Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1866

Last change on this file since 1866 was 1866, checked in by raasch, 9 years ago
scripts adjusted for lcocean
Property svn:keywords set to `Id Rev`
File size: 57.8 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	# adjusted for lcocean
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1866 2016-04-15 06:50:59Z raasch $
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	(ocean) local_address="ocean"; local_host=lcocean;;
234	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
235	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
236	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
237	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
238	(pingui) local_address=134.106.74.118; local_host=lcfor;;
239	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
240	(rte*) local_address=133.5.185.60; local_host=lcrte;;
241	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
242	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
243	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
244	(solano) local_address=130.75.105.110; local_host=lcmuk;;
245	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
246	(tc*) local_address="ocean"; local_host=lcocean;;
247	(t2a*) local_address=10.1.6.165; local_host=lctit;;
248	(urban*) local_address=147.46.30.151 local_host=lcsb;;
249	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
250	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
251	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
252	(*) printf "\n +++ \"$local_host\" unknown";
253	printf "\n please contact the PALM group at IMUK";
254	locat=parameter; exit;;
255	esac
256
257
258
259	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
260	remote_host=$local_host
261
262
263
264
265	# READ THE SHELLSCRIPT-OPTIONS
266	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
267	do
268	case $option in
269	(c) job_catalog=$OPTARG;;
270	(d) delete_dayfile=true;;
271	(D) no_submit=true;;
272	(e) email_notification=$OPTARG;;
273	(g) group_number=$OPTARG;;
274	(h) remote_host=$OPTARG;;
275	(m) memory=$OPTARG;;
276	(n) job_name=$OPTARG;;
277	(N) node_usage=$OPTARG;;
278	(O) threads_per_task=$OPTARG;;
279	(P) scp_port=$OPTARG;;
280	(q) no_default_queue=$OPTARG;;
281	(t) cputime=$OPTARG;;
282	(T) tasks_per_node=$OPTARG;;
283	(u) remote_user=$OPTARG;;
284	(v) verify=false;;
285	(X) numprocs=$OPTARG;;
286	(\?) printf "\n +++ Option $OPTARG unknown \n";
287	locat=option; exit;;
288	esac
289	done
290
291
292	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
293	(( to_shift = $OPTIND - 1 ))
294	shift $to_shift; file_to_send=$1
295
296
297	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
298	if [ "$1" = "?" ]
299	then
300	(printf "\n *** subjob can be called as follows:\n"
301	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
302	printf "\n Description of available options:\n"
303	printf "\n Option Description Default-Value"
304	printf "\n -c job-input- and output-catalog ~/job_queue"
305	printf "\n -d no job-protocol will be created ---"
306	printf "\n -D only the job-file will be created ---"
307	printf "\n -h execution host, available hosts: $remote_host"
308	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
309	printf "\n lckiaps, lctit, nech"
310	printf "\n -m memory demand per process in MByte ---"
311	printf "\n -n jobname <jobdatei>"
312	printf "\n -O threads per task (for OpenMP usage) 1"
313	printf "\n -P ssh/scp port default port"
314	printf "\n -q job-queue to be used default"
315	printf "\n -t allowed cpu-time in seconds ---"
316	printf "\n -T tasks per node (on parallel hosts) ---"
317	printf "\n -u username on execution host from .netrc"
318	printf "\n -v no prompt for confirmation ---"
319	printf "\n -X # of processors (on parallel hosts) 1"
320	printf "\n "
321	printf "\n The only possible positional parameter is <jobfile>:"
322	printf "\n The complete NQS-job must be provided here."
