Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1943

Last change on this file since 1943 was 1941, checked in by raasch, 8 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 57.9 KB

Line
1	#!/bin/bash
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2014 Leibniz Universitaet Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1941 2016-06-14 05:17:11Z suehring $
30	#
31	# 1940 2016-06-14 05:15:20Z raasch
32	# adjustments for lckiaps
33	#
34	# 1866 2016-04-15 06:50:59Z raasch
35	# adjusted for lcocean
36	#
37	# 1841 2016-04-07 19:14:06Z raasch
38	# script now running under bash
39	#
40	# 1701 2015-11-02 07:43:04Z maronga
41	# Bugfix: added missing init_cmds for lccrayh/lccrayb
42	#
43	# 1621 2015-07-17 11:39:33Z heinze
44	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
45	#
46	# 1575 2015-03-27 09:56:27Z raasch
47	# mpp2-queues added to lccrayh
48	#
49	# 1547 2015-01-29 15:09:12Z witha
50	# adjustments for ForWind computing cluster (lcflow)
51	#
52	# 1545 2015-01-29 06:52:23Z heinze
53	# local host name for blizzard further specified
54	#
55	# 1480 2014-10-17 14:41:49Z raasch
56	# adjustments for 2nd stage of HLRNIII
57	#
58	# 1468 2014-09-24 14:06:57Z maronga
59	# Typo removed (addres->address)
60	# Adjustments for lcxe6
61	#
62	# 1452 2014-08-22 09:41:06Z heinze
63	# local hosts for blizzard added
64	#
65	# 1450 2014-08-21 07:31:51Z heinze
66	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
67	#
68	# 1442 2014-07-28 07:09:10Z raasch
69	# HLRN-III (lccrayb/lccrayh) queues adjusted
70	#
71	# 1378 2014-04-28 06:04:58Z raasch
72	# -et option added for lctit
73	#
74	# 1350 2014-04-04 13:01:30Z maronga
75	# location of qsub updated for lcxe6
76	#
77	# 1289 2014-03-04 07:12:34Z raasch
78	# German comments translated to English
79	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
80	#
81	# 1279 2014-01-28 12:10:14Z raasch
82	# node calculation modified due to changes in mrun (tasks_per_node must not be
83	# an integral divisor of numprocs any more)
84	#
85	# 1274 2014-01-09 13:14:54Z heinze
86	# adjustments for lccrayh
87	#
88	# 1266 2013-12-11 12:07:34Z heinze
89	# further adjustments for lccrayb (use msub instead of qsub)
90	#
91	# 1264 2013-12-09 12:46:09Z fricke
92	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
93	#
94	# 1262 2013-12-09 10:57:20Z fricke
95	# further adjustments for lccrayb
96	#
97	# 1260 2013-12-04 12:48:04Z raasch
98	# jaboticaba admitted
99	#
100	# 1255 2013-11-07 14:43:35Z raasch
101	# further adjustments for lccrayb
102	#
103	# 1224 2013-09-16 07:27:23Z raasch
104	# first adjustments for lccrayb
105	#
106	# 1202 2013-07-10 16:22:07Z witha
107	# adjustments for Forwind cluster (lcflow)
108	#
109	# 1199 2013-07-05 14:52:22Z raasch
110	# adjustments for CSC Helsinki (lccrayf)
111	#
112	# use of cluster/express queue enabled (ibmh)
113	# vinessa added (imuk)
114	#
115	# 1103 2013-02-20 02:15:53Z raasch
116	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
117	# instead typeset -L/R),
118	# further adjustments for lckyuh
119	#
120	# 2013-02-10 01:47:43Z raasch
121	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
122	# and for Forwind cluster (lcflow)
123	#
124	# 1094 2013-02-03 01:52:12Z raasch
125	# new option -P for explicit setting of ssh/scp port,
126	# decalpha parts (yonsei) removed
127	#
128	# 2013-02-02 07:06:13Z raasch
129	# adjustments for Kyushu-University computing center (lckyut - tatara)
130	# old changelog messages removed
131	#
132	# 1046 2012-11-09 14:38:45Z maronga
133	# code put under GPL (PALM 3.9)
134	#
135	# 08/07/94 - Siggi - first version finished
136	# 29/06/94 - Siggi - script development started
137	#--------------------------------------------------------------------------------#
138	# subjob - script for automatic generation and submission of batch-job files
139	# for various batch queuing systems
140	#--------------------------------------------------------------------------------#
141
142
143	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
144	delete_dayfile=false
145	email_notification=none
146	group_number=none
147	locat=normal
148	no_default_queue=none
149	no_submit=false
150	job_catalog="~/job_queue"
151	job_name=none
152	local_user=$LOGNAME
153	node_usage=shared
154	numprocs=0
155	punkte="..........................................................."
