Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1207

Last change on this file since 1207 was 1204, checked in by witha, 11 years ago
further script adjustments for Forwind cluster (lcflow)
Property svn:keywords set to `Id Rev`
File size: 55.8 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	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1204 2013-07-11 12:50:09Z witha $
30	#
31	# 1202 2013-07-10 16:22:07Z witha
32	# adjustments for Forwind cluster (lcflow)
33	#
34	# 1199 2013-07-05 14:52:22Z raasch
35	# adjustments for CSC Helsinki (lccrayf)
36	#
37	# use of cluster/express queue enabled (ibmh)
38	# vinessa added (imuk)
39	#
40	# 1103 2013-02-20 02:15:53Z raasch
41	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
42	# instead typeset -L/R),
43	# further adjustments for lckyuh
44	#
45	# 2013-02-10 01:47:43Z raasch
46	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
47	# and for Forwind cluster (lcflow)
48	#
49	# 1094 2013-02-03 01:52:12Z raasch
50	# new option -P for explicit setting of ssh/scp port,
51	# decalpha parts (yonsei) removed
52	#
53	# 2013-02-02 07:06:13Z raasch
54	# adjustments for Kyushu-University computing center (lckyut - tatara)
55	# old changelog messages removed
56	#
57	# 1046 2012-11-09 14:38:45Z maronga
58	# code put under GPL (PALM 3.9)
59	#
60	# 08/07/94 - Siggi - first version finished
61	# 29/06/94 - Siggi - script development started
62	#--------------------------------------------------------------------------------#
63	# subjob - script for automatic generation and submission of batch-job files
64	# for various batch queuing systems
65	#--------------------------------------------------------------------------------#
66
67
68	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
69	delete_dayfile=false
70	email_notification=none
71	group_number=none
72	locat=normal
73	no_default_queue=none
74	no_submit=false
75	job_catalog="~/job_queue"
76	job_name=none
77	local_user=$LOGNAME
78	node_usage=shared
79	numprocs=0
80	punkte="..........................................................."
81	submcom=qsub
82	queue=default
83	remote_host=none
84	remote_user=""
85	verify=true
86
87	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
88	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
89
90
91
92	# FEHLERBEHANDLUNG
93	# BEI EXIT:
94	trap 'if [[ $locat != normal ]]
95	then
96	case $locat in
97	(option) printf "\n --> available optios can be displayed"
98	printf " by typing:"
99	printf "\n \"subjob ?\" \n";;
100	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
101	(*) printf "\n +++ unknown error"
102	printf "\n please inform S. Raasch!\n"
103	esac
104	[[ -f $job_to_send ]] && rm $job_to_send
105	printf "\n\n+++ SUBJOB killed \n\n"
106	fi' exit
107
108
109	# BEI TERMINAL-BREAK:
110	trap '[[ -f $job_to_send ]] && rm $job_to_send
111	printf "\n\n+++ SUBJOB killed \n\n"
112	exit
113	' 2
114
115
116
117
118	# LOKALEN HOSTNAMEN FESTSTELLEN
119	local_host=$(hostname)
120
121
122
123	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
124	# UEBERHAUPT ZULAESSIG IST
125	# Note: One of the entries for "lck" or "lckordi" always has to be
126	# comment out, because the hostname (node*) is same for both machines
127	case $local_host in
128	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
129	(atmos) local_addres=172.20.25.35; local_host=lcide;;
130	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
131	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
132	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
133	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
134	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
135	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
136	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
137	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
138	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
139	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
140	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
141	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
142	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
143	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
144	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
145	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
146	(cs*) local_addres=136.172.44.131; local_host=nech;;
147	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
148	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
149	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
150	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
151	(node*) local_addres=165.132.26.61 local_host=lck;;
152	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
153	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
154	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
155	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
156	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
157	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
158	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
159	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
160	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
161	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
162	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
163	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
164	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
165	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
166	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
167	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
168	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
169	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
170	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
171	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
172	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
173	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
174	(meller) local_addres=134.106.74.155; local_host=lcfor;;
175	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
176	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
177	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
178	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
179	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
180	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
181	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
182	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
183	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
184	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
185	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
186	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
187	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
188	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
189	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
190	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
191	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
192	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
193	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
194	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
195	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
196	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
197	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
198	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
199	(*) printf "\n +++ \"$local_host\" unknown";
200	printf "\n please inform S. Raasch!";
201	locat=parameter; exit;;
202	esac
203
204
205
206	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
207	remote_host=$local_host
208
209
210
211
212	# PROZEDUROPTIONEN EINLESEN
213	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
214	do
215	case $option in
216	(c) job_catalog=$OPTARG;;
217	(d) delete_dayfile=true;;
218	(D) no_submit=true;;
219	(e) email_notification=$OPTARG;;
220	(g) group_number=$OPTARG;;
221	(h) remote_host=$OPTARG;;
222	(m) memory=$OPTARG;;
223	(n) job_name=$OPTARG;;
224	(N) node_usage=$OPTARG;;
225	(O) threads_per_task=$OPTARG;;
226	(P) scp_port=$OPTARG;;
227	(q) no_default_queue=$OPTARG;;
228	(t) cputime=$OPTARG;;
229	(T) tasks_per_node=$OPTARG;;
230	(u) remote_user=$OPTARG;;
231	(v) verify=false;;
232	(X) numprocs=$OPTARG;;
233	(\?) printf "\n +++ Option $OPTARG unknown \n";
234	locat=option; exit;;
235	esac
236	done
237
238
239	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
240	(( to_shift = $OPTIND - 1 ))
241	shift $to_shift; file_to_send=$1
242
243
244	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
245	if [ "$1" = "?" ]
246	then
247	(printf "\n *** subjob can be called as follows:\n"
248	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
249	printf "\n Description of available options:\n"
250	printf "\n Option Description Default-Value"
251	printf "\n -c job-input- and output-catalog ~/job_queue"
252	printf "\n -d no job-protocol will be created ---"
253	printf "\n -D only the job-file will be created ---"
254	printf "\n -h execution host, available hosts: $remote_host"
255	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
256	printf "\n lckiaps, lctit, nech, necriam, unics"
257	printf "\n -m memory demand per process in MByte ---"
258	printf "\n -n jobname <jobdatei>"
259	printf "\n -O threads per task (for OpenMP usage) 1"
260	printf "\n -P ssh/scp port default port"
261	printf "\n -q job-queue to be used default"
262	printf "\n -t allowed cpu-time in seconds ---"
263	printf "\n -T tasks per node (on parallel hosts) ---"
264	printf "\n -u username on execution host from .netrc"
265	printf "\n -v no prompt for confirmation ---"
266	printf "\n -X # of processors (on parallel hosts) 1"
267	printf "\n "
268	printf "\n The only possible positional parameter is <jobfile>:"
269	printf "\n The complete NQS-job must be provided here."