323	printf "\n <jobfile>=? creates this outline\n\n") \| more
324	exit
325	fi
326
327
328
329	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
330	if [[ "$file_to_send" = "" ]]
331	then
332	printf "\n +++ job-file missing"
333	locat=parameter; exit
334	else
335	if [[ -f $file_to_send ]]
336	then
337	true
338	else
339	printf "\n +++ job-file: "
340	printf "\n $file_to_send"
341	printf "\n does not exist"
342	locat=parameter; exit
343	fi
344	fi
345
346
347
348	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
349	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
350	if [[ $job_name = none ]]
351	then
352	job_name=$file_to_send
353	fi
354	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
355	then
356	printf "\n +++ job-file name: "
357	printf "\n $job_name"
358	printf "\n must not contain \"/\"-characters"
359	locat=parameter; exit
360	fi
361
362
363
364
365	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
366	# OR IF NO HOST HAS BEEN GIVEN
367	if [[ $remote_host = none ]]
368	then
369	printf "\n +++ host missing"
370	locat=option; exit
371	else
372	case $remote_host in
373	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
374	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
375	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
376	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
377	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
378	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
379	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
380	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
381	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
382	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
383	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
384	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
385	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
386	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
387	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
388	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
389	(lcocean) remote_address="ocean"; submcom=qsub;;
390	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
391	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
392	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
393	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
394	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
395	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
396	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
397	locat=parameter; exit;;
398	esac
399	fi
400
401
402	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
403	if [[ $no_default_queue != none ]]
404	then
405	error=false
406	ndq=$no_default_queue
407	case $remote_host in
408	(ibm) case $ndq in
409	(p690_express\|p690_standard\|p690_long) error=false;;
410	(*) error=true;;
411	esac;;
412	(ibmh) case $ndq in
413	(cluster\|express) error=false;;
414	(*) error=true;;
415	esac;;
416	(ibmkisti) case $ndq in
417	(class.32plus\|class.1-2\|class.2-32) error=false;;
418	(*) error=true;;
419	esac;;
420	(ibmku) case $ndq in
421	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
422	(*) error=true;;
423	esac;;
424	(ibms) case $ndq in
425	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
426	(*) error=true;;
427	esac;;
428	(lcbullhh) case $ndq in
429	(compute\|shared) error=false;;
430	(*) error=true;;
431	esac;;
432	(lccrayb) case $ndq in
433	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
434	(*) error=true;;
435	esac;;
436	(lccrayh) case $ndq in
437	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
438	(*) error=true;;
439	esac;;
440	(lccrayf) case $ndq in
441	(usup\|test*\|small\|large) error=false;;
442	(*) error=true;;
443	esac;;
444	(lcflow) case $ndq in
445	(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;;
446	(*) error=true;;
447	esac;;
448	(lckiaps) case $ndq in
449	(express\|normal) error=false;;
450	(*) error=true;;
451	esac;;
452	(lckyoto) case $ndq in
453	(eh\|ph) error=false;;
454	(*) error=true;;
455	esac;;
456	(lckyuh) case $ndq in
457	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
458	(*) error=true;;
459	esac;;
460	(lckyut) case $ndq in
461	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
462	(*) error=true;;
463	esac;;
464	(lctit) case $ndq in
465	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
466	(*) error=true;;
467	esac;;
468	(t3eb) case $ndq in
469	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
470	(*) error=true;;
471	esac;;
472	(t3eh) case $ndq in
473	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
474	(*) error=true;;
475	esac;;
476	(t3ej2\|t3ej5) case $ndq in
477	(low\|normal\|high) error=false;;
478	(*) error=true;;
479	esac;;
480	(t3es) case $ndq in
481	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
482	(*) error=true;;
483	esac;;
484	esac
485	if [[ $error = true ]]
486	then
487	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
488	locat=parameter; exit
489	else
490	queue=$no_default_queue
491	fi
492	fi
493
494
495
496	# CHECK THE CPU-TIME
497	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
498	done=false
499	while [[ $done = false ]]
500	do
501	if (( $cputime <= 0 ))
502	then
