" ]] && disc_space=50
906
907	cat > $job_to_send << %%END%%
908	#!/bin/bash
909	#$ -S /bin/bash
910	#$ -N $job_name
911	#$ -cwd
912	#$ -l h_rt=$timestring
913	#$ -l h_vmem=${memory}M
914	#$ -o $remote_dayfile
915	#$ -j y
916	$pe_set
917	#$ -R y
918	#$ -l h_fsize=${disc_space}G
919	$queue_set
920
921	%%END%%
922
923	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
924	then
925
926	if [[ $numprocs != 0 ]]
927	then
928	cat > $job_to_send << %%END%%
929	#!/bin/ksh
930	#PBS -N $job_name
931	#PBS -l walltime=$timestring
932	#PBS -l ncpus=$numprocs
933	#PBS -l pmem=${memory}mb
934	#PBS -o $remote_dayfile
935	#PBS -l nodes=$nodes:ppn=${processes_per_node}
936	#PBS -j oe
937
938	mpd &
939
940	%%END%%
941
942	else
943	cat > $job_to_send << %%END%%
944	#!/bin/ksh
945	#PBS -N $job_name
946	#PBS -l walltime=$timestring
947	#PBS -l ncpus=1
948	#PBS -l pmem=${memory}mb
949	#PBS -o $remote_dayfile
950	#PBS -j oe
951
952	%%END%%
953
954	fi
955
956	elif [[ $remote_host = lckiaps ]]
957	then
958
959	if [[ $numprocs != 0 ]]
960	then
961	cat > $job_to_send << %%END%%
962	#!/bin/ksh
963	#PBS -N $job_name
964	#PBS -l walltime=$timestring
965	#PBS -l select=1:ncpus=$numprocs
966	#PBS -l pmem=${memory}mb
967	#PBS -q $queue
968	#PBS -o $remote_dayfile
969	#PBS -j oe
970	#PBS -V
971
972	%%END%%
973
974	else
975	cat > $job_to_send << %%END%%
976	#!/bin/ksh
977	#PBS -N $job_name
978	#PBS -l walltime=$timestring
979	#PBS -l ncpus=1
980	#PBS -l pmem=${memory}mb
981	#PBS -o $remote_dayfile
982	#PBS -j oe
983
984	%%END%%
985
986	fi
987
988	elif [[ $remote_host = lcyon ]]
989	then
990
991	if [[ $numprocs != 0 ]]
992	then
993	cat > $job_to_send << %%END%%
994	#!/bin/ksh
995	#PBS -N $job_name
996	#PBS -l walltime=$timestring
997	#PBS -l ncpus=$numprocs
998	#PBS -l pmem=${memory}mb
999	#PBS -o $remote_dayfile
1000	#PBS -j oe
1001
1002	%%END%%
1003
1004	else
1005	cat > $job_to_send << %%END%%
1006	#!/bin/ksh
1007	#PBS -N $job_name
1008	#PBS -l walltime=$timestring
1009	#PBS -l ncpus=1
1010	#PBS -l pmem=${memory}mb
1011	#PBS -o $remote_dayfile
1012	#PBS -j oe
1013
1014	%%END%%
1015
1016	fi
1017
1018	elif [[ $remote_host = lcxe6 ]]
1019	then
1020
1021	if [[ $numprocs != 0 ]]
1022	then
1023	cat > $job_to_send << %%END%%
1024	#!/bin/ksh
1025	#PBS -S /bin/ksh
1026	#PBS -N $job_name
1027	#PBS -A $project_account
1028	#PBS -j oe
1029	#PBS -l walltime=$timestring
1030	#PBS -l mppwidth=${numprocs}
1031	#PBS -l mppnppn=${processes_per_node}
1032	#PBS -m abe
1033	#PBS -o $remote_dayfile
1034	$email_directive
1035
1036	$init_cmds
1037	$module_calls
1038
1039	%%END%%
1040
1041	else
1042	cat > $job_to_send << %%END%%
1043	#!/bin/ksh
1044	#PBS -S /bin/ksh
1045	#PBS -N $job_name
1046	#PBS -A $project_account
1047	#PBS -j oe
1048	#PBS -l walltime=$timestring
1049	#PBS -l ncpus=1
1050	#PBS -l pmem=${memory}mb
1051	#PBS -m abe
1052	$email_directive
1053	#PBS -o $remote_dayfile
1054
1055	$init_cmds
1056	$module_calls
1057
1058	%%END%%
1059
1060	fi
1061
1062	elif [[ $remote_host = lckyoto ]]
1063	then
1064
1065	cat > $job_to_send << %%END%%
1066	#!/bin/ksh
1067	# @\$-o $remote_dayfile
1068	# @\$-eo -oi
1069	# @\$-lP 16
1070	# @\$-lp 1
1071	# @\$-lm 28gb -llm unlimited -ls unlimited
1072	# @\$-q $queue
1073	# @\$-Pvn abs_pack
1074	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1075	#. /thin/local/etc/setprofile/intel-11.0.sh
1076	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1077	. ~/.myprofile
1078	#. /home2/t/t51254/palm/current_version/myprofile
1079	#. /thin/apps/pgi/mpi.sh
1080	#
1081	env
1082	#
1083	set -x
1084
1085	%%END%%
1086
1087	elif [[ $remote_host = lcxt5m ]]
1088	then
1089
1090	if [[ $numprocs != 0 ]]
1091	then
1092	cat > $job_to_send << %%END%%
1093	#!/bin/ksh
1094	#PBS -S /bin/ksh
1095	#PBS -N $job_name
1096	#PBS -j oe
1097	#PBS -l walltime=$timestring
1098	#PBS -l mppwidth=${numprocs}
1099	#PBS -l mppnppn=${processes_per_node}
1100	#PBS -m abe
1101	#PBS -o $remote_dayfile
1102
1103	$init_cmds
1104	$module_calls
1105
1106	%%END%%
1107
1108	else
1109	cat > $job_to_send << %%END%%
1110	#!/bin/ksh
1111	#PBS -S /bin/ksh
1112	#PBS -N $job_name
1113	#PBS -j oe
1114	#PBS -l walltime=$timestring
1115	#PBS -l ncpus=1
1116	#PBS -l pmem=${memory}mb
1117	#PBS -m abe
