Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2185

Last change on this file since 2185 was 2184, checked in by raasch, 8 years ago
script bugfix and adjustment for lckyu*-hosts
Property svn:keywords set to `Id Rev`
File size: 58.4 KB

Line
1	#!/bin/bash
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2014 Leibniz Universitaet Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	# bugfix: localhost renamed local_host
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 2184 2017-03-21 04:31:22Z raasch $
30	#
31	# 2148 2017-02-09 16:56:42Z scharf
32	# added kuma and gharbi to the list of known hosts
33	#
34	# 2134 2017-02-02 07:33:46Z raasch
35	# option -E added to msub commands on HLRN-III machines to allow output of more
36	# job informations in the job protocol files
37	#
38	# 1944 2016-06-15 06:29:00Z raasch
39	# adjustments for using HLRN ssh-keys
40	#
41	# 1940 2016-06-14 05:15:20Z raasch
42	# adjustments for lckiaps
43	#
44	# 1866 2016-04-15 06:50:59Z raasch
45	# adjusted for lcocean
46	#
47	# 1841 2016-04-07 19:14:06Z raasch
48	# script now running under bash
49	#
50	# 1701 2015-11-02 07:43:04Z maronga
51	# Bugfix: added missing init_cmds for lccrayh/lccrayb
52	#
53	# 1621 2015-07-17 11:39:33Z heinze
54	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
55	#
56	# 1575 2015-03-27 09:56:27Z raasch
57	# mpp2-queues added to lccrayh
58	#
59	# 1547 2015-01-29 15:09:12Z witha
60	# adjustments for ForWind computing cluster (lcflow)
61	#
62	# 1545 2015-01-29 06:52:23Z heinze
63	# local host name for blizzard further specified
64	#
65	# 1480 2014-10-17 14:41:49Z raasch
66	# adjustments for 2nd stage of HLRNIII
67	#
68	# 1468 2014-09-24 14:06:57Z maronga
69	# Typo removed (addres->address)
70	# Adjustments for lcxe6
71	#
72	# 1452 2014-08-22 09:41:06Z heinze
73	# local hosts for blizzard added
74	#
75	# 1450 2014-08-21 07:31:51Z heinze
76	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
77	#
78	# 1442 2014-07-28 07:09:10Z raasch
79	# HLRN-III (lccrayb/lccrayh) queues adjusted
80	#
81	# 1378 2014-04-28 06:04:58Z raasch
82	# -et option added for lctit
83	#
84	# 1350 2014-04-04 13:01:30Z maronga
85	# location of qsub updated for lcxe6
86	#
87	# 1289 2014-03-04 07:12:34Z raasch
88	# German comments translated to English
89	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
90	#
91	# 1279 2014-01-28 12:10:14Z raasch
92	# node calculation modified due to changes in mrun (tasks_per_node must not be
93	# an integral divisor of numprocs any more)
94	#
95	# 1274 2014-01-09 13:14:54Z heinze
96	# adjustments for lccrayh
97	#
98	# 1266 2013-12-11 12:07:34Z heinze
99	# further adjustments for lccrayb (use msub instead of qsub)
100	#
101	# 1264 2013-12-09 12:46:09Z fricke
102	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
103	#
104	# 1262 2013-12-09 10:57:20Z fricke
105	# further adjustments for lccrayb
106	#
107	# 1260 2013-12-04 12:48:04Z raasch
108	# jaboticaba admitted
109	#
110	# 1255 2013-11-07 14:43:35Z raasch
111	# further adjustments for lccrayb
112	#
113	# 1224 2013-09-16 07:27:23Z raasch
114	# first adjustments for lccrayb
115	#
116	# 1202 2013-07-10 16:22:07Z witha
117	# adjustments for Forwind cluster (lcflow)
118	#
119	# 1199 2013-07-05 14:52:22Z raasch
120	# adjustments for CSC Helsinki (lccrayf)
121	#
122	# use of cluster/express queue enabled (ibmh)
123	# vinessa added (imuk)
124	#
125	# 1103 2013-02-20 02:15:53Z raasch
126	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
127	# instead typeset -L/R),
128	# further adjustments for lckyuh
129	#
130	# 2013-02-10 01:47:43Z raasch
131	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
132	# and for Forwind cluster (lcflow)
133	#
134	# 1094 2013-02-03 01:52:12Z raasch
135	# new option -P for explicit setting of ssh/scp port,
136	# decalpha parts (yonsei) removed
137	#
138	# 2013-02-02 07:06:13Z raasch
139	# adjustments for Kyushu-University computing center (lckyut - tatara)
140	# old changelog messages removed
141	#
142	# 1046 2012-11-09 14:38:45Z maronga
143	# code put under GPL (PALM 3.9)
144	#
145	# 08/07/94 - Siggi - first version finished
146	# 29/06/94 - Siggi - script development started
147	#--------------------------------------------------------------------------------#
148	# subjob - script for automatic generation and submission of batch-job files
149	# for various batch queuing systems
150	#--------------------------------------------------------------------------------#
151
152
153	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
154	delete_dayfile=false
155	email_notification=none
156	group_number=none
157	locat=normal
158	no_default_queue=none
159	no_submit=false
160	job_catalog="~/job_queue"
161	job_name=none
162	local_user=$LOGNAME
163	node_usage=shared
164	numprocs=0
165	punkte="..........................................................."