503	printf "\n +++ wrong cpu-time or cpu-time missing"
504	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
505	printf "\n >>> "
506	read cputime 1>/dev/null 2>&1
507	else
508	done=true
509	fi
510	done
511	if [[ $remote_host = nech ]]
512	then
513	if (( tasks_per_node != 0 ))
514	then
515	(( cputime = cputime * tasks_per_node ))
516	elif [[ $numprocs != 0 ]]
517	then
518	(( cputime = cputime * numprocs ))
519	fi
520	fi
521	(( stunden = cputime / 3600 ))
522	(( resttime = cputime - stunden * 3600 ))
523	(( minuten = resttime / 60 ))
524	(( sekunden = resttime - minuten * 60 ))
525	timestring=${stunden}:${minuten}:${sekunden}
526
527
528
529	# CHECK THE MEMORY DEMAND
530	done=false
531	while [[ $done = false ]]
532	do
533	if (( memory <= 0 ))
534	then
535	printf "\n +++ wrong memory demand or memory demand missing"
536	printf "\n >>> Please type memory in MByte per process as INTEGER:"
537	printf "\n >>> "
538	read memory 1>/dev/null 2>&1
539	else
540	done=true
541	fi
542	done
543
544	if [[ $remote_host = nech ]]
545	then
546	if (( tasks_per_node != 0 ))
547	then
548	(( Memory = memory * tasks_per_node / 1000 ))
549	elif [[ $numprocs != 0 ]]
550	then
551	(( Memory = memory * numprocs / 1000 ))
552	else
553	(( Memory = memory / 1000 ))
554	fi
555	elif [[ $remote_host = lctit ]]
556	then
557	(( Memory = memory * tasks_per_node / 1000 ))
558	fi
559
560
561	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
562	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
563	then
564	(( memory = memory * threads_per_task ))
565	fi
566
567
568	# CALCULATE NUMBER OF REQUIRED NODES
569	if (( tasks_per_node != 0 ))
570	then
571	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
572	fi
573
574
575	# CALCULATE NUMBER OF PROCESSES PER NODE
576	(( processes_per_node = tasks_per_node * threads_per_task ))
577
578
579	# CALCULATE NUMBER OF MPI TASKS
580	(( mpi_tasks = numprocs / threads_per_task ))
581
582
583	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
584	if [[ "$scp_port" != "" ]]
585	then
586	PORTOPT="-P $scp_port"
587	SSH_PORTOPT="-p $scp_port"
588	fi
589
590
591	# HEADER-OUTPUT
592	if [[ $verify = true ]]
593	then
594	printf "\n\n"
595	printf "#--------------------------------------------------------------# \n"
596	spalte1=SUBJOB;spalte2=$(date)
597	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
598	printf "\| \| \n"
599	printf "\| values of parameters/options: \| \n"
600	spalte1=$(echo local_host$punkte \| cut -c-20)
601	spalte2=$punkte$local_host
602	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
603	spalte1=$(echo remote_host$punkte \| cut -c-20)
604	spalte2=$punkte$remote_host
605	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
606	spalte1=$(echo queue$punkte \| cut -c-20)
607	spalte2=$punkte$queue
608	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
609	spalte1=$(echo memory$punkte \| cut -c-20)
610	spalte2="$punkte$memory mb"
611	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
612	spalte1=$(echo cputime$punkte \| cut -c-20)
613	spalte2="$punkte$cputime sec"
614	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
615	spalte1=$(echo job_name$punkte \| cut -c-20)
616	spalte2="$punkte$job_name"
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	printf "#--------------------------------------------------------------# \n\n"
619
620
621	# QUERY CHECK
622	antwort="dummy"
623	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
624	do
625	read antwort?" >>> continue (y/n) ? "
626	done
627	if [[ $antwort = n \|\| $antwort = N ]]
628	then
629	locat=verify; exit
630	fi
631	printf "\n"
632	fi
633
634	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
635	identifier=$RANDOM
636	job_on_remhost=${job_name}_${identifier}_$local_host
637	job_to_send=job_to_send_$identifier
638	if [[ $delete_dayfile = false ]]
639	then
640	remote_dayfile=${local_host}_${job_name}_result_$identifier
641	local_dayfile=${remote_host}_${job_name}
642	else
643	remote_dayfile=/dev/null
644	fi
645
646
647	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
648	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
649	then
650
651	# GENERAL LOADLEVELER SETTINGS
652	execute_in_shell="#!/bin/ksh"
653	use_shell="# @ shell = /bin/ksh"
654	consumable_memory="ConsumableMemory($memory mb)"
655	class="# @ class = $queue"
656	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
657	network_to_use="# @ network.mpi = sn_all,shared,us"
658	data_limit="# @ data_limit = 1.76gb"
659	image_size="# @ image_size = 50"
660	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
661
662	if [[ $email_notification = none ]]
663	then
664	notify_user=""
665	else
666	notify_user="# @ notify_user = $email_notification"
667	if [[ $delete_dayfile = true ]]
668	then
669	notification='# @ notification = never'
670	fi
671	fi
672
673	if [[ $remote_host = ibmh ]]
674	then
675	data_limit=""
676	network_to_use=""
677	class="# @ class = $queue"
678	environment=""
679	rset="# @ rset = RSET_MCM_AFFINITY"
680	task_affinity="# @ task_affinity = core(1)"
681	elif [[ $remote_host = ibmkisti ]]
682	then
683	network_to_use="# @ network.MPI = sn_all,shared,US"
684	wall_clock_limit="# @ wall_clock_limit = $timestring"