156	submcom=qsub
157	queue=default
158	remote_host=none
159	remote_user=""
160	verify=true
161
162	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
163	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
164
165
166
167	# ERROR HANDLING
168	# IN CASE OF EXIT:
169	trap 'if [[ $locat != normal ]]
170	then
171	case $locat in
172	(option) printf "\n --> available optios can be displayed"
173	printf " by typing:"
174	printf "\n \"subjob ?\" \n";;
175	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
176	(*) printf "\n +++ unknown error"
177	printf "\n please inform S. Raasch!\n"
178	esac
179	[[ -f $job_to_send ]] && rm $job_to_send
180	printf "\n\n+++ SUBJOB killed \n\n"
181	fi' exit
182
183
184	# IN CASE OF TERMINAL-BREAK:
185	trap '[[ -f $job_to_send ]] && rm $job_to_send
186	printf "\n\n+++ SUBJOB killed \n\n"
187	exit
188	' 2
189
190
191	# DETERMINE NAME OF LOCAL HOST
192	local_host=$(hostname)
193
194	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
195	# IS ADMITTED AT ALL)
196	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
197	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
198	case $local_host in
199	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
200	(atmos) local_address=172.20.25.35; local_host=lcide;;
201	(austru) local_address=130.75.105.128; local_host=lcmuk;;
202	(autan) local_address=130.75.105.57; local_host=lcmuk;;
203	(bora) local_address=130.75.105.103; local_host=lcmuk;;
204	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
205	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
206	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
207	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
208	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
209	(breva) local_address=130.75.105.98; local_host=lcmuk;;
210	(buran) local_address=130.75.105.58; local_host=lcmuk;;
211	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
212	(climate*) local_address=165.132.26.68; local_host=lcyon;;
213	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
214	(cs*) local_address=136.172.44.131; local_host=nech;;
215	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
216	(flow01) local_address=10.141.255.71; local_host=lcflow;;
217	(flow02) local_address=10.141.255.72; local_host=lcflow;;
218	(node*) local_address=165.132.26.61 local_host=lck;;
219	# (node*) local_address=210.219.61.8 local_host=lckordi;;
220	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
221	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
222	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
223	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
224	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
225	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
226	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
227	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
228	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
229	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
230	(sno) local_address=130.75.105.113; local_host=lcmuk;;
231	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
232	(login*) local_address=118.128.66.201; local_host=lckiaps;;
233	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
234	(meller) local_address=134.106.74.155; local_host=lcfor;;
235	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
236	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
237	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
238	(nobel*) local_address=150.183.5.101; local_host=ibms;;
239	(ocean) local_address="ocean"; local_host=lcocean;;
240	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
241	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
242	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
243	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
244	(pingui) local_address=134.106.74.118; local_host=lcfor;;
245	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
246	(rte*) local_address=133.5.185.60; local_host=lcrte;;
247	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
248	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
249	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
250	(solano) local_address=130.75.105.110; local_host=lcmuk;;
251	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
252	(tc*) local_address="ocean"; local_host=lcocean;;
253	(t2a*) local_address=10.1.6.165; local_host=lctit;;
254	(urban*) local_address=147.46.30.151 local_host=lcsb;;
255	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
256	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
257	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
258	(*) printf "\n +++ \"$local_host\" unknown";
259	printf "\n please contact the PALM group at IMUK";
260	locat=parameter; exit;;
261	esac
262
263
264
265	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
266	remote_host=$local_host
267
268
269
270
271	# READ THE SHELLSCRIPT-OPTIONS
272	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
273	do
274	case $option in
275	(c) job_catalog=$OPTARG;;
276	(d) delete_dayfile=true;;
277	(D) no_submit=true;;
278	(e) email_notification=$OPTARG;;
279	(g) group_number=$OPTARG;;
280	(h) remote_host=$OPTARG;;
281	(m) memory=$OPTARG;;
282	(n) job_name=$OPTARG;;
283	(N) node_usage=$OPTARG;;
284	(O) threads_per_task=$OPTARG;;
285	(P) scp_port=$OPTARG;;
286	(q) no_default_queue=$OPTARG;;
287	(t) cputime=$OPTARG;;
288	(T) tasks_per_node=$OPTARG;;
289	(u) remote_user=$OPTARG;;
290	(v) verify=false;;
291	(X) numprocs=$OPTARG;;
292	(\?) printf "\n +++ Option $OPTARG unknown \n";
293	locat=option; exit;;
294	esac
295	done
296
297
298	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
299	(( to_shift = $OPTIND - 1 ))
300	shift $to_shift; file_to_send=$1
301
302
303	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
304	if [ "$1" = "?" ]
305	then
306	(printf "\n *** subjob can be called as follows:\n"
307	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
308	printf "\n Description of available options:\n"
309	printf "\n Option Description Default-Value"
310	printf "\n -c job-input- and output-catalog ~/job_queue"
311	printf "\n -d no job-protocol will be created ---"
312	printf "\n -D only the job-file will be created ---"
313	printf "\n -h execution host, available hosts: $remote_host"
314	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
315	printf "\n lckiaps, lctit, nech"
316	printf "\n -m memory demand per process in MByte ---"
317	printf "\n -n jobname <jobdatei>"
318	printf "\n -O threads per task (for OpenMP usage) 1"
319	printf "\n -P ssh/scp port default port"
320	printf "\n -q job-queue to be used default"
321	printf "\n -t allowed cpu-time in seconds ---"
322	printf "\n -T tasks per node (on parallel hosts) ---"
323	printf "\n -u username on execution host from .netrc"
324	printf "\n -v no prompt for confirmation ---"
325	printf "\n -X # of processors (on parallel hosts) 1"
326	printf "\n "
327	printf "\n The only possible positional parameter is <jobfile>:"
328	printf "\n The complete NQS-job must be provided here."
