Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1249

Last change on this file since 1249 was 1235, checked in by keck, 11 years ago
added new configuration file for cuhk
Property svn:keywords set to `Id Rev`
File size: 56.7 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 1235 2013-09-25 07:10:22Z heinze $
30	#
31	# 1224 2013-09-16 07:27:23Z raasch
32	# first adjustments for lccrayb
33	#
34	# 1202 2013-07-10 16:22:07Z witha
35	# adjustments for Forwind cluster (lcflow)
36	#
37	# 1199 2013-07-05 14:52:22Z raasch
38	# adjustments for CSC Helsinki (lccrayf)
39	#
40	# use of cluster/express queue enabled (ibmh)
41	# vinessa added (imuk)
42	#
43	# 1103 2013-02-20 02:15:53Z raasch
44	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
45	# instead typeset -L/R),
46	# further adjustments for lckyuh
47	#
48	# 2013-02-10 01:47:43Z raasch
49	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
50	# and for Forwind cluster (lcflow)
51	#
52	# 1094 2013-02-03 01:52:12Z raasch
53	# new option -P for explicit setting of ssh/scp port,
54	# decalpha parts (yonsei) removed
55	#
56	# 2013-02-02 07:06:13Z raasch
57	# adjustments for Kyushu-University computing center (lckyut - tatara)
58	# old changelog messages removed
59	#
60	# 1046 2012-11-09 14:38:45Z maronga
61	# code put under GPL (PALM 3.9)
62	#
63	# 08/07/94 - Siggi - first version finished
64	# 29/06/94 - Siggi - script development started
65	#--------------------------------------------------------------------------------#
66	# subjob - script for automatic generation and submission of batch-job files
67	# for various batch queuing systems
68	#--------------------------------------------------------------------------------#
69
70
71	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
72	delete_dayfile=false
73	email_notification=none
74	group_number=none
75	locat=normal
76	no_default_queue=none
77	no_submit=false
78	job_catalog="~/job_queue"
79	job_name=none
80	local_user=$LOGNAME
81	node_usage=shared
82	numprocs=0
83	punkte="..........................................................."
84	submcom=qsub
85	queue=default
86	remote_host=none
87	remote_user=""
88	verify=true
89
90	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
91	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
92
93
94
95	# FEHLERBEHANDLUNG
96	# BEI EXIT:
97	trap 'if [[ $locat != normal ]]
98	then
99	case $locat in
100	(option) printf "\n --> available optios can be displayed"
101	printf " by typing:"
102	printf "\n \"subjob ?\" \n";;
103	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
104	(*) printf "\n +++ unknown error"
105	printf "\n please inform S. Raasch!\n"
106	esac
107	[[ -f $job_to_send ]] && rm $job_to_send
108	printf "\n\n+++ SUBJOB killed \n\n"
109	fi' exit
110
111
112	# BEI TERMINAL-BREAK:
113	trap '[[ -f $job_to_send ]] && rm $job_to_send
114	printf "\n\n+++ SUBJOB killed \n\n"
115	exit
116	' 2
117
118
119
120
121	# LOKALEN HOSTNAMEN FESTSTELLEN
122	local_host=$(hostname)
123
124
125
126	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
127	# UEBERHAUPT ZULAESSIG IST
128	# Note: One of the entries for "lck" or "lckordi" always has to be
129	# comment out, because the hostname (node*) is same for both machines
130	case $local_host in
131	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
132	(atmos) local_addres=172.20.25.35; local_host=lcide;;
133	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
134	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
135	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
136	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
137	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
138	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
139	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
140	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
141	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
142	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
143	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
144	(blogin*) local_addres=130.73.233.1; local_host=lccrayb;;
145	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
146	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
147	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
148	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
149	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
150	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
151	(cs*) local_addres=136.172.44.131; local_host=nech;;
152	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
153	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
154	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
155	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
156	(node*) local_addres=165.132.26.61 local_host=lck;;
157	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
158	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
159	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
160	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
161	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
162	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
163	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
164	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
165	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
166	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
167	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
168	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
169	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
170	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
171	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
172	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
173	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
174	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
175	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
176	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
177	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
178	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
179	(meller) local_addres=134.106.74.155; local_host=lcfor;;
180	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
181	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
182	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
183	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
184	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
185	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
186	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
187	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
188	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
189	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
190	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
191	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
192	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
193	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
194	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
195	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
196	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
197	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
198	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
199	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
200	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
201	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
202	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
203	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
204	(*) printf "\n +++ \"$local_host\" unknown";
205	printf "\n please inform S. Raasch!";
206	locat=parameter; exit;;
207	esac
208
209
210
211	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
212	remote_host=$local_host
213
214
215
216
217	# PROZEDUROPTIONEN EINLESEN
218	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
219	do
220	case $option in
221	(c) job_catalog=$OPTARG;;
222	(d) delete_dayfile=true;;
223	(D) no_submit=true;;
224	(e) email_notification=$OPTARG;;
225	(g) group_number=$OPTARG;;
226	(h) remote_host=$OPTARG;;
227	(m) memory=$OPTARG;;
228	(n) job_name=$OPTARG;;
229	(N) node_usage=$OPTARG;;
230	(O) threads_per_task=$OPTARG;;
231	(P) scp_port=$OPTARG;;
232	(q) no_default_queue=$OPTARG;;
233	(t) cputime=$OPTARG;;
234	(T) tasks_per_node=$OPTARG;;
235	(u) remote_user=$OPTARG;;
236	(v) verify=false;;
237	(X) numprocs=$OPTARG;;
238	(\?) printf "\n +++ Option $OPTARG unknown \n";
239	locat=option; exit;;
240	esac
241	done
242
243
244	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
245	(( to_shift = $OPTIND - 1 ))
246	shift $to_shift; file_to_send=$1
247
248
249	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
250	if [ "$1" = "?" ]
251	then
252	(printf "\n *** subjob can be called as follows:\n"
253	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
254	printf "\n Description of available options:\n"
255	printf "\n Option Description Default-Value"
256	printf "\n -c job-input- and output-catalog ~/job_queue"
257	printf "\n -d no job-protocol will be created ---"
258	printf "\n -D only the job-file will be created ---"
259	printf "\n -h execution host, available hosts: $remote_host"
260	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
261	printf "\n lckiaps, lctit, nech, necriam, unics"
262	printf "\n -m memory demand per process in MByte ---"
263	printf "\n -n jobname <jobdatei>"
264	printf "\n -O threads per task (for OpenMP usage) 1"
265	printf "\n -P ssh/scp port default port"
266	printf "\n -q job-queue to be used default"
267	printf "\n -t allowed cpu-time in seconds ---"
268	printf "\n -T tasks per node (on parallel hosts) ---"
269	printf "\n -u username on execution host from .netrc"
270	printf "\n -v no prompt for confirmation ---"
271	printf "\n -X # of processors (on parallel hosts) 1"
272	printf "\n "
273	printf "\n The only possible positional parameter is <jobfile>:"
274	printf "\n The complete NQS-job must be provided here."