1118	#PBS -o $remote_dayfile
1119
1120	$init_cmds
1121	$module_calls
1122
1123	%%END%%
1124
1125	fi
1126
1127	elif [[ $remote_host = lckyuh ]]
1128	then
1129	cat > $job_to_send << %%END%%
1130	#!/bin/bash
1131	#PJM -L "rscgrp=$queue"
1132	#PJM -L "node=$nodes"
1133	#PJM --mpi "proc=$numprocs"
1134	#PJM -L "elapse=$timestring"
1135	#PJM -o $remote_dayfile
1136	#PJM -j
1137	#PJM -X
1138	#PJM --no-stging
1139
1140	export LANG=en_US.UTF-8
1141	%%END%%
1142
1143	elif [[ $remote_host = lckyut ]]
1144	then
1145	cat > $job_to_send << %%END%%
1146	#!/bin/bash
1147	#PJM -L "rscgrp=$queue"
1148	#PJM -L "vnode=$numprocs"
1149	#PJM -L "vnode-core=1"
1150	#PJM -L "elapse=$timestring"
1151	#PJM --mpi proc=$numprocs
1152	#PJM -o $remote_dayfile
1153	#PJM -j
1154	#PJM -X
1155	#PJM --no-stging
1156
1157	export LANG=en_US.UTF-8
1158	%%END%%
1159
1160	elif [[ $remote_host = nech ]]
1161	then
1162
1163	if (( nodes > 1 ))
1164	then
1165	cat > $job_to_send << %%END%%
1166	#!/bin/ksh
1167	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1168	#PBS -l ${qsubmem}=${Memory}gb
1169	#PBS -b $nodes
1170	#PBS -o $remote_dayfile
1171	#PBS -N palm
1172	#PBS -j o
1173	#PBS -T mpisx
1174
1175	%%END%%
1176
1177	elif [[ $numprocs != 0 ]]
1178	then
1179	cat > $job_to_send << %%END%%
1180	#!/bin/ksh
1181	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1182	#PBS -l ${qsubmem}=${Memory}gb
1183	#PBS -o $remote_dayfile
1184	#PBS -N palm
1185	#PBS -j o
1186
1187	%%END%%
1188
1189	else
1190	cat > $job_to_send << %%END%%
1191	#!/bin/ksh
1192	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1193	#PBS -o $remote_dayfile
1194	#PBS -j o
1195
1196	%%END%%
1197
1198	fi
1199
1200	elif [[ $remote_host = lctit ]]
1201	then
1202	cat > $job_to_send << %%END%%
1203	#!/bin/ksh
1204	$init_cmds
1205	$module_calls
1206
1207	%%END%%
1208
1209	# SET OPTIONS FOR SUBMIT-COMMAND
1210	if [[ $tasks_per_node != $processes_per_node ]]
1211	then
1212	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -o $remote_dayfile -j oe -et 1 -q $queue "
1213	else
1214	submcom="$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 "
1215	fi
1216
1217	else
1218
1219	cat > $job_to_send << %%END%%
1220	# @\$-q ${queue}
1221	# @\$-l${qsubtime} $timestring
1222	# @\$-l${qsubmem} ${memory}mb
1223	# @\$-o $remote_dayfile
1224	# @\$-eo
1225
1226	%%END%%
1227
1228	fi
1229
1230
1231	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1232	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1233	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1234	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1235	then
1236	echo "set +vx" >> $job_to_send
1237	echo "trap '" >> $job_to_send
1238	echo "set +vx" >> $job_to_send
1239	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 ]]
1240	then
1241	if [[ $remote_host = ibmh ]]
1242	then
1243	return_queue=c1
1244	elif [[ $remote_host = ibmkisti ]]
1245	then
1246	return_queue=class.1-2
1247	elif [[ $remote_host = ibmku ]]
1248	then
1249	return_queue=sdbg2
1250	elif [[ $remote_host = ibms ]]
1251	then
1252	return_queue=p_normal
1253	elif [[ $remote_host = lcbullhh ]]
1254	then
1255	return_queue=shared
1256	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1257	then
1258	return_queue=dataq
1259	elif [[ $remote_host = lcxe6 ]]
1260	then
1261	return_queue=debug
1262	elif [[ $remote_host = lckiaps ]]
1263	then
1264	return_queue=express
1265	elif [[ $remote_host = lckyuh ]]
1266	then
1267	return_queue=cx-single
1268	elif [[ $remote_host = lckyut ]]
1269	then
1270	return_queue=cx-single
1271	else
1272	return_queue=unknown
1273	fi
1274
1275	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1276	then
1277
1278	if [[ $remote_host = ibmku ]]
1279	then
1280	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1281	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1282	else
1283	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1284	fi
1285	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1286	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1287	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1288	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1289	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1290	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1291	if [[ $host != "ibmh" ]]