270	printf "\n <jobfile>=? creates this outline\n\n") \| more
271	exit
272	fi
273
274
275
276	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
277	if [[ "$file_to_send" = "" ]]
278	then
279	printf "\n +++ job-file missing"
280	locat=parameter; exit
281	else
282	if [[ -f $file_to_send ]]
283	then
284	true
285	else
286	printf "\n +++ job-file: "
287	printf "\n $file_to_send"
288	printf "\n does not exist"
289	locat=parameter; exit
290	fi
291	fi
292
293
294
295	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
296	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
297	if [[ $job_name = none ]]
298	then
299	job_name=$file_to_send
300	fi
301	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
302	then
303	printf "\n +++ job-file name: "
304	printf "\n $job_name"
305	printf "\n must not contain \"/\"-characters"
306	locat=parameter; exit
307	fi
308
309
310
311
312	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
313	# ODER WENN HOST NICHT ANGEGEBEN WURDE
314	if [[ $remote_host = none ]]
315	then
316	printf "\n +++ host missing"
317	locat=option; exit
318	else
319	case $remote_host in
320	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
321	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
322	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
323	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
324	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
325	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
327	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
328	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
329	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
330	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
331	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
332	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
333	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
334	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
335	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
336	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
337	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
338	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
339	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
340	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
341	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
342	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
343	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
344	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
345	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
346	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
347	locat=parameter; exit;;
348	esac
349	fi
350
351
352	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
353	if [[ $no_default_queue != none ]]
354	then
355	error=false
356	ndq=$no_default_queue
357	case $remote_host in
358	(ibm) case $ndq in
359	(p690_express\|p690_standard\|p690_long) error=false;;
360	(*) error=true;;
361	esac;;
362	(ibmh) case $ndq in
363	(cluster\|express) error=false;;
364	(*) error=true;;
365	esac;;
366	(ibmkisti) case $ndq in
367	(class.32plus\|class.1-2\|class.2-32) error=false;;
368	(*) error=true;;
369	esac;;
370	(ibmku) case $ndq in
371	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
372	(*) error=true;;
373	esac;;
374	(ibms) case $ndq in
375	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
376	(*) error=true;;
377	esac;;
378	(ibmy) case $ndq in
379	(parallel) error=false;;
380	(*) error=true;;
381	esac;;
382	(lccrayf) case $ndq in
383	(usup\|test*\|small\|large) error=false;;
384	(*) error=true;;
385	esac;;
386	(lcflow) case $ndq in
387	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
388	(*) error=true;;
389	esac;;
390	(lckiaps) case $ndq in
391	(express\|normal) error=false;;
392	(*) error=true;;
393	esac;;
394	(lckyoto) case $ndq in
395	(eh\|ph) error=false;;
396	(*) error=true;;
397	esac;;
398	(lckyuh) case $ndq in
399	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
400	(*) error=true;;
401	esac;;
402	(lckyut) case $ndq in
403	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
404	(*) error=true;;
405	esac;;
406	(lcsgib\|lcsgih) case $ndq in
407	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
408	(*) error=true;;
409	esac;;
410	(lctit) case $ndq in
411	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
412	(*) error=true;;
413	esac;;
414	(t3eb) case $ndq in
415	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
416	(*) error=true;;
417	esac;;
418	(necriam) case $ndq in
419	(SP\|SS\|P6) error=false;;
420	(*) error=true;;
421	esac;;
422	(t3eh) case $ndq in
423	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
424	(*) error=true;;
425	esac;;
426	(t3ej2\|t3ej5) case $ndq in
427	(low\|normal\|high) error=false;;
428	(*) error=true;;
429	esac;;
430	(t3es) case $ndq in
431	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
432	(*) error=true;;
433	esac;;
434	(unics) case $ndq in
435	(unics\|ht) error=false;;
436	(*) error=true;;
437	esac;;
438	esac
439	if [[ $error = true ]]
440	then
441	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
442	locat=parameter; exit
443	else
444	queue=$no_default_queue
445	fi
446	fi
447
448
449
450	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