166	submcom=qsub
167	queue=default
168	remote_host=none
169	remote_user=""
170	verify=true
171
172	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
173	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
174
175
176
177	# ERROR HANDLING
178	# IN CASE OF EXIT:
179	trap 'if [[ $locat != normal ]]
180	then
181	case $locat in
182	(option) printf "\n --> available optios can be displayed"
183	printf " by typing:"
184	printf "\n \"subjob ?\" \n";;
185	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
186	(*) printf "\n +++ unknown error"
187	printf "\n please inform S. Raasch!\n"
188	esac
189	[[ -f $job_to_send ]] && rm $job_to_send
190	printf "\n\n+++ SUBJOB killed \n\n"
191	fi' exit
192
193
194	# IN CASE OF TERMINAL-BREAK:
195	trap '[[ -f $job_to_send ]] && rm $job_to_send
196	printf "\n\n+++ SUBJOB killed \n\n"
197	exit
198	' 2
199
200
201	# DETERMINE NAME OF LOCAL HOST
202	local_host=$(hostname)
203
204	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
205	# IS ADMITTED AT ALL)
206	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
207	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
208	case $local_host in
209	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
210	(atmos) local_address=172.20.25.35; local_host=lcide;;
211	(austru) local_address=130.75.105.128; local_host=lcmuk;;
212	(autan) local_address=130.75.105.57; local_host=lcmuk;;
213	(bora) local_address=130.75.105.103; local_host=lcmuk;;
214	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
215	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
216	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
217	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
218	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
219	(breva) local_address=130.75.105.98; local_host=lcmuk;;
220	(buran) local_address=130.75.105.58; local_host=lcmuk;;
221	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
222	(climate*) local_address=165.132.26.68; local_host=lcyon;;
223	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
224	(cs*) local_address=136.172.44.131; local_host=nech;;
225	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
226	(flow01) local_address=10.141.255.71; local_host=lcflow;;
227	(flow02) local_address=10.141.255.72; local_host=lcflow;;
228	(node*) local_address=165.132.26.61 local_host=lck;;
229	# (node*) local_address=210.219.61.8 local_host=lckordi;;
230	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
231	(gharbi) local_address=130.75.105.47; local_host=lcmuk;;
232	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
233	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
234	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
235	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
236	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
237	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
238	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
239	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
240	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
241	(sno) local_address=130.75.105.113; local_host=lcmuk;;
242	(kuma) local_address=130.75.105.115; local_host=lcmuk;;
243	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
244	(login*) local_address=118.128.66.201; local_host=lckiaps;;
245	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
246	(meller) local_address=134.106.74.155; local_host=lcfor;;
247	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
248	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
249	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
250	(nobel*) local_address=150.183.5.101; local_host=ibms;;
251	(ocean) local_address="ocean"; local_host=lcocean;;
252	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
253	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
254	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
255	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
256	(pingui) local_address=134.106.74.118; local_host=lcfor;;
257	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
258	(rte*) local_address=133.5.185.60; local_host=lcrte;;
259	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
260	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
261	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
262	(solano) local_address=130.75.105.110; local_host=lcmuk;;
263	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
264	(tc*) local_address="ocean"; local_host=lcocean;;
265	(t2a*) local_address=10.1.6.165; local_host=lctit;;
266	(urban*) local_address=147.46.30.151 local_host=lcsb;;
267	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
268	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
269	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
270	(*) printf "\n +++ \"$local_host\" unknown";
271	printf "\n please contact the PALM group at IMUK";
272	locat=parameter; exit;;
273	esac
274
275
276
277	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
278	remote_host=$local_host
279
280
281
282
283	# READ THE SHELLSCRIPT-OPTIONS
284	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
285	do
286	case $option in
287	(c) job_catalog=$OPTARG;;
288	(d) delete_dayfile=true;;
289	(D) no_submit=true;;
290	(e) email_notification=$OPTARG;;
291	(g) group_number=$OPTARG;;
292	(h) remote_host=$OPTARG;;
293	(m) memory=$OPTARG;;
294	(n) job_name=$OPTARG;;
295	(N) node_usage=$OPTARG;;
296	(O) threads_per_task=$OPTARG;;
297	(P) scp_port=$OPTARG;;
298	(q) no_default_queue=$OPTARG;;
299	(t) cputime=$OPTARG;;
300	(T) tasks_per_node=$OPTARG;;
301	(u) remote_user=$OPTARG;;
302	(v) verify=false;;
303	(X) numprocs=$OPTARG;;
304	(\?) printf "\n +++ Option $OPTARG unknown \n";
305	locat=option; exit;;
306	esac
307	done
308
309
310	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
311	(( to_shift = $OPTIND - 1 ))
312	shift $to_shift; file_to_send=$1
313
314
315	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
316	if [ "$1" = "?" ]
317	then
318	(printf "\n *** subjob can be called as follows:\n"
319	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
320	printf "\n Description of available options:\n"
321	printf "\n Option Description Default-Value"
322	printf "\n -c job-input- and output-catalog ~/job_queue"
323	printf "\n -d no job-protocol will be created ---"
324	printf "\n -D only the job-file will be created ---"
325	printf "\n -h execution host, available hosts: $remote_host"
326	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
327	printf "\n lckiaps, lctit, nech"
328	printf "\n -m memory demand per process in MByte ---"
329	printf "\n -n jobname <jobdatei>"
330	printf "\n -O threads per task (for OpenMP usage) 1"
331	printf "\n -P ssh/scp port default port"
332	printf "\n -q job-queue to be used default"
333	printf "\n -t allowed cpu-time in seconds ---"
334	printf "\n -T tasks per node (on parallel hosts) ---"
335	printf "\n -u username on execution host from .netrc"
336	printf "\n -v no prompt for confirmation ---"
337	printf "\n -X # of processors (on parallel hosts) 1"
338	printf "\n "
339	printf "\n The only possible positional parameter is <jobfile>:"
340	printf "\n The complete NQS-job must be provided here."