275	printf "\n <jobfile>=? creates this outline\n\n") \| more
276	exit
277	fi
278
279
280
281	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
282	if [[ "$file_to_send" = "" ]]
283	then
284	printf "\n +++ job-file missing"
285	locat=parameter; exit
286	else
287	if [[ -f $file_to_send ]]
288	then
289	true
290	else
291	printf "\n +++ job-file: "
292	printf "\n $file_to_send"
293	printf "\n does not exist"
294	locat=parameter; exit
295	fi
296	fi
297
298
299
300	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
301	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
302	if [[ $job_name = none ]]
303	then
304	job_name=$file_to_send
305	fi
306	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
307	then
308	printf "\n +++ job-file name: "
309	printf "\n $job_name"
310	printf "\n must not contain \"/\"-characters"
311	locat=parameter; exit
312	fi
313
314
315
316
317	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
318	# ODER WENN HOST NICHT ANGEGEBEN WURDE
319	if [[ $remote_host = none ]]
320	then
321	printf "\n +++ host missing"
322	locat=option; exit
323	else
324	case $remote_host in
325	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
327	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
328	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
329	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
331	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
332	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
333	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
334	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
335	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
336	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
337	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
338	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
339	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
340	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
341	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
342	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
343	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
344	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
345	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
346	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
347	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
348	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
349	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
350	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
351	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
352	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
353	locat=parameter; exit;;
354	esac
355	fi
356
357
358	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
359	if [[ $no_default_queue != none ]]
360	then
361	error=false
362	ndq=$no_default_queue
363	case $remote_host in
364	(ibm) case $ndq in
365	(p690_express\|p690_standard\|p690_long) error=false;;
366	(*) error=true;;
367	esac;;
368	(ibmh) case $ndq in
369	(cluster\|express) error=false;;
370	(*) error=true;;
371	esac;;
372	(ibmkisti) case $ndq in
373	(class.32plus\|class.1-2\|class.2-32) error=false;;
374	(*) error=true;;
375	esac;;
376	(ibmku) case $ndq in
377	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
378	(*) error=true;;
379	esac;;
380	(ibms) case $ndq in
381	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
382	(*) error=true;;
383	esac;;
384	(ibmy) case $ndq in
385	(parallel) error=false;;
386	(*) error=true;;
387	esac;;
388	(lccrayb) case $ndq in
389	(mpp1q\|mpp1testq\|testq) error=false;;
390	(*) error=true;;
391	esac;;
392	(lccrayf) case $ndq in
393	(usup\|test*\|small\|large) error=false;;
394	(*) error=true;;
395	esac;;
396	(lcflow) case $ndq in
397	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
398	(*) error=true;;
399	esac;;
400	(lckiaps) case $ndq in
401	(express\|normal) error=false;;
402	(*) error=true;;
403	esac;;
404	(lckyoto) case $ndq in
405	(eh\|ph) error=false;;
406	(*) error=true;;
407	esac;;
408	(lckyuh) case $ndq in
409	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
410	(*) error=true;;
411	esac;;
412	(lckyut) case $ndq in
413	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
414	(*) error=true;;
415	esac;;
416	(lcsgib\|lcsgih) case $ndq in
417	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
418	(*) error=true;;
419	esac;;
420	(lctit) case $ndq in
421	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
422	(*) error=true;;
423	esac;;
424	(t3eb) case $ndq in
425	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
426	(*) error=true;;
427	esac;;
428	(necriam) case $ndq in
429	(SP\|SS\|P6) error=false;;
430	(*) error=true;;
431	esac;;
432	(t3eh) case $ndq in
433	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
434	(*) error=true;;
435	esac;;
436	(t3ej2\|t3ej5) case $ndq in
437	(low\|normal\|high) error=false;;
438	(*) error=true;;
439	esac;;
440	(t3es) case $ndq in
441	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
442	(*) error=true;;
443	esac;;
444	(unics) case $ndq in
445	(unics\|ht) error=false;;
446	(*) error=true;;
447	esac;;
448	esac
449	if [[ $error = true ]]
450	then
451	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
452	locat=parameter; exit
453	else
454	queue=$no_default_queue
455	fi
456	fi
457
458
459
460	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
