Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1227

Last change on this file since 1227 was 1225, checked in by raasch, 11 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 56.6 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 1225 2013-09-16 07:33:10Z fuhrmann $
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	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
134	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
135	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
136	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
137	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
138	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
139	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
140	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
141	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
142	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
143	(blogin*) local_addres=130.73.233.1; local_host=lccrayb;;
144	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
145	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
146	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
147	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
148	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
149	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
150	(cs*) local_addres=136.172.44.131; local_host=nech;;
151	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
152	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
153	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
154	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
155	(node*) local_addres=165.132.26.61 local_host=lck;;
156	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
157	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
158	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
159	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
160	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
161	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
162	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
163	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
164	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
165	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
166	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
167	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
168	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
169	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
170	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
171	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
172	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
173	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
174	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
175	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
176	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
177	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
178	(meller) local_addres=134.106.74.155; local_host=lcfor;;
179	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
180	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
181	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
182	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
183	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
184	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
185	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
186	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
187	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
188	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
189	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
190	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
191	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
192	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
193	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
194	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
195	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
196	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
197	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
198	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
199	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
200	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
201	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
202	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
203	(*) printf "\n +++ \"$local_host\" unknown";
204	printf "\n please inform S. Raasch!";
205	locat=parameter; exit;;
206	esac
207
208
209
210	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
211	remote_host=$local_host
212
213
214
215
216	# PROZEDUROPTIONEN EINLESEN
217	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
218	do
219	case $option in
220	(c) job_catalog=$OPTARG;;
221	(d) delete_dayfile=true;;
222	(D) no_submit=true;;
223	(e) email_notification=$OPTARG;;
224	(g) group_number=$OPTARG;;
225	(h) remote_host=$OPTARG;;
226	(m) memory=$OPTARG;;
227	(n) job_name=$OPTARG;;
228	(N) node_usage=$OPTARG;;
229	(O) threads_per_task=$OPTARG;;
230	(P) scp_port=$OPTARG;;
231	(q) no_default_queue=$OPTARG;;
232	(t) cputime=$OPTARG;;
233	(T) tasks_per_node=$OPTARG;;
234	(u) remote_user=$OPTARG;;
235	(v) verify=false;;
236	(X) numprocs=$OPTARG;;
237	(\?) printf "\n +++ Option $OPTARG unknown \n";
238	locat=option; exit;;
239	esac
240	done
241
242
243	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
244	(( to_shift = $OPTIND - 1 ))
245	shift $to_shift; file_to_send=$1
246
247
248	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
249	if [ "$1" = "?" ]
250	then
251	(printf "\n *** subjob can be called as follows:\n"
252	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
253	printf "\n Description of available options:\n"
254	printf "\n Option Description Default-Value"
255	printf "\n -c job-input- and output-catalog ~/job_queue"
256	printf "\n -d no job-protocol will be created ---"
257	printf "\n -D only the job-file will be created ---"
258	printf "\n -h execution host, available hosts: $remote_host"
259	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
260	printf "\n lckiaps, lctit, nech, necriam, unics"
261	printf "\n -m memory demand per process in MByte ---"
262	printf "\n -n jobname <jobdatei>"
263	printf "\n -O threads per task (for OpenMP usage) 1"
264	printf "\n -P ssh/scp port default port"
265	printf "\n -q job-queue to be used default"
266	printf "\n -t allowed cpu-time in seconds ---"
267	printf "\n -T tasks per node (on parallel hosts) ---"
268	printf "\n -u username on execution host from .netrc"
269	printf "\n -v no prompt for confirmation ---"
270	printf "\n -X # of processors (on parallel hosts) 1"
271	printf "\n "
272	printf "\n The only possible positional parameter is <jobfile>:"
273	printf "\n The complete NQS-job must be provided here."