341	printf "\n <jobfile>=? creates this outline\n\n") \| more
342	exit
343	fi
344
345
346
347	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
348	if [[ "$file_to_send" = "" ]]
349	then
350	printf "\n +++ job-file missing"
351	locat=parameter; exit
352	else
353	if [[ -f $file_to_send ]]
354	then
355	true
356	else
357	printf "\n +++ job-file: "
358	printf "\n $file_to_send"
359	printf "\n does not exist"
360	locat=parameter; exit
361	fi
362	fi
363
364
365
366	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
367	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
368	if [[ $job_name = none ]]
369	then
370	job_name=$file_to_send
371	fi
372	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
373	then
374	printf "\n +++ job-file name: "
375	printf "\n $job_name"
376	printf "\n must not contain \"/\"-characters"
377	locat=parameter; exit
378	fi
379
380
381
382
383	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
384	# OR IF NO HOST HAS BEEN GIVEN
385	if [[ $remote_host = none ]]
386	then
387	printf "\n +++ host missing"
388	locat=option; exit
389	else
390	case $remote_host in
391	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
392	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
393	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
394	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
395	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
396	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
397	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom="/opt/moab/default/bin/msub -E";;
398	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom="/opt/moab/default/bin/msub -E";;
399	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
400	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
401	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
402	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
403	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
404	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
405	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
406	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
407	(lcocean) remote_address="ocean"; submcom=qsub;;
408	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
409	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
410	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
411	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
412	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
413	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
414	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
415	locat=parameter; exit;;
416	esac
417	fi
418
419
420	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
421	if [[ $no_default_queue != none ]]
422	then
423	error=false
424	ndq=$no_default_queue
425	case $remote_host in
426	(ibm) case $ndq in
427	(p690_express\|p690_standard\|p690_long) error=false;;
428	(*) error=true;;
429	esac;;
430	(ibmh) case $ndq in
431	(cluster\|express) error=false;;
432	(*) error=true;;
433	esac;;
434	(ibmkisti) case $ndq in
435	(class.32plus\|class.1-2\|class.2-32) error=false;;
436	(*) error=true;;
437	esac;;
438	(ibmku) case $ndq in
439	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
440	(*) error=true;;
441	esac;;
442	(ibms) case $ndq in
443	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
444	(*) error=true;;
445	esac;;
446	(lcbullhh) case $ndq in
447	(compute\|compute2\|shared) error=false;;
448	(*) error=true;;
449	esac;;
450	(lccrayb) case $ndq in
451	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
452	(*) error=true;;
453	esac;;
454	(lccrayh) case $ndq in
455	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
456	(*) error=true;;
457	esac;;
458	(lccrayf) case $ndq in
459	(usup\|test*\|small\|large) error=false;;
460	(*) error=true;;
461	esac;;
462	(lcflow) case $ndq in
463	(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;;
464	(*) error=true;;
465	esac;;
466	(lckiaps) case $ndq in
467	(express\|normal\|normal20\|quickq) error=false;;
468	(*) error=true;;
469	esac;;
470	(lckyoto) case $ndq in
471	(eh\|ph) error=false;;
472	(*) error=true;;
473	esac;;
474	(lckyuh) case $ndq in
475	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
476	(*) error=true;;
477	esac;;
478	(lckyut) case $ndq in
479	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
480	(*) error=true;;
481	esac;;
482	(lctit) case $ndq in
483	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
484	(*) error=true;;
485	esac;;
486	(t3eb) case $ndq in
487	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
488	(*) error=true;;
489	esac;;
490	(t3eh) case $ndq in
491	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
492	(*) error=true;;
493	esac;;
494	(t3ej2\|t3ej5) case $ndq in
495	(low\|normal\|high) error=false;;
496	(*) error=true;;
497	esac;;
498	(t3es) case $ndq in
499	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
500	(*) error=true;;
501	esac;;
502	esac
503	if [[ $error = true ]]
504	then
505	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
506	locat=parameter; exit
507	else
508	queue=$no_default_queue
509	fi
510	fi
511
512
513
514	# CHECK THE CPU-TIME
515	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
516	done=false
517	while [[ $done = false ]]
518	do
519	if (( $cputime <= 0 ))
520	then
521	printf "\n +++ wrong cpu-time or cpu-time missing"
522	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
523	printf "\n >>> "
524	read cputime 1>/dev/null 2>&1
525	else
526	done=true
527	fi
528	done
529	if [[ $remote_host = nech ]]
530	then
531	if (( tasks_per_node != 0 ))
532	then
533	(( cputime = cputime * tasks_per_node ))
534	elif [[ $numprocs != 0 ]]
535	then
536	(( cputime = cputime * numprocs ))
537	fi
538	fi
539	(( stunden = cputime / 3600 ))
540	(( resttime = cputime - stunden * 3600 ))
541	(( minuten = resttime / 60 ))
542	(( sekunden = resttime - minuten * 60 ))
543	timestring=${stunden}:${minuten}:${sekunden}
544
545
546
547	# CHECK THE MEMORY DEMAND
548	done=false
549	while [[ $done = false ]]
550	do
551	if (( memory <= 0 ))
552	then