451	# AUFTEILEN
452	done=false
453	while [[ $done = false ]]
454	do
455	if (( $cputime <= 0 ))
456	then
457	printf "\n +++ wrong cpu-time or cpu-time missing"
458	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
459	printf "\n >>> "
460	read cputime 1>/dev/null 2>&1
461	else
462	done=true
463	fi
464	done
465	if [[ $remote_host = nech ]]
466	then
467	if (( tasks_per_node != 0 ))
468	then
469	(( cputime = cputime * tasks_per_node ))
470	elif [[ $numprocs != 0 ]]
471	then
472	(( cputime = cputime * numprocs ))
473	fi
474	fi
475	(( stunden = cputime / 3600 ))
476	(( resttime = cputime - stunden * 3600 ))
477	(( minuten = resttime / 60 ))
478	(( sekunden = resttime - minuten * 60 ))
479	timestring=${stunden}:${minuten}:${sekunden}
480
481
482
483	# PRUEFEN DER KERNSPEICHERANFORDERUNG
484	done=false
485	while [[ $done = false ]]
486	do
487	if (( memory <= 0 ))
488	then
489	printf "\n +++ wrong memory demand or memory demand missing"
490	printf "\n >>> Please type memory in MByte per process as INTEGER:"
491	printf "\n >>> "
492	read memory 1>/dev/null 2>&1
493	else
494	done=true
495	fi
496	done
497
498	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
499	then
500	if (( tasks_per_node != 0 ))
501	then
502	(( Memory = memory * tasks_per_node / 1000 ))
503	elif [[ $numprocs != 0 ]]
504	then
505	(( Memory = memory * numprocs / 1000 ))
506	else
507	(( Memory = memory / 1000 ))
508	fi
509	elif [[ $remote_host = lctit ]]
510	then
511	(( Memory = memory * tasks_per_node / 1000 ))
512	fi
513
514
515	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
516	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
517	then
518	(( memory = memory * threads_per_task ))
519	fi
520
521
522	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
523	if (( tasks_per_node != 0 ))
524	then
525	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
526	fi
527
528
529	# Calculate number of processes per node
530	(( processes_per_node = tasks_per_node * threads_per_task ))
531
532
533	# Calculate number of MPI tasks
534	(( mpi_tasks = numprocs / threads_per_task ))
535
536
537	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
538	if [[ "$scp_port" != "" ]]
539	then
540	PORTOPT="-P $scp_port"
541	SSH_PORTOPT="-p $scp_port"
542	fi
543
544
545	# HEADER-AUSGABE
546	if [[ $verify = true ]]
547	then
548	printf "\n\n"
549	printf "#--------------------------------------------------------------# \n"
550	spalte1=SUBJOB;spalte2=$(date)
551	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
552	printf "\| \| \n"
553	printf "\| values of parameters/options: \| \n"
554	spalte1=$(echo local_host$punkte \| cut -c-20)
555	spalte2=$punkte$local_host
556	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
557	spalte1=$(echo remote_host$punkte \| cut -c-20)
558	spalte2=$punkte$remote_host
559	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
560	spalte1=$(echo queue$punkte \| cut -c-20)
561	spalte2=$punkte$queue
562	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
563	spalte1=$(echo memory$punkte \| cut -c-20)
564	spalte2="$punkte$memory mb"
565	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
566	spalte1=$(echo cputime$punkte \| cut -c-20)
567	spalte2="$punkte$cputime sec"
568	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
569	spalte1=$(echo job_name$punkte \| cut -c-20)
570	spalte2="$punkte$job_name"
571	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
572	printf "#--------------------------------------------------------------# \n\n"
573
574
575	# KONTROLLABFRAGE, OB ALLES O.K.
576	antwort="dummy"
577	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
578	do
579	read antwort?" >>> continue (y/n) ? "
580	done
581	if [[ $antwort = n \|\| $antwort = N ]]
582	then
583	locat=verify; exit
584	fi
585	printf "\n"
586	fi
587
588
589
590	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
591	kennung=$RANDOM
592	job_on_remhost=${job_name}_${kennung}_$local_host
593	job_to_send=job_to_send_$kennung
594	if [[ $delete_dayfile = false ]]
595	then
596	remote_dayfile=${local_host}_${job_name}_result_$kennung
597	local_dayfile=${remote_host}_${job_name}
598	else
599	remote_dayfile=/dev/null
600	fi
601
602
603
604	# Generate the batch job scripts (qsub/msub/LoadLeveler)
605	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
606	then
607
608	# General LoadLeveler settings
609	execute_in_shell="#!/bin/ksh"
610	use_shell="# @ shell = /bin/ksh"
611	consumable_memory="ConsumableMemory($memory mb)"
612	class="# @ class = $queue"
613	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
614	network_to_use="# @ network.mpi = sn_all,shared,us"
615	data_limit="# @ data_limit = 1.76gb"
616	image_size="# @ image_size = 50"
617	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
618
619	if [[ $email_notification = none ]]
620	then
621	notify_user=""
622	else
623	notify_user="# @ notify_user = $email_notification"
624	if [[ $delete_dayfile = true ]]
625	then
626	notification='# @ notification = never'
627	fi
628	fi
629
630	if [[ $remote_host = ibmh ]]
631	then
632	data_limit=""
633	network_to_use=""
634	class="# @ class = $queue"
635	environment=""
636	rset="# @ rset = RSET_MCM_AFFINITY"
637	task_affinity="# @ task_affinity = core(1)"
638	elif [[ $remote_host = ibmkisti ]]