274	printf "\n <jobfile>=? creates this outline\n\n") \| more
275	exit
276	fi
277
278
279
280	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
281	if [[ "$file_to_send" = "" ]]
282	then
283	printf "\n +++ job-file missing"
284	locat=parameter; exit
285	else
286	if [[ -f $file_to_send ]]
287	then
288	true
289	else
290	printf "\n +++ job-file: "
291	printf "\n $file_to_send"
292	printf "\n does not exist"
293	locat=parameter; exit
294	fi
295	fi
296
297
298
299	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
300	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
301	if [[ $job_name = none ]]
302	then
303	job_name=$file_to_send
304	fi
305	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
306	then
307	printf "\n +++ job-file name: "
308	printf "\n $job_name"
309	printf "\n must not contain \"/\"-characters"
310	locat=parameter; exit
311	fi
312
313
314
315
316	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
317	# ODER WENN HOST NICHT ANGEGEBEN WURDE
318	if [[ $remote_host = none ]]
319	then
320	printf "\n +++ host missing"
321	locat=option; exit
322	else
323	case $remote_host in
324	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
325	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
327	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
328	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
329	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
331	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
332	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
333	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
334	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
335	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
336	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
337	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
338	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
339	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
340	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
341	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
342	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
343	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
344	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
345	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
346	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
347	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
348	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
349	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
350	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
351	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
352	locat=parameter; exit;;
353	esac
354	fi
355
356
357	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
358	if [[ $no_default_queue != none ]]
359	then
360	error=false
361	ndq=$no_default_queue
362	case $remote_host in
363	(ibm) case $ndq in
364	(p690_express\|p690_standard\|p690_long) error=false;;
365	(*) error=true;;
366	esac;;
367	(ibmh) case $ndq in
368	(cluster\|express) error=false;;
369	(*) error=true;;
370	esac;;
371	(ibmkisti) case $ndq in
372	(class.32plus\|class.1-2\|class.2-32) error=false;;
373	(*) error=true;;
374	esac;;
375	(ibmku) case $ndq in
376	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
377	(*) error=true;;
378	esac;;
379	(ibms) case $ndq in
380	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
381	(*) error=true;;
382	esac;;
383	(ibmy) case $ndq in
384	(parallel) error=false;;
385	(*) error=true;;
386	esac;;
387	(lccrayb) case $ndq in
388	(mpp1q\|mpp1testq\|testq) error=false;;
389	(*) error=true;;
390	esac;;
391	(lccrayf) case $ndq in
392	(usup\|test*\|small\|large) error=false;;
393	(*) error=true;;
394	esac;;
395	(lcflow) case $ndq in
396	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
397	(*) error=true;;
398	esac;;
399	(lckiaps) case $ndq in
400	(express\|normal) error=false;;
401	(*) error=true;;
402	esac;;
403	(lckyoto) case $ndq in
404	(eh\|ph) error=false;;
405	(*) error=true;;
406	esac;;
407	(lckyuh) case $ndq in
408	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
409	(*) error=true;;
410	esac;;
411	(lckyut) case $ndq in
412	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
413	(*) error=true;;
414	esac;;
415	(lcsgib\|lcsgih) case $ndq in
416	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
417	(*) error=true;;
418	esac;;
419	(lctit) case $ndq in
420	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
421	(*) error=true;;
422	esac;;
423	(t3eb) case $ndq in
424	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
425	(*) error=true;;
426	esac;;
427	(necriam) case $ndq in
428	(SP\|SS\|P6) error=false;;
429	(*) error=true;;
430	esac;;
431	(t3eh) case $ndq in
432	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
433	(*) error=true;;
434	esac;;
435	(t3ej2\|t3ej5) case $ndq in
436	(low\|normal\|high) error=false;;
437	(*) error=true;;