553	printf "\n +++ wrong memory demand or memory demand missing"
554	printf "\n >>> Please type memory in MByte per process as INTEGER:"
555	printf "\n >>> "
556	read memory 1>/dev/null 2>&1
557	else
558	done=true
559	fi
560	done
561
562	if [[ $remote_host = nech ]]
563	then
564	if (( tasks_per_node != 0 ))
565	then
566	(( Memory = memory * tasks_per_node / 1000 ))
567	elif [[ $numprocs != 0 ]]
568	then
569	(( Memory = memory * numprocs / 1000 ))
570	else
571	(( Memory = memory / 1000 ))
572	fi
573	elif [[ $remote_host = lctit ]]
574	then
575	(( Memory = memory * tasks_per_node / 1000 ))
576	fi
577
578
579	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
580	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
581	then
582	(( memory = memory * threads_per_task ))
583	fi
584
585
586	# CALCULATE NUMBER OF REQUIRED NODES
587	if (( tasks_per_node != 0 ))
588	then
589	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
590	fi
591
592
593	# CALCULATE NUMBER OF PROCESSES PER NODE
594	(( processes_per_node = tasks_per_node * threads_per_task ))
595
596
597	# CALCULATE NUMBER OF MPI TASKS
598	(( mpi_tasks = numprocs / threads_per_task ))
599
600
601	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
602	if [[ "$scp_port" != "" ]]
603	then
604	PORTOPT="-P $scp_port"
605	SSH_PORTOPT="-p $scp_port"
606	fi
607
608
609	# HEADER-OUTPUT
610	if [[ $verify = true ]]
611	then
612	printf "\n\n"
613	printf "#--------------------------------------------------------------# \n"
614	spalte1=SUBJOB;spalte2=$(date)
615	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
616	printf "\| \| \n"
617	printf "\| values of parameters/options: \| \n"
618	spalte1=$(echo local_host$punkte \| cut -c-20)
619	spalte2=$punkte$local_host
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	spalte1=$(echo remote_host$punkte \| cut -c-20)
622	spalte2=$punkte$remote_host
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	spalte1=$(echo queue$punkte \| cut -c-20)
625	spalte2=$punkte$queue
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	spalte1=$(echo memory$punkte \| cut -c-20)
628	spalte2="$punkte$memory mb"
629	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
630	spalte1=$(echo cputime$punkte \| cut -c-20)
631	spalte2="$punkte$cputime sec"
632	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
633	spalte1=$(echo job_name$punkte \| cut -c-20)
634	spalte2="$punkte$job_name"
635	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
636	printf "#--------------------------------------------------------------# \n\n"
637
638
639	# QUERY CHECK
640	antwort="dummy"
641	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
642	do
643	read antwort?" >>> continue (y/n) ? "
644	done
645	if [[ $antwort = n \|\| $antwort = N ]]
646	then
647	locat=verify; exit
648	fi
649	printf "\n"
650	fi
651
652	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
653	identifier=$RANDOM
654	job_on_remhost=${job_name}_${identifier}_$local_host
655	job_to_send=job_to_send_$identifier
656	if [[ $delete_dayfile = false ]]
657	then
658	remote_dayfile=${local_host}_${job_name}_result_$identifier
659	local_dayfile=${remote_host}_${job_name}
660	else
661	remote_dayfile=/dev/null
662	fi
663
664
665	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
666	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
667	then
668
669	# GENERAL LOADLEVELER SETTINGS
670	execute_in_shell="#!/bin/ksh"
671	use_shell="# @ shell = /bin/ksh"
672	consumable_memory="ConsumableMemory($memory mb)"
673	class="# @ class = $queue"
674	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
675	network_to_use="# @ network.mpi = sn_all,shared,us"
676	data_limit="# @ data_limit = 1.76gb"
677	image_size="# @ image_size = 50"
678	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
679
680	if [[ $email_notification = none ]]
681	then
682	notify_user=""
683	else
684	notify_user="# @ notify_user = $email_notification"
685	if [[ $delete_dayfile = true ]]
686	then
687	notification='# @ notification = never'
688	fi
689	fi
690
691	if [[ $remote_host = ibmh ]]
692	then
693	data_limit=""
694	network_to_use=""
695	class="# @ class = $queue"
696	environment=""
697	rset="# @ rset = RSET_MCM_AFFINITY"
698	task_affinity="# @ task_affinity = core(1)"
699	elif [[ $remote_host = ibmkisti ]]
700	then
701	network_to_use="# @ network.MPI = sn_all,shared,US"
702	wall_clock_limit="# @ wall_clock_limit = $timestring"
703	if [[ $threads_per_task = 1 ]]
704	then
705	rset="# @ rset = RSET_MCM_AFFINITY"
706	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
707	fi
708	environment=""
709	use_shell=""
710	data_limit=""
711	image_size=""
712	elif [[ $remote_host = ibmku ]]
713	then
714	execute_in_shell="#!/usr/bin/ksh"
715	use_shell="# @ shell = /usr/bin/ksh"
716	consumable_memory=""
717	environment=""
718	network_to_use="# @ network.mpi = sn_all,shared,us"
719	data_limit=""
720	image_size=""
721	elif [[ $remote_host = ibms ]]
722	then
723	network_to_use="# @ network.mpi = csss,shared,us"
724	fi
725
726	cat > $job_to_send << %%END%%
727	$execute_in_shell
728	$use_shell
729
730	# @ job_type = parallel
731	# @ job_name = $job_name
732	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
733	# @ output = $remote_dayfile
734	# @ error = $remote_dayfile
735	$wall_clock_limit
736	$image_size
737	$class
738	$environment
739	$network_to_use
740	$data_limit
741	$rset
742	$mcm_affinity_options
743	$task_affinity
744	$notification
745	$notify_user
746
747	%%END%%
748
749	if (( nodes > 0 ))
750	then
751
752	if [[ $remote_host != ibmkisti ]]
753	then
754
755	cat >> $job_to_send << %%END%%
756	# @ node = $nodes
757	# @ tasks_per_node = $processes_per_node
758	# @ node_usage = $node_usage
759	# @ queue
760
761	%%END%%
762
763	else
764
765	cat >> $job_to_send << %%END%%
766	# @ total_tasks = $mpi_tasks
767	# @ blocking = unlimited
768	# @ queue
769
770	%%END%%
771
772	fi
773
774	else
775
776	cat >> $job_to_send << %%END%%
777	# @ blocking = unlimited
778	# @ total_tasks = $numprocs
779	# @ node_usage = $node_usage
780	# @ queue
781
782	%%END%%
783
784	fi
785