329	printf "\n <jobfile>=? creates this outline\n\n") \| more
330	exit
331	fi
332
333
334
335	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
336	if [[ "$file_to_send" = "" ]]
337	then
338	printf "\n +++ job-file missing"
339	locat=parameter; exit
340	else
341	if [[ -f $file_to_send ]]
342	then
343	true
344	else
345	printf "\n +++ job-file: "
346	printf "\n $file_to_send"
347	printf "\n does not exist"
348	locat=parameter; exit
349	fi
350	fi
351
352
353
354	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
355	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
356	if [[ $job_name = none ]]
357	then
358	job_name=$file_to_send
359	fi
360	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
361	then
362	printf "\n +++ job-file name: "
363	printf "\n $job_name"
364	printf "\n must not contain \"/\"-characters"
365	locat=parameter; exit
366	fi
367
368
369
370
371	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
372	# OR IF NO HOST HAS BEEN GIVEN
373	if [[ $remote_host = none ]]
374	then
375	printf "\n +++ host missing"
376	locat=option; exit
377	else
378	case $remote_host in
379	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
380	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
381	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
382	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
383	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
384	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
385	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
386	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
387	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
388	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
389	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
390	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
391	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
392	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
393	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
394	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
395	(lcocean) remote_address="ocean"; submcom=qsub;;
396	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
397	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
398	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
399	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
400	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
401	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
402	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
403	locat=parameter; exit;;
404	esac
405	fi
406
407
408	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
409	if [[ $no_default_queue != none ]]
410	then
411	error=false
412	ndq=$no_default_queue
413	case $remote_host in
414	(ibm) case $ndq in
415	(p690_express\|p690_standard\|p690_long) error=false;;
416	(*) error=true;;
417	esac;;
418	(ibmh) case $ndq in
419	(cluster\|express) error=false;;
420	(*) error=true;;
421	esac;;
422	(ibmkisti) case $ndq in
423	(class.32plus\|class.1-2\|class.2-32) error=false;;
424	(*) error=true;;
425	esac;;
426	(ibmku) case $ndq in
427	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
428	(*) error=true;;
429	esac;;
430	(ibms) case $ndq in
431	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
432	(*) error=true;;
433	esac;;
434	(lcbullhh) case $ndq in
435	(compute\|shared) error=false;;
436	(*) error=true;;
437	esac;;
438	(lccrayb) case $ndq in
439	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
440	(*) error=true;;
441	esac;;
442	(lccrayh) case $ndq in
443	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
444	(*) error=true;;
445	esac;;
446	(lccrayf) case $ndq in
447	(usup\|test*\|small\|large) error=false;;
448	(*) error=true;;
449	esac;;
450	(lcflow) case $ndq in
451	(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;;
452	(*) error=true;;
453	esac;;
454	(lckiaps) case $ndq in
455	(express\|normal\|normal20\|quickq) error=false;;
456	(*) error=true;;
457	esac;;
458	(lckyoto) case $ndq in
459	(eh\|ph) error=false;;
460	(*) error=true;;
461	esac;;
462	(lckyuh) case $ndq in
463	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
464	(*) error=true;;
465	esac;;
466	(lckyut) case $ndq in
467	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
468	(*) error=true;;
469	esac;;
470	(lctit) case $ndq in
471	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
472	(*) error=true;;
473	esac;;
474	(t3eb) case $ndq in
475	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
476	(*) error=true;;
477	esac;;
478	(t3eh) case $ndq in
479	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
480	(*) error=true;;
481	esac;;
482	(t3ej2\|t3ej5) case $ndq in
483	(low\|normal\|high) error=false;;
484	(*) error=true;;
485	esac;;
486	(t3es) case $ndq in
487	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
488	(*) error=true;;
489	esac;;
490	esac
491	if [[ $error = true ]]
492	then
493	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
494	locat=parameter; exit
495	else
496	queue=$no_default_queue
497	fi
498	fi
499
500
501
502	# CHECK THE CPU-TIME
503	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
504	done=false
505	while [[ $done = false ]]
506	do
507	if (( $cputime <= 0 ))
508	then
509	printf "\n +++ wrong cpu-time or cpu-time missing"
510	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
511	printf "\n >>> "
512	read cputime 1>/dev/null 2>&1
513	else
514	done=true
515	fi
516	done
517	if [[ $remote_host = nech ]]
518	then
519	if (( tasks_per_node != 0 ))
520	then
521	(( cputime = cputime * tasks_per_node ))
522	elif [[ $numprocs != 0 ]]
523	then
524	(( cputime = cputime * numprocs ))
525	fi
526	fi
527	(( stunden = cputime / 3600 ))
528	(( resttime = cputime - stunden * 3600 ))
529	(( minuten = resttime / 60 ))
530	(( sekunden = resttime - minuten * 60 ))
531	timestring=${stunden}:${minuten}:${sekunden}
532
533
534
535	# CHECK THE MEMORY DEMAND
536	done=false
537	while [[ $done = false ]]
538	do
539	if (( memory <= 0 ))
540	then
541	printf "\n +++ wrong memory demand or memory demand missing"