461	# AUFTEILEN
462	done=false
463	while [[ $done = false ]]
464	do
465	if (( $cputime <= 0 ))
466	then
467	printf "\n +++ wrong cpu-time or cpu-time missing"
468	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
469	printf "\n >>> "
470	read cputime 1>/dev/null 2>&1
471	else
472	done=true
473	fi
474	done
475	if [[ $remote_host = nech ]]
476	then
477	if (( tasks_per_node != 0 ))
478	then
479	(( cputime = cputime * tasks_per_node ))
480	elif [[ $numprocs != 0 ]]
481	then
482	(( cputime = cputime * numprocs ))
483	fi
484	fi
485	(( stunden = cputime / 3600 ))
486	(( resttime = cputime - stunden * 3600 ))
487	(( minuten = resttime / 60 ))
488	(( sekunden = resttime - minuten * 60 ))
489	timestring=${stunden}:${minuten}:${sekunden}
490
491
492
493	# PRUEFEN DER KERNSPEICHERANFORDERUNG
494	done=false
495	while [[ $done = false ]]
496	do
497	if (( memory <= 0 ))
498	then
499	printf "\n +++ wrong memory demand or memory demand missing"
500	printf "\n >>> Please type memory in MByte per process as INTEGER:"
501	printf "\n >>> "
502	read memory 1>/dev/null 2>&1
503	else
504	done=true
505	fi
506	done
507
508	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
509	then
510	if (( tasks_per_node != 0 ))
511	then
512	(( Memory = memory * tasks_per_node / 1000 ))
513	elif [[ $numprocs != 0 ]]
514	then
515	(( Memory = memory * numprocs / 1000 ))
516	else
517	(( Memory = memory / 1000 ))
518	fi
519	elif [[ $remote_host = lctit ]]
520	then
521	(( Memory = memory * tasks_per_node / 1000 ))
522	fi
523
524
525	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
526	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
527	then
528	(( memory = memory * threads_per_task ))
529	fi
530
531
532	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
533	if (( tasks_per_node != 0 ))
534	then
535	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
536	fi
537
538
539	# Calculate number of processes per node
540	(( processes_per_node = tasks_per_node * threads_per_task ))
541
542
543	# Calculate number of MPI tasks
544	(( mpi_tasks = numprocs / threads_per_task ))
545
546
547	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
548	if [[ "$scp_port" != "" ]]
549	then
550	PORTOPT="-P $scp_port"
551	SSH_PORTOPT="-p $scp_port"
552	fi
553
554
555	# HEADER-AUSGABE
556	if [[ $verify = true ]]
557	then
558	printf "\n\n"
559	printf "#--------------------------------------------------------------# \n"
560	spalte1=SUBJOB;spalte2=$(date)
561	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
562	printf "\| \| \n"
563	printf "\| values of parameters/options: \| \n"
564	spalte1=$(echo local_host$punkte \| cut -c-20)
565	spalte2=$punkte$local_host
566	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
567	spalte1=$(echo remote_host$punkte \| cut -c-20)
568	spalte2=$punkte$remote_host
569	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
570	spalte1=$(echo queue$punkte \| cut -c-20)
571	spalte2=$punkte$queue
572	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
573	spalte1=$(echo memory$punkte \| cut -c-20)
574	spalte2="$punkte$memory mb"
575	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
576	spalte1=$(echo cputime$punkte \| cut -c-20)
577	spalte2="$punkte$cputime sec"
578	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
579	spalte1=$(echo job_name$punkte \| cut -c-20)
580	spalte2="$punkte$job_name"
581	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
582	printf "#--------------------------------------------------------------# \n\n"
583
584
585	# KONTROLLABFRAGE, OB ALLES O.K.
586	antwort="dummy"
587	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
588	do
589	read antwort?" >>> continue (y/n) ? "
590	done
591	if [[ $antwort = n \|\| $antwort = N ]]
592	then
593	locat=verify; exit
594	fi
595	printf "\n"
596	fi
597
598
599
600	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
601	kennung=$RANDOM
602	job_on_remhost=${job_name}_${kennung}_$local_host
603	job_to_send=job_to_send_$kennung
604	if [[ $delete_dayfile = false ]]
605	then
606	remote_dayfile=${local_host}_${job_name}_result_$kennung
607	local_dayfile=${remote_host}_${job_name}
608	else
609	remote_dayfile=/dev/null
610	fi
611
612
613
614	# Generate the batch job scripts (qsub/msub/LoadLeveler)
615	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
616	then
617
618	# General LoadLeveler settings
619	execute_in_shell="#!/bin/ksh"
620	use_shell="# @ shell = /bin/ksh"
621	consumable_memory="ConsumableMemory($memory mb)"
622	class="# @ class = $queue"
623	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
624	network_to_use="# @ network.mpi = sn_all,shared,us"
625	data_limit="# @ data_limit = 1.76gb"
626	image_size="# @ image_size = 50"
627	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
628
629	if [[ $email_notification = none ]]
630	then
631	notify_user=""
632	else
633	notify_user="# @ notify_user = $email_notification"
634	if [[ $delete_dayfile = true ]]
635	then
636	notification='# @ notification = never'
637	fi
638	fi
639
640	if [[ $remote_host = ibmh ]]
641	then
642	data_limit=""
643	network_to_use=""
644	class="# @ class = $queue"
645	environment=""
646	rset="# @ rset = RSET_MCM_AFFINITY"
647	task_affinity="# @ task_affinity = core(1)"
648	elif [[ $remote_host = ibmkisti ]]
649	then