786	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
787	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
788	then
789	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
790	fi
791
792	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
793	then
794
795	cat > $job_to_send << %%END%%
796	#!/bin/ksh
797
798	# @ job_type = serial
799	# @ node_usage = $node_usage
800	# @ job_name = palm
801	# @ wall_clock_limit = ${timestring},$timestring
802	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
803	# @ output = $remote_dayfile
804	# @ error = $remote_dayfile
805	$class
806	$notification
807
808	# @ queue
809
810	%%END%%
811
812	elif [[ $remote_host = lcbullhh ]]
813	then
814	if [[ $numprocs != 0 ]]
815	then
816	cat > $job_to_send << %%END%%
817	#!/bin/bash -l
818	#SBATCH -J $job_name
819	#SBATCH -t $timestring
820	#SBATCH -N $nodes
821	#SBATCH --ntasks-per-node=$processes_per_node
822	#SBATCH -p $queue
823	#SBATCH -o $remote_dayfile
824	#SBATCH -e $remote_dayfile
825	#SBATCH -A $project_account
826
827	$init_cmds
828	$module_calls
829
830	%%END%%
831
832	else
833	cat > $job_to_send << %%END%%
834	#!/bin/bash -l
835	#SBATCH -J $job_name
836	#SBATCH -t $timestring
837	#SBATCH -l ncpus=1
838	#SBATCH -l pmem=${memory}mb
839	#SBATCH -m abe
840	#SBATCH -o $remote_dayfile
841	#SBATCH -e $remote_dayfile
842	#SBATCH -A $project_account
843
844	$init_cmds
845	$module_calls
846
847	%%END%%
848
849	fi
850
851	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
852	then
853
854	if [[ "$feature" != "" ]]
855	then
856	featuredir="#PBS -l feature=$feature"
857	fi
858
859	if [[ $numprocs != 0 ]]
860	then
861	cat > $job_to_send << %%END%%
862	#!/bin/bash
863	#PBS -N $job_name
864	#PBS -l walltime=$timestring
865	#PBS -l nodes=$nodes:ppn=$processes_per_node
866	#PBS -o $remote_dayfile
867	#PBS -j oe
868	#PBS -q $queue
869	$featuredir
870
871	$init_cmds
872	$module_calls
873
874	%%END%%
875
876	else
877
878	continue
879
880	fi
881
882	elif [[ $remote_host = lccrayf ]]
883	then
884
885	if [[ $numprocs != 0 ]]
886	then
887	cat > $job_to_send << %%END%%
888	#!/bin/bash -l
889	#SBATCH -J $job_name
890	#SBATCH -t $timestring
891	#SBATCH -N $nodes
892	#SBATCH --ntasks-per-node=$processes_per_node
893	#SBATCH -p $queue
894	#SBATCH -o $remote_dayfile
895	#SBATCH -e $remote_dayfile
896
897	$init_cmds
898	$module_calls
899
900	%%END%%
901
902	else
903	cat > $job_to_send << %%END%%
904	#!/bin/bash -l
905	#SBATCH -J $job_name
906	#SBATCH -t $timestring
907	#SBATCH -l ncpus=1
908	#SBATCH -l pmem=${memory}mb
909	#SBATCH -m abe
910	#SBATCH -o $remote_dayfile
911	#SBATCH -e $remote_dayfile
912
913	$init_cmds
914	$module_calls
915
916	%%END%%
917
918	fi
919
920	elif [[ $remote_host = lcflow ]]
921	then
922	if [[ $numprocs != 0 ]]
923	then
924	pe_set="#$ -pe impi $numprocs"
925	else
926	pe_set="#$ -pe impi 1"
927	fi
928	if [[ $queue = default ]]
929	then
930	queue_set=""
931	else
932	queue_set="#$ -q $queue"
933	fi
934	[[ "$disc_space" = "" ]] && disc_space=50
935
936	cat > $job_to_send << %%END%%
937	#!/bin/bash
938	#$ -S /bin/bash
939	#$ -N $job_name
940	#$ -cwd
941	#$ -l h_rt=$timestring
942	#$ -l h_vmem=${memory}M
943	#$ -o $remote_dayfile
944	#$ -j y
945	$pe_set
946	#$ -R y
947	#$ -l h_fsize=${disc_space}G
948	$queue_set
949
950	%%END%%
951
952	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
953	then
954
955	if [[ $numprocs != 0 ]]
956	then
957	cat > $job_to_send << %%END%%
958	#!/bin/ksh
959	#PBS -N $job_name
960	#PBS -l walltime=$timestring
961	#PBS -l ncpus=$numprocs
962	#PBS -l pmem=${memory}mb
963	#PBS -o $remote_dayfile
964	#PBS -l nodes=$nodes:ppn=${processes_per_node}
965	#PBS -j oe
966
967	mpd &
968
969	%%END%%
970
971	else
972	cat > $job_to_send << %%END%%
973	#!/bin/ksh
974	#PBS -N $job_name
975	#PBS -l walltime=$timestring
976	#PBS -l ncpus=1
977	#PBS -l pmem=${memory}mb
978	#PBS -o $remote_dayfile
979	#PBS -j oe
980
981	%%END%%
982
983	fi
984
985	elif [[ $remote_host = lckiaps ]]
986	then
987
988	if [[ $numprocs != 0 ]]
989	then
990	cat > $job_to_send << %%END%%
991	#!/bin/bash
992	#PBS -N $job_name
993	#PBS -l walltime=$timestring
994	#PBS -l select=1:ncpus=$numprocs
995	#PBS -l pmem=${memory}mb
996	#PBS -q $queue
997	#PBS -o $remote_dayfile
998	#PBS -j oe
999	#PBS -V
1000
1001	%%END%%
1002
1003	else
1004	cat > $job_to_send << %%END%%
1005	#!/bin/bash
1006	#PBS -N $job_name
1007	#PBS -l walltime=$timestring
1008	#PBS -l ncpus=1
1009	#PBS -l pmem=${memory}mb
1010	#PBS -o $remote_dayfile
1011	#PBS -j oe
1012
1013	%%END%%
1014
1015	fi
1016
1017	elif [[ $remote_host = lcyon ]]
1018	then
1019
1020	if [[ $numprocs != 0 ]]
1021	then
1022	cat > $job_to_send << %%END%%
1023	#!/bin/ksh
1024	#PBS -N $job_name
1025	#PBS -l walltime=$timestring
1026	#PBS -l ncpus=$numprocs
1027	#PBS -l pmem=${memory}mb
1028	#PBS -o $remote_dayfile
1029	#PBS -j oe
1030
1031	%%END%%
1032
1033	else
1034	cat > $job_to_send << %%END%%
1035	#!/bin/ksh
1036	#PBS -N $job_name
1037	#PBS -l walltime=$timestring
1038	#PBS -l ncpus=1
1039	#PBS -l pmem=${memory}mb
1040	#PBS -o $remote_dayfile
1041	#PBS -j oe
1042
1043	%%END%%
1044
1045	fi
1046
1047	elif [[ $remote_host = lcxe6 ]]
1048	then
1049
1050	if [[ $numprocs != 0 ]]
1051	then
1052	cat > $job_to_send << %%END%%
1053	#!/bin/ksh
1054	#PBS -S /bin/ksh
1055	#PBS -N $job_name
1056	#PBS -A $project_account
1057	#PBS -j oe
1058	#PBS -l walltime=$timestring
1059	#PBS -l mppwidth=${numprocs}
1060	#PBS -l mppnppn=${processes_per_node}
1061	#PBS -m abe
1062	#PBS -o $remote_dayfile
1063	$email_directive
1064
1065	$init_cmds
1066	$module_calls
1067
1068	%%END%%
1069
1070	else
1071	cat > $job_to_send << %%END%%
1072	#!/bin/ksh
1073	#PBS -S /bin/ksh
1074	#PBS -N $job_name
1075	#PBS -A $project_account
1076	#PBS -j oe
1077	#PBS -l walltime=$timestring
1078	#PBS -l ncpus=1
1079	#PBS -l pmem=${memory}mb
1080	#PBS -m abe
1081	$email_directive
1082	#PBS -o $remote_dayfile
1083
1084	$init_cmds
1085	$module_calls
1086
1087	%%END%%
1088
1089	fi
1090
1091	elif [[ $remote_host = lckyoto ]]
1092	then
1093
1094	cat > $job_to_send << %%END%%
1095	#!/bin/ksh
