Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1199

Last change on this file since 1199 was 1199, checked in by raasch, 11 years ago
further script adjustments for CSC Helsinki (lccrayf), executables for batch jobs can be created in advance, in order to avoid calling the compiler within the batch job
Property svn:keywords set to `Id Rev`
File size: 55.4 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2012 Leibniz University Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	# adjustments for CSC Helsinki (lccrayf)
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1199 2013-07-05 14:52:22Z raasch $
30	# use of cluster/express queue enabled (ibmh)
31	# vinessa added (imuk)
32	#
33	# 1103 2013-02-20 02:15:53Z raasch
34	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
35	# instead typeset -L/R),
36	# further adjustments for lckyuh
37	#
38	# 2013-02-10 01:47:43Z raasch
39	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
40	# and for Forwind cluster (lcflow)
41	#
42	# 1094 2013-02-03 01:52:12Z raasch
43	# new option -P for explicit setting of ssh/scp port,
44	# decalpha parts (yonsei) removed
45	#
46	# 2013-02-02 07:06:13Z raasch
47	# adjustments for Kyushu-University computing center (lckyut - tatara)
48	# old changelog messages removed
49	#
50	# 1046 2012-11-09 14:38:45Z maronga
51	# code put under GPL (PALM 3.9)
52	#
53	# 08/07/94 - Siggi - first version finished
54	# 29/06/94 - Siggi - script development started
55	#--------------------------------------------------------------------------------#
56	# subjob - script for automatic generation and submission of batch-job files
57	# for various batch queuing systems
58	#--------------------------------------------------------------------------------#
59
60
61	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
62	delete_dayfile=false
63	email_notification=none
64	group_number=none
65	locat=normal
66	no_default_queue=none
67	no_submit=false
68	job_catalog="~/job_queue"
69	job_name=none
70	local_user=$LOGNAME
71	node_usage=shared
72	numprocs=0
73	punkte="..........................................................."
74	submcom=qsub
75	queue=default
76	remote_host=none
77	remote_user=""
78	verify=true
79
80	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
81	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
82
83
84
85	# FEHLERBEHANDLUNG
86	# BEI EXIT:
87	trap 'if [[ $locat != normal ]]
88	then
89	case $locat in
90	(option) printf "\n --> available optios can be displayed"
91	printf " by typing:"
92	printf "\n \"subjob ?\" \n";;
93	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
94	(*) printf "\n +++ unknown error"
95	printf "\n please inform S. Raasch!\n"
96	esac
97	[[ -f $job_to_send ]] && rm $job_to_send
98	printf "\n\n+++ SUBJOB killed \n\n"
99	fi' exit
100
101
102	# BEI TERMINAL-BREAK:
103	trap '[[ -f $job_to_send ]] && rm $job_to_send
104	printf "\n\n+++ SUBJOB killed \n\n"
105	exit
106	' 2
107
108
109
110
111	# LOKALEN HOSTNAMEN FESTSTELLEN
112	local_host=$(hostname)
113
114
115
116	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
117	# UEBERHAUPT ZULAESSIG IST
118	# Note: One of the entries for "lck" or "lckordi" always has to be
119	# comment out, because the hostname (node*) is same for both machines
120	case $local_host in
121	(atmos) local_addres=172.20.25.35; local_host=lcide;;
122	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
123	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
124	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
125	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
126	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
127	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
128	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
129	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
130	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
131	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
132	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
133	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
134	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
135	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
136	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
137	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
138	(cs*) local_addres=136.172.44.131; local_host=nech;;
139	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
140	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
141	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
142	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
143	(node*) local_addres=165.132.26.61 local_host=lck;;
144	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
145	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
146	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
147	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
148	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
149	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
150	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
151	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
152	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
153	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
154	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
155	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
156	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
157	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
158	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
159	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
160	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
161	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
162	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
163	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
164	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
165	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
166	(meller) local_addres=134.106.74.155; local_host=lcfor;;
167	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
168	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
169	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
170	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
171	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
172	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
173	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
174	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
175	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
176	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
177	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
178	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
179	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
180	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
181	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
182	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
183	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
184	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
185	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
186	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
187	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
188	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
189	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
190	(*) printf "\n +++ \"$local_host\" unknown";
191	printf "\n please inform S. Raasch!";
192	locat=parameter; exit;;
193	esac
194
195
196
197	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
198	remote_host=$local_host
199
200
201
202
203	# PROZEDUROPTIONEN EINLESEN
204	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
205	do
206	case $option in
207	(c) job_catalog=$OPTARG;;
208	(d) delete_dayfile=true;;
209	(D) no_submit=true;;
210	(e) email_notification=$OPTARG;;
211	(g) group_number=$OPTARG;;
212	(h) remote_host=$OPTARG;;
213	(m) memory=$OPTARG;;
214	(n) job_name=$OPTARG;;
215	(N) node_usage=$OPTARG;;
216	(O) threads_per_task=$OPTARG;;
217	(P) scp_port=$OPTARG;;
218	(q) no_default_queue=$OPTARG;;
219	(t) cputime=$OPTARG;;
220	(T) tasks_per_node=$OPTARG;;
221	(u) remote_user=$OPTARG;;
222	(v) verify=false;;
223	(X) numprocs=$OPTARG;;
224	(\?) printf "\n +++ Option $OPTARG unknown \n";
225	locat=option; exit;;
226	esac
227	done
228
229
230	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
231	(( to_shift = $OPTIND - 1 ))
232	shift $to_shift; file_to_send=$1
233
234
235	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
236	if [ "$1" = "?" ]
237	then
238	(printf "\n *** subjob can be called as follows:\n"
239	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
240	printf "\n Description of available options:\n"
241	printf "\n Option Description Default-Value"
242	printf "\n -c job-input- and output-catalog ~/job_queue"
243	printf "\n -d no job-protocol will be created ---"
244	printf "\n -D only the job-file will be created ---"
245	printf "\n -h execution host, available hosts: $remote_host"
246	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
247	printf "\n lckiaps, lctit, nech, necriam, unics"
248	printf "\n -m memory demand per process in MByte ---"
249	printf "\n -n jobname <jobdatei>"
250	printf "\n -O threads per task (for OpenMP usage) 1"
251	printf "\n -P ssh/scp port default port"
252	printf "\n -q job-queue to be used default"
253	printf "\n -t allowed cpu-time in seconds ---"
254	printf "\n -T tasks per node (on parallel hosts) ---"
255	printf "\n -u username on execution host from .netrc"
256	printf "\n -v no prompt for confirmation ---"
257	printf "\n -X # of processors (on parallel hosts) 1"
258	printf "\n "
259	printf "\n The only possible positional parameter is <jobfile>:"
260	printf "\n The complete NQS-job must be provided here."