650	network_to_use="# @ network.MPI = sn_all,shared,US"
651	wall_clock_limit="# @ wall_clock_limit = $timestring"
652	if [[ $threads_per_task = 1 ]]
653	then
654	rset="# @ rset = RSET_MCM_AFFINITY"
655	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
656	fi
657	environment=""
658	use_shell=""
659	data_limit=""
660	image_size=""
661	elif [[ $remote_host = ibmku ]]
662	then
663	execute_in_shell="#!/usr/bin/ksh"
664	use_shell="# @ shell = /usr/bin/ksh"
665	consumable_memory=""
666	environment=""
667	network_to_use="# @ network.mpi = sn_all,shared,us"
668	data_limit=""
669	image_size=""
670	elif [[ $remote_host = ibms ]]
671	then
672	network_to_use="# @ network.mpi = csss,shared,us"
673	elif [[ $remote_host = ibmy ]]
674	then
675	consumable_memory=""
676	network_to_use=""
677	fi
678
679	cat > $job_to_send << %%END%%
680	$execute_in_shell
681	$use_shell
682
683	# @ job_type = parallel
684	# @ job_name = $job_name
685	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
686	# @ output = $remote_dayfile
687	# @ error = $remote_dayfile
688	$wall_clock_limit
689	$image_size
690	$class
691	$environment
692	$network_to_use
693	$data_limit
694	$rset
695	$mcm_affinity_options
696	$task_affinity
697	$notification
698	$notify_user
699
700	%%END%%
701
702	if (( nodes > 0 ))
703	then
704
705	if [[ $remote_host != ibmkisti ]]
706	then
707
708	cat >> $job_to_send << %%END%%
709	# @ node = $nodes
710	# @ tasks_per_node = $processes_per_node
711	# @ node_usage = $node_usage
712	# @ queue
713
714	%%END%%
715
716	else
717
718	cat >> $job_to_send << %%END%%
719	# @ total_tasks = $mpi_tasks
720	# @ blocking = unlimited
721	# @ queue
722
723	%%END%%
724
725	fi
726
727	else
728
729	if [[ $remote_host != ibmy ]]
730	then
731
732	cat >> $job_to_send << %%END%%
733	# @ blocking = unlimited
734	# @ total_tasks = $numprocs
735	# @ node_usage = $node_usage
736	# @ queue
737
738	%%END%%
739
740	else
741
742	cat >> $job_to_send << %%END%%
743	# @ node = 1
744	# @ total_tasks = $numprocs
745	# @ queue
746
747	%%END%%
748
749	fi
750
751	fi
752
753	# workaround because of silly job filter on ibmkisti
754	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
755	then
756	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
757	fi
758
759	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
760	then
761
762	cat > $job_to_send << %%END%%
763	#!/bin/ksh
764
765	# @ job_type = serial
766	# @ node_usage = $node_usage
767	# @ job_name = palm
768	# @ wall_clock_limit = ${timestring},$timestring
769	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
770	# @ output = $remote_dayfile
771	# @ error = $remote_dayfile
772	$class
773	$notification
774
775	# @ queue
776
777	%%END%%
778
779	elif [[ $remote_host = lccrayb ]]
780	then
781
782	if [[ $numprocs != 0 ]]
783	then
784	cat > $job_to_send << %%END%%
785	#!/bin/bash --login
786	#PBS -N $job_name
787	#PBS -l walltime=$timestring
788	#PBS -l mppwidth=$numprocs
789	#PBS -l mppnppn=$processes_per_node
790	#PBS -o $remote_dayfile
791	#PBS -j oe
792	#PBS -q $queue
793
794	$module_calls
795	export ATP_ENABLED=0
796	export MPICH_GNI_BTE_MULTI_CHANNEL=disabled
797
798	%%END%%
799
800	else
801
802	continue
803
804	fi
805
806	elif [[ $remote_host = lccrayf ]]
807	then
808
809	if [[ $numprocs != 0 ]]
810	then
811	cat > $job_to_send << %%END%%
812	#!/bin/bash -l
813	#SBATCH -J $job_name
814	#SBATCH -t $timestring
815	#SBATCH -N $nodes
816	#SBATCH --ntasks-per-node=$processes_per_node
817	#SBATCH -p $queue
818	#SBATCH -o $remote_dayfile
819	#SBATCH -e $remote_dayfile
820
821	$init_cmds
822	$module_calls
823
824	%%END%%
825
826	else
827	cat > $job_to_send << %%END%%
828	#!/bin/bash -l
829	#SBATCH -J $job_name
830	#SBATCH -t $timestring
831	#SBATCH -l ncpus=1
832	#SBATCH -l pmem=${memory}mb
833	#SBATCH -m abe
834	#SBATCH -o $remote_dayfile
835	#SBATCH -e $remote_dayfile
836
837	$init_cmds
838	$module_calls
839
840	%%END%%
841
842	fi
843
844	elif [[ $remote_host = lcfimm ]]
845	then
846
847	if [[ $numprocs != 0 ]]
848	then
849	cat > $job_to_send << %%END%%
850	#!/bin/ksh
851	#PBS -N $job_name
852	#PBS -A $project_account
853	#PBS -l walltime=$timestring
854	#PBS -l nodes=${nodes}:ppn=$processes_per_node
855	#PBS -l pmem=${memory}mb
856	#PBS -m abe
857	#PBS -o $remote_dayfile
858	#PBS -j oe
859	mpd &
860
861	%%END%%
862
863	else
864	cat > $job_to_send << %%END%%
865	#!/bin/ksh
866	#PBS -N $job_name
867	#PBS -A $project_account
868	#PBS -l walltime=$timestring
869	#PBS -l ncpus=1
870	#PBS -l pmem=${memory}mb
871	#PBS -m abe
872	#PBS -o $remote_dayfile
873	#PBS -j oe
874
875	%%END%%
876
877	fi
878
879	elif [[ $remote_host = lcflow ]]
880	then
881	if [ $memory -gt 1800 ]; then
882	use_himem=""
883	else
884	use_himem="#"
885	fi
886	if [[ $numprocs != 0 ]]
887	then
888	pe_set="#$ -pe impi41 $numprocs"
889	else
890	pe_set="#$ -pe impi41 1"
891	fi
892	if [[ $queue = default ]]
893	then
894	queue_set=""
895	else
896	queue_set="#$ -q $queue"
897	fi
898	[[ "$disc_space" = "" ]] && disc_space=50
899
900	cat > $job_to_send << %%END%%
901	#!/bin/bash
902	#$ -S /bin/bash
903	#$ -N $job_name
904	#$ -cwd
905	#$ -l h_rt=$timestring
906	#$ -l h_vmem=${memory}M
907	#$ -o $remote_dayfile
908	#$ -j y
909	$pe_set
910	#$ -R y
911	${use_himem}#$ -l highmem=true
912	#$ -l h_fsize=${disc_space}G
913	$queue_set
914
915	%%END%%
916