1096	# @\$-o $remote_dayfile
1097	# @\$-eo -oi
1098	# @\$-lP 16
1099	# @\$-lp 1
1100	# @\$-lm 28gb -llm unlimited -ls unlimited
1101	# @\$-q $queue
1102	# @\$-Pvn abs_pack
1103	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1104	#. /thin/local/etc/setprofile/intel-11.0.sh
1105	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1106	. ~/.myprofile
1107	#. /home2/t/t51254/palm/current_version/myprofile
1108	#. /thin/apps/pgi/mpi.sh
1109	#
1110	env
1111	#
1112	set -x
1113
1114	%%END%%
1115
1116	elif [[ $remote_host = lcxt5m ]]
1117	then
1118
1119	if [[ $numprocs != 0 ]]
1120	then
1121	cat > $job_to_send << %%END%%
1122	#!/bin/ksh
1123	#PBS -S /bin/ksh
1124	#PBS -N $job_name
1125	#PBS -j oe
1126	#PBS -l walltime=$timestring
1127	#PBS -l mppwidth=${numprocs}
1128	#PBS -l mppnppn=${processes_per_node}
1129	#PBS -m abe
1130	#PBS -o $remote_dayfile
1131
1132	$init_cmds
1133	$module_calls
1134
1135	%%END%%
1136
1137	else
1138	cat > $job_to_send << %%END%%
1139	#!/bin/ksh
1140	#PBS -S /bin/ksh
1141	#PBS -N $job_name
1142	#PBS -j oe
1143	#PBS -l walltime=$timestring
1144	#PBS -l ncpus=1
1145	#PBS -l pmem=${memory}mb
1146	#PBS -m abe
1147	#PBS -o $remote_dayfile
1148
1149	$init_cmds
1150	$module_calls
1151
1152	%%END%%
1153
1154	fi
1155
1156	elif [[ $remote_host = lckyuh ]]
1157	then
1158	cat > $job_to_send << %%END%%
1159	#!/bin/bash
1160	#PJM -L "rscgrp=$queue"
1161	#PJM -L "node=$nodes"
1162	#PJM --mpi "proc=$numprocs"
1163	#PJM -L "elapse=$timestring"
1164	#PJM -o $remote_dayfile
1165	#PJM -j
1166	#PJM -X
1167	#PJM --no-stging
1168
1169	export LANG=en_US.UTF-8
1170	%%END%%
1171
1172	elif [[ $remote_host = lckyut ]]
1173	then
1174	cat > $job_to_send << %%END%%
1175	#!/bin/bash
1176	#PJM -L "rscgrp=$queue"
1177	#PJM -L "vnode=$numprocs"
1178	#PJM -L "vnode-core=1"
1179	#PJM -L "elapse=$timestring"
1180	#PJM --mpi proc=$numprocs
1181	#PJM -o $remote_dayfile
1182	#PJM -j
1183	#PJM -X
1184	#PJM --no-stging
1185
1186	export LANG=en_US.UTF-8
1187	%%END%%
1188
1189	elif [[ $remote_host = lcocean ]]
1190	then
1191	cat > $job_to_send << %%END%%
1192	#!/bin/bash
1193	#$ -cwd
1194	#$ -V
1195	#$ -N $job_name
1196	#$ -pe orte $numprocs
1197	#$ -o $remote_dayfile
1198	#$ -j y
1199	#$ -R y
1200	$init_cmds
1201	$module_calls
1202
1203	%%END%%
1204
1205	elif [[ $remote_host = nech ]]
1206	then
1207
1208	if (( nodes > 1 ))
1209	then
1210	cat > $job_to_send << %%END%%
1211	#!/bin/ksh
1212	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1213	#PBS -l ${qsubmem}=${Memory}gb
1214	#PBS -b $nodes
1215	#PBS -o $remote_dayfile
1216	#PBS -N palm
1217	#PBS -j o
1218	#PBS -T mpisx
1219
1220	%%END%%
1221
1222	elif [[ $numprocs != 0 ]]
1223	then
1224	cat > $job_to_send << %%END%%
1225	#!/bin/ksh
1226	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1227	#PBS -l ${qsubmem}=${Memory}gb
1228	#PBS -o $remote_dayfile
1229	#PBS -N palm
1230	#PBS -j o
1231
1232	%%END%%
1233
1234	else
1235	cat > $job_to_send << %%END%%
1236	#!/bin/ksh
1237	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1238	#PBS -o $remote_dayfile
1239	#PBS -j o
1240
1241	%%END%%
1242
1243	fi
1244
1245	elif [[ $remote_host = lctit ]]
1246	then
1247	cat > $job_to_send << %%END%%
1248	#!/bin/ksh
1249	$init_cmds
1250	$module_calls
1251
1252	%%END%%
1253
1254	# SET OPTIONS FOR SUBMIT-COMMAND
1255	if [[ $tasks_per_node != $processes_per_node ]]
1256	then
1257	submcom="$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 $queue "
1258	else
1259	submcom="$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 "
1260	fi
1261
1262	else
1263
1264	cat > $job_to_send << %%END%%
1265	# @\$-q ${queue}
1266	# @\$-l${qsubtime} $timestring
1267	# @\$-l${qsubmem} ${memory}mb
1268	# @\$-o $remote_dayfile
1269	# @\$-eo
1270
1271	%%END%%
1272
1273	fi
1274
1275
1276	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1277	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1278	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1279	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1280	then
1281	echo "set +vx" >> $job_to_send
1282	echo "trap '" >> $job_to_send
1283	echo "set +vx" >> $job_to_send
1284	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 ]]
1285	then
1286	if [[ $remote_host = ibmh ]]
1287	then
1288	return_queue=c1
1289	elif [[ $remote_host = ibmkisti ]]
1290	then
1291	return_queue=class.1-2
1292	elif [[ $remote_host = ibmku ]]
1293	then
1294	return_queue=sdbg2
1295	elif [[ $remote_host = ibms ]]
1296	then
1297	return_queue=p_normal
1298	elif [[ $remote_host = lcbullhh ]]
1299	then
1300	return_queue=shared
1301	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1302	then
1303	return_queue=dataq
1304	elif [[ $remote_host = lcxe6 ]]
1305	then
1306	return_queue=debug
1307	elif [[ $remote_host = lckiaps ]]
1308	then
1309	return_queue=express
1310	elif [[ $remote_host = lckyuh ]]
1311	then
1312	return_queue=cx-single
1313	elif [[ $remote_host = lckyut ]]
1314	then
1315	return_queue=cx-single
1316	else
1317	return_queue=unknown
1318	fi
1319
1320	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1321	then
1322
1323	if [[ $remote_host = ibmku ]]
1324	then
1325	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1326	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1327	else
1328	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1329	fi
1330	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1331	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1333	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1334	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1335	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1336	if [[ $host != "ibmh" ]]
1337	then
1338	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1339	fi
1340	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1341	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1342
1343	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1344	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1345
1346	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1347	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