438	esac;;
439	(t3es) case $ndq in
440	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
441	(*) error=true;;
442	esac;;
443	(unics) case $ndq in
444	(unics\|ht) error=false;;
445	(*) error=true;;
446	esac;;
447	esac
448	if [[ $error = true ]]
449	then
450	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
451	locat=parameter; exit
452	else
453	queue=$no_default_queue
454	fi
455	fi
456
457
458
459	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
460	# AUFTEILEN
461	done=false
462	while [[ $done = false ]]
463	do
464	if (( $cputime <= 0 ))
465	then
466	printf "\n +++ wrong cpu-time or cpu-time missing"
467	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
468	printf "\n >>> "
469	read cputime 1>/dev/null 2>&1
470	else
471	done=true
472	fi
473	done
474	if [[ $remote_host = nech ]]
475	then
476	if (( tasks_per_node != 0 ))
477	then
478	(( cputime = cputime * tasks_per_node ))
479	elif [[ $numprocs != 0 ]]
480	then
481	(( cputime = cputime * numprocs ))
482	fi
483	fi
484	(( stunden = cputime / 3600 ))
485	(( resttime = cputime - stunden * 3600 ))
486	(( minuten = resttime / 60 ))
487	(( sekunden = resttime - minuten * 60 ))
488	timestring=${stunden}:${minuten}:${sekunden}
489
490
491
492	# PRUEFEN DER KERNSPEICHERANFORDERUNG
493	done=false
494	while [[ $done = false ]]
495	do
496	if (( memory <= 0 ))
497	then
498	printf "\n +++ wrong memory demand or memory demand missing"
499	printf "\n >>> Please type memory in MByte per process as INTEGER:"
500	printf "\n >>> "
501	read memory 1>/dev/null 2>&1
502	else
503	done=true
504	fi
505	done
506
507	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
508	then
509	if (( tasks_per_node != 0 ))
510	then
511	(( Memory = memory * tasks_per_node / 1000 ))
512	elif [[ $numprocs != 0 ]]
513	then
514	(( Memory = memory * numprocs / 1000 ))
515	else
516	(( Memory = memory / 1000 ))
517	fi
518	elif [[ $remote_host = lctit ]]
519	then
520	(( Memory = memory * tasks_per_node / 1000 ))
521	fi
522
523
524	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
525	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
526	then
527	(( memory = memory * threads_per_task ))
528	fi
529
530
531	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
532	if (( tasks_per_node != 0 ))
533	then
534	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
535	fi
536
537
538	# Calculate number of processes per node
539	(( processes_per_node = tasks_per_node * threads_per_task ))
540
541
542	# Calculate number of MPI tasks
543	(( mpi_tasks = numprocs / threads_per_task ))
544
545
546	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
547	if [[ "$scp_port" != "" ]]
548	then
549	PORTOPT="-P $scp_port"
550	SSH_PORTOPT="-p $scp_port"
551	fi
552
553
554	# HEADER-AUSGABE
555	if [[ $verify = true ]]
556	then
557	printf "\n\n"
558	printf "#--------------------------------------------------------------# \n"
559	spalte1=SUBJOB;spalte2=$(date)
560	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
561	printf "\| \| \n"
562	printf "\| values of parameters/options: \| \n"
563	spalte1=$(echo local_host$punkte \| cut -c-20)
564	spalte2=$punkte$local_host
565	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
566	spalte1=$(echo remote_host$punkte \| cut -c-20)
567	spalte2=$punkte$remote_host
568	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
569	spalte1=$(echo queue$punkte \| cut -c-20)
570	spalte2=$punkte$queue
571	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
572	spalte1=$(echo memory$punkte \| cut -c-20)
573	spalte2="$punkte$memory mb"
574	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
575	spalte1=$(echo cputime$punkte \| cut -c-20)
576	spalte2="$punkte$cputime sec"
577	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
578	spalte1=$(echo job_name$punkte \| cut -c-20)
579	spalte2="$punkte$job_name"
580	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
581	printf "#--------------------------------------------------------------# \n\n"
582
583
584	# KONTROLLABFRAGE, OB ALLES O.K.
585	antwort="dummy"
586	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
587	do
588	read antwort?" >>> continue (y/n) ? "
589	done
590	if [[ $antwort = n \|\| $antwort = N ]]
591	then
592	locat=verify; exit
593	fi
594	printf "\n"
595	fi
596
597
598
599	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
600	kennung=$RANDOM
601	job_on_remhost=${job_name}_${kennung}_$local_host
602	job_to_send=job_to_send_$kennung
603	if [[ $delete_dayfile = false ]]
604	then
605	remote_dayfile=${local_host}_${job_name}_result_$kennung
606	local_dayfile=${remote_host}_${job_name}
607	else
608	remote_dayfile=/dev/null
609	fi
610
611
612
613	# Generate the batch job scripts (qsub/msub/LoadLeveler)
614	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
615	then
616
617	# General LoadLeveler settings
618	execute_in_shell="#!/bin/ksh"
619	use_shell="# @ shell = /bin/ksh"