261	printf "\n <jobfile>=? creates this outline\n\n") \| more
262	exit
263	fi
264
265
266
267	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
268	if [[ "$file_to_send" = "" ]]
269	then
270	printf "\n +++ job-file missing"
271	locat=parameter; exit
272	else
273	if [[ -f $file_to_send ]]
274	then
275	true
276	else
277	printf "\n +++ job-file: "
278	printf "\n $file_to_send"
279	printf "\n does not exist"
280	locat=parameter; exit
281	fi
282	fi
283
284
285
286	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
287	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
288	if [[ $job_name = none ]]
289	then
290	job_name=$file_to_send
291	fi
292	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
293	then
294	printf "\n +++ job-file name: "
295	printf "\n $job_name"
296	printf "\n must not contain \"/\"-characters"
297	locat=parameter; exit
298	fi
299
300
301
302
303	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
304	# ODER WENN HOST NICHT ANGEGEBEN WURDE
305	if [[ $remote_host = none ]]
306	then
307	printf "\n +++ host missing"
308	locat=option; exit
309	else
310	case $remote_host in
311	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
312	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
313	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
314	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
315	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
316	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
317	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
318	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
319	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
320	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
321	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
322	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
323	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
324	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
325	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
326	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
327	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
328	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
329	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
330	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
331	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
332	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
333	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
334	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
335	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
336	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
337	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
338	locat=parameter; exit;;
339	esac
340	fi
341
342
343	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
344	if [[ $no_default_queue != none ]]
345	then
346	error=false
347	ndq=$no_default_queue
348	case $remote_host in
349	(ibm) case $ndq in
350	(p690_express\|p690_standard\|p690_long) error=false;;
351	(*) error=true;;
352	esac;;
353	(ibmh) case $ndq in
354	(cluster\|express) error=false;;
355	(*) error=true;;
356	esac;;
357	(ibmkisti) case $ndq in
358	(class.32plus\|class.1-2\|class.2-32) error=false;;
359	(*) error=true;;
360	esac;;
361	(ibmku) case $ndq in
362	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
363	(*) error=true;;
364	esac;;
365	(ibms) case $ndq in
366	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
367	(*) error=true;;
368	esac;;
369	(ibmy) case $ndq in
370	(parallel) error=false;;
371	(*) error=true;;
372	esac;;
373	(lccrayf) case $ndq in
374	(usup\|test*\|small\|large) error=false;;
375	(*) error=true;;
376	esac;;
377	(lcflow) case $ndq in
378	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
379	(*) error=true;;
380	esac;;
381	(lckiaps) case $ndq in
382	(express\|normal) error=false;;
383	(*) error=true;;
384	esac;;
385	(lckyoto) case $ndq in
386	(eh\|ph) error=false;;
387	(*) error=true;;
388	esac;;
389	(lckyuh) case $ndq in
390	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
391	(*) error=true;;
392	esac;;
393	(lckyut) case $ndq in
394	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
395	(*) error=true;;
396	esac;;
397	(lcsgib\|lcsgih) case $ndq in
398	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
399	(*) error=true;;
400	esac;;
401	(lctit) case $ndq in
402	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
403	(*) error=true;;
404	esac;;
405	(t3eb) case $ndq in
406	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
407	(*) error=true;;
408	esac;;
409	(necriam) case $ndq in
410	(SP\|SS\|P6) error=false;;
411	(*) error=true;;
412	esac;;
413	(t3eh) case $ndq in
414	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
415	(*) error=true;;
416	esac;;
417	(t3ej2\|t3ej5) case $ndq in
418	(low\|normal\|high) error=false;;
419	(*) error=true;;
420	esac;;
421	(t3es) case $ndq in
422	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
423	(*) error=true;;
424	esac;;
425	(unics) case $ndq in
426	(unics\|ht) error=false;;
427	(*) error=true;;
428	esac;;
429	esac
430	if [[ $error = true ]]
431	then
432	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
433	locat=parameter; exit
434	else
435	queue=$no_default_queue