1348	if [[ $remote_host = ibmku ]]
1349	then
1350	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1351	fi
1352	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1353
1354	elif [[ $remote_host = nech ]]
1355	then
1356	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1357	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1358	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1359	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1360	echo "#PBS -j o" >> $job_to_send
1361	echo " " >> $job_to_send
1362	echo "set -x" >> $job_to_send
1363	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1364	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1365	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1366	echo "%%END%%" >> $job_to_send
1367
1368	elif [[ $remote_host = lcbullhh ]]
1369	then
1370	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1371	echo "#!/bin/bash" >> $job_to_send
1372	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1373	echo "#SBATCH -t 00:20:00" >> $job_to_send
1374	echo "#SBATCH -N 1" >> $job_to_send
1375	echo "#SBATCH -n 1" >> $job_to_send
1376	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1377	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1378	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1379	echo "#SBATCH -A $project_account" >> $job_to_send
1380	echo "#SBATCH -p $return_queue" >> $job_to_send
1381	echo " " >> $job_to_send
1382	echo "set -x" >> $job_to_send
1383	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1384	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1385	echo "%%END%%" >> $job_to_send
1386
1387	elif [[ $remote_host = lckyuh ]]
1388	then
1389	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1390	echo "#!/bin/bash" >> $job_to_send
1391	echo "#PJM -L \"node=1\"" >> $job_to_send
1392	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1393	echo "#PJM --no-stging" >> $job_to_send
1394	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1395	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1396	echo "#PJM -j" >> $job_to_send
1397	echo " " >> $job_to_send
1398	echo "export LANG=en_US.UTF-8" >> $job_to_send
1399	echo "set -x" >> $job_to_send
1400	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1401	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1402	echo "%%END%%" >> $job_to_send
1403
1404	elif [[ $remote_host = lckyut ]]
1405	then
1406	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1407	echo "#!/bin/bash" >> $job_to_send
1408	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1409	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1410	echo "#PJM --no-stging" >> $job_to_send
1411	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1412	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1413	echo "#PJM -j" >> $job_to_send
1414	echo " " >> $job_to_send
1415	echo "export LANG=en_US.UTF-8" >> $job_to_send
1416	echo "set -x" >> $job_to_send
1417	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1418	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1419	echo "%%END%%" >> $job_to_send
1420
1421	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1422	then
1423	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1424	echo "#!/bin/bash" >> $job_to_send
1425	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1426	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1427	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1428	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1429	echo "#PBS -j oe" >> $job_to_send
1430	echo " " >> $job_to_send
1431	echo "set -x" >> $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 "%%END%%" >> $job_to_send
1435
1436	elif [[ $remote_host = lcocean ]]
1437	then
1438	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1439	echo "#!/bin/bash" >> $job_to_send
1440	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1441	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1442	echo "SGEPREFIX -cwd" >> $job_to_send
1443	echo "SGEPREFIX -j y" >> $job_to_send
1444	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1445	echo " " >> $job_to_send
1446	echo "set -x" >> $job_to_send
1447	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1448	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1449	echo "" >> $job_to_send
1450	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1451	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1452	echo "rm -f scpjob.${identifier}" >> $job_to_send
1453	echo "%%END%%" >> $job_to_send
1454	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1455	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1456
1457	elif [[ $remote_host = lcflow ]]
1458	then
1459	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1460	echo "#!/bin/bash" >> $job_to_send
1461	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1462	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1463	echo "SGEPREFIX -cwd" >> $job_to_send
1464	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1465	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1466	echo "SGEPREFIX -j y" >> $job_to_send
1467	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1468	echo " " >> $job_to_send
1469	echo "set -x" >> $job_to_send
1470	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1471	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1472	echo "" >> $job_to_send
1473	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1474	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1475	echo "rm -f scpjob.${identifier}" >> $job_to_send
1476	echo "%%END%%" >> $job_to_send
1477	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1478	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1479	elif [[ $remote_host = lcxe6 ]]
1480	then
1481	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1482	echo "#!/bin/ksh" >> $job_to_send
1483	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1484	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1485	echo "#PBS -A $project_account" >> $job_to_send
1486	echo "#PBS -l mppwidth=1" >> $job_to_send
1487	echo "#PBS -l mppnppn=1" >> $job_to_send