685	if [[ $threads_per_task = 1 ]]
686	then
687	rset="# @ rset = RSET_MCM_AFFINITY"
688	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
689	fi
690	environment=""
691	use_shell=""
692	data_limit=""
693	image_size=""
694	elif [[ $remote_host = ibmku ]]
695	then
696	execute_in_shell="#!/usr/bin/ksh"
697	use_shell="# @ shell = /usr/bin/ksh"
698	consumable_memory=""
699	environment=""
700	network_to_use="# @ network.mpi = sn_all,shared,us"
701	data_limit=""
702	image_size=""
703	elif [[ $remote_host = ibms ]]
704	then
705	network_to_use="# @ network.mpi = csss,shared,us"
706	fi
707
708	cat > $job_to_send << %%END%%
709	$execute_in_shell
710	$use_shell
711
712	# @ job_type = parallel
713	# @ job_name = $job_name
714	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
715	# @ output = $remote_dayfile
716	# @ error = $remote_dayfile
717	$wall_clock_limit
718	$image_size
719	$class
720	$environment
721	$network_to_use
722	$data_limit
723	$rset
724	$mcm_affinity_options
725	$task_affinity
726	$notification
727	$notify_user
728
729	%%END%%
730
731	if (( nodes > 0 ))
732	then
733
734	if [[ $remote_host != ibmkisti ]]
735	then
736
737	cat >> $job_to_send << %%END%%
738	# @ node = $nodes
739	# @ tasks_per_node = $processes_per_node
740	# @ node_usage = $node_usage
741	# @ queue
742
743	%%END%%
744
745	else
746
747	cat >> $job_to_send << %%END%%
748	# @ total_tasks = $mpi_tasks
749	# @ blocking = unlimited
750	# @ queue
751
752	%%END%%
753
754	fi
755
756	else
757
758	cat >> $job_to_send << %%END%%
759	# @ blocking = unlimited
760	# @ total_tasks = $numprocs
761	# @ node_usage = $node_usage
762	# @ queue
763
764	%%END%%
765
766	fi
767
768	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
769	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
770	then
771	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
772	fi
773
774	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
775	then
776
777	cat > $job_to_send << %%END%%
778	#!/bin/ksh
779
780	# @ job_type = serial
781	# @ node_usage = $node_usage
782	# @ job_name = palm
783	# @ wall_clock_limit = ${timestring},$timestring
784	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
785	# @ output = $remote_dayfile
786	# @ error = $remote_dayfile
787	$class
788	$notification
789
790	# @ queue
791
792	%%END%%
793
794	elif [[ $remote_host = lcbullhh ]]
795	then
796	if [[ $numprocs != 0 ]]
797	then
798	cat > $job_to_send << %%END%%
799	#!/bin/bash -l
800	#SBATCH -J $job_name
801	#SBATCH -t $timestring
802	#SBATCH -N $nodes
803	#SBATCH --ntasks-per-node=$processes_per_node
804	#SBATCH -p $queue
805	#SBATCH -o $remote_dayfile
806	#SBATCH -e $remote_dayfile
807	#SBATCH -A $project_account
808
809	$init_cmds
810	$module_calls
811
812	%%END%%
813
814	else
815	cat > $job_to_send << %%END%%
816	#!/bin/bash -l
817	#SBATCH -J $job_name
818	#SBATCH -t $timestring
819	#SBATCH -l ncpus=1
820	#SBATCH -l pmem=${memory}mb
821	#SBATCH -m abe
822	#SBATCH -o $remote_dayfile
823	#SBATCH -e $remote_dayfile
824	#SBATCH -A $project_account
825
826	$init_cmds
827	$module_calls
828
829	%%END%%
830
831	fi
832
833	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
834	then
835
836	if [[ "$feature" != "" ]]
837	then
838	featuredir="#PBS -l feature=$feature"
839	fi
840
841	if [[ $numprocs != 0 ]]
842	then
843	cat > $job_to_send << %%END%%
844	#!/bin/bash
845	#PBS -N $job_name
846	#PBS -l walltime=$timestring
847	#PBS -l nodes=$nodes:ppn=$processes_per_node
848	#PBS -o $remote_dayfile
849	#PBS -j oe
850	#PBS -q $queue
851	$featuredir
852
853	$init_cmds
854	$module_calls
855
856	%%END%%
857
858	else
859
860	continue
861
862	fi
863
864	elif [[ $remote_host = lccrayf ]]
865	then
866
867	if [[ $numprocs != 0 ]]
868	then
869	cat > $job_to_send << %%END%%
870	#!/bin/bash -l
871	#SBATCH -J $job_name
872	#SBATCH -t $timestring
873	#SBATCH -N $nodes
874	#SBATCH --ntasks-per-node=$processes_per_node
875	#SBATCH -p $queue
876	#SBATCH -o $remote_dayfile
877	#SBATCH -e $remote_dayfile
878
879	$init_cmds
880	$module_calls
881
882	%%END%%
883
884	else
885	cat > $job_to_send << %%END%%
886	#!/bin/bash -l
887	#SBATCH -J $job_name
888	#SBATCH -t $timestring
889	#SBATCH -l ncpus=1
890	#SBATCH -l pmem=${memory}mb
891	#SBATCH -m abe
892	#SBATCH -o $remote_dayfile
893	#SBATCH -e $remote_dayfile
894
895	$init_cmds
896	$module_calls
897
898	%%END%%
899
900	fi
901
902	elif [[ $remote_host = lcflow ]]
903	then
904	if [[ $numprocs != 0 ]]
905	then
906	pe_set="#$ -pe impi $numprocs"
907	else
908	pe_set="#$ -pe impi 1"
909	fi
910	if [[ $queue = default ]]
911	then
912	queue_set=""
913	else
914	queue_set="#$ -q $queue"
915	fi
916	[[ "$disc_space" = "" ]] && disc_space=50
917
918	cat > $job_to_send << %%END%%
919	#!/bin/bash
920	#$ -S /bin/bash
921	#$ -N $job_name
922	#$ -cwd
923	#$ -l h_rt=$timestring
924	#$ -l h_vmem=${memory}M
925	#$ -o $remote_dayfile
926	#$ -j y
927	$pe_set
928	#$ -R y
929	#$ -l h_fsize=${disc_space}G
930	$queue_set
931
932	%%END%%
933
934	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
935	then
936
937	if [[ $numprocs != 0 ]]
938	then
939	cat > $job_to_send << %%END%%
940	#!/bin/ksh
941	#PBS -N $job_name
942	#PBS -l walltime=$timestring