620	consumable_memory="ConsumableMemory($memory mb)"
621	class="# @ class = $queue"
622	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
623	network_to_use="# @ network.mpi = sn_all,shared,us"
624	data_limit="# @ data_limit = 1.76gb"
625	image_size="# @ image_size = 50"
626	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
627
628	if [[ $email_notification = none ]]
629	then
630	notify_user=""
631	else
632	notify_user="# @ notify_user = $email_notification"
633	if [[ $delete_dayfile = true ]]
634	then
635	notification='# @ notification = never'
636	fi
637	fi
638
639	if [[ $remote_host = ibmh ]]
640	then
641	data_limit=""
642	network_to_use=""
643	class="# @ class = $queue"
644	environment=""
645	rset="# @ rset = RSET_MCM_AFFINITY"
646	task_affinity="# @ task_affinity = core(1)"
647	elif [[ $remote_host = ibmkisti ]]
648	then
649	network_to_use="# @ network.MPI = sn_all,shared,US"
650	wall_clock_limit="# @ wall_clock_limit = $timestring"
651	if [[ $threads_per_task = 1 ]]
652	then
653	rset="# @ rset = RSET_MCM_AFFINITY"
654	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
655	fi
656	environment=""
657	use_shell=""
658	data_limit=""
659	image_size=""
660	elif [[ $remote_host = ibmku ]]
661	then
662	execute_in_shell="#!/usr/bin/ksh"
663	use_shell="# @ shell = /usr/bin/ksh"
664	consumable_memory=""
665	environment=""
666	network_to_use="# @ network.mpi = sn_all,shared,us"
667	data_limit=""
668	image_size=""
669	elif [[ $remote_host = ibms ]]
670	then
671	network_to_use="# @ network.mpi = csss,shared,us"
672	elif [[ $remote_host = ibmy ]]
673	then
674	consumable_memory=""
675	network_to_use=""
676	fi
677
678	cat > $job_to_send << %%END%%
679	$execute_in_shell
680	$use_shell
681
682	# @ job_type = parallel
683	# @ job_name = $job_name
684	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
685	# @ output = $remote_dayfile
686	# @ error = $remote_dayfile
687	$wall_clock_limit
688	$image_size
689	$class
690	$environment
691	$network_to_use
692	$data_limit
693	$rset
694	$mcm_affinity_options
695	$task_affinity
696	$notification
697	$notify_user
698
699	%%END%%
700
701	if (( nodes > 0 ))
702	then
703
704	if [[ $remote_host != ibmkisti ]]
705	then
706
707	cat >> $job_to_send << %%END%%
708	# @ node = $nodes
709	# @ tasks_per_node = $processes_per_node
710	# @ node_usage = $node_usage
711	# @ queue
712
713	%%END%%
714
715	else
716
717	cat >> $job_to_send << %%END%%
718	# @ total_tasks = $mpi_tasks
719	# @ blocking = unlimited
720	# @ queue
721
722	%%END%%
723
724	fi
725
726	else
727
728	if [[ $remote_host != ibmy ]]
729	then
730
731	cat >> $job_to_send << %%END%%
732	# @ blocking = unlimited
733	# @ total_tasks = $numprocs
734	# @ node_usage = $node_usage
735	# @ queue
736
737	%%END%%
738
739	else
740
741	cat >> $job_to_send << %%END%%
742	# @ node = 1
743	# @ total_tasks = $numprocs
744	# @ queue
745
746	%%END%%
747
748	fi
749
750	fi
751
752	# workaround because of silly job filter on ibmkisti
753	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
754	then
755	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
756	fi
757
758	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
759	then
760
761	cat > $job_to_send << %%END%%
762	#!/bin/ksh
763
764	# @ job_type = serial
765	# @ node_usage = $node_usage
766	# @ job_name = palm
767	# @ wall_clock_limit = ${timestring},$timestring
768	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
769	# @ output = $remote_dayfile
770	# @ error = $remote_dayfile
771	$class
772	$notification
773
774	# @ queue
775
776	%%END%%
777
778	elif [[ $remote_host = lccrayb ]]
779	then
780
781	if [[ $numprocs != 0 ]]
782	then
783	cat > $job_to_send << %%END%%
784	#!/bin/bash --login
785	#PBS -N $job_name
786	#PBS -l walltime=$timestring
787	#PBS -l mppwidth=$numprocs
788	#PBS -l mppnppn=$processes_per_node
789	#PBS -o $remote_dayfile
790	#PBS -j oe
791	#PBS -q $queue
792
793	$module_calls
794	export ATP_ENABLED=0
795	export MPICH_GNI_BTE_MULTI_CHANNEL=disabled
796
797	%%END%%
798
799	else
800
801	continue
802
803	fi
804
805	elif [[ $remote_host = lccrayf ]]
806	then
807
808	if [[ $numprocs != 0 ]]
809	then
810	cat > $job_to_send << %%END%%
811	#!/bin/bash -l
812	#SBATCH -J $job_name
813	#SBATCH -t $timestring
814	#SBATCH -N $nodes
815	#SBATCH --ntasks-per-node=$processes_per_node
816	#SBATCH -p $queue
817	#SBATCH -o $remote_dayfile
818	#SBATCH -e $remote_dayfile
819
820	$init_cmds
821	$module_calls
822
823	%%END%%
824
825	else
826	cat > $job_to_send << %%END%%
827	#!/bin/bash -l
828	#SBATCH -J $job_name
829	#SBATCH -t $timestring
830	#SBATCH -l ncpus=1
831	#SBATCH -l pmem=${memory}mb
832	#SBATCH -m abe
833	#SBATCH -o $remote_dayfile
834	#SBATCH -e $remote_dayfile
835
836	$init_cmds
837	$module_calls
838
839	%%END%%
840
841	fi
842
843	elif [[ $remote_host = lcfimm ]]
844	then
845
846	if [[ $numprocs != 0 ]]
847	then
848	cat > $job_to_send << %%END%%