639	then
640	network_to_use="# @ network.MPI = sn_all,shared,US"
641	wall_clock_limit="# @ wall_clock_limit = $timestring"
642	if [[ $threads_per_task = 1 ]]
643	then
644	rset="# @ rset = RSET_MCM_AFFINITY"
645	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
646	fi
647	environment=""
648	use_shell=""
649	data_limit=""
650	image_size=""
651	elif [[ $remote_host = ibmku ]]
652	then
653	execute_in_shell="#!/usr/bin/ksh"
654	use_shell="# @ shell = /usr/bin/ksh"
655	consumable_memory=""
656	environment=""
657	network_to_use="# @ network.mpi = sn_all,shared,us"
658	data_limit=""
659	image_size=""
660	elif [[ $remote_host = ibms ]]
661	then
662	network_to_use="# @ network.mpi = csss,shared,us"
663	elif [[ $remote_host = ibmy ]]
664	then
665	consumable_memory=""
666	network_to_use=""
667	fi
668
669	cat > $job_to_send << %%END%%
670	$execute_in_shell
671	$use_shell
672
673	# @ job_type = parallel
674	# @ job_name = $job_name
675	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
676	# @ output = $remote_dayfile
677	# @ error = $remote_dayfile
678	$wall_clock_limit
679	$image_size
680	$class
681	$environment
682	$network_to_use
683	$data_limit
684	$rset
685	$mcm_affinity_options
686	$task_affinity
687	$notification
688	$notify_user
689
690	%%END%%
691
692	if (( nodes > 0 ))
693	then
694
695	if [[ $remote_host != ibmkisti ]]
696	then
697
698	cat >> $job_to_send << %%END%%
699	# @ node = $nodes
700	# @ tasks_per_node = $processes_per_node
701	# @ node_usage = $node_usage
702	# @ queue
703
704	%%END%%
705
706	else
707
708	cat >> $job_to_send << %%END%%
709	# @ total_tasks = $mpi_tasks
710	# @ blocking = unlimited
711	# @ queue
712
713	%%END%%
714
715	fi
716
717	else
718
719	if [[ $remote_host != ibmy ]]
720	then
721
722	cat >> $job_to_send << %%END%%
723	# @ blocking = unlimited
724	# @ total_tasks = $numprocs
725	# @ node_usage = $node_usage
726	# @ queue
727
728	%%END%%
729
730	else
731
732	cat >> $job_to_send << %%END%%
733	# @ node = 1
734	# @ total_tasks = $numprocs
735	# @ queue
736
737	%%END%%
738
739	fi
740
741	fi
742
743	# workaround because of silly job filter on ibmkisti
744	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
745	then
746	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
747	fi
748
749	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
750	then
751
752	cat > $job_to_send << %%END%%
753	#!/bin/ksh
754
755	# @ job_type = serial
756	# @ node_usage = $node_usage
757	# @ job_name = palm
758	# @ wall_clock_limit = ${timestring},$timestring
759	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
760	# @ output = $remote_dayfile
761	# @ error = $remote_dayfile
762	$class
763	$notification
764
765	# @ queue
766
767	%%END%%
768
769	elif [[ $remote_host = lccrayf ]]
770	then
771
772	if [[ $numprocs != 0 ]]
773	then
774	cat > $job_to_send << %%END%%
775	#!/bin/bash -l
776	#SBATCH -J $job_name
777	#SBATCH -t $timestring
778	#SBATCH -N $nodes
779	#SBATCH --ntasks-per-node=$processes_per_node
780	#SBATCH -p $queue
781	#SBATCH -o $remote_dayfile
782	#SBATCH -e $remote_dayfile
783
784	$init_cmds
785	$module_calls
786
787	%%END%%
788
789	else
790	cat > $job_to_send << %%END%%
791	#!/bin/bash -l
792	#SBATCH -J $job_name
793	#SBATCH -t $timestring
794	#SBATCH -l ncpus=1
795	#SBATCH -l pmem=${memory}mb
796	#SBATCH -m abe
797	#SBATCH -o $remote_dayfile
798	#SBATCH -e $remote_dayfile
799
800	$init_cmds
801	$module_calls
802
803	%%END%%
804
805	fi
806
807	elif [[ $remote_host = lcfimm ]]
808	then
809
810	if [[ $numprocs != 0 ]]
811	then
812	cat > $job_to_send << %%END%%
813	#!/bin/ksh
814	#PBS -N $job_name
815	#PBS -A $project_account
816	#PBS -l walltime=$timestring
817	#PBS -l nodes=${nodes}:ppn=$processes_per_node
818	#PBS -l pmem=${memory}mb
819	#PBS -m abe
820	#PBS -o $remote_dayfile
821	#PBS -j oe
822	mpd &
823
824	%%END%%
825
826	else
827	cat > $job_to_send << %%END%%
828	#!/bin/ksh
829	#PBS -N $job_name
830	#PBS -A $project_account
831	#PBS -l walltime=$timestring
832	#PBS -l ncpus=1
833	#PBS -l pmem=${memory}mb
834	#PBS -m abe
835	#PBS -o $remote_dayfile
836	#PBS -j oe
837
838	%%END%%
839
840	fi
841
842	elif [[ $remote_host = lcflow ]]
843	then
844	if [ $memory -gt 1800 ]; then
845	use_himem=""
846	else
847	use_himem="#"
848	fi
849	if [[ $numprocs != 0 ]]
850	then
851	pe_set="#$ -pe impi41 $numprocs"
852	else
853	pe_set="#$ -pe impi41 1"
854	fi
855	if [[ $queue = default ]]
856	then
857	queue_set=""
858	else
859	queue_set="#$ -q $queue"
860	fi
861	[[ "$disc_space" = "" ]] && disc_space=50
862
863	cat > $job_to_send << %%END%%
864	#!/bin/bash
865	#$ -S /bin/bash
866	#$ -N $job_name
867	#$ -cwd
868	#$ -l h_rt=$timestring
869	#$ -l h_vmem=${memory}M
870	#$ -o $remote_dayfile
871	#$ -j y
872	$pe_set
873	#$ -R y
874	${use_himem}#$ -l highmem=true
875	#$ -l h_fsize=${disc_space}G
876	$queue_set
877
878	%%END%%
879
880	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
881	then
882
883	if [[ $numprocs != 0 ]]
884	then
885	cat > $job_to_send << %%END%%
886	#!/bin/ksh
887	#PBS -N $job_name
888	#PBS -l walltime=$timestring
889	#PBS -l ncpus=$numprocs
890	#PBS -l pmem=${memory}mb
891	#PBS -o $remote_dayfile
892	#PBS -l nodes=$nodes:ppn=${processes_per_node}
893	#PBS -j oe
894
895	mpd &
896
897	%%END%%
898
899	else