943	#PBS -l ncpus=$numprocs
944	#PBS -l pmem=${memory}mb
945	#PBS -o $remote_dayfile
946	#PBS -l nodes=$nodes:ppn=${processes_per_node}
947	#PBS -j oe
948
949	mpd &
950
951	%%END%%
952
953	else
954	cat > $job_to_send << %%END%%
955	#!/bin/ksh
956	#PBS -N $job_name
957	#PBS -l walltime=$timestring
958	#PBS -l ncpus=1
959	#PBS -l pmem=${memory}mb
960	#PBS -o $remote_dayfile
961	#PBS -j oe
962
963	%%END%%
964
965	fi
966
967	elif [[ $remote_host = lckiaps ]]
968	then
969
970	if [[ $numprocs != 0 ]]
971	then
972	cat > $job_to_send << %%END%%
973	#!/bin/ksh
974	#PBS -N $job_name
975	#PBS -l walltime=$timestring
976	#PBS -l select=1:ncpus=$numprocs
977	#PBS -l pmem=${memory}mb
978	#PBS -q $queue
979	#PBS -o $remote_dayfile
980	#PBS -j oe
981	#PBS -V
982
983	%%END%%
984
985	else
986	cat > $job_to_send << %%END%%
987	#!/bin/ksh
988	#PBS -N $job_name
989	#PBS -l walltime=$timestring
990	#PBS -l ncpus=1
991	#PBS -l pmem=${memory}mb
992	#PBS -o $remote_dayfile
993	#PBS -j oe
994
995	%%END%%
996
997	fi
998
999	elif [[ $remote_host = lcyon ]]
1000	then
1001
1002	if [[ $numprocs != 0 ]]
1003	then
1004	cat > $job_to_send << %%END%%
1005	#!/bin/ksh
1006	#PBS -N $job_name
1007	#PBS -l walltime=$timestring
1008	#PBS -l ncpus=$numprocs
1009	#PBS -l pmem=${memory}mb
1010	#PBS -o $remote_dayfile
1011	#PBS -j oe
1012
1013	%%END%%
1014
1015	else
1016	cat > $job_to_send << %%END%%
1017	#!/bin/ksh
1018	#PBS -N $job_name
1019	#PBS -l walltime=$timestring
1020	#PBS -l ncpus=1
1021	#PBS -l pmem=${memory}mb
1022	#PBS -o $remote_dayfile
1023	#PBS -j oe
1024
1025	%%END%%
1026
1027	fi
1028
1029	elif [[ $remote_host = lcxe6 ]]
1030	then
1031
1032	if [[ $numprocs != 0 ]]
1033	then
1034	cat > $job_to_send << %%END%%
1035	#!/bin/ksh
1036	#PBS -S /bin/ksh
1037	#PBS -N $job_name
1038	#PBS -A $project_account
1039	#PBS -j oe
1040	#PBS -l walltime=$timestring
1041	#PBS -l mppwidth=${numprocs}
1042	#PBS -l mppnppn=${processes_per_node}
1043	#PBS -m abe
1044	#PBS -o $remote_dayfile
1045	$email_directive
1046
1047	$init_cmds
1048	$module_calls
1049
1050	%%END%%
1051
1052	else
1053	cat > $job_to_send << %%END%%
1054	#!/bin/ksh
1055	#PBS -S /bin/ksh
1056	#PBS -N $job_name
1057	#PBS -A $project_account
1058	#PBS -j oe
1059	#PBS -l walltime=$timestring
1060	#PBS -l ncpus=1
1061	#PBS -l pmem=${memory}mb
1062	#PBS -m abe
1063	$email_directive
1064	#PBS -o $remote_dayfile
1065
1066	$init_cmds
1067	$module_calls
1068
1069	%%END%%
1070
1071	fi
1072
1073	elif [[ $remote_host = lckyoto ]]
1074	then
1075
1076	cat > $job_to_send << %%END%%
1077	#!/bin/ksh
1078	# @\$-o $remote_dayfile
1079	# @\$-eo -oi
1080	# @\$-lP 16
1081	# @\$-lp 1
1082	# @\$-lm 28gb -llm unlimited -ls unlimited
1083	# @\$-q $queue
1084	# @\$-Pvn abs_pack
1085	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1086	#. /thin/local/etc/setprofile/intel-11.0.sh
1087	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1088	. ~/.myprofile
1089	#. /home2/t/t51254/palm/current_version/myprofile
1090	#. /thin/apps/pgi/mpi.sh
1091	#
1092	env
1093	#
1094	set -x
1095
1096	%%END%%
1097
1098	elif [[ $remote_host = lcxt5m ]]
1099	then
1100
1101	if [[ $numprocs != 0 ]]
1102	then
1103	cat > $job_to_send << %%END%%
1104	#!/bin/ksh
1105	#PBS -S /bin/ksh
1106	#PBS -N $job_name
1107	#PBS -j oe
1108	#PBS -l walltime=$timestring
1109	#PBS -l mppwidth=${numprocs}
1110	#PBS -l mppnppn=${processes_per_node}
1111	#PBS -m abe
1112	#PBS -o $remote_dayfile
1113
1114	$init_cmds
1115	$module_calls
1116
1117	%%END%%
1118
1119	else
1120	cat > $job_to_send << %%END%%
1121	#!/bin/ksh
1122	#PBS -S /bin/ksh
1123	#PBS -N $job_name
1124	#PBS -j oe
1125	#PBS -l walltime=$timestring
1126	#PBS -l ncpus=1
1127	#PBS -l pmem=${memory}mb
1128	#PBS -m abe
1129	#PBS -o $remote_dayfile
1130
1131	$init_cmds
1132	$module_calls
1133
1134	%%END%%
1135
1136	fi
1137
1138	elif [[ $remote_host = lckyuh ]]
1139	then
1140	cat > $job_to_send << %%END%%
1141	#!/bin/bash
1142	#PJM -L "rscgrp=$queue"
1143	#PJM -L "node=$nodes"
1144	#PJM --mpi "proc=$numprocs"
1145	#PJM -L "elapse=$timestring"
1146	#PJM -o $remote_dayfile
1147	#PJM -j
1148	#PJM -X
1149	#PJM --no-stging
1150
1151	export LANG=en_US.UTF-8
1152	%%END%%
1153
1154	elif [[ $remote_host = lckyut ]]
1155	then
1156	cat > $job_to_send << %%END%%
1157	#!/bin/bash
1158	#PJM -L "rscgrp=$queue"
1159	#PJM -L "vnode=$numprocs"
1160	#PJM -L "vnode-core=1"
1161	#PJM -L "elapse=$timestring"
1162	#PJM --mpi proc=$numprocs
1163	#PJM -o $remote_dayfile
1164	#PJM -j
1165	#PJM -X
1166	#PJM --no-stging
1167
1168	export LANG=en_US.UTF-8
1169	%%END%%
1170
1171	elif [[ $remote_host = lcocean ]]
1172	then
1173	cat > $job_to_send << %%END%%
1174	#!/bin/bash
1175	#$ -cwd
1176	#$ -V
1177	#$ -N $job_name
1178	#$ -pe orte $numprocs
1179	#$ -o $remote_dayfile
1180	#$ -j y
1181	#$ -R y
1182	$init_cmds
1183	$module_calls
1184
1185	%%END%%
1186
1187	elif [[ $remote_host = nech ]]
1188	then
1189
1190	if (( nodes > 1 ))
1191	then
1192	cat > $job_to_send << %%END%%
1193	#!/bin/ksh
1194	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1195	#PBS -l ${qsubmem}=${Memory}gb
1196	#PBS -b $nodes
1197	#PBS -o $remote_dayfile
1198	#PBS -N palm
1199	#PBS -j o
1200	#PBS -T mpisx
1201