849	#!/bin/ksh
850	#PBS -N $job_name
851	#PBS -A $project_account
852	#PBS -l walltime=$timestring
853	#PBS -l nodes=${nodes}:ppn=$processes_per_node
854	#PBS -l pmem=${memory}mb
855	#PBS -m abe
856	#PBS -o $remote_dayfile
857	#PBS -j oe
858	mpd &
859
860	%%END%%
861
862	else
863	cat > $job_to_send << %%END%%
864	#!/bin/ksh
865	#PBS -N $job_name
866	#PBS -A $project_account
867	#PBS -l walltime=$timestring
868	#PBS -l ncpus=1
869	#PBS -l pmem=${memory}mb
870	#PBS -m abe
871	#PBS -o $remote_dayfile
872	#PBS -j oe
873
874	%%END%%
875
876	fi
877
878	elif [[ $remote_host = lcflow ]]
879	then
880	if [ $memory -gt 1800 ]; then
881	use_himem=""
882	else
883	use_himem="#"
884	fi
885	if [[ $numprocs != 0 ]]
886	then
887	pe_set="#$ -pe impi41 $numprocs"
888	else
889	pe_set="#$ -pe impi41 1"
890	fi
891	if [[ $queue = default ]]
892	then
893	queue_set=""
894	else
895	queue_set="#$ -q $queue"
896	fi
897	[[ "$disc_space" = "" ]] && disc_space=50
898
899	cat > $job_to_send << %%END%%
900	#!/bin/bash
901	#$ -S /bin/bash
902	#$ -N $job_name
903	#$ -cwd
904	#$ -l h_rt=$timestring
905	#$ -l h_vmem=${memory}M
906	#$ -o $remote_dayfile
907	#$ -j y
908	$pe_set
909	#$ -R y
910	${use_himem}#$ -l highmem=true
911	#$ -l h_fsize=${disc_space}G
912	$queue_set
913
914	%%END%%
915
916	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
917	then
918
919	if [[ $numprocs != 0 ]]
920	then
921	cat > $job_to_send << %%END%%
922	#!/bin/ksh
923	#PBS -N $job_name
924	#PBS -l walltime=$timestring
925	#PBS -l ncpus=$numprocs
926	#PBS -l pmem=${memory}mb
927	#PBS -o $remote_dayfile
928	#PBS -l nodes=$nodes:ppn=${processes_per_node}
929	#PBS -j oe
930
931	mpd &
932
933	%%END%%
934
935	else
936	cat > $job_to_send << %%END%%
937	#!/bin/ksh
938	#PBS -N $job_name
939	#PBS -l walltime=$timestring
940	#PBS -l ncpus=1
941	#PBS -l pmem=${memory}mb
942	#PBS -o $remote_dayfile
943	#PBS -j oe
944
945	%%END%%
946
947	fi
948
949	elif [[ $remote_host = lckiaps ]]
950	then
951
952	if [[ $numprocs != 0 ]]
953	then
954	cat > $job_to_send << %%END%%
955	#!/bin/ksh
956	#PBS -N $job_name
957	#PBS -l walltime=$timestring
958	#PBS -l select=1:ncpus=$numprocs
959	#PBS -l pmem=${memory}mb
960	#PBS -q $queue
961	#PBS -o $remote_dayfile
962	#PBS -j oe
963	#PBS -V
964
965	%%END%%
966
967	else
968	cat > $job_to_send << %%END%%
969	#!/bin/ksh
970	#PBS -N $job_name
971	#PBS -l walltime=$timestring
972	#PBS -l ncpus=1
973	#PBS -l pmem=${memory}mb
974	#PBS -o $remote_dayfile
975	#PBS -j oe
976
977	%%END%%
978
979	fi
980
981	elif [[ $remote_host = lcyon ]]
982	then
983
984	if [[ $numprocs != 0 ]]
985	then
986	cat > $job_to_send << %%END%%
987	#!/bin/ksh
988	#PBS -N $job_name
989	#PBS -l walltime=$timestring
990	#PBS -l ncpus=$numprocs
991	#PBS -l pmem=${memory}mb
992	#PBS -o $remote_dayfile
993	#PBS -j oe
994
995	%%END%%
996
997	else
998	cat > $job_to_send << %%END%%
999	#!/bin/ksh
1000	#PBS -N $job_name
1001	#PBS -l walltime=$timestring
1002	#PBS -l ncpus=1
1003	#PBS -l pmem=${memory}mb
1004	#PBS -o $remote_dayfile
1005	#PBS -j oe
1006
1007	%%END%%
1008
1009	fi
1010
1011	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1012	then
1013
1014	if [[ "$sgi_feature" != "" ]]
1015	then
1016	feature_directive="#PBS -l feature=$sgi_feature"
1017	else
1018	feature_directive=""
1019	fi
1020
1021	if [[ $queue = dataq \|\| $queue = permq ]]
1022	then
1023	feature_directive="#PBS -l feature=data"
1024	fi
1025
1026	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1027	then
1028	queue_directive="#PBS -q $queue"
1029	else
1030	queue_directive=""
1031	fi
1032
1033	if [[ $email_notification = none ]]
1034	then
1035	email_directive=""
1036	else
1037	email_directive="#PBS -M $email_notification"
1038	fi
1039
1040	if [[ $numprocs != 0 ]]
1041	then
1042	cat > $job_to_send << %%END%%
1043	#!/bin/bash
1044	#PBS -N $job_name
1045	#PBS -l walltime=$timestring
1046	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1047	#PBS -l naccesspolicy=$node_usage
1048	#PBS -o $remote_dayfile
1049	#PBS -j oe
1050	$feature_directive
1051	$queue_directive
1052	$email_directive
1053
1054	eval \`/sw/swdist/bin/modulesinit\`
1055	#. /usr/share/modules/init/bash
1056	$init_cmds
1057	$module_calls
1058
1059	echo ld_library_path=\$LD_LIBRARY_PATH
1060
1061	%%END%%
1062
1063	else
1064	cat > $job_to_send << %%END%%
1065	#PBS -S /bin/bash
1066	#PBS -N $job_name
1067	#PBS -l walltime=$timestring
1068	#PBS -l ncpus=1
1069	#PBS -o $remote_dayfile
1070	#PBS -j oe
1071	$queue_directive
1072	$email_directive
1073
1074	eval \`/sw/swdist/bin/modulesinit\`
1075	#. /usr/share/modules/init/bash
1076	$init_cmds
1077	$module_calls
1078
1079	%%END%%
1080
1081	fi
1082
1083	elif [[ $remote_host = lcxe6 ]]
1084	then
1085
1086	if [[ $numprocs != 0 ]]
1087	then
1088	cat > $job_to_send << %%END%%
1089	#!/bin/ksh
1090	#PBS -S /bin/ksh
1091	#PBS -N $job_name
1092	#PBS -A $project_account
1093	#PBS -j oe