900	cat > $job_to_send << %%END%%
901	#!/bin/ksh
902	#PBS -N $job_name
903	#PBS -l walltime=$timestring
904	#PBS -l ncpus=1
905	#PBS -l pmem=${memory}mb
906	#PBS -o $remote_dayfile
907	#PBS -j oe
908
909	%%END%%
910
911	fi
912
913	elif [[ $remote_host = lckiaps ]]
914	then
915
916	if [[ $numprocs != 0 ]]
917	then
918	cat > $job_to_send << %%END%%
919	#!/bin/ksh
920	#PBS -N $job_name
921	#PBS -l walltime=$timestring
922	#PBS -l select=1:ncpus=$numprocs
923	#PBS -l pmem=${memory}mb
924	#PBS -q $queue
925	#PBS -o $remote_dayfile
926	#PBS -j oe
927	#PBS -V
928
929	%%END%%
930
931	else
932	cat > $job_to_send << %%END%%
933	#!/bin/ksh
934	#PBS -N $job_name
935	#PBS -l walltime=$timestring
936	#PBS -l ncpus=1
937	#PBS -l pmem=${memory}mb
938	#PBS -o $remote_dayfile
939	#PBS -j oe
940
941	%%END%%
942
943	fi
944
945	elif [[ $remote_host = lcyon ]]
946	then
947
948	if [[ $numprocs != 0 ]]
949	then
950	cat > $job_to_send << %%END%%
951	#!/bin/ksh
952	#PBS -N $job_name
953	#PBS -l walltime=$timestring
954	#PBS -l ncpus=$numprocs
955	#PBS -l pmem=${memory}mb
956	#PBS -o $remote_dayfile
957	#PBS -j oe
958
959	%%END%%
960
961	else
962	cat > $job_to_send << %%END%%
963	#!/bin/ksh
964	#PBS -N $job_name
965	#PBS -l walltime=$timestring
966	#PBS -l ncpus=1
967	#PBS -l pmem=${memory}mb
968	#PBS -o $remote_dayfile
969	#PBS -j oe
970
971	%%END%%
972
973	fi
974
975	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
976	then
977
978	if [[ "$sgi_feature" != "" ]]
979	then
980	feature_directive="#PBS -l feature=$sgi_feature"
981	else
982	feature_directive=""
983	fi
984
985	if [[ $queue = dataq \|\| $queue = permq ]]
986	then
987	feature_directive="#PBS -l feature=data"
988	fi
989
990	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
991	then
992	queue_directive="#PBS -q $queue"
993	else
994	queue_directive=""
995	fi
996
997	if [[ $email_notification = none ]]
998	then
999	email_directive=""
1000	else
1001	email_directive="#PBS -M $email_notification"
1002	fi
1003
1004	if [[ $numprocs != 0 ]]
1005	then
1006	cat > $job_to_send << %%END%%
1007	#!/bin/bash
1008	#PBS -N $job_name
1009	#PBS -l walltime=$timestring
1010	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1011	#PBS -l naccesspolicy=$node_usage
1012	#PBS -o $remote_dayfile
1013	#PBS -j oe
1014	$feature_directive
1015	$queue_directive
1016	$email_directive
1017
1018	eval \`/sw/swdist/bin/modulesinit\`
1019	#. /usr/share/modules/init/bash
1020	$init_cmds
1021	$module_calls
1022
1023	echo ld_library_path=\$LD_LIBRARY_PATH
1024
1025	%%END%%
1026
1027	else
1028	cat > $job_to_send << %%END%%
1029	#PBS -S /bin/bash
1030	#PBS -N $job_name
1031	#PBS -l walltime=$timestring
1032	#PBS -l ncpus=1
1033	#PBS -o $remote_dayfile
1034	#PBS -j oe
1035	$queue_directive
1036	$email_directive
1037
1038	eval \`/sw/swdist/bin/modulesinit\`
1039	#. /usr/share/modules/init/bash
1040	$init_cmds
1041	$module_calls
1042
1043	%%END%%
1044
1045	fi
1046
1047	elif [[ $remote_host = lcxe6 ]]
1048	then
1049
1050	if [[ $numprocs != 0 ]]
1051	then
1052	cat > $job_to_send << %%END%%
1053	#!/bin/ksh
1054	#PBS -S /bin/ksh
1055	#PBS -N $job_name
1056	#PBS -A $project_account
1057	#PBS -j oe
1058	#PBS -l walltime=$timestring
1059	#PBS -l mppwidth=${numprocs}
1060	#PBS -l mppnppn=${processes_per_node}
1061	#PBS -m abe
1062	#PBS -o $remote_dayfile
1063	$email_directive
1064
1065	$init_cmds
1066	$module_calls
1067
1068	%%END%%
1069
1070	else
1071	cat > $job_to_send << %%END%%
1072	#!/bin/ksh
1073	#PBS -S /bin/ksh
1074	#PBS -N $job_name
1075	#PBS -A $project_account
1076	#PBS -j oe
1077	#PBS -l walltime=$timestring
1078	#PBS -l ncpus=1
1079	#PBS -l pmem=${memory}mb
1080	#PBS -m abe
1081	$email_directive
1082	#PBS -o $remote_dayfile
1083
1084	$init_cmds
1085	$module_calls
1086
1087	%%END%%
1088
1089	fi
1090
1091	elif [[ $remote_host = lckyoto ]]
1092	then
1093
1094	cat > $job_to_send << %%END%%
1095	#!/bin/ksh
1096	# @\$-o $remote_dayfile
1097	# @\$-eo -oi
1098	# @\$-lP 16
1099	# @\$-lp 1
1100	# @\$-lm 28gb -llm unlimited -ls unlimited
1101	# @\$-q $queue
1102	# @\$-Pvn abs_pack
1103	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1104	#. /thin/local/etc/setprofile/intel-11.0.sh
1105	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1106	. ~/.myprofile
1107	#. /home2/t/t51254/palm/current_version/myprofile
1108	#. /thin/apps/pgi/mpi.sh
1109	#
1110	env
1111	#
1112	set -x
1113
1114	%%END%%
1115
1116	elif [[ $remote_host = lcxt5m ]]
1117	then
1118
1119	if [[ $numprocs != 0 ]]
1120	then
1121	cat > $job_to_send << %%END%%
1122	#!/bin/ksh
1123	#PBS -S /bin/ksh
1124	#PBS -N $job_name
1125	#PBS -j oe
1126	#PBS -l walltime=$timestring
1127	#PBS -l mppwidth=${numprocs}
1128	#PBS -l mppnppn=${processes_per_node}
1129	#PBS -m abe
1130	#PBS -o $remote_dayfile
1131
1132	$init_cmds
1133	$module_calls
1134
1135	%%END%%
1136
1137	else
1138	cat > $job_to_send << %%END%%
1139	#!/bin/ksh
1140	#PBS -S /bin/ksh
1141	#PBS -N $job_name
1142	#PBS -j oe
1143	#PBS -l walltime=$timestring
1144	#PBS -l ncpus=1
1145	#PBS -l pmem=${memory}mb
1146	#PBS -m abe
1147	#PBS -o $remote_dayfile
1148
1149	$init_cmds
1150	$module_calls
1151
1152	%%END%%
1153
1154	fi
1155
1156	elif [[ $remote_host = lckyuh ]]
1157	then
1158	cat > $job_to_send << %%END%%
1159	#!/bin/bash