1292	then
1293	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1294	fi
1295	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1296	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1297
1298	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1299	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1300
1301	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1302	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
1303	if [[ $remote_host = ibmku ]]
1304	then
1305	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1306	fi
1307	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1308
1309	elif [[ $remote_host = nech ]]
1310	then
1311	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1312	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1313	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1314	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1315	echo "#PBS -j o" >> $job_to_send
1316	echo " " >> $job_to_send
1317	echo "set -x" >> $job_to_send
1318	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1319	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1320	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1321	echo "%%END%%" >> $job_to_send
1322
1323	elif [[ $remote_host = lcbullhh ]]
1324	then
1325	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1326	echo "#!/bin/bash" >> $job_to_send
1327	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1328	echo "#SBATCH -t 00:20:00" >> $job_to_send
1329	echo "#SBATCH -N 1" >> $job_to_send
1330	echo "#SBATCH -n 1" >> $job_to_send
1331	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1332	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1333	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1334	echo "#SBATCH -A $project_account" >> $job_to_send
1335	echo "#SBATCH -p $return_queue" >> $job_to_send
1336	echo " " >> $job_to_send
1337	echo "set -x" >> $job_to_send
1338	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1339	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1340	echo "%%END%%" >> $job_to_send
1341
1342	elif [[ $remote_host = lckyuh ]]
1343	then
1344	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1345	echo "#!/bin/bash" >> $job_to_send
1346	echo "#PJM -L \"node=1\"" >> $job_to_send
1347	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1348	echo "#PJM --no-stging" >> $job_to_send
1349	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1350	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1351	echo "#PJM -j" >> $job_to_send
1352	echo " " >> $job_to_send
1353	echo "export LANG=en_US.UTF-8" >> $job_to_send
1354	echo "set -x" >> $job_to_send
1355	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1356	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1357	echo "%%END%%" >> $job_to_send
1358
1359	elif [[ $remote_host = lckyut ]]
1360	then
1361	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1362	echo "#!/bin/bash" >> $job_to_send
1363	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1364	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1365	echo "#PJM --no-stging" >> $job_to_send
1366	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1367	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1368	echo "#PJM -j" >> $job_to_send
1369	echo " " >> $job_to_send
1370	echo "export LANG=en_US.UTF-8" >> $job_to_send
1371	echo "set -x" >> $job_to_send
1372	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1373	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1374	echo "%%END%%" >> $job_to_send
1375
1376	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1377	then
1378	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1379	echo "#!/bin/bash" >> $job_to_send
1380	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1381	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1382	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1383	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1384	echo "#PBS -j oe" >> $job_to_send
1385	echo " " >> $job_to_send
1386	echo "set -x" >> $job_to_send
1387	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1388	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1389	echo "%%END%%" >> $job_to_send