1094	#PBS -l walltime=$timestring
1095	#PBS -l mppwidth=${numprocs}
1096	#PBS -l mppnppn=${processes_per_node}
1097	#PBS -m abe
1098	#PBS -o $remote_dayfile
1099	$email_directive
1100
1101	$init_cmds
1102	$module_calls
1103
1104	%%END%%
1105
1106	else
1107	cat > $job_to_send << %%END%%
1108	#!/bin/ksh
1109	#PBS -S /bin/ksh
1110	#PBS -N $job_name
1111	#PBS -A $project_account
1112	#PBS -j oe
1113	#PBS -l walltime=$timestring
1114	#PBS -l ncpus=1
1115	#PBS -l pmem=${memory}mb
1116	#PBS -m abe
1117	$email_directive
1118	#PBS -o $remote_dayfile
1119
1120	$init_cmds
1121	$module_calls
1122
1123	%%END%%
1124
1125	fi
1126
1127	elif [[ $remote_host = lckyoto ]]
1128	then
1129
1130	cat > $job_to_send << %%END%%
1131	#!/bin/ksh
1132	# @\$-o $remote_dayfile
1133	# @\$-eo -oi
1134	# @\$-lP 16
1135	# @\$-lp 1
1136	# @\$-lm 28gb -llm unlimited -ls unlimited
1137	# @\$-q $queue
1138	# @\$-Pvn abs_pack
1139	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1140	#. /thin/local/etc/setprofile/intel-11.0.sh
1141	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1142	. ~/.myprofile
1143	#. /home2/t/t51254/palm/current_version/myprofile
1144	#. /thin/apps/pgi/mpi.sh
1145	#
1146	env
1147	#
1148	set -x
1149
1150	%%END%%
1151
1152	elif [[ $remote_host = lcxt5m ]]
1153	then
1154
1155	if [[ $numprocs != 0 ]]
1156	then
1157	cat > $job_to_send << %%END%%
1158	#!/bin/ksh
1159	#PBS -S /bin/ksh
1160	#PBS -N $job_name
1161	#PBS -j oe
1162	#PBS -l walltime=$timestring
1163	#PBS -l mppwidth=${numprocs}
1164	#PBS -l mppnppn=${processes_per_node}
1165	#PBS -m abe
1166	#PBS -o $remote_dayfile
1167
1168	$init_cmds
1169	$module_calls
1170
1171	%%END%%
1172
1173	else
1174	cat > $job_to_send << %%END%%
1175	#!/bin/ksh
1176	#PBS -S /bin/ksh
1177	#PBS -N $job_name
1178	#PBS -j oe
1179	#PBS -l walltime=$timestring
1180	#PBS -l ncpus=1
1181	#PBS -l pmem=${memory}mb
1182	#PBS -m abe
1183	#PBS -o $remote_dayfile
1184
1185	$init_cmds
1186	$module_calls
1187
1188	%%END%%
1189
1190	fi
1191
1192	elif [[ $remote_host = lckyuh ]]
1193	then
1194	cat > $job_to_send << %%END%%
1195	#!/bin/bash
1196	#PJM -L "rscgrp=$queue"
1197	#PJM -L "node=$nodes"
1198	#PJM --mpi "proc=$numprocs"
1199	#PJM -L "elapse=$timestring"
1200	#PJM -o $remote_dayfile
1201	#PJM -j
1202	#PJM -X
1203	#PJM --no-stging
1204
1205	export LANG=en_US.UTF-8
1206	%%END%%
1207
1208	elif [[ $remote_host = lckyut ]]
1209	then
1210	cat > $job_to_send << %%END%%
1211	#!/bin/bash
1212	#PJM -L "rscgrp=$queue"
1213	#PJM -L "vnode=$numprocs"
1214	#PJM -L "vnode-core=1"
1215	#PJM -L "elapse=$timestring"
1216	#PJM --mpi proc=$numprocs
1217	#PJM -o $remote_dayfile
1218	#PJM -j
1219	#PJM -X
1220	#PJM --no-stging
1221
1222	export LANG=en_US.UTF-8
1223	%%END%%
1224
1225	elif [[ $remote_host = nech ]]
1226	then
1227
1228	if (( nodes > 1 ))
1229	then
1230	cat > $job_to_send << %%END%%
1231	#!/bin/ksh
1232	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1233	#PBS -l ${qsubmem}=${Memory}gb
1234	#PBS -b $nodes
1235	#PBS -o $remote_dayfile
1236	#PBS -N palm
1237	#PBS -j o
1238	#PBS -T mpisx
1239
1240	%%END%%
1241
1242	elif [[ $numprocs != 0 ]]
1243	then
1244	cat > $job_to_send << %%END%%
1245	#!/bin/ksh
1246	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1247	#PBS -l ${qsubmem}=${Memory}gb
1248	#PBS -o $remote_dayfile
1249	#PBS -N palm
1250	#PBS -j o
1251
1252	%%END%%
1253
1254	else
1255	cat > $job_to_send << %%END%%
1256	#!/bin/ksh
1257	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1258	#PBS -o $remote_dayfile
1259	#PBS -j o
1260
1261	%%END%%
1262
1263	fi
1264
1265	elif [[ $remote_host = necriam ]]
1266	then
1267
1268	if (( nodes > 1 ))
1269	then
1270	cat > $job_to_send << %%END%%
1271	#!/bin/ksh
1272	#PBS -b $nodes
1273	#PBS -o $remote_dayfile
1274	#PBS -N $job_name
1275	#PBS -j o
1276	#PBS -v MPIPROGINV=YES
1277
1278	%%END%%
1279
1280	elif [[ $numprocs != 0 ]]
1281	then
1282	cat > $job_to_send << %%END%%
1283	#!/bin/ksh
1284	#PBS -q ${queue}
1285	#PBS -o $remote_dayfile
1286	#PBS -N $job_name
1287	#PBS -j o
1288	#PBS -v MPIPROGINV=YES
1289
1290	%%END%%
1291
1292	else
1293	cat > $job_to_send << %%END%%
1294	#!/bin/ksh
1295	#PBS -o $remote_dayfile
1296	#PBS -j o
1297
1298	%%END%%
1299
1300	fi
1301
1302	elif [[ $remote_host = lctit ]]
1303	then
1304	cat > $job_to_send << %%END%%
1305	#!/bin/ksh
1306	$init_cmds
1307	$module_calls
1308
1309	%%END%%
1310
1311	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1312	if [[ $tasks_per_node != $processes_per_node ]]
1313	then
1314	submcom="$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 $queue "
1315	else
1316	submcom="$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 "
1317	fi
1318
1319	else
1320
1321	cat > $job_to_send << %%END%%
1322	# @\$-q ${queue}
1323	# @\$-l${qsubtime} $timestring
1324	# @\$-l${qsubmem} ${memory}mb
1325	# @\$-o $remote_dayfile
1326	# @\$-eo
1327
1328	%%END%%
1329
1330	fi
1331
1332