1160	#PJM -L "rscgrp=$queue"
1161	#PJM -L "node=$nodes"
1162	#PJM --mpi "proc=$numprocs"
1163	#PJM -L "elapse=$timestring"
1164	#PJM -o $remote_dayfile
1165	#PJM -j
1166	#PJM -X
1167	#PJM --no-stging
1168
1169	export LANG=en_US.UTF-8
1170	%%END%%
1171
1172	elif [[ $remote_host = lckyut ]]
1173	then
1174	cat > $job_to_send << %%END%%
1175	#!/bin/bash
1176	#PJM -L "rscgrp=$queue"
1177	#PJM -L "vnode=$numprocs"
1178	#PJM -L "vnode-core=1"
1179	#PJM -L "elapse=$timestring"
1180	#PJM --mpi proc=$numprocs
1181	#PJM -o $remote_dayfile
1182	#PJM -j
1183	#PJM -X
1184	#PJM --no-stging
1185
1186	export LANG=en_US.UTF-8
1187	%%END%%
1188
1189	elif [[ $remote_host = nech ]]
1190	then
1191
1192	if (( nodes > 1 ))
1193	then
1194	cat > $job_to_send << %%END%%
1195	#!/bin/ksh
1196	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1197	#PBS -l ${qsubmem}=${Memory}gb
1198	#PBS -b $nodes
1199	#PBS -o $remote_dayfile
1200	#PBS -N palm
1201	#PBS -j o
1202	#PBS -T mpisx
1203
1204	%%END%%
1205
1206	elif [[ $numprocs != 0 ]]
1207	then
1208	cat > $job_to_send << %%END%%
1209	#!/bin/ksh
1210	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1211	#PBS -l ${qsubmem}=${Memory}gb
1212	#PBS -o $remote_dayfile
1213	#PBS -N palm
1214	#PBS -j o
1215
1216	%%END%%
1217
1218	else
1219	cat > $job_to_send << %%END%%
1220	#!/bin/ksh
1221	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1222	#PBS -o $remote_dayfile
1223	#PBS -j o
1224
1225	%%END%%
1226
1227	fi
1228
1229	elif [[ $remote_host = necriam ]]
1230	then
1231
1232	if (( nodes > 1 ))
1233	then
1234	cat > $job_to_send << %%END%%
1235	#!/bin/ksh
1236	#PBS -b $nodes
1237	#PBS -o $remote_dayfile
1238	#PBS -N $job_name
1239	#PBS -j o
1240	#PBS -v MPIPROGINV=YES
1241
1242	%%END%%
1243
1244	elif [[ $numprocs != 0 ]]
1245	then
1246	cat > $job_to_send << %%END%%
1247	#!/bin/ksh
1248	#PBS -q ${queue}
1249	#PBS -o $remote_dayfile
1250	#PBS -N $job_name
1251	#PBS -j o
1252	#PBS -v MPIPROGINV=YES
1253
1254	%%END%%
1255
1256	else
1257	cat > $job_to_send << %%END%%
1258	#!/bin/ksh
1259	#PBS -o $remote_dayfile
1260	#PBS -j o
1261
1262	%%END%%
1263
1264	fi
1265
1266	elif [[ $remote_host = lctit ]]
1267	then
1268	cat > $job_to_send << %%END%%
1269	#!/bin/ksh
1270	$init_cmds
1271	$module_calls
1272
1273	%%END%%
1274
1275	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1276	if [[ $tasks_per_node != $processes_per_node ]]
1277	then
1278	submcom="$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 $queue "
1279	else
1280	submcom="$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 "
1281	fi
1282
1283	else
1284
1285	cat > $job_to_send << %%END%%
1286	# @\$-q ${queue}
1287	# @\$-l${qsubtime} $timestring
1288	# @\$-l${qsubmem} ${memory}mb
1289	# @\$-o $remote_dayfile
1290	# @\$-eo
1291
1292	%%END%%
1293
1294	fi
1295
1296
1297	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1298	# BEI EXIT VERANLASSEN
1299	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1300	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1301	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1302	then
1303	echo "set +vx" >> $job_to_send
1304	echo "trap '" >> $job_to_send
1305	echo "set +vx" >> $job_to_send
1306	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* ]]
1307	then
1308	if [[ $remote_host = ibmh ]]
1309	then
1310	return_queue=c1
1311	elif [[ $remote_host = ibmkisti ]]
1312	then
1313	return_queue=class.1-2
1314	elif [[ $remote_host = ibmku ]]
1315	then
1316	return_queue=sdbg2
1317	elif [[ $remote_host = ibms ]]
1318	then
1319	return_queue=p_normal
1320	elif [[ $remote_host = ibmy ]]
1321	then
1322	return_queue=serial
1323	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1324	then
1325	return_queue=serialq
1326	elif [[ $remote_host = necriam ]]
1327	then
1328	return_queue=SP
1329	elif [[ $remote_host = lckiaps ]]
1330	then
1331	return_queue=express
1332	elif [[ $remote_host = lckyuh ]]
1333	then
1334	return_queue=cx-single
1335	elif [[ $remote_host = lckyut ]]
1336	then
1337	return_queue=cx-single
1338	else
1339	return_queue=unknown
1340	fi
1341
1342	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1343	then
1344
1345	if [[ $remote_host = ibmku ]]
1346	then
1347	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1348	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1349	else
1350	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1351	fi
1352	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1353	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1354	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1355	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1356	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1357	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1358	if [[ $host != "ibmh" ]]
1359	then
1360	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1361	fi
1362	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1363	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1364
1365	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1366	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1367
1368	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1369	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
1370	if [[ $remote_host = ibmku ]]
1371	then
1372	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1373	fi