917	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
918	then
919
920	if [[ $numprocs != 0 ]]
921	then
922	cat > $job_to_send << %%END%%
923	#!/bin/ksh
924	#PBS -N $job_name
925	#PBS -l walltime=$timestring
926	#PBS -l ncpus=$numprocs
927	#PBS -l pmem=${memory}mb
928	#PBS -o $remote_dayfile
929	#PBS -l nodes=$nodes:ppn=${processes_per_node}
930	#PBS -j oe
931
932	mpd &
933
934	%%END%%
935
936	else
937	cat > $job_to_send << %%END%%
938	#!/bin/ksh
939	#PBS -N $job_name
940	#PBS -l walltime=$timestring
941	#PBS -l ncpus=1
942	#PBS -l pmem=${memory}mb
943	#PBS -o $remote_dayfile
944	#PBS -j oe
945
946	%%END%%
947
948	fi
949
950	elif [[ $remote_host = lckiaps ]]
951	then
952
953	if [[ $numprocs != 0 ]]
954	then
955	cat > $job_to_send << %%END%%
956	#!/bin/ksh
957	#PBS -N $job_name
958	#PBS -l walltime=$timestring
959	#PBS -l select=1:ncpus=$numprocs
960	#PBS -l pmem=${memory}mb
961	#PBS -q $queue
962	#PBS -o $remote_dayfile
963	#PBS -j oe
964	#PBS -V
965
966	%%END%%
967
968	else
969	cat > $job_to_send << %%END%%
970	#!/bin/ksh
971	#PBS -N $job_name
972	#PBS -l walltime=$timestring
973	#PBS -l ncpus=1
974	#PBS -l pmem=${memory}mb
975	#PBS -o $remote_dayfile
976	#PBS -j oe
977
978	%%END%%
979
980	fi
981
982	elif [[ $remote_host = lcyon ]]
983	then
984
985	if [[ $numprocs != 0 ]]
986	then
987	cat > $job_to_send << %%END%%
988	#!/bin/ksh
989	#PBS -N $job_name
990	#PBS -l walltime=$timestring
991	#PBS -l ncpus=$numprocs
992	#PBS -l pmem=${memory}mb
993	#PBS -o $remote_dayfile
994	#PBS -j oe
995
996	%%END%%
997
998	else
999	cat > $job_to_send << %%END%%
1000	#!/bin/ksh
1001	#PBS -N $job_name
1002	#PBS -l walltime=$timestring
1003	#PBS -l ncpus=1
1004	#PBS -l pmem=${memory}mb
1005	#PBS -o $remote_dayfile
1006	#PBS -j oe
1007
1008	%%END%%
1009
1010	fi
1011
1012	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1013	then
1014
1015	if [[ "$sgi_feature" != "" ]]
1016	then
1017	feature_directive="#PBS -l feature=$sgi_feature"
1018	else
1019	feature_directive=""
1020	fi
1021
1022	if [[ $queue = dataq \|\| $queue = permq ]]
1023	then
1024	feature_directive="#PBS -l feature=data"
1025	fi
1026
1027	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1028	then
1029	queue_directive="#PBS -q $queue"
1030	else
1031	queue_directive=""
1032	fi
1033
1034	if [[ $email_notification = none ]]
1035	then
1036	email_directive=""
1037	else
1038	email_directive="#PBS -M $email_notification"
1039	fi
1040
1041	if [[ $numprocs != 0 ]]
1042	then
1043	cat > $job_to_send << %%END%%
1044	#!/bin/bash
1045	#PBS -N $job_name
1046	#PBS -l walltime=$timestring
1047	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1048	#PBS -l naccesspolicy=$node_usage
1049	#PBS -o $remote_dayfile
1050	#PBS -j oe
1051	$feature_directive
1052	$queue_directive
1053	$email_directive
1054
1055	eval \`/sw/swdist/bin/modulesinit\`
1056	#. /usr/share/modules/init/bash
1057	$init_cmds
1058	$module_calls
1059
1060	echo ld_library_path=\$LD_LIBRARY_PATH
1061
1062	%%END%%
1063
1064	else
1065	cat > $job_to_send << %%END%%
1066	#PBS -S /bin/bash
1067	#PBS -N $job_name
1068	#PBS -l walltime=$timestring
1069	#PBS -l ncpus=1
1070	#PBS -o $remote_dayfile
1071	#PBS -j oe
1072	$queue_directive
1073	$email_directive
1074
1075	eval \`/sw/swdist/bin/modulesinit\`
1076	#. /usr/share/modules/init/bash
1077	$init_cmds
1078	$module_calls
1079
1080	%%END%%
1081
1082	fi
1083
1084	elif [[ $remote_host = lcxe6 ]]
1085	then
1086
1087	if [[ $numprocs != 0 ]]
1088	then
1089	cat > $job_to_send << %%END%%
1090	#!/bin/ksh
1091	#PBS -S /bin/ksh
1092	#PBS -N $job_name
1093	#PBS -A $project_account
1094	#PBS -j oe
1095	#PBS -l walltime=$timestring
1096	#PBS -l mppwidth=${numprocs}
1097	#PBS -l mppnppn=${processes_per_node}
1098	#PBS -m abe
1099	#PBS -o $remote_dayfile
1100	$email_directive
1101
1102	$init_cmds
1103	$module_calls
1104
1105	%%END%%
1106
1107	else
1108	cat > $job_to_send << %%END%%
1109	#!/bin/ksh
1110	#PBS -S /bin/ksh
1111	#PBS -N $job_name
1112	#PBS -A $project_account
1113	#PBS -j oe
1114	#PBS -l walltime=$timestring
1115	#PBS -l ncpus=1
1116	#PBS -l pmem=${memory}mb
1117	#PBS -m abe
1118	$email_directive
1119	#PBS -o $remote_dayfile
1120
1121	$init_cmds
1122	$module_calls
1123
1124	%%END%%
1125
1126	fi
1127
1128	elif [[ $remote_host = lckyoto ]]
1129	then
1130
1131	cat > $job_to_send << %%END%%
1132	#!/bin/ksh
1133	# @\$-o $remote_dayfile
1134	# @\$-eo -oi
1135	# @\$-lP 16
1136	# @\$-lp 1
1137	# @\$-lm 28gb -llm unlimited -ls unlimited
1138	# @\$-q $queue
1139	# @\$-Pvn abs_pack
1140	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1141	#. /thin/local/etc/setprofile/intel-11.0.sh
1142	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1143	. ~/.myprofile
1144	#. /home2/t/t51254/palm/current_version/myprofile
1145	#. /thin/apps/pgi/mpi.sh
1146	#
1147	env
1148	#
1149	set -x
1150
1151	%%END%%
1152
1153	elif [[ $remote_host = lcxt5m ]]
1154	then
1155
1156	if [[ $numprocs != 0 ]]
1157	then
1158	cat > $job_to_send << %%END%%
1159	#!/bin/ksh
1160	#PBS -S /bin/ksh
1161	#PBS -N $job_name
1162	#PBS -j oe
1163	#PBS -l walltime=$timestring
1164	#PBS -l mppwidth=${numprocs}
1165	#PBS -l mppnppn=${processes_per_node}
1166	#PBS -m abe
1167	#PBS -o $remote_dayfile
1168
1169	$init_cmds
1170	$module_calls
1171
1172	%%END%%
1173
1174	else
1175	cat > $job_to_send << %%END%%
1176	#!/bin/ksh
1177	#PBS -S /bin/ksh