542	printf "\n >>> Please type memory in MByte per process as INTEGER:"
543	printf "\n >>> "
544	read memory 1>/dev/null 2>&1
545	else
546	done=true
547	fi
548	done
549
550	if [[ $remote_host = nech ]]
551	then
552	if (( tasks_per_node != 0 ))
553	then
554	(( Memory = memory * tasks_per_node / 1000 ))
555	elif [[ $numprocs != 0 ]]
556	then
557	(( Memory = memory * numprocs / 1000 ))
558	else
559	(( Memory = memory / 1000 ))
560	fi
561	elif [[ $remote_host = lctit ]]
562	then
563	(( Memory = memory * tasks_per_node / 1000 ))
564	fi
565
566
567	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
568	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
569	then
570	(( memory = memory * threads_per_task ))
571	fi
572
573
574	# CALCULATE NUMBER OF REQUIRED NODES
575	if (( tasks_per_node != 0 ))
576	then
577	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
578	fi
579
580
581	# CALCULATE NUMBER OF PROCESSES PER NODE
582	(( processes_per_node = tasks_per_node * threads_per_task ))
583
584
585	# CALCULATE NUMBER OF MPI TASKS
586	(( mpi_tasks = numprocs / threads_per_task ))
587
588
589	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
590	if [[ "$scp_port" != "" ]]
591	then
592	PORTOPT="-P $scp_port"
593	SSH_PORTOPT="-p $scp_port"
594	fi
595
596
597	# HEADER-OUTPUT
598	if [[ $verify = true ]]
599	then
600	printf "\n\n"
601	printf "#--------------------------------------------------------------# \n"
602	spalte1=SUBJOB;spalte2=$(date)
603	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
604	printf "\| \| \n"
605	printf "\| values of parameters/options: \| \n"
606	spalte1=$(echo local_host$punkte \| cut -c-20)
607	spalte2=$punkte$local_host
608	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
609	spalte1=$(echo remote_host$punkte \| cut -c-20)
610	spalte2=$punkte$remote_host
611	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
612	spalte1=$(echo queue$punkte \| cut -c-20)
613	spalte2=$punkte$queue
614	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
615	spalte1=$(echo memory$punkte \| cut -c-20)
616	spalte2="$punkte$memory mb"
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	spalte1=$(echo cputime$punkte \| cut -c-20)
619	spalte2="$punkte$cputime sec"
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	spalte1=$(echo job_name$punkte \| cut -c-20)
622	spalte2="$punkte$job_name"
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	printf "#--------------------------------------------------------------# \n\n"
625
626
627	# QUERY CHECK
628	antwort="dummy"
629	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
630	do
631	read antwort?" >>> continue (y/n) ? "
632	done
633	if [[ $antwort = n \|\| $antwort = N ]]
634	then
635	locat=verify; exit
636	fi
637	printf "\n"
638	fi
639
640	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
641	identifier=$RANDOM
642	job_on_remhost=${job_name}_${identifier}_$local_host
643	job_to_send=job_to_send_$identifier
644	if [[ $delete_dayfile = false ]]
645	then
646	remote_dayfile=${local_host}_${job_name}_result_$identifier
647	local_dayfile=${remote_host}_${job_name}
648	else
649	remote_dayfile=/dev/null
650	fi
651
652
653	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
654	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
655	then
656
657	# GENERAL LOADLEVELER SETTINGS
658	execute_in_shell="#!/bin/ksh"
659	use_shell="# @ shell = /bin/ksh"
660	consumable_memory="ConsumableMemory($memory mb)"
661	class="# @ class = $queue"
662	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
663	network_to_use="# @ network.mpi = sn_all,shared,us"
664	data_limit="# @ data_limit = 1.76gb"
665	image_size="# @ image_size = 50"
666	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
667
668	if [[ $email_notification = none ]]
669	then
670	notify_user=""
671	else
672	notify_user="# @ notify_user = $email_notification"
673	if [[ $delete_dayfile = true ]]
674	then
675	notification='# @ notification = never'
676	fi
677	fi
678
679	if [[ $remote_host = ibmh ]]
680	then
681	data_limit=""
682	network_to_use=""
683	class="# @ class = $queue"
684	environment=""
685	rset="# @ rset = RSET_MCM_AFFINITY"
686	task_affinity="# @ task_affinity = core(1)"
687	elif [[ $remote_host = ibmkisti ]]
688	then
689	network_to_use="# @ network.MPI = sn_all,shared,US"
690	wall_clock_limit="# @ wall_clock_limit = $timestring"
691	if [[ $threads_per_task = 1 ]]
692	then
693	rset="# @ rset = RSET_MCM_AFFINITY"
694	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
695	fi
696	environment=""
697	use_shell=""
698	data_limit=""
699	image_size=""
700	elif [[ $remote_host = ibmku ]]
701	then
702	execute_in_shell="#!/usr/bin/ksh"
703	use_shell="# @ shell = /usr/bin/ksh"
704	consumable_memory=""
705	environment=""
706	network_to_use="# @ network.mpi = sn_all,shared,us"
707	data_limit=""
708	image_size=""
709	elif [[ $remote_host = ibms ]]
710	then
711	network_to_use="# @ network.mpi = csss,shared,us"
712	fi
713
714	cat > $job_to_send << %%END%%
715	$execute_in_shell
716	$use_shell
717
718	# @ job_type = parallel
719	# @ job_name = $job_name
720	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
721	# @ output = $remote_dayfile
722	# @ error = $remote_dayfile
723	$wall_clock_limit
724	$image_size
725	$class
726	$environment
727	$network_to_use
728	$data_limit
729	$rset
730	$mcm_affinity_options
731	$task_affinity
732	$notification
733	$notify_user
734
735	%%END%%
736
737	if (( nodes > 0 ))
738	then
739
740	if [[ $remote_host != ibmkisti ]]
741	then
742
743	cat >> $job_to_send << %%END%%
744	# @ node = $nodes
745	# @ tasks_per_node = $processes_per_node
746	# @ node_usage = $node_usage
747	# @ queue
748
749	%%END%%
750
751	else
752
753	cat >> $job_to_send << %%END%%
754	# @ total_tasks = $mpi_tasks
755	# @ blocking = unlimited
756	# @ queue
757
758	%%END%%
759
760	fi
761
762	else
763
764	cat >> $job_to_send << %%END%%
765	# @ blocking = unlimited
766	# @ total_tasks = $numprocs
767	# @ node_usage = $node_usage
768	# @ queue
769
770	%%END%%
771
772	fi
773
774	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