1202	%%END%%
1203
1204	elif [[ $numprocs != 0 ]]
1205	then
1206	cat > $job_to_send << %%END%%
1207	#!/bin/ksh
1208	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1209	#PBS -l ${qsubmem}=${Memory}gb
1210	#PBS -o $remote_dayfile
1211	#PBS -N palm
1212	#PBS -j o
1213
1214	%%END%%
1215
1216	else
1217	cat > $job_to_send << %%END%%
1218	#!/bin/ksh
1219	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1220	#PBS -o $remote_dayfile
1221	#PBS -j o
1222
1223	%%END%%
1224
1225	fi
1226
1227	elif [[ $remote_host = lctit ]]
1228	then
1229	cat > $job_to_send << %%END%%
1230	#!/bin/ksh
1231	$init_cmds
1232	$module_calls
1233
1234	%%END%%
1235
1236	# SET OPTIONS FOR SUBMIT-COMMAND
1237	if [[ $tasks_per_node != $processes_per_node ]]
1238	then
1239	submcom="$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 $queue "
1240	else
1241	submcom="$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 "
1242	fi
1243
1244	else
1245
1246	cat > $job_to_send << %%END%%
1247	# @\$-q ${queue}
1248	# @\$-l${qsubtime} $timestring
1249	# @\$-l${qsubmem} ${memory}mb
1250	# @\$-o $remote_dayfile
1251	# @\$-eo
1252
1253	%%END%%
1254
1255	fi
1256
1257
1258	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1259	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1260	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1261	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1262	then
1263	echo "set +vx" >> $job_to_send
1264	echo "trap '" >> $job_to_send
1265	echo "set +vx" >> $job_to_send
1266	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 \|\| $remote_host = lcocean ]]
1267	then
1268	if [[ $remote_host = ibmh ]]
1269	then
1270	return_queue=c1
1271	elif [[ $remote_host = ibmkisti ]]
1272	then
1273	return_queue=class.1-2
1274	elif [[ $remote_host = ibmku ]]
1275	then
1276	return_queue=sdbg2
1277	elif [[ $remote_host = ibms ]]
1278	then
1279	return_queue=p_normal
1280	elif [[ $remote_host = lcbullhh ]]
1281	then
1282	return_queue=shared
1283	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1284	then
1285	return_queue=dataq
1286	elif [[ $remote_host = lcxe6 ]]
1287	then
1288	return_queue=debug
1289	elif [[ $remote_host = lckiaps ]]
1290	then
1291	return_queue=express
1292	elif [[ $remote_host = lckyuh ]]
1293	then
1294	return_queue=cx-single
1295	elif [[ $remote_host = lckyut ]]
1296	then
1297	return_queue=cx-single
1298	else
1299	return_queue=unknown
1300	fi
1301
1302	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1303	then
1304
1305	if [[ $remote_host = ibmku ]]
1306	then
1307	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1308	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1309	else
1310	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1311	fi
1312	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1313	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1314	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1315	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1316	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1317	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1318	if [[ $host != "ibmh" ]]
1319	then
1320	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1321	fi
1322	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1323	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1324
1325	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1326	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1327
1328	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1329	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
1330	if [[ $remote_host = ibmku ]]
1331	then
1332	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1333	fi
1334	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1335
1336	elif [[ $remote_host = nech ]]
1337	then
1338	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1339	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1340	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1341	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1342	echo "#PBS -j o" >> $job_to_send
1343	echo " " >> $job_to_send
1344	echo "set -x" >> $job_to_send
1345	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1346	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1347	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1348	echo "%%END%%" >> $job_to_send
1349
1350	elif [[ $remote_host = lcbullhh ]]
1351	then
1352	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1353	echo "#!/bin/bash" >> $job_to_send
1354	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1355	echo "#SBATCH -t 00:20:00" >> $job_to_send
1356	echo "#SBATCH -N 1" >> $job_to_send
1357	echo "#SBATCH -n 1" >> $job_to_send
1358	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1359	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1360	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1361	echo "#SBATCH -A $project_account" >> $job_to_send
1362	echo "#SBATCH -p $return_queue" >> $job_to_send
1363	echo " " >> $job_to_send
1364	echo "set -x" >> $job_to_send
1365	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1366	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1367	echo "%%END%%" >> $job_to_send
1368
1369	elif [[ $remote_host = lckyuh ]]