1178	#PBS -N $job_name
1179	#PBS -j oe
1180	#PBS -l walltime=$timestring
1181	#PBS -l ncpus=1
1182	#PBS -l pmem=${memory}mb
1183	#PBS -m abe
1184	#PBS -o $remote_dayfile
1185
1186	$init_cmds
1187	$module_calls
1188
1189	%%END%%
1190
1191	fi
1192
1193	elif [[ $remote_host = lckyuh ]]
1194	then
1195	cat > $job_to_send << %%END%%
1196	#!/bin/bash
1197	#PJM -L "rscgrp=$queue"
1198	#PJM -L "node=$nodes"
1199	#PJM --mpi "proc=$numprocs"
1200	#PJM -L "elapse=$timestring"
1201	#PJM -o $remote_dayfile
1202	#PJM -j
1203	#PJM -X
1204	#PJM --no-stging
1205
1206	export LANG=en_US.UTF-8
1207	%%END%%
1208
1209	elif [[ $remote_host = lckyut ]]
1210	then
1211	cat > $job_to_send << %%END%%
1212	#!/bin/bash
1213	#PJM -L "rscgrp=$queue"
1214	#PJM -L "vnode=$numprocs"
1215	#PJM -L "vnode-core=1"
1216	#PJM -L "elapse=$timestring"
1217	#PJM --mpi proc=$numprocs
1218	#PJM -o $remote_dayfile
1219	#PJM -j
1220	#PJM -X
1221	#PJM --no-stging
1222
1223	export LANG=en_US.UTF-8
1224	%%END%%
1225
1226	elif [[ $remote_host = nech ]]
1227	then
1228
1229	if (( nodes > 1 ))
1230	then
1231	cat > $job_to_send << %%END%%
1232	#!/bin/ksh
1233	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1234	#PBS -l ${qsubmem}=${Memory}gb
1235	#PBS -b $nodes
1236	#PBS -o $remote_dayfile
1237	#PBS -N palm
1238	#PBS -j o
1239	#PBS -T mpisx
1240
1241	%%END%%
1242
1243	elif [[ $numprocs != 0 ]]
1244	then
1245	cat > $job_to_send << %%END%%
1246	#!/bin/ksh
1247	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1248	#PBS -l ${qsubmem}=${Memory}gb
1249	#PBS -o $remote_dayfile
1250	#PBS -N palm
1251	#PBS -j o
1252
1253	%%END%%
1254
1255	else
1256	cat > $job_to_send << %%END%%
1257	#!/bin/ksh
1258	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1259	#PBS -o $remote_dayfile
1260	#PBS -j o
1261
1262	%%END%%
1263
1264	fi
1265
1266	elif [[ $remote_host = necriam ]]
1267	then
1268
1269	if (( nodes > 1 ))
1270	then
1271	cat > $job_to_send << %%END%%
1272	#!/bin/ksh
1273	#PBS -b $nodes
1274	#PBS -o $remote_dayfile
1275	#PBS -N $job_name
1276	#PBS -j o
1277	#PBS -v MPIPROGINV=YES
1278
1279	%%END%%
1280
1281	elif [[ $numprocs != 0 ]]
1282	then
1283	cat > $job_to_send << %%END%%
1284	#!/bin/ksh
1285	#PBS -q ${queue}
1286	#PBS -o $remote_dayfile
1287	#PBS -N $job_name
1288	#PBS -j o
1289	#PBS -v MPIPROGINV=YES
1290
1291	%%END%%
1292
1293	else
1294	cat > $job_to_send << %%END%%
1295	#!/bin/ksh
1296	#PBS -o $remote_dayfile
1297	#PBS -j o
1298
1299	%%END%%
1300
1301	fi
1302
1303	elif [[ $remote_host = lctit ]]
1304	then
1305	cat > $job_to_send << %%END%%
1306	#!/bin/ksh
1307	$init_cmds
1308	$module_calls
1309
1310	%%END%%
1311
1312	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1313	if [[ $tasks_per_node != $processes_per_node ]]
1314	then
1315	submcom="$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 $queue "
1316	else
1317	submcom="$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 "
1318	fi
1319
1320	else
1321
1322	cat > $job_to_send << %%END%%
1323	# @\$-q ${queue}
1324	# @\$-l${qsubtime} $timestring
1325	# @\$-l${qsubmem} ${memory}mb
1326	# @\$-o $remote_dayfile
1327	# @\$-eo
1328
1329	%%END%%
1330
1331	fi
1332
1333
1334	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1335	# BEI EXIT VERANLASSEN
1336	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1337	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1338	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1339	then
1340	echo "set +vx" >> $job_to_send
1341	echo "trap '" >> $job_to_send
1342	echo "set +vx" >> $job_to_send
1343	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* ]]
1344	then
1345	if [[ $remote_host = ibmh ]]
1346	then
1347	return_queue=c1
1348	elif [[ $remote_host = ibmkisti ]]
1349	then
1350	return_queue=class.1-2
1351	elif [[ $remote_host = ibmku ]]
1352	then
1353	return_queue=sdbg2
1354	elif [[ $remote_host = ibms ]]
1355	then
1356	return_queue=p_normal
1357	elif [[ $remote_host = ibmy ]]
1358	then
1359	return_queue=serial
1360	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1361	then
1362	return_queue=serialq
1363	elif [[ $remote_host = necriam ]]
1364	then
1365	return_queue=SP
1366	elif [[ $remote_host = lckiaps ]]
1367	then
1368	return_queue=express
1369	elif [[ $remote_host = lckyuh ]]
1370	then
1371	return_queue=cx-single
1372	elif [[ $remote_host = lckyut ]]
1373	then
1374	return_queue=cx-single
1375	else
1376	return_queue=unknown
1377	fi
1378
1379	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1380	then
1381
1382	if [[ $remote_host = ibmku ]]
1383	then
1384	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1385	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1386	else
1387	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1388	fi
1389	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1390	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1391	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1392	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1393	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1394	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1395	if [[ $host != "ibmh" ]]
1396	then
1397	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1398	fi
1399	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1400	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1401