775	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
776	then
777	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
778	fi
779
780	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
781	then
782
783	cat > $job_to_send << %%END%%
784	#!/bin/ksh
785
786	# @ job_type = serial
787	# @ node_usage = $node_usage
788	# @ job_name = palm
789	# @ wall_clock_limit = ${timestring},$timestring
790	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
791	# @ output = $remote_dayfile
792	# @ error = $remote_dayfile
793	$class
794	$notification
795
796	# @ queue
797
798	%%END%%
799
800	elif [[ $remote_host = lcbullhh ]]
801	then
802	if [[ $numprocs != 0 ]]
803	then
804	cat > $job_to_send << %%END%%
805	#!/bin/bash -l
806	#SBATCH -J $job_name
807	#SBATCH -t $timestring
808	#SBATCH -N $nodes
809	#SBATCH --ntasks-per-node=$processes_per_node
810	#SBATCH -p $queue
811	#SBATCH -o $remote_dayfile
812	#SBATCH -e $remote_dayfile
813	#SBATCH -A $project_account
814
815	$init_cmds
816	$module_calls
817
818	%%END%%
819
820	else
821	cat > $job_to_send << %%END%%
822	#!/bin/bash -l
823	#SBATCH -J $job_name
824	#SBATCH -t $timestring
825	#SBATCH -l ncpus=1
826	#SBATCH -l pmem=${memory}mb
827	#SBATCH -m abe
828	#SBATCH -o $remote_dayfile
829	#SBATCH -e $remote_dayfile
830	#SBATCH -A $project_account
831
832	$init_cmds
833	$module_calls
834
835	%%END%%
836
837	fi
838
839	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
840	then
841
842	if [[ "$feature" != "" ]]
843	then
844	featuredir="#PBS -l feature=$feature"
845	fi
846
847	if [[ $numprocs != 0 ]]
848	then
849	cat > $job_to_send << %%END%%
850	#!/bin/bash
851	#PBS -N $job_name
852	#PBS -l walltime=$timestring
853	#PBS -l nodes=$nodes:ppn=$processes_per_node
854	#PBS -o $remote_dayfile
855	#PBS -j oe
856	#PBS -q $queue
857	$featuredir
858
859	$init_cmds
860	$module_calls
861
862	%%END%%
863
864	else
865
866	continue
867
868	fi
869
870	elif [[ $remote_host = lccrayf ]]
871	then
872
873	if [[ $numprocs != 0 ]]
874	then
875	cat > $job_to_send << %%END%%
876	#!/bin/bash -l
877	#SBATCH -J $job_name
878	#SBATCH -t $timestring
879	#SBATCH -N $nodes
880	#SBATCH --ntasks-per-node=$processes_per_node
881	#SBATCH -p $queue
882	#SBATCH -o $remote_dayfile
883	#SBATCH -e $remote_dayfile
884
885	$init_cmds
886	$module_calls
887
888	%%END%%
889
890	else
891	cat > $job_to_send << %%END%%
892	#!/bin/bash -l
893	#SBATCH -J $job_name
894	#SBATCH -t $timestring
895	#SBATCH -l ncpus=1
896	#SBATCH -l pmem=${memory}mb
897	#SBATCH -m abe
898	#SBATCH -o $remote_dayfile
899	#SBATCH -e $remote_dayfile
900
901	$init_cmds
902	$module_calls
903
904	%%END%%
905
906	fi
907
908	elif [[ $remote_host = lcflow ]]
909	then
910	if [[ $numprocs != 0 ]]
911	then
912	pe_set="#$ -pe impi $numprocs"
913	else
914	pe_set="#$ -pe impi 1"
915	fi
916	if [[ $queue = default ]]
917	then
918	queue_set=""
919	else
920	queue_set="#$ -q $queue"
921	fi
922	[[ "$disc_space" = "" ]] && disc_space=50
923
924	cat > $job_to_send << %%END%%
925	#!/bin/bash
926	#$ -S /bin/bash
927	#$ -N $job_name
928	#$ -cwd
929	#$ -l h_rt=$timestring
930	#$ -l h_vmem=${memory}M
931	#$ -o $remote_dayfile
932	#$ -j y
933	$pe_set
934	#$ -R y
935	#$ -l h_fsize=${disc_space}G
936	$queue_set
937
938	%%END%%
939
940	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
941	then
942
943	if [[ $numprocs != 0 ]]
944	then
945	cat > $job_to_send << %%END%%
946	#!/bin/ksh
947	#PBS -N $job_name
948	#PBS -l walltime=$timestring
949	#PBS -l ncpus=$numprocs
950	#PBS -l pmem=${memory}mb
951	#PBS -o $remote_dayfile
952	#PBS -l nodes=$nodes:ppn=${processes_per_node}
953	#PBS -j oe
954
955	mpd &
956
957	%%END%%
958
959	else
960	cat > $job_to_send << %%END%%
961	#!/bin/ksh
962	#PBS -N $job_name
963	#PBS -l walltime=$timestring
964	#PBS -l ncpus=1
965	#PBS -l pmem=${memory}mb
966	#PBS -o $remote_dayfile
967	#PBS -j oe
968
969	%%END%%
970
971	fi
972
973	elif [[ $remote_host = lckiaps ]]
974	then
975
976	if [[ $numprocs != 0 ]]
977	then
978	cat > $job_to_send << %%END%%
979	#!/bin/bash
980	#PBS -N $job_name
981	#PBS -l walltime=$timestring
982	#PBS -l select=1:ncpus=$numprocs
983	#PBS -l pmem=${memory}mb
984	#PBS -q $queue
985	#PBS -o $remote_dayfile
986	#PBS -j oe
987	#PBS -V
988
989	%%END%%
990
991	else
992	cat > $job_to_send << %%END%%
993	#!/bin/bash
994	#PBS -N $job_name
995	#PBS -l walltime=$timestring
996	#PBS -l ncpus=1
997	#PBS -l pmem=${memory}mb
998	#PBS -o $remote_dayfile
999	#PBS -j oe
1000
1001	%%END%%
1002
1003	fi
1004
1005	elif [[ $remote_host = lcyon ]]
1006	then
1007
1008	if [[ $numprocs != 0 ]]
1009	then
1010	cat > $job_to_send << %%END%%
1011	#!/bin/ksh
1012	#PBS -N $job_name
1013	#PBS -l walltime=$timestring
1014	#PBS -l ncpus=$numprocs
1015	#PBS -l pmem=${memory}mb
1016	#PBS -o $remote_dayfile
1017	#PBS -j oe
1018
1019	%%END%%
1020
1021	else
1022	cat > $job_to_send << %%END%%
1023	#!/bin/ksh
1024	#PBS -N $job_name
1025	#PBS -l walltime=$timestring
1026	#PBS -l ncpus=1
1027	#PBS -l pmem=${memory}mb
1028	#PBS -o $remote_dayfile
1029	#PBS -j oe
1030
1031	%%END%%
1032
1033	fi
1034
1035	elif [[ $remote_host = lcxe6 ]]
1036	then
1037
1038	if [[ $numprocs != 0 ]]
1039	then
1040	cat > $job_to_send << %%END%%
1041	#!/bin/ksh
1042	#PBS -S /bin/ksh
1043	#PBS -N $job_name
1044	#PBS -A $project_account
1045	#PBS -j oe
1046	#PBS -l walltime=$timestring
1047	#PBS -l mppwidth=${numprocs}
1048	#PBS -l mppnppn=${processes_per_node}
1049	#PBS -m abe
1050	#PBS -o $remote_dayfile
1051	$email_directive
1052
1053	$init_cmds
1054	$module_calls
1055
1056	%%END%%
1057
1058	else
1059	cat > $job_to_send << %%END%%
1060	#!/bin/ksh
1061	#PBS -S /bin/ksh
1062	#PBS -N $job_name
1063	#PBS -A $project_account
1064	#PBS -j oe
1065	#PBS -l walltime=$timestring
1066	#PBS -l ncpus=1
1067	#PBS -l pmem=${memory}mb
1068	#PBS -m abe
1069	$email_directive
1070	#PBS -o $remote_dayfile
1071
1072	$init_cmds
1073	$module_calls
1074
1075	%%END%%
1076
1077	fi
1078
1079	elif [[ $remote_host = lckyoto ]]
1080	then
1081
1082	cat > $job_to_send << %%END%%
1083	#!/bin/ksh
1084	# @\$-o $remote_dayfile