1370	then
1371	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1372	echo "#!/bin/bash" >> $job_to_send
1373	echo "#PJM -L \"node=1\"" >> $job_to_send
1374	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1375	echo "#PJM --no-stging" >> $job_to_send
1376	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1377	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1378	echo "#PJM -j" >> $job_to_send
1379	echo " " >> $job_to_send
1380	echo "export LANG=en_US.UTF-8" >> $job_to_send
1381	echo "set -x" >> $job_to_send
1382	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1383	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1384	echo "%%END%%" >> $job_to_send
1385
1386	elif [[ $remote_host = lckyut ]]
1387	then
1388	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1389	echo "#!/bin/bash" >> $job_to_send
1390	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1391	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1392	echo "#PJM --no-stging" >> $job_to_send
1393	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1394	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1395	echo "#PJM -j" >> $job_to_send
1396	echo " " >> $job_to_send
1397	echo "export LANG=en_US.UTF-8" >> $job_to_send
1398	echo "set -x" >> $job_to_send
1399	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1400	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1401	echo "%%END%%" >> $job_to_send
1402
1403	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1404	then
1405	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1406	echo "#!/bin/bash" >> $job_to_send
1407	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1408	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1409	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1410	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1411	echo "#PBS -j oe" >> $job_to_send
1412	echo " " >> $job_to_send
1413	echo "set -x" >> $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 "%%END%%" >> $job_to_send
1417
1418	elif [[ $remote_host = lcocean ]]
1419	then
1420	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1421	echo "#!/bin/bash" >> $job_to_send
1422	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1423	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1424	echo "SGEPREFIX -cwd" >> $job_to_send
1425	echo "SGEPREFIX -j y" >> $job_to_send
1426	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1427	echo " " >> $job_to_send
1428	echo "set -x" >> $job_to_send
1429	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1430	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1431	echo "" >> $job_to_send
1432	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1433	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1434	echo "rm -f scpjob.${identifier}" >> $job_to_send
1435	echo "%%END%%" >> $job_to_send
1436	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1437	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1438
1439	elif [[ $remote_host = lcflow ]]
1440	then
1441	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1442	echo "#!/bin/bash" >> $job_to_send
1443	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1444	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1445	echo "SGEPREFIX -cwd" >> $job_to_send
1446	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1447	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1448	echo "SGEPREFIX -j y" >> $job_to_send
1449	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1450	echo " " >> $job_to_send
1451	echo "set -x" >> $job_to_send
1452	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1453	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1454	echo "" >> $job_to_send
1455	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1456	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1457	echo "rm -f scpjob.${identifier}" >> $job_to_send
1458	echo "%%END%%" >> $job_to_send
1459	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1460	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1461	elif [[ $remote_host = lcxe6 ]]
1462	then
1463	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1464	echo "#!/bin/ksh" >> $job_to_send
1465	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1466	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1467	echo "#PBS -A $project_account" >> $job_to_send
1468	echo "#PBS -l mppwidth=1" >> $job_to_send
1469	echo "#PBS -l mppnppn=1" >> $job_to_send
1470	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1471	echo "#PBS -j oe" >> $job_to_send
1472	echo " " >> $job_to_send
1473	echo "set -x" >> $job_to_send
1474	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1475	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1476	echo "%%END%%" >> $job_to_send
1477	else
1478
1479	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1480	echo "# @\\\$-q $return_queue" >> $job_to_send
1481	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1482	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1483	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1484	then
1485	echo "# @\$-l mpp_p=0" >> $job_to_send
1486	fi
1487	echo '# @\$-lF 10mb' >> $job_to_send