1374	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1375
1376	elif [[ $remote_host = nech ]]
1377	then
1378	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1379	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1380	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1381	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1382	echo "#PBS -j o" >> $job_to_send
1383	echo " " >> $job_to_send
1384	echo "set -x" >> $job_to_send
1385	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1386	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1387	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1388	echo "%%END%%" >> $job_to_send
1389
1390	elif [[ $remote_host = necriam ]]
1391	then
1392	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1393	echo "#PBS -q $return_queue" >> $job_to_send
1394	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1395	echo "#PBS -j o" >> $job_to_send
1396	echo " " >> $job_to_send
1397	echo "set -x" >> $job_to_send
1398	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1399	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1400	echo "%%END%%" >> $job_to_send
1401
1402	elif [[ $remote_host = lckyuh ]]
1403	then
1404	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1405	echo "#!/bin/bash" >> $job_to_send
1406	echo "#PJM -L \"node=1\"" >> $job_to_send
1407	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1408	echo "#PJM --no-stging" >> $job_to_send
1409	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1410	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1411	echo "#PJM -j" >> $job_to_send
1412	echo " " >> $job_to_send
1413	echo "export LANG=en_US.UTF-8" >> $job_to_send
1414	echo "set -x" >> $job_to_send
1415	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1416	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1417	echo "%%END%%" >> $job_to_send
1418
1419	elif [[ $remote_host = lckyut ]]
1420	then
1421	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1422	echo "#!/bin/bash" >> $job_to_send
1423	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1424	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1425	echo "#PJM --no-stging" >> $job_to_send
1426	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1427	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1428	echo "#PJM -j" >> $job_to_send
1429	echo " " >> $job_to_send
1430	echo "export LANG=en_US.UTF-8" >> $job_to_send
1431	echo "set -x" >> $job_to_send
1432	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1433	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1434	echo "%%END%%" >> $job_to_send
1435
1436	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1437	then
1438	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1439	echo "#!/bin/bash" >> $job_to_send
1440	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1441	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1442	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1443	echo "#PBS -l feature=data" >> $job_to_send
1444	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1445	echo "#PBS -j oe" >> $job_to_send
1446	echo " " >> $job_to_send
1447	echo ". /usr/share/modules/init/bash" >> $job_to_send
1448	echo "set -x" >> $job_to_send
1449	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1450	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1451	echo "%%END%%" >> $job_to_send
1452
1453	elif [[ $remote_host = lcflow ]]
1454	then
1455	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1456	echo "#!/bin/bash" >> $job_to_send
1457	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1458	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1459	echo "SGEPREFIX -cwd" >> $job_to_send
1460	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1461	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1462	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1463	echo "SGEPREFIX -j y" >> $job_to_send
1464	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1465	echo " " >> $job_to_send
1466	echo "set -x" >> $job_to_send
1467	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1468	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1469	echo "" >> $job_to_send
1470	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1471	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1472	echo "rm -f scpjob.${kennung}" >> $job_to_send
1473	echo "%%END%%" >> $job_to_send
1474	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1475	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1476
1477	else
1478
1479	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1480	echo "# @\\\$-q $return_queue" >> $job_to_send
1481	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1482	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1483	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1484	then
1485	echo "# @\$-l mpp_p=0" >> $job_to_send
1486	fi
1487	echo '# @\$-lF 10mb' >> $job_to_send
1488	# echo '# @\$-o /dev/null' >> $job_to_send
1489	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1490	echo '# @\\\$-eo' >> $job_to_send
1491	echo " " >> $job_to_send
1492	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1493	then
1494	echo "set +vx" >> $job_to_send
1495	echo ". .profile" >> $job_to_send
1496	fi