436	fi
437	fi
438
439
440
441	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
442	# AUFTEILEN
443	done=false
444	while [[ $done = false ]]
445	do
446	if (( $cputime <= 0 ))
447	then
448	printf "\n +++ wrong cpu-time or cpu-time missing"
449	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
450	printf "\n >>> "
451	read cputime 1>/dev/null 2>&1
452	else
453	done=true
454	fi
455	done
456	if [[ $remote_host = nech ]]
457	then
458	if (( tasks_per_node != 0 ))
459	then
460	(( cputime = cputime * tasks_per_node ))
461	elif [[ $numprocs != 0 ]]
462	then
463	(( cputime = cputime * numprocs ))
464	fi
465	fi
466	(( stunden = cputime / 3600 ))
467	(( resttime = cputime - stunden * 3600 ))
468	(( minuten = resttime / 60 ))
469	(( sekunden = resttime - minuten * 60 ))
470	timestring=${stunden}:${minuten}:${sekunden}
471
472
473
474	# PRUEFEN DER KERNSPEICHERANFORDERUNG
475	done=false
476	while [[ $done = false ]]
477	do
478	if (( memory <= 0 ))
479	then
480	printf "\n +++ wrong memory demand or memory demand missing"
481	printf "\n >>> Please type memory in MByte per process as INTEGER:"
482	printf "\n >>> "
483	read memory 1>/dev/null 2>&1
484	else
485	done=true
486	fi
487	done
488
489	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
490	then
491	if (( tasks_per_node != 0 ))
492	then
493	(( Memory = memory * tasks_per_node / 1000 ))
494	elif [[ $numprocs != 0 ]]
495	then
496	(( Memory = memory * numprocs / 1000 ))
497	else
498	(( Memory = memory / 1000 ))
499	fi
500	elif [[ $remote_host = lctit ]]
501	then
502	(( Memory = memory * tasks_per_node / 1000 ))
503	fi
504
505
506	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
507	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
508	then
509	(( memory = memory * threads_per_task ))
510	fi
511
512
513	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
514	if (( tasks_per_node != 0 ))
515	then
516	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
517	fi
518
519
520	# Calculate number of processes per node
521	(( processes_per_node = tasks_per_node * threads_per_task ))
522
523
524	# Calculate number of MPI tasks
525	(( mpi_tasks = numprocs / threads_per_task ))
526
527
528	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
529	if [[ "$scp_port" != "" ]]
530	then
531	PORTOPT="-P $scp_port"
532	SSH_PORTOPT="-p $scp_port"
533	fi
534
535
536	# HEADER-AUSGABE
537	if [[ $verify = true ]]
538	then
539	printf "\n\n"
540	printf "#--------------------------------------------------------------# \n"
541	spalte1=SUBJOB;spalte2=$(date)
542	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
543	printf "\| \| \n"
544	printf "\| values of parameters/options: \| \n"
545	spalte1=$(echo local_host$punkte \| cut -c-20)
546	spalte2=$punkte$local_host
547	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
548	spalte1=$(echo remote_host$punkte \| cut -c-20)
549	spalte2=$punkte$remote_host
550	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
551	spalte1=$(echo queue$punkte \| cut -c-20)
552	spalte2=$punkte$queue
553	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
554	spalte1=$(echo memory$punkte \| cut -c-20)
555	spalte2="$punkte$memory mb"
556	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
557	spalte1=$(echo cputime$punkte \| cut -c-20)
558	spalte2="$punkte$cputime sec"
559	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
560	spalte1=$(echo job_name$punkte \| cut -c-20)
561	spalte2="$punkte$job_name"
562	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
563	printf "#--------------------------------------------------------------# \n\n"
564
565
566	# KONTROLLABFRAGE, OB ALLES O.K.
567	antwort="dummy"
568	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
569	do
570	read antwort?" >>> continue (y/n) ? "
571	done
572	if [[ $antwort = n \|\| $antwort = N ]]
573	then
574	locat=verify; exit
575	fi
576	printf "\n"
577	fi
578
579
580
581	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
582	kennung=$RANDOM
583	job_on_remhost=${job_name}_${kennung}_$local_host
584	job_to_send=job_to_send_$kennung
585	if [[ $delete_dayfile = false ]]
586	then
587	remote_dayfile=${local_host}_${job_name}_result_$kennung
588	local_dayfile=${remote_host}_${job_name}
589	else
590	remote_dayfile=/dev/null
591	fi
592
593
594
595	# Generate the batch job scripts (qsub/msub/LoadLeveler)
596	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
597	then
598
599	# General LoadLeveler settings
600	execute_in_shell="#!/bin/ksh"
601	use_shell="# @ shell = /bin/ksh"
602	consumable_memory="ConsumableMemory($memory mb)"
603	class="# @ class = $queue"
604	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
605	network_to_use="# @ network.mpi = sn_all,shared,us"
606	data_limit="# @ data_limit = 1.76gb"
607	image_size="# @ image_size = 50"
608	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
609
610	if [[ $email_notification = none ]]
611	then
612	notify_user=""
613	else
614	notify_user="# @ notify_user = $email_notification"
615	if [[ $delete_dayfile = true ]]
616	then
617	notification='# @ notification = never'
618	fi
619	fi
620
621	if [[ $remote_host = ibmh ]]
622	then
623	data_limit=""
624	network_to_use=""
625	class="# @ class = $queue"