1333	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1334	# BEI EXIT VERANLASSEN
1335	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1336	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1337	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1338	then
1339	echo "set +vx" >> $job_to_send
1340	echo "trap '" >> $job_to_send
1341	echo "set +vx" >> $job_to_send
1342	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* ]]
1343	then
1344	if [[ $remote_host = ibmh ]]
1345	then
1346	return_queue=c1
1347	elif [[ $remote_host = ibmkisti ]]
1348	then
1349	return_queue=class.1-2
1350	elif [[ $remote_host = ibmku ]]
1351	then
1352	return_queue=sdbg2
1353	elif [[ $remote_host = ibms ]]
1354	then
1355	return_queue=p_normal
1356	elif [[ $remote_host = ibmy ]]
1357	then
1358	return_queue=serial
1359	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1360	then
1361	return_queue=serialq
1362	elif [[ $remote_host = necriam ]]
1363	then
1364	return_queue=SP
1365	elif [[ $remote_host = lckiaps ]]
1366	then
1367	return_queue=express
1368	elif [[ $remote_host = lckyuh ]]
1369	then
1370	return_queue=cx-single
1371	elif [[ $remote_host = lckyut ]]
1372	then
1373	return_queue=cx-single
1374	else
1375	return_queue=unknown
1376	fi
1377
1378	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1379	then
1380
1381	if [[ $remote_host = ibmku ]]
1382	then
1383	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1384	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1385	else
1386	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1387	fi
1388	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1389	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1390	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1391	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1392	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1393	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1394	if [[ $host != "ibmh" ]]
1395	then
1396	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1397	fi
1398	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1399	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1400
1401	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1402	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1403
1404	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1405	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
1406	if [[ $remote_host = ibmku ]]
1407	then
1408	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1409	fi
1410	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1411
1412	elif [[ $remote_host = nech ]]
1413	then
1414	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1415	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1416	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1417	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1418	echo "#PBS -j o" >> $job_to_send
1419	echo " " >> $job_to_send
1420	echo "set -x" >> $job_to_send
1421	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1422	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1423	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1424	echo "%%END%%" >> $job_to_send
1425
1426	elif [[ $remote_host = necriam ]]
1427	then
1428	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1429	echo "#PBS -q $return_queue" >> $job_to_send
1430	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1431	echo "#PBS -j o" >> $job_to_send
1432	echo " " >> $job_to_send
1433	echo "set -x" >> $job_to_send
1434	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1435	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1436	echo "%%END%%" >> $job_to_send
1437
1438	elif [[ $remote_host = lckyuh ]]
1439	then
1440	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1441	echo "#!/bin/bash" >> $job_to_send
1442	echo "#PJM -L \"node=1\"" >> $job_to_send
1443	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1444	echo "#PJM --no-stging" >> $job_to_send
1445	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1446	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1447	echo "#PJM -j" >> $job_to_send
1448	echo " " >> $job_to_send
1449	echo "export LANG=en_US.UTF-8" >> $job_to_send
1450	echo "set -x" >> $job_to_send
1451	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1452	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1453	echo "%%END%%" >> $job_to_send
1454
1455	elif [[ $remote_host = lckyut ]]
1456	then
1457	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1458	echo "#!/bin/bash" >> $job_to_send
1459	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1460	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1461	echo "#PJM --no-stging" >> $job_to_send
1462	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1463	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1464	echo "#PJM -j" >> $job_to_send
1465	echo " " >> $job_to_send
1466	echo "export LANG=en_US.UTF-8" >> $job_to_send