1085	# @\$-eo -oi
1086	# @\$-lP 16
1087	# @\$-lp 1
1088	# @\$-lm 28gb -llm unlimited -ls unlimited
1089	# @\$-q $queue
1090	# @\$-Pvn abs_pack
1091	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1092	#. /thin/local/etc/setprofile/intel-11.0.sh
1093	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1094	. ~/.myprofile
1095	#. /home2/t/t51254/palm/current_version/myprofile
1096	#. /thin/apps/pgi/mpi.sh
1097	#
1098	env
1099	#
1100	set -x
1101
1102	%%END%%
1103
1104	elif [[ $remote_host = lcxt5m ]]
1105	then
1106
1107	if [[ $numprocs != 0 ]]
1108	then
1109	cat > $job_to_send << %%END%%
1110	#!/bin/ksh
1111	#PBS -S /bin/ksh
1112	#PBS -N $job_name
1113	#PBS -j oe
1114	#PBS -l walltime=$timestring
1115	#PBS -l mppwidth=${numprocs}
1116	#PBS -l mppnppn=${processes_per_node}
1117	#PBS -m abe
1118	#PBS -o $remote_dayfile
1119
1120	$init_cmds
1121	$module_calls
1122
1123	%%END%%
1124
1125	else
1126	cat > $job_to_send << %%END%%
1127	#!/bin/ksh
1128	#PBS -S /bin/ksh
1129	#PBS -N $job_name
1130	#PBS -j oe
1131	#PBS -l walltime=$timestring
1132	#PBS -l ncpus=1
1133	#PBS -l pmem=${memory}mb
1134	#PBS -m abe
1135	#PBS -o $remote_dayfile
1136
1137	$init_cmds
1138	$module_calls
1139
1140	%%END%%
1141
1142	fi
1143
1144	elif [[ $remote_host = lckyuh ]]
1145	then
1146	cat > $job_to_send << %%END%%
1147	#!/bin/bash
1148	#PJM -L "rscgrp=$queue"
1149	#PJM -L "node=$nodes"
1150	#PJM --mpi "proc=$numprocs"
1151	#PJM -L "elapse=$timestring"
1152	#PJM -o $remote_dayfile
1153	#PJM -j
1154	#PJM -X
1155	#PJM --no-stging
1156
1157	export LANG=en_US.UTF-8
1158	%%END%%
1159
1160	elif [[ $remote_host = lckyut ]]
1161	then
1162	cat > $job_to_send << %%END%%
1163	#!/bin/bash
1164	#PJM -L "rscgrp=$queue"
1165	#PJM -L "vnode=$numprocs"
1166	#PJM -L "vnode-core=1"
1167	#PJM -L "elapse=$timestring"
1168	#PJM --mpi proc=$numprocs
1169	#PJM -o $remote_dayfile
1170	#PJM -j
1171	#PJM -X
1172	#PJM --no-stging
1173
1174	export LANG=en_US.UTF-8
1175	%%END%%
1176
1177	elif [[ $remote_host = lcocean ]]
1178	then
1179	cat > $job_to_send << %%END%%
1180	#!/bin/bash
1181	#$ -cwd
1182	#$ -V
1183	#$ -N $job_name
1184	#$ -pe orte $numprocs
1185	#$ -o $remote_dayfile
1186	#$ -j y
1187	#$ -R y
1188	$init_cmds
1189	$module_calls
1190
1191	%%END%%
1192
1193	elif [[ $remote_host = nech ]]
1194	then
1195
1196	if (( nodes > 1 ))
1197	then
1198	cat > $job_to_send << %%END%%
1199	#!/bin/ksh
1200	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1201	#PBS -l ${qsubmem}=${Memory}gb
1202	#PBS -b $nodes
1203	#PBS -o $remote_dayfile
1204	#PBS -N palm
1205	#PBS -j o
1206	#PBS -T mpisx
1207
1208	%%END%%
1209
1210	elif [[ $numprocs != 0 ]]
1211	then
1212	cat > $job_to_send << %%END%%
1213	#!/bin/ksh
1214	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1215	#PBS -l ${qsubmem}=${Memory}gb
1216	#PBS -o $remote_dayfile
1217	#PBS -N palm
1218	#PBS -j o
1219
1220	%%END%%
1221
1222	else
1223	cat > $job_to_send << %%END%%
1224	#!/bin/ksh
1225	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1226	#PBS -o $remote_dayfile
1227	#PBS -j o
1228
1229	%%END%%
1230
1231	fi
1232
1233	elif [[ $remote_host = lctit ]]
1234	then
1235	cat > $job_to_send << %%END%%
1236	#!/bin/ksh
1237	$init_cmds
1238	$module_calls
1239
1240	%%END%%
1241
1242	# SET OPTIONS FOR SUBMIT-COMMAND
1243	if [[ $tasks_per_node != $processes_per_node ]]
1244	then
1245	submcom="$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 $queue "
1246	else
1247	submcom="$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 "
1248	fi
1249
1250	else
1251
1252	cat > $job_to_send << %%END%%
1253	# @\$-q ${queue}
1254	# @\$-l${qsubtime} $timestring
1255	# @\$-l${qsubmem} ${memory}mb
1256	# @\$-o $remote_dayfile
1257	# @\$-eo
1258
1259	%%END%%
1260
1261	fi
1262
1263
1264	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1265	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1266	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1267	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1268	then
1269	echo "set +vx" >> $job_to_send
1270	echo "trap '" >> $job_to_send
1271	echo "set +vx" >> $job_to_send
1272	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 ]]
1273	then
1274	if [[ $remote_host = ibmh ]]
1275	then
1276	return_queue=c1
1277	elif [[ $remote_host = ibmkisti ]]
1278	then
1279	return_queue=class.1-2
1280	elif [[ $remote_host = ibmku ]]
1281	then
1282	return_queue=sdbg2
1283	elif [[ $remote_host = ibms ]]
1284	then
1285	return_queue=p_normal
1286	elif [[ $remote_host = lcbullhh ]]
1287	then
1288	return_queue=shared
1289	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1290	then
1291	return_queue=dataq
1292	elif [[ $remote_host = lcxe6 ]]
1293	then
1294	return_queue=debug
1295	elif [[ $remote_host = lckiaps ]]
1296	then
1297	return_queue=express
1298	elif [[ $remote_host = lckyuh ]]
1299	then
1300	return_queue=cx-single
1301	elif [[ $remote_host = lckyut ]]
1302	then
1303	return_queue=cx-single
1304	else
1305	return_queue=unknown
1306	fi
1307
1308	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1309	then
1310
1311	if [[ $remote_host = ibmku ]]
1312	then
1313	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1314	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1315	else
1316	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1317	fi
1318	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1319	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1320	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1321	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1322	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1323	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1324	if [[ $host != "ibmh" ]]
1325	then
1326	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1327	fi
1328	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1329	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1330
1331	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1333
1334	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1335	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