1488	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1489	echo '# @\\\$-eo' >> $job_to_send
1490	echo " " >> $job_to_send
1491	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1492	then
1493	echo "set +vx" >> $job_to_send
1494	echo ". .profile" >> $job_to_send
1495	fi
1496	echo "set -x" >> $job_to_send
1497	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
1498	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1499	echo "%%END%%" >> $job_to_send
1500
1501	fi
1502
1503	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1504	then
1505	echo "llsubmit scpjob.$identifier" >> $job_to_send
1506	elif [[ $remote_host = lcbullhh ]]
1507	then
1508	echo "sbatch scpjob.$identifier" >> $job_to_send
1509	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1510	then
1511	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1512	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1513	then
1514	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1515	elif [[ $remote_host = t3es ]]
1516	then
1517	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1518	elif [[ $remote_host = lckiaps ]]
1519	then
1520	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1521	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1522	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1523	elif [[ $remote_host = lckyu* ]]
1524	then
1525	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1526	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1527	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1528	then
1529	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1530	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1531	else
1532	echo "$submcom scpjob.$identifier" >> $job_to_send
1533	fi
1534	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1535	then
1536	echo "rm scpjob.$identifier" >> $job_to_send
1537	fi
1538	if [[ $remote_host = nech ]]
1539	then
1540	echo "cd -" >> $job_to_send
1541	fi
1542	else
1543	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1544	# ??? funktioniert das ÃŒberhaupt noch ???
1545	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1546	fi
1547	echo "set -x" >> $job_to_send
1548	echo " ' exit" >> $job_to_send
1549	echo "set -x" >> $job_to_send
1550	fi
1551
1552
1553	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1554	cat $file_to_send >> $job_to_send
1555
1556	if [[ $remote_host = ibm ]]
1557	then
1558	echo " " >> $job_to_send
1559	echo "exit" >> $job_to_send
1560	fi
1561
1562	# REMOVE JOB-FILE
1563	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1564	then
1565	echo " " >> $job_to_send
1566	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1567	fi
1568
1569
1570	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1571	if [[ $no_submit = false ]]
1572	then
1573	if [[ $remote_host != $local_host ]]
1574	then
1575	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1576	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1577	then
1578	job_catalog_save=$job_catalog
1579	job_catalog=job_queue
1580	elif [[ $remote_host = nech ]]
1581	then
1582	job_catalog_save=$job_catalog
1583	job_catalog=/hpf/b/${remote_user}/job_queue
1584	fi
1585	if [[ $remote_host = nech ]]
1586	then
1587	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1588	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1589	else
1590	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1591	fi
1592	if [[ $? = 1 ]]
1593	then
1594	locat=scp; exit
1595	fi
1596	if [[ $remote_host = ibms ]]
1597	then
1598	job_catalog=$job_catalog_save
1599	fi
1600	[[ $verify = true ]] && printf "\n >>> finished\n"
1601	else
1602	eval job_catalog=$job_catalog
1603	cp $job_to_send ${job_catalog}/$job_on_remhost
1604	fi
1605
1606
1607
1608	# START NQS- / LOADLEVELER-JOB
1609	if [[ $remote_host != $local_host ]]
1610	then
1611	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1612
1613	if [[ $remote_host = ibmku ]]
1614	then
1615	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1616	elif [[ $remote_host = lcflow ]]
1617	then
1618	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1619	else
1620	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1621	fi
1622
1623	[[ $verify = true ]] && printf " >>> o.k.\n"
1624	else
1625	cd $job_catalog
1626	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1627	then
1628	eval $submcom $job_on_remhost
1629	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1630	then
1631	chmod u+x $job_on_remhost
1632	eval $submcom $job_on_remhost
1633	elif [[ $local_host = nech ]]
1634	then
1635	if [[ $queue = default ]]
1636	then
1637	eval $submcom $job_on_remhost
1638	else
1639	eval $submcom -q $queue $job_on_remhost
1640	fi
1641	elif [[ $local_host = lcbullhh ]]
1642	then
1643	if [[ $queue = default ]]
1644	then
1645	eval $submcom $job_on_remhost
1646	fi
1647	else
1648	qsub $job_on_remhost
1649	fi
1650
1651	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1652	# AT THE END OF THE JOB
1653	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1654	then
1655	rm $job_on_remhost
1656	fi
1657	cd - > /dev/null
1658	fi
1659	fi
1660
1661	# FINAL ACTIONS
1662	if [[ $no_submit = false ]]
1663	then
1664	rm -f $job_to_send
1665	fi
1666	[[ $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 |