1488	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1489	echo "#PBS -j oe" >> $job_to_send
1490	echo " " >> $job_to_send
1491	echo "set -x" >> $job_to_send
1492	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1493	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1494	echo "%%END%%" >> $job_to_send
1495	else
1496
1497	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1498	echo "# @\\\$-q $return_queue" >> $job_to_send
1499	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1500	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1501	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1502	then
1503	echo "# @\$-l mpp_p=0" >> $job_to_send
1504	fi
1505	echo '# @\$-lF 10mb' >> $job_to_send
1506	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1507	echo '# @\\\$-eo' >> $job_to_send
1508	echo " " >> $job_to_send
1509	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1510	then
1511	echo "set +vx" >> $job_to_send
1512	echo ". .profile" >> $job_to_send
1513	fi
1514	echo "set -x" >> $job_to_send
1515	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
1516	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1517	echo "%%END%%" >> $job_to_send
1518
1519	fi
1520
1521	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1522	then
1523	echo "llsubmit scpjob.$identifier" >> $job_to_send
1524	elif [[ $remote_host = lcbullhh ]]
1525	then
1526	echo "sbatch scpjob.$identifier" >> $job_to_send
1527	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1528	then
1529	echo "msub -E -q $return_queue scpjob.$identifier" >> $job_to_send
1530	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1531	then
1532	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1533	elif [[ $remote_host = t3es ]]
1534	then
1535	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1536	elif [[ $remote_host = lckiaps ]]
1537	then
1538	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1539	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1540	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1541	elif [[ $remote_host = lckyu* ]]
1542	then
1543	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1544	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1545	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1546	then
1547	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1548	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1549	else
1550	echo "$submcom scpjob.$identifier" >> $job_to_send
1551	fi
1552	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1553	then
1554	echo "rm scpjob.$identifier" >> $job_to_send
1555	fi
1556	if [[ $remote_host = nech ]]
1557	then
1558	echo "cd -" >> $job_to_send
1559	fi
1560	else
1561	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1562	# ??? funktioniert das ÃŒberhaupt noch ???
1563	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1564	fi
1565	echo "set -x" >> $job_to_send
1566	echo " ' exit" >> $job_to_send
1567	echo "set -x" >> $job_to_send
1568	fi
1569
1570
1571	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1572	cat $file_to_send >> $job_to_send
1573
1574	if [[ $remote_host = ibm ]]
1575	then
1576	echo " " >> $job_to_send
1577	echo "exit" >> $job_to_send
1578	fi
1579
1580	# REMOVE JOB-FILE
1581	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1582	then
1583	echo " " >> $job_to_send
1584	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1585	fi
1586
1587
1588	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1589	if [[ $no_submit = false ]]
1590	then
1591	if [[ $remote_host != $local_host ]]
1592	then
1593	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1594	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1595	then
1596	job_catalog_save=$job_catalog
1597	job_catalog=job_queue
1598	elif [[ $remote_host = nech ]]
1599	then
1600	job_catalog_save=$job_catalog
1601	job_catalog=/hpf/b/${remote_user}/job_queue
1602	fi
1603	if [[ $remote_host = nech ]]
1604	then
1605	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1606	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1607	else
1608	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1609	fi
1610	if [[ $? = 1 ]]
1611	then
1612	locat=scp; exit
1613	fi
1614	if [[ $remote_host = ibms ]]
1615	then
1616	job_catalog=$job_catalog_save
1617	fi
1618	[[ $verify = true ]] && printf "\n >>> finished\n"
1619	else
1620	eval job_catalog=$job_catalog
1621	cp $job_to_send ${job_catalog}/$job_on_remhost
1622	fi
1623
1624
1625
1626	# START NQS- / LOADLEVELER-JOB
1627	if [[ $remote_host != $local_host ]]
1628	then
1629	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1630
1631	if [[ $remote_host = ibmku ]]
1632	then
1633	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1634	elif [[ $remote_host = lcflow ]]
1635	then
1636	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1637	else
1638	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1639	fi
1640
1641	[[ $verify = true ]] && printf " >>> o.k.\n"
1642	else
1643	cd $job_catalog
1644	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1645	then
1646	eval $submcom $job_on_remhost
1647	elif [[ $local_host = lctit \|\| $local_host = lcxe6 \|\| $local_host = lck \|\| $local_host = lckordi \|\| $local_host = lcyon \|\| $local_host = lcsb \|\| $local_host = lckyu* ]]
1648	then
1649	chmod u+x $job_on_remhost
1650	eval $submcom $job_on_remhost
1651	elif [[ $local_host = nech ]]
1652	then
1653	if [[ $queue = default ]]
1654	then
1655	eval $submcom $job_on_remhost
1656	else
1657	eval $submcom -q $queue $job_on_remhost
1658	fi
1659	elif [[ $local_host = lcbullhh ]]
1660	then
1661	if [[ $queue = default ]]
1662	then
1663	eval $submcom $job_on_remhost
1664	fi
1665	else
1666	qsub $job_on_remhost
1667	fi
1668
1669	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1670	# AT THE END OF THE JOB
1671	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1672	then
1673	rm $job_on_remhost
1674	fi
1675	cd - > /dev/null
1676	fi
1677	fi
1678
1679	# FINAL ACTIONS
1680	if [[ $no_submit = false ]]
1681	then
1682	rm -f $job_to_send
1683	fi
1684	[[ $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 |