626	environment=""
627	rset="# @ rset = RSET_MCM_AFFINITY"
628	task_affinity="# @ task_affinity = core(1)"
629	elif [[ $remote_host = ibmkisti ]]
630	then
631	network_to_use="# @ network.MPI = sn_all,shared,US"
632	wall_clock_limit="# @ wall_clock_limit = $timestring"
633	if [[ $threads_per_task = 1 ]]
634	then
635	rset="# @ rset = RSET_MCM_AFFINITY"
636	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
637	fi
638	environment=""
639	use_shell=""
640	data_limit=""
641	image_size=""
642	elif [[ $remote_host = ibmku ]]
643	then
644	execute_in_shell="#!/usr/bin/ksh"
645	use_shell="# @ shell = /usr/bin/ksh"
646	consumable_memory=""
647	environment=""
648	network_to_use="# @ network.mpi = sn_all,shared,us"
649	data_limit=""
650	image_size=""
651	elif [[ $remote_host = ibms ]]
652	then
653	network_to_use="# @ network.mpi = csss,shared,us"
654	elif [[ $remote_host = ibmy ]]
655	then
656	consumable_memory=""
657	network_to_use=""
658	fi
659
660	cat > $job_to_send << %%END%%
661	$execute_in_shell
662	$use_shell
663
664	# @ job_type = parallel
665	# @ job_name = $job_name
666	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
667	# @ output = $remote_dayfile
668	# @ error = $remote_dayfile
669	$wall_clock_limit
670	$image_size
671	$class
672	$environment
673	$network_to_use
674	$data_limit
675	$rset
676	$mcm_affinity_options
677	$task_affinity
678	$notification
679	$notify_user
680
681	%%END%%
682
683	if (( nodes > 0 ))
684	then
685
686	if [[ $remote_host != ibmkisti ]]
687	then
688
689	cat >> $job_to_send << %%END%%
690	# @ node = $nodes
691	# @ tasks_per_node = $processes_per_node
692	# @ node_usage = $node_usage
693	# @ queue
694
695	%%END%%
696
697	else
698
699	cat >> $job_to_send << %%END%%
700	# @ total_tasks = $mpi_tasks
701	# @ blocking = unlimited
702	# @ queue
703
704	%%END%%
705
706	fi
707
708	else
709
710	if [[ $remote_host != ibmy ]]
711	then
712
713	cat >> $job_to_send << %%END%%
714	# @ blocking = unlimited
715	# @ total_tasks = $numprocs
716	# @ node_usage = $node_usage
717	# @ queue
718
719	%%END%%
720
721	else
722
723	cat >> $job_to_send << %%END%%
724	# @ node = 1
725	# @ total_tasks = $numprocs
726	# @ queue
727
728	%%END%%
729
730	fi
731
732	fi
733
734	# workaround because of silly job filter on ibmkisti
735	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
736	then
737	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
738	fi
739
740	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
741	then
742
743	cat > $job_to_send << %%END%%
744	#!/bin/ksh
745
746	# @ job_type = serial
747	# @ node_usage = $node_usage
748	# @ job_name = palm
749	# @ wall_clock_limit = ${timestring},$timestring
750	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
751	# @ output = $remote_dayfile
752	# @ error = $remote_dayfile
753	$class
754	$notification
755
756	# @ queue
757
758	%%END%%
759
760	elif [[ $remote_host = lccrayf ]]
761	then
762
763	if [[ $numprocs != 0 ]]
764	then
765	cat > $job_to_send << %%END%%
766	#!/bin/bash -l
767	#SBATCH -J $job_name
768	#SBATCH -t $timestring
769	#SBATCH -N $nodes
770	#SBATCH --ntasks-per-node=$processes_per_node
771	#SBATCH -p $queue
772	#SBATCH -o $remote_dayfile
773	#SBATCH -e $remote_dayfile
774
775	$init_cmds
776	$module_calls
777
778	%%END%%
779
780	else
781	cat > $job_to_send << %%END%%
782	#!/bin/bash -l
783	#SBATCH -J $job_name
784	#SBATCH -t $timestring
785	#SBATCH -l ncpus=1
786	#SBATCH -l pmem=${memory}mb
787	#SBATCH -m abe
788	#SBATCH -o $remote_dayfile
789	#SBATCH -e $remote_dayfile
790
791	$init_cmds
792	$module_calls
793
794	%%END%%
795
796	fi
797
798	elif [[ $remote_host = lcfimm ]]
799	then
800
801	if [[ $numprocs != 0 ]]
802	then
803	cat > $job_to_send << %%END%%
804	#!/bin/ksh
805	#PBS -N $job_name
806	#PBS -A $project_account
807	#PBS -l walltime=$timestring
808	#PBS -l nodes=${nodes}:ppn=$processes_per_node
809	#PBS -l pmem=${memory}mb
810	#PBS -m abe
811	#PBS -o $remote_dayfile
812	#PBS -j oe
813	mpd &
814
815	%%END%%
816
817	else
818	cat > $job_to_send << %%END%%
819	#!/bin/ksh
820	#PBS -N $job_name
821	#PBS -A $project_account
822	#PBS -l walltime=$timestring
823	#PBS -l ncpus=1
824	#PBS -l pmem=${memory}mb
825	#PBS -m abe
826	#PBS -o $remote_dayfile
827	#PBS -j oe
828
829	%%END%%
830
831	fi
832
833	elif [[ $remote_host = lcflow ]]
834	then
835	if [ $memory -gt 1900 ]; then
836	use_himem=""
837	else
838	use_himem="#"
839	fi
840	if [[ $numprocs != 0 ]]
841	then
842	pe_set="#$ -pe impi $numprocs"
843	else
844	pe_set="#$ -pe impi 1"
845	fi
846	if [[ $queue = default ]]
847	then
848	queue_set=""
849	else
850	queue_set="#$ -q $queue"
851	fi
852	[[ "$disc_space" = "" ]] && disc_space=50
853
854	cat > $job_to_send << %%END%%
855	#!/bin/bash
856	#$ -S /bin/bash
857	#$ -N $job_name
858	#$ -cwd
859	#$ -l h_rt=$timestring
860	#$ -l h_vmem=${memory}M
861	#$ -o $remote_dayfile
862	#$ -j y
863	$pe_set
864	#$ -R y
865	${use_himem}#$ -l highmem=true
866	#$ -l h_fsize=${disc_space}G
867	$queue_set
868
869	%%END%%
870
871	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
872	then
873
874	if [[ $numprocs != 0 ]]
875	then
876	cat > $job_to_send << %%END%%
877	#!/bin/ksh
878	#PBS -N $job_name
879	#PBS -l walltime=$timestring
880	#PBS -l ncpus=$numprocs