1336	if [[ $remote_host = ibmku ]]
1337	then
1338	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1339	fi
1340	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1341
1342	elif [[ $remote_host = nech ]]
1343	then
1344	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1345	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1346	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1347	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1348	echo "#PBS -j o" >> $job_to_send
1349	echo " " >> $job_to_send
1350	echo "set -x" >> $job_to_send
1351	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1352	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1353	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1354	echo "%%END%%" >> $job_to_send
1355
1356	elif [[ $remote_host = lcbullhh ]]
1357	then
1358	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1359	echo "#!/bin/bash" >> $job_to_send
1360	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1361	echo "#SBATCH -t 00:20:00" >> $job_to_send
1362	echo "#SBATCH -N 1" >> $job_to_send
1363	echo "#SBATCH -n 1" >> $job_to_send
1364	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1365	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1366	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1367	echo "#SBATCH -A $project_account" >> $job_to_send
1368	echo "#SBATCH -p $return_queue" >> $job_to_send
1369	echo " " >> $job_to_send
1370	echo "set -x" >> $job_to_send
1371	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1372	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1373	echo "%%END%%" >> $job_to_send
1374
1375	elif [[ $remote_host = lckyuh ]]
1376	then
1377	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1378	echo "#!/bin/bash" >> $job_to_send
1379	echo "#PJM -L \"node=1\"" >> $job_to_send
1380	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1381	echo "#PJM --no-stging" >> $job_to_send
1382	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1383	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1384	echo "#PJM -j" >> $job_to_send
1385	echo " " >> $job_to_send
1386	echo "export LANG=en_US.UTF-8" >> $job_to_send
1387	echo "set -x" >> $job_to_send
1388	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1389	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1390	echo "%%END%%" >> $job_to_send
1391
1392	elif [[ $remote_host = lckyut ]]
1393	then
1394	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1395	echo "#!/bin/bash" >> $job_to_send
1396	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1397	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1398	echo "#PJM --no-stging" >> $job_to_send
1399	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1400	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1401	echo "#PJM -j" >> $job_to_send
1402	echo " " >> $job_to_send
1403	echo "export LANG=en_US.UTF-8" >> $job_to_send
1404	echo "set -x" >> $job_to_send
1405	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1406	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1407	echo "%%END%%" >> $job_to_send
1408
1409	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1410	then
1411	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1412	echo "#!/bin/bash" >> $job_to_send
1413	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1414	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1415	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1416	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1417	echo "#PBS -j oe" >> $job_to_send
1418	echo " " >> $job_to_send
1419	echo "set -x" >> $job_to_send
1420	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1421	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1422	echo "%%END%%" >> $job_to_send
1423
1424	elif [[ $remote_host = lcocean ]]
1425	then
1426	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1427	echo "#!/bin/bash" >> $job_to_send
1428	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1429	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1430	echo "SGEPREFIX -cwd" >> $job_to_send
1431	echo "SGEPREFIX -j y" >> $job_to_send
1432	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1433	echo " " >> $job_to_send
1434	echo "set -x" >> $job_to_send
1435	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1436	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1437	echo "" >> $job_to_send
1438	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1439	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1440	echo "rm -f scpjob.${identifier}" >> $job_to_send
1441	echo "%%END%%" >> $job_to_send
1442	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1443	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1444
1445	elif [[ $remote_host = lcflow ]]
1446	then
1447	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1448	echo "#!/bin/bash" >> $job_to_send
1449	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1450	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1451	echo "SGEPREFIX -cwd" >> $job_to_send
1452	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1453	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1454	echo "SGEPREFIX -j y" >> $job_to_send
1455	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1456	echo " " >> $job_to_send
1457	echo "set -x" >> $job_to_send
1458	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1459	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1460	echo "" >> $job_to_send
1461	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1462	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1463	echo "rm -f scpjob.${identifier}" >> $job_to_send
1464	echo "%%END%%" >> $job_to_send
1465	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1466	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1467	elif [[ $remote_host = lcxe6 ]]
1468	then
1469	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1470	echo "#!/bin/ksh" >> $job_to_send
1471	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1472	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1473	echo "#PBS -A $project_account" >> $job_to_send