1402	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1403	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1404
1405	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1406	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
1407	if [[ $remote_host = ibmku ]]
1408	then
1409	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1410	fi
1411	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1412
1413	elif [[ $remote_host = nech ]]
1414	then
1415	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1416	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1417	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1418	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1419	echo "#PBS -j o" >> $job_to_send
1420	echo " " >> $job_to_send
1421	echo "set -x" >> $job_to_send
1422	echo "cd /pf/b/${remote_user}/job_queue" >> $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 [[ $remote_host = necriam ]]
1428	then
1429	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1430	echo "#PBS -q $return_queue" >> $job_to_send
1431	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1432	echo "#PBS -j o" >> $job_to_send
1433	echo " " >> $job_to_send
1434	echo "set -x" >> $job_to_send
1435	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1436	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1437	echo "%%END%%" >> $job_to_send
1438
1439	elif [[ $remote_host = lckyuh ]]
1440	then
1441	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1442	echo "#!/bin/bash" >> $job_to_send
1443	echo "#PJM -L \"node=1\"" >> $job_to_send
1444	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1445	echo "#PJM --no-stging" >> $job_to_send
1446	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1447	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1448	echo "#PJM -j" >> $job_to_send
1449	echo " " >> $job_to_send
1450	echo "export LANG=en_US.UTF-8" >> $job_to_send
1451	echo "set -x" >> $job_to_send
1452	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1453	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1454	echo "%%END%%" >> $job_to_send
1455
1456	elif [[ $remote_host = lckyut ]]
1457	then
1458	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1459	echo "#!/bin/bash" >> $job_to_send
1460	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1461	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1462	echo "#PJM --no-stging" >> $job_to_send
1463	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1464	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1465	echo "#PJM -j" >> $job_to_send
1466	echo " " >> $job_to_send
1467	echo "export LANG=en_US.UTF-8" >> $job_to_send
1468	echo "set -x" >> $job_to_send
1469	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1470	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1471	echo "%%END%%" >> $job_to_send
1472
1473	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1474	then
1475	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1476	echo "#!/bin/bash" >> $job_to_send
1477	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1478	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1479	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1480	echo "#PBS -l feature=data" >> $job_to_send
1481	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1482	echo "#PBS -j oe" >> $job_to_send
1483	echo " " >> $job_to_send
1484	echo ". /usr/share/modules/init/bash" >> $job_to_send
1485	echo "set -x" >> $job_to_send
1486	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1487	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1488	echo "%%END%%" >> $job_to_send
1489
1490	elif [[ $remote_host = lcflow ]]
1491	then
1492	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1493	echo "#!/bin/bash" >> $job_to_send
1494	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1495	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1496	echo "SGEPREFIX -cwd" >> $job_to_send
1497	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1498	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1499	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1500	echo "SGEPREFIX -j y" >> $job_to_send
1501	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1502	echo " " >> $job_to_send
1503	echo "set -x" >> $job_to_send
1504	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1505	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1506	echo "" >> $job_to_send
1507	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1508	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1509	echo "rm -f scpjob.${kennung}" >> $job_to_send
1510	echo "%%END%%" >> $job_to_send
1511	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1512	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1513
1514	else
1515
1516	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1517	echo "# @\\\$-q $return_queue" >> $job_to_send
1518	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1519	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1520	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1521	then
1522	echo "# @\$-l mpp_p=0" >> $job_to_send
1523	fi
1524	echo '# @\$-lF 10mb' >> $job_to_send
1525	# echo '# @\$-o /dev/null' >> $job_to_send
1526	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1527	echo '# @\\\$-eo' >> $job_to_send
1528	echo " " >> $job_to_send
1529	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1530	then