881	#PBS -l pmem=${memory}mb
882	#PBS -o $remote_dayfile
883	#PBS -l nodes=$nodes:ppn=${processes_per_node}
884	#PBS -j oe
885
886	mpd &
887
888	%%END%%
889
890	else
891	cat > $job_to_send << %%END%%
892	#!/bin/ksh
893	#PBS -N $job_name
894	#PBS -l walltime=$timestring
895	#PBS -l ncpus=1
896	#PBS -l pmem=${memory}mb
897	#PBS -o $remote_dayfile
898	#PBS -j oe
899
900	%%END%%
901
902	fi
903
904	elif [[ $remote_host = lckiaps ]]
905	then
906
907	if [[ $numprocs != 0 ]]
908	then
909	cat > $job_to_send << %%END%%
910	#!/bin/ksh
911	#PBS -N $job_name
912	#PBS -l walltime=$timestring
913	#PBS -l select=1:ncpus=$numprocs
914	#PBS -l pmem=${memory}mb
915	#PBS -q $queue
916	#PBS -o $remote_dayfile
917	#PBS -j oe
918	#PBS -V
919
920	%%END%%
921
922	else
923	cat > $job_to_send << %%END%%
924	#!/bin/ksh
925	#PBS -N $job_name
926	#PBS -l walltime=$timestring
927	#PBS -l ncpus=1
928	#PBS -l pmem=${memory}mb
929	#PBS -o $remote_dayfile
930	#PBS -j oe
931
932	%%END%%
933
934	fi
935
936	elif [[ $remote_host = lcyon ]]
937	then
938
939	if [[ $numprocs != 0 ]]
940	then
941	cat > $job_to_send << %%END%%
942	#!/bin/ksh
943	#PBS -N $job_name
944	#PBS -l walltime=$timestring
945	#PBS -l ncpus=$numprocs
946	#PBS -l pmem=${memory}mb
947	#PBS -o $remote_dayfile
948	#PBS -j oe
949
950	%%END%%
951
952	else
953	cat > $job_to_send << %%END%%
954	#!/bin/ksh
955	#PBS -N $job_name
956	#PBS -l walltime=$timestring
957	#PBS -l ncpus=1
958	#PBS -l pmem=${memory}mb
959	#PBS -o $remote_dayfile
960	#PBS -j oe
961
962	%%END%%
963
964	fi
965
966	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
967	then
968
969	if [[ "$sgi_feature" != "" ]]
970	then
971	feature_directive="#PBS -l feature=$sgi_feature"
972	else
973	feature_directive=""
974	fi
975
976	if [[ $queue = dataq \|\| $queue = permq ]]
977	then
978	feature_directive="#PBS -l feature=data"
979	fi
980
981	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
982	then
983	queue_directive="#PBS -q $queue"
984	else
985	queue_directive=""
986	fi
987
988	if [[ $email_notification = none ]]
989	then
990	email_directive=""
991	else
992	email_directive="#PBS -M $email_notification"
993	fi
994
995	if [[ $numprocs != 0 ]]
996	then
997	cat > $job_to_send << %%END%%
998	#!/bin/bash
999	#PBS -N $job_name
1000	#PBS -l walltime=$timestring
1001	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1002	#PBS -l naccesspolicy=$node_usage
1003	#PBS -o $remote_dayfile
1004	#PBS -j oe
1005	$feature_directive
1006	$queue_directive
1007	$email_directive
1008
1009	eval \`/sw/swdist/bin/modulesinit\`
1010	#. /usr/share/modules/init/bash
1011	$init_cmds
1012	$module_calls
1013
1014	echo ld_library_path=\$LD_LIBRARY_PATH
1015
1016	%%END%%
1017
1018	else
1019	cat > $job_to_send << %%END%%
1020	#PBS -S /bin/bash
1021	#PBS -N $job_name
1022	#PBS -l walltime=$timestring
1023	#PBS -l ncpus=1
1024	#PBS -o $remote_dayfile
1025	#PBS -j oe
1026	$queue_directive
1027	$email_directive
1028
1029	eval \`/sw/swdist/bin/modulesinit\`
1030	#. /usr/share/modules/init/bash
1031	$init_cmds
1032	$module_calls
1033
1034	%%END%%
1035
1036	fi
1037
1038	elif [[ $remote_host = lcxe6 ]]
1039	then
1040
1041	if [[ $numprocs != 0 ]]
1042	then
1043	cat > $job_to_send << %%END%%
1044	#!/bin/ksh
1045	#PBS -S /bin/ksh
1046	#PBS -N $job_name
1047	#PBS -A $project_account
1048	#PBS -j oe
1049	#PBS -l walltime=$timestring
1050	#PBS -l mppwidth=${numprocs}
1051	#PBS -l mppnppn=${processes_per_node}
1052	#PBS -m abe
1053	#PBS -o $remote_dayfile
1054	$email_directive
1055
1056	$init_cmds
1057	$module_calls
1058
1059	%%END%%
1060
1061	else
1062	cat > $job_to_send << %%END%%
1063	#!/bin/ksh
1064	#PBS -S /bin/ksh
1065	#PBS -N $job_name
1066	#PBS -A $project_account
1067	#PBS -j oe
1068	#PBS -l walltime=$timestring
1069	#PBS -l ncpus=1
1070	#PBS -l pmem=${memory}mb
1071	#PBS -m abe
1072	$email_directive
1073	#PBS -o $remote_dayfile
1074
1075	$init_cmds
1076	$module_calls
1077
1078	%%END%%
1079
1080	fi
1081
1082	elif [[ $remote_host = lckyoto ]]
1083	then
1084
1085	cat > $job_to_send << %%END%%
1086	#!/bin/ksh
1087	# @\$-o $remote_dayfile
1088	# @\$-eo -oi
1089	# @\$-lP 16
1090	# @\$-lp 1
1091	# @\$-lm 28gb -llm unlimited -ls unlimited
1092	# @\$-q $queue
1093	# @\$-Pvn abs_pack
1094	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1095	#. /thin/local/etc/setprofile/intel-11.0.sh
1096	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1097	. ~/.myprofile
1098	#. /home2/t/t51254/palm/current_version/myprofile
1099	#. /thin/apps/pgi/mpi.sh
1100	#
1101	env
1102	#
1103	set -x
1104
1105	%%END%%
1106
1107	elif [[ $remote_host = lcxt5m ]]
1108	then
1109
1110	if [[ $numprocs != 0 ]]
1111	then
1112	cat > $job_to_send << %%END%%
1113	#!/bin/ksh
1114	#PBS -S /bin/ksh
1115	#PBS -N $job_name
1116	#PBS -j oe
1117	#PBS -l walltime=$timestring
1118	#PBS -l mppwidth=${numprocs}
1119	#PBS -l mppnppn=${processes_per_node}
1120	#PBS -m abe
1121	#PBS -o $remote_dayfile
1122
1123	$init_cmds
1124	$module_calls
1125
1126	%%END%%
1127
1128	else
1129	cat > $job_to_send << %%END%%
1130	#!/bin/ksh
1131	#PBS -S /bin/ksh
1132	#PBS -N $job_name
1133	#PBS -j oe
1134	#PBS -l walltime=$timestring
1135	#PBS -l ncpus=1
1136	#PBS -l pmem=${memory}mb
1137	#PBS -m abe