1390
1391	elif [[ $remote_host = lcflow ]]
1392	then
1393	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1394	echo "#!/bin/bash" >> $job_to_send
1395	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1396	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1397	echo "SGEPREFIX -cwd" >> $job_to_send
1398	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1399	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1400	echo "SGEPREFIX -j y" >> $job_to_send
1401	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1402	echo " " >> $job_to_send
1403	echo "set -x" >> $job_to_send
1404	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1405	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1406	echo "" >> $job_to_send
1407	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1408	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1409	echo "rm -f scpjob.${identifier}" >> $job_to_send
1410	echo "%%END%%" >> $job_to_send
1411	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1412	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1413	elif [[ $remote_host = lcxe6 ]]
1414	then
1415	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1416	echo "#!/bin/ksh" >> $job_to_send
1417	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1418	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1419	echo "#PBS -A $project_account" >> $job_to_send
1420	echo "#PBS -l mppwidth=1" >> $job_to_send
1421	echo "#PBS -l mppnppn=1" >> $job_to_send
1422	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1423	echo "#PBS -j oe" >> $job_to_send
1424	echo " " >> $job_to_send
1425	echo "set -x" >> $job_to_send
1426	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1427	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1428	echo "%%END%%" >> $job_to_send
1429	else
1430
1431	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1432	echo "# @\\\$-q $return_queue" >> $job_to_send
1433	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1434	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1435	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1436	then
1437	echo "# @\$-l mpp_p=0" >> $job_to_send
1438	fi
1439	echo '# @\$-lF 10mb' >> $job_to_send
1440	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1441	echo '# @\\\$-eo' >> $job_to_send
1442	echo " " >> $job_to_send
1443	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1444	then
1445	echo "set +vx" >> $job_to_send
1446	echo ". .profile" >> $job_to_send
1447	fi
1448	echo "set -x" >> $job_to_send
1449	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
1450	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1451	echo "%%END%%" >> $job_to_send
1452
1453	fi
1454
1455	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1456	then
1457	echo "llsubmit scpjob.$identifier" >> $job_to_send
1458	elif [[ $remote_host = lcbullhh ]]
1459	then
1460	echo "sbatch scpjob.$identifier" >> $job_to_send
1461	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1462	then
1463	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1464	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1465	then
1466	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1467	elif [[ $remote_host = t3es ]]
1468	then
1469	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1470	elif [[ $remote_host = lckiaps ]]
1471	then
1472	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1473	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1474	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1475	elif [[ $remote_host = lckyu* ]]
1476	then
1477	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1478	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1479	elif [[ $remote_host = lcflow ]]
1480	then
1481	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1482	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1483	else
1484	echo "$submcom scpjob.$identifier" >> $job_to_send
1485	fi
1486	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1487	then
1488	echo "rm scpjob.$identifier" >> $job_to_send
1489	fi
1490	if [[ $remote_host = nech ]]
1491	then
1492	echo "cd -" >> $job_to_send
1493	fi
1494	else
1495	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1496	# ??? funktioniert das ÃŒberhaupt noch ???