1531	echo "set +vx" >> $job_to_send
1532	echo ". .profile" >> $job_to_send
1533	fi
1534	echo "set -x" >> $job_to_send
1535	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
1536	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1537	echo "%%END%%" >> $job_to_send
1538	fi
1539
1540	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1541	then
1542	echo "llsubmit scpjob.$kennung" >> $job_to_send
1543	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1544	then
1545	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1546	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1547	echo "msub scpjob.$kennung" >> $job_to_send
1548	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1549	then
1550	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1551	elif [[ $remote_host = t3es ]]
1552	then
1553	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1554	elif [[ $remote_host = lckiaps ]]
1555	then
1556	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1557	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1558	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1559	elif [[ $remote_host = lckyu* ]]
1560	then
1561	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1562	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1563	elif [[ $remote_host = lcflow ]]
1564	then
1565	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1566	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1567	else
1568	echo "$submcom scpjob.$kennung" >> $job_to_send
1569	fi
1570	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1571	then
1572	echo "rm scpjob.$kennung" >> $job_to_send
1573	fi
1574	if [[ $remote_host = nech ]]
1575	then
1576	echo "cd -" >> $job_to_send
1577	fi
1578	else
1579	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1580	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1581	fi
1582	echo "set -x" >> $job_to_send
1583	echo " ' exit" >> $job_to_send
1584	echo "set -x" >> $job_to_send
1585	fi
1586
1587
1588
1589	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1590	cat $file_to_send >> $job_to_send
1591
1592	if [[ $remote_host = ibm ]]
1593	then
1594	echo " " >> $job_to_send
1595	echo "exit" >> $job_to_send
1596	fi
1597
1598	# remove job file
1599	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1600	then
1601	echo " " >> $job_to_send
1602	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1603	fi
1604
1605
1606
1607	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1608	if [[ -z $remote_user ]]
1609	then
1610	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1611	then
1612	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1613	fi
1614	fi
1615
1616
1617
1618	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1619	if [[ $no_submit = false ]]
1620	then
1621	if [[ $remote_host != $local_host ]]
1622	then
1623	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1624	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1625	then
1626	job_catalog_save=$job_catalog
1627	job_catalog=job_queue
1628	elif [[ $remote_host = nech ]]
1629	then
1630	job_catalog_save=$job_catalog
1631	job_catalog=/hpf/b/${remote_user}/job_queue
1632	fi
1633	if [[ $remote_host = nech ]]
1634	then
1635	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1636	# TRANSFERIERT WERDEN
1637	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1638	else
1639	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1640	fi
1641	if [[ $? = 1 ]]
1642	then
1643	locat=scp; exit
1644	fi
1645	if [[ $remote_host = ibms ]]
1646	then
1647	job_catalog=$job_catalog_save
1648	fi
1649	[[ $verify = true ]] && printf "\n >>> finished\n"
1650	else
1651	eval job_catalog=$job_catalog
1652	cp $job_to_send ${job_catalog}/$job_on_remhost
1653	fi
1654
1655
1656
1657	# NQS- BZW. LOADLEVELER-JOB STARTEN
1658	if [[ $remote_host != $local_host ]]
1659	then
1660	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1661
1662	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1663	then
1664	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1665	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1666	elif [[ $remote_host = ibmku ]]
1667	then
1668	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1669	elif [[ $remote_host = lcflow ]]
1670	then
1671	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1672	else
1673	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1674	fi
1675
1676	[[ $verify = true ]] && printf " >>> o.k.\n"
1677	else
1678	cd $job_catalog
1679	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1680	then
1681	eval $submcom $job_on_remhost
1682	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1683	then
1684	chmod u+x $job_on_remhost
1685	eval $submcom $job_on_remhost
1686	elif [[ $local_host = nech ]]
1687	then
1688	if [[ $queue = default ]]
1689	then
1690	eval $submcom $job_on_remhost
1691	else
1692	eval $submcom -q $queue $job_on_remhost
1693	fi
1694	else
1695	qsub $job_on_remhost
1696	fi
1697
1698	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1699	# only at the end of the job.
1700	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1701	then
1702	rm $job_on_remhost
1703	fi
1704	cd - > /dev/null
1705	fi
1706	fi
1707
1708
1709
1710	# ABSCHLUSSARBEITEN
1711
1712	if [[ $no_submit = false ]]
1713	then
1714	rm -f $job_to_send
1715	fi
1716	[[ $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 |