1138	#PBS -o $remote_dayfile
1139
1140	$init_cmds
1141	$module_calls
1142
1143	%%END%%
1144
1145	fi
1146
1147	elif [[ $remote_host = lckyuh ]]
1148	then
1149	cat > $job_to_send << %%END%%
1150	#!/bin/bash
1151	#PJM -L "rscgrp=$queue"
1152	#PJM -L "node=$nodes"
1153	#PJM --mpi "proc=$numprocs"
1154	#PJM -L "elapse=$timestring"
1155	#PJM -o $remote_dayfile
1156	#PJM -j
1157	#PJM -X
1158	#PJM --no-stging
1159
1160	export LANG=en_US.UTF-8
1161	%%END%%
1162
1163	elif [[ $remote_host = lckyut ]]
1164	then
1165	cat > $job_to_send << %%END%%
1166	#!/bin/bash
1167	#PJM -L "rscgrp=$queue"
1168	#PJM -L "vnode=$numprocs"
1169	#PJM -L "vnode-core=1"
1170	#PJM -L "elapse=$timestring"
1171	#PJM --mpi proc=$numprocs
1172	#PJM -o $remote_dayfile
1173	#PJM -j
1174	#PJM -X
1175	#PJM --no-stging
1176
1177	export LANG=en_US.UTF-8
1178	%%END%%
1179
1180	elif [[ $remote_host = nech ]]
1181	then
1182
1183	if (( nodes > 1 ))
1184	then
1185	cat > $job_to_send << %%END%%
1186	#!/bin/ksh
1187	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1188	#PBS -l ${qsubmem}=${Memory}gb
1189	#PBS -b $nodes
1190	#PBS -o $remote_dayfile
1191	#PBS -N palm
1192	#PBS -j o
1193	#PBS -T mpisx
1194
1195	%%END%%
1196
1197	elif [[ $numprocs != 0 ]]
1198	then
1199	cat > $job_to_send << %%END%%
1200	#!/bin/ksh
1201	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1202	#PBS -l ${qsubmem}=${Memory}gb
1203	#PBS -o $remote_dayfile
1204	#PBS -N palm
1205	#PBS -j o
1206
1207	%%END%%
1208
1209	else
1210	cat > $job_to_send << %%END%%
1211	#!/bin/ksh
1212	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1213	#PBS -o $remote_dayfile
1214	#PBS -j o
1215
1216	%%END%%
1217
1218	fi
1219
1220	elif [[ $remote_host = necriam ]]
1221	then
1222
1223	if (( nodes > 1 ))
1224	then
1225	cat > $job_to_send << %%END%%
1226	#!/bin/ksh
1227	#PBS -b $nodes
1228	#PBS -o $remote_dayfile
1229	#PBS -N $job_name
1230	#PBS -j o
1231	#PBS -v MPIPROGINV=YES
1232
1233	%%END%%
1234
1235	elif [[ $numprocs != 0 ]]
1236	then
1237	cat > $job_to_send << %%END%%
1238	#!/bin/ksh
1239	#PBS -q ${queue}
1240	#PBS -o $remote_dayfile
1241	#PBS -N $job_name
1242	#PBS -j o
1243	#PBS -v MPIPROGINV=YES
1244
1245	%%END%%
1246
1247	else
1248	cat > $job_to_send << %%END%%
1249	#!/bin/ksh
1250	#PBS -o $remote_dayfile
1251	#PBS -j o
1252
1253	%%END%%
1254
1255	fi
1256
1257	elif [[ $remote_host = lctit ]]
1258	then
1259	cat > $job_to_send << %%END%%
1260	#!/bin/ksh
1261	$init_cmds
1262	$module_calls
1263
1264	%%END%%
1265
1266	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1267	if [[ $tasks_per_node != $processes_per_node ]]
1268	then
1269	submcom="$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 -q $queue "
1270	else
1271	submcom="$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 -q $queue "
1272	fi
1273
1274	else
1275
1276	cat > $job_to_send << %%END%%
1277	# @\$-q ${queue}
1278	# @\$-l${qsubtime} $timestring
1279	# @\$-l${qsubmem} ${memory}mb
1280	# @\$-o $remote_dayfile
1281	# @\$-eo
1282
1283	%%END%%
1284
1285	fi
1286
1287
1288	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1289	# BEI EXIT VERANLASSEN
1290	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1291	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1292	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1293	then
1294	echo "set +vx" >> $job_to_send
1295	echo "trap '" >> $job_to_send
1296	echo "set +vx" >> $job_to_send
1297	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* ]]
1298	then
1299	if [[ $remote_host = ibmh ]]
1300	then
1301	return_queue=c1
1302	elif [[ $remote_host = ibmkisti ]]
1303	then
1304	return_queue=class.1-2
1305	elif [[ $remote_host = ibmku ]]
1306	then
1307	return_queue=sdbg2
1308	elif [[ $remote_host = ibms ]]
1309	then
1310	return_queue=p_normal
1311	elif [[ $remote_host = ibmy ]]
1312	then
1313	return_queue=serial
1314	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1315	then
1316	return_queue=serialq
1317	elif [[ $remote_host = necriam ]]
1318	then
1319	return_queue=SP
1320	elif [[ $remote_host = lckiaps ]]
1321	then
1322	return_queue=express
1323	elif [[ $remote_host = lckyuh ]]
1324	then
1325	return_queue=cx-single
1326	elif [[ $remote_host = lckyut ]]
1327	then
1328	return_queue=cx-single
1329	else
1330	return_queue=unknown
1331	fi
1332
1333	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1334	then
1335
1336	if [[ $remote_host = ibmku ]]
1337	then
1338	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1339	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1340	else
1341	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1342	fi
1343	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1344	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1345	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1346	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1347	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1348	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1349	if [[ $host != "ibmh" ]]
1350	then
1351	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1352	fi
1353	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1354	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1355