1474	echo "#PBS -l mppwidth=1" >> $job_to_send
1475	echo "#PBS -l mppnppn=1" >> $job_to_send
1476	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1477	echo "#PBS -j oe" >> $job_to_send
1478	echo " " >> $job_to_send
1479	echo "set -x" >> $job_to_send
1480	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1481	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1482	echo "%%END%%" >> $job_to_send
1483	else
1484
1485	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1486	echo "# @\\\$-q $return_queue" >> $job_to_send
1487	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1488	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1489	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1490	then
1491	echo "# @\$-l mpp_p=0" >> $job_to_send
1492	fi
1493	echo '# @\$-lF 10mb' >> $job_to_send
1494	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1495	echo '# @\\\$-eo' >> $job_to_send
1496	echo " " >> $job_to_send
1497	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1498	then
1499	echo "set +vx" >> $job_to_send
1500	echo ". .profile" >> $job_to_send
1501	fi
1502	echo "set -x" >> $job_to_send
1503	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
1504	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1505	echo "%%END%%" >> $job_to_send
1506
1507	fi
1508
1509	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1510	then
1511	echo "llsubmit scpjob.$identifier" >> $job_to_send
1512	elif [[ $remote_host = lcbullhh ]]
1513	then
1514	echo "sbatch scpjob.$identifier" >> $job_to_send
1515	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1516	then
1517	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1518	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1519	then
1520	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1521	elif [[ $remote_host = t3es ]]
1522	then
1523	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1524	elif [[ $remote_host = lckiaps ]]
1525	then
1526	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1527	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1528	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1529	elif [[ $remote_host = lckyu* ]]
1530	then
1531	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1532	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1533	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1534	then
1535	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1536	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1537	else
1538	echo "$submcom scpjob.$identifier" >> $job_to_send
1539	fi
1540	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1541	then
1542	echo "rm scpjob.$identifier" >> $job_to_send
1543	fi
1544	if [[ $remote_host = nech ]]
1545	then
1546	echo "cd -" >> $job_to_send
1547	fi
1548	else
1549	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1550	# ??? funktioniert das ÃŒberhaupt noch ???
1551	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1552	fi
1553	echo "set -x" >> $job_to_send
1554	echo " ' exit" >> $job_to_send
1555	echo "set -x" >> $job_to_send
1556	fi
1557
1558
1559	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1560	cat $file_to_send >> $job_to_send
1561
1562	if [[ $remote_host = ibm ]]
1563	then
1564	echo " " >> $job_to_send
1565	echo "exit" >> $job_to_send
1566	fi
1567
1568	# REMOVE JOB-FILE
1569	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1570	then
1571	echo " " >> $job_to_send
1572	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1573	fi
1574
1575
1576	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1577	if [[ $no_submit = false ]]
1578	then
1579	if [[ $remote_host != $local_host ]]
1580	then
1581	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1582	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1583	then
1584	job_catalog_save=$job_catalog
1585	job_catalog=job_queue
1586	elif [[ $remote_host = nech ]]
1587	then
1588	job_catalog_save=$job_catalog
1589	job_catalog=/hpf/b/${remote_user}/job_queue
1590	fi
1591	if [[ $remote_host = nech ]]
1592	then
1593	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1594	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1595	else
1596	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1597	fi
1598	if [[ $? = 1 ]]
1599	then
1600	locat=scp; exit
1601	fi
1602	if [[ $remote_host = ibms ]]
1603	then
1604	job_catalog=$job_catalog_save
1605	fi
1606	[[ $verify = true ]] && printf "\n >>> finished\n"
1607	else
1608	eval job_catalog=$job_catalog
1609	cp $job_to_send ${job_catalog}/$job_on_remhost
1610	fi
1611
1612
1613
1614	# START NQS- / LOADLEVELER-JOB
1615	if [[ $remote_host != $local_host ]]
1616	then
1617	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1618
1619	if [[ $remote_host = ibmku ]]
1620	then
1621	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1622	elif [[ $remote_host = lcflow ]]
1623	then
1624	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1625	else
1626	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1627	fi
1628
1629	[[ $verify = true ]] && printf " >>> o.k.\n"
1630	else
1631	cd $job_catalog
1632	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1633	then
1634	eval $submcom $job_on_remhost
1635	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1636	then
1637	chmod u+x $job_on_remhost
1638	eval $submcom $job_on_remhost
1639	elif [[ $local_host = nech ]]
1640	then
1641	if [[ $queue = default ]]
1642	then
1643	eval $submcom $job_on_remhost
1644	else
1645	eval $submcom -q $queue $job_on_remhost
1646	fi
1647	elif [[ $local_host = lcbullhh ]]
1648	then
1649	if [[ $queue = default ]]
1650	then
1651	eval $submcom $job_on_remhost
1652	fi
1653	else
1654	qsub $job_on_remhost
1655	fi
1656
1657	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1658	# AT THE END OF THE JOB
1659	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1660	then
1661	rm $job_on_remhost
1662	fi
1663	cd - > /dev/null
1664	fi
1665	fi
1666
1667	# FINAL ACTIONS
1668	if [[ $no_submit = false ]]
1669	then
1670	rm -f $job_to_send
1671	fi
1672	[[ $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 |