1497	echo "set -x" >> $job_to_send
1498	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
1499	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1500	echo "%%END%%" >> $job_to_send
1501	fi
1502
1503	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1504	then
1505	echo "llsubmit scpjob.$kennung" >> $job_to_send
1506	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1507	then
1508	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1509	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1510	echo "msub scpjob.$kennung" >> $job_to_send
1511	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1512	then
1513	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1514	elif [[ $remote_host = t3es ]]
1515	then
1516	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1517	elif [[ $remote_host = lckiaps ]]
1518	then
1519	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1520	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1521	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1522	elif [[ $remote_host = lckyu* ]]
1523	then
1524	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1525	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1526	elif [[ $remote_host = lcflow ]]
1527	then
1528	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1529	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1530	else
1531	echo "$submcom scpjob.$kennung" >> $job_to_send
1532	fi
1533	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1534	then
1535	echo "rm scpjob.$kennung" >> $job_to_send
1536	fi
1537	if [[ $remote_host = nech ]]
1538	then
1539	echo "cd -" >> $job_to_send
1540	fi
1541	else
1542	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1543	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1544	fi
1545	echo "set -x" >> $job_to_send
1546	echo " ' exit" >> $job_to_send
1547	echo "set -x" >> $job_to_send
1548	fi
1549
1550
1551
1552	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1553	cat $file_to_send >> $job_to_send
1554
1555	if [[ $remote_host = ibm ]]
1556	then
1557	echo " " >> $job_to_send
1558	echo "exit" >> $job_to_send
1559	fi
1560
1561	# remove job file
1562	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1563	then
1564	echo " " >> $job_to_send
1565	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1566	fi
1567
1568
1569
1570	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1571	if [[ -z $remote_user ]]
1572	then
1573	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1574	then
1575	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1576	fi
1577	fi
1578
1579
1580
1581	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1582	if [[ $no_submit = false ]]
1583	then
1584	if [[ $remote_host != $local_host ]]
1585	then
1586	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1587	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1588	then
1589	job_catalog_save=$job_catalog
1590	job_catalog=job_queue
1591	elif [[ $remote_host = nech ]]
1592	then
1593	job_catalog_save=$job_catalog
1594	job_catalog=/hpf/b/${remote_user}/job_queue
1595	fi
1596	if [[ $remote_host = nech ]]
1597	then
1598	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1599	# TRANSFERIERT WERDEN
1600	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1601	else
1602	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1603	fi
1604	if [[ $? = 1 ]]
1605	then
1606	locat=scp; exit
1607	fi
1608	if [[ $remote_host = ibms ]]
1609	then
1610	job_catalog=$job_catalog_save
1611	fi
1612	[[ $verify = true ]] && printf "\n >>> finished\n"
1613	else
1614	eval job_catalog=$job_catalog
1615	cp $job_to_send ${job_catalog}/$job_on_remhost
1616	fi
1617
1618
1619
1620	# NQS- BZW. LOADLEVELER-JOB STARTEN
1621	if [[ $remote_host != $local_host ]]
1622	then
1623	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1624
1625	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1626	then
1627	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1628	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1629	elif [[ $remote_host = ibmku ]]
1630	then
1631	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1632	elif [[ $remote_host = lcflow ]]
1633	then
1634	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1635	else
1636	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1637	fi
1638
1639	[[ $verify = true ]] && printf " >>> o.k.\n"
1640	else
1641	cd $job_catalog
1642	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $local_host = lccrayf ]]
1643	then
1644	eval $submcom $job_on_remhost
1645	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1646	then
1647	chmod u+x $job_on_remhost
1648	eval $submcom $job_on_remhost
1649	elif [[ $local_host = nech ]]
1650	then
1651	if [[ $queue = default ]]
1652	then
1653	eval $submcom $job_on_remhost
1654	else
1655	eval $submcom -q $queue $job_on_remhost
1656	fi
1657	else
1658	qsub $job_on_remhost
1659	fi
1660
1661	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1662	# only at the end of the job.
1663	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1664	then
1665	rm $job_on_remhost
1666	fi
1667	cd - > /dev/null
1668	fi
1669	fi
1670
1671
1672
1673	# ABSCHLUSSARBEITEN
1674
1675	if [[ $no_submit = false ]]
1676	then
1677	rm -f $job_to_send
1678	fi
1679	[[ $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 |