1356	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1357	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1358
1359	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1360	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1361	if [[ $remote_host = ibmku ]]
1362	then
1363	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1364	fi
1365	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1366
1367	elif [[ $remote_host = nech ]]
1368	then
1369	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1370	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1371	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1372	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1373	echo "#PBS -j o" >> $job_to_send
1374	echo " " >> $job_to_send
1375	echo "set -x" >> $job_to_send
1376	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1377	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1378	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1379	echo "%%END%%" >> $job_to_send
1380
1381	elif [[ $remote_host = necriam ]]
1382	then
1383	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1384	echo "#PBS -q $return_queue" >> $job_to_send
1385	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1386	echo "#PBS -j o" >> $job_to_send
1387	echo " " >> $job_to_send
1388	echo "set -x" >> $job_to_send
1389	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1390	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1391	echo "%%END%%" >> $job_to_send
1392
1393	elif [[ $remote_host = lckyuh ]]
1394	then
1395	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1396	echo "#!/bin/bash" >> $job_to_send
1397	echo "#PJM -L \"node=1\"" >> $job_to_send
1398	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1399	echo "#PJM --no-stging" >> $job_to_send
1400	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1401	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1402	echo "#PJM -j" >> $job_to_send
1403	echo " " >> $job_to_send
1404	echo "export LANG=en_US.UTF-8" >> $job_to_send
1405	echo "set -x" >> $job_to_send
1406	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1407	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1408	echo "%%END%%" >> $job_to_send
1409
1410	elif [[ $remote_host = lckyut ]]
1411	then
1412	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1413	echo "#!/bin/bash" >> $job_to_send
1414	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1415	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1416	echo "#PJM --no-stging" >> $job_to_send
1417	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1418	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1419	echo "#PJM -j" >> $job_to_send
1420	echo " " >> $job_to_send
1421	echo "export LANG=en_US.UTF-8" >> $job_to_send
1422	echo "set -x" >> $job_to_send
1423	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1424	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1425	echo "%%END%%" >> $job_to_send
1426
1427	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1428	then
1429	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1430	echo "#!/bin/bash" >> $job_to_send
1431	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1432	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1433	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1434	echo "#PBS -l feature=data" >> $job_to_send
1435	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1436	echo "#PBS -j oe" >> $job_to_send
1437	echo " " >> $job_to_send
1438	echo ". /usr/share/modules/init/bash" >> $job_to_send
1439	echo "set -x" >> $job_to_send
1440	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1441	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1442	echo "%%END%%" >> $job_to_send
1443
1444	elif [[ $remote_host = lcflow ]]
1445	then
1446	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1447	echo "#!/bin/bash" >> $job_to_send
1448	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1449	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1450	echo "SGEPREFIX -cwd" >> $job_to_send
1451	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1452	echo "SGEPREFIX -l h_vmem=100M" >> $job_to_send
1453	echo "SGEPREFIX -j y" >> $job_to_send
1454	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1455	echo " " >> $job_to_send
1456	echo "set -x" >> $job_to_send
1457	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1458	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1459	echo "" >> $job_to_send
1460	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1461	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1462	echo "rm -f scpjob.${kennung}" >> $job_to_send
1463	echo "%%END%%" >> $job_to_send
1464	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1465	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1466
1467	else
1468
1469	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1470	echo "# @\\\$-q $return_queue" >> $job_to_send
1471	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1472	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1473	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1474	then
1475	echo "# @\$-l mpp_p=0" >> $job_to_send
1476	fi
1477	echo '# @\$-lF 10mb' >> $job_to_send
1478	# echo '# @\$-o /dev/null' >> $job_to_send