1497	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1498	fi
1499	echo "set -x" >> $job_to_send
1500	echo " ' exit" >> $job_to_send
1501	echo "set -x" >> $job_to_send
1502	fi
1503
1504
1505	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1506	cat $file_to_send >> $job_to_send
1507
1508	if [[ $remote_host = ibm ]]
1509	then
1510	echo " " >> $job_to_send
1511	echo "exit" >> $job_to_send
1512	fi
1513
1514	# REMOVE JOB-FILE
1515	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1516	then
1517	echo " " >> $job_to_send
1518	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1519	fi
1520
1521
1522	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1523	if [[ $no_submit = false ]]
1524	then
1525	if [[ $remote_host != $local_host ]]
1526	then
1527	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1528	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1529	then
1530	job_catalog_save=$job_catalog
1531	job_catalog=job_queue
1532	elif [[ $remote_host = nech ]]
1533	then
1534	job_catalog_save=$job_catalog
1535	job_catalog=/hpf/b/${remote_user}/job_queue
1536	fi
1537	if [[ $remote_host = nech ]]
1538	then
1539	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1540	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1541	else
1542	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1543	fi
1544	if [[ $? = 1 ]]
1545	then
1546	locat=scp; exit
1547	fi
1548	if [[ $remote_host = ibms ]]
1549	then
1550	job_catalog=$job_catalog_save
1551	fi
1552	[[ $verify = true ]] && printf "\n >>> finished\n"
1553	else
1554	eval job_catalog=$job_catalog
1555	cp $job_to_send ${job_catalog}/$job_on_remhost
1556	fi
1557
1558
1559
1560	# START NQS- / LOADLEVELER-JOB
1561	if [[ $remote_host != $local_host ]]
1562	then
1563	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1564
1565	if [[ $remote_host = ibmku ]]
1566	then
1567	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1568	elif [[ $remote_host = lcflow ]]
1569	then
1570	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1571	else
1572	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1573	fi
1574
1575	[[ $verify = true ]] && printf " >>> o.k.\n"
1576	else
1577	cd $job_catalog
1578	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1579	then
1580	eval $submcom $job_on_remhost
1581	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1582	then
1583	chmod u+x $job_on_remhost
1584	eval $submcom $job_on_remhost
1585	elif [[ $local_host = nech ]]
1586	then
1587	if [[ $queue = default ]]
1588	then
1589	eval $submcom $job_on_remhost
1590	else
1591	eval $submcom -q $queue $job_on_remhost
1592	fi
1593	elif [[ $local_host = lcbullhh ]]
1594	then
1595	if [[ $queue = default ]]
1596	then
1597	eval $submcom $job_on_remhost
1598	fi
1599	else
1600	qsub $job_on_remhost
1601	fi
1602
1603	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1604	# AT THE END OF THE JOB
1605	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1606	then
1607	rm $job_on_remhost
1608	fi
1609	cd - > /dev/null
1610	fi
1611	fi
1612
1613	# FINAL ACTIONS
1614	if [[ $no_submit = false ]]
1615	then
1616	rm -f $job_to_send
1617	fi
1618	[[ $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 |