1479	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1480	echo '# @\\\$-eo' >> $job_to_send
1481	echo " " >> $job_to_send
1482	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1483	then
1484	echo "set +vx" >> $job_to_send
1485	echo ". .profile" >> $job_to_send
1486	fi
1487	echo "set -x" >> $job_to_send
1488	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1489	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1490	echo "%%END%%" >> $job_to_send
1491	fi
1492
1493	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1494	then
1495	echo "llsubmit scpjob.$kennung" >> $job_to_send
1496	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1497	then
1498	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1499	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1500	echo "msub scpjob.$kennung" >> $job_to_send
1501	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1502	then
1503	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1504	elif [[ $remote_host = t3es ]]
1505	then
1506	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1507	elif [[ $remote_host = lckiaps ]]
1508	then
1509	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1510	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1511	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1512	elif [[ $remote_host = lckyu* ]]
1513	then
1514	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1515	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1516	elif [[ $remote_host = lcflow ]]
1517	then
1518	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1519	echo "ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1520	else
1521	echo "$submcom scpjob.$kennung" >> $job_to_send
1522	fi
1523	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1524	then
1525	echo "rm scpjob.$kennung" >> $job_to_send
1526	fi
1527	if [[ $remote_host = nech ]]
1528	then
1529	echo "cd -" >> $job_to_send
1530	fi
1531	else
1532	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1533	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1534	fi
1535	echo "set -x" >> $job_to_send
1536	echo " ' exit" >> $job_to_send
1537	echo "set -x" >> $job_to_send
1538	fi
1539
1540
1541
1542	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1543	cat $file_to_send >> $job_to_send
1544
1545	if [[ $remote_host = ibm ]]
1546	then
1547	echo " " >> $job_to_send
1548	echo "exit" >> $job_to_send
1549	fi
1550
1551	# remove job file
1552	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1553	then
1554	echo " " >> $job_to_send
1555	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1556	fi
1557
1558
1559
1560	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1561	if [[ -z $remote_user ]]
1562	then
1563	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1564	then
1565	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1566	fi
1567	fi
1568
1569
1570
1571	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1572	if [[ $no_submit = false ]]
1573	then
1574	if [[ $remote_host != $local_host ]]
1575	then
1576	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1577	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1578	then
1579	job_catalog_save=$job_catalog
1580	job_catalog=job_queue
1581	elif [[ $remote_host = nech ]]
1582	then
1583	job_catalog_save=$job_catalog
1584	job_catalog=/hpf/b/${remote_user}/job_queue
1585	fi
1586	if [[ $remote_host = nech ]]
1587	then
1588	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1589	# TRANSFERIERT WERDEN
1590	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1591	else
1592	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1593	fi
1594	if [[ $? = 1 ]]
1595	then
1596	locat=scp; exit
1597	fi
1598	if [[ $remote_host = ibms ]]
1599	then
1600	job_catalog=$job_catalog_save
1601	fi
1602	[[ $verify = true ]] && printf "\n >>> finished\n"
1603	else
1604	eval job_catalog=$job_catalog
1605	cp $job_to_send ${job_catalog}/$job_on_remhost
1606	fi
1607
1608
1609
1610	# NQS- BZW. LOADLEVELER-JOB STARTEN
1611	if [[ $remote_host != $local_host ]]
1612	then
1613	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1614
1615	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1616	then
1617	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1618	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1619	elif [[ $remote_host = ibmku ]]
1620	then
1621	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1622	elif [[ $remote_host = lcflow ]]
1623	then
1624	ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1625	else
1626	ssh $SSH_PORTOPT $remote_addres -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 -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $local_host = lccrayf ]]
1633	then
1634	eval $submcom $job_on_remhost
1635	elif [[ $local_host = lcfimm \|\| $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	else
1648	qsub $job_on_remhost
1649	fi
1650
1651	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1652	# only at the end of the job.
1653	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1654	then
1655	rm $job_on_remhost
1656	fi
1657	cd - > /dev/null
1658	fi
1659	fi
1660
1661
1662
1663	# ABSCHLUSSARBEITEN
1664
1665	if [[ $no_submit = false ]]
1666	then
1667	rm -f $job_to_send
1668	fi
1669	[[ $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 |