1467	echo "set -x" >> $job_to_send
1468	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1469	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1470	echo "%%END%%" >> $job_to_send
1471
1472	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1473	then
1474	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1475	echo "#!/bin/bash" >> $job_to_send
1476	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1477	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1478	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1479	echo "#PBS -l feature=data" >> $job_to_send
1480	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1481	echo "#PBS -j oe" >> $job_to_send
1482	echo " " >> $job_to_send
1483	echo ". /usr/share/modules/init/bash" >> $job_to_send
1484	echo "set -x" >> $job_to_send
1485	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1486	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1487	echo "%%END%%" >> $job_to_send
1488
1489	elif [[ $remote_host = lcflow ]]
1490	then
1491	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1492	echo "#!/bin/bash" >> $job_to_send
1493	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1494	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1495	echo "SGEPREFIX -cwd" >> $job_to_send
1496	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1497	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1498	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1499	echo "SGEPREFIX -j y" >> $job_to_send
1500	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1501	echo " " >> $job_to_send
1502	echo "set -x" >> $job_to_send
1503	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1504	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1505	echo "" >> $job_to_send
1506	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1507	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1508	echo "rm -f scpjob.${kennung}" >> $job_to_send
1509	echo "%%END%%" >> $job_to_send
1510	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1511	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1512
1513	else
1514
1515	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1516	echo "# @\\\$-q $return_queue" >> $job_to_send
1517	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1518	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1519	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1520	then
1521	echo "# @\$-l mpp_p=0" >> $job_to_send
1522	fi
1523	echo '# @\$-lF 10mb' >> $job_to_send
1524	# echo '# @\$-o /dev/null' >> $job_to_send
1525	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1526	echo '# @\\\$-eo' >> $job_to_send
1527	echo " " >> $job_to_send
1528	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1529	then
1530	echo "set +vx" >> $job_to_send
1531	echo ". .profile" >> $job_to_send
1532	fi
1533	echo "set -x" >> $job_to_send
1534	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
1535	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1536	echo "%%END%%" >> $job_to_send
1537	fi
1538
1539	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1540	then
1541	echo "llsubmit scpjob.$kennung" >> $job_to_send
1542	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1543	then
1544	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1545	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1546	echo "msub scpjob.$kennung" >> $job_to_send
1547	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1548	then
1549	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1550	elif [[ $remote_host = t3es ]]
1551	then
1552	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1553	elif [[ $remote_host = lckiaps ]]
1554	then
1555	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1556	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1557	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1558	elif [[ $remote_host = lckyu* ]]
1559	then
1560	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1561	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1562	elif [[ $remote_host = lcflow ]]
1563	then
1564	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1565	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1566	else
1567	echo "$submcom scpjob.$kennung" >> $job_to_send
1568	fi
1569	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1570	then
1571	echo "rm scpjob.$kennung" >> $job_to_send
1572	fi
1573	if [[ $remote_host = nech ]]
1574	then
1575	echo "cd -" >> $job_to_send
1576	fi
1577	else
1578	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1579	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1580	fi
1581	echo "set -x" >> $job_to_send
1582	echo " ' exit" >> $job_to_send
1583	echo "set -x" >> $job_to_send
1584	fi
1585
1586
1587
1588	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1589	cat $file_to_send >> $job_to_send
1590
1591	if [[ $remote_host = ibm ]]
1592	then
1593	echo " " >> $job_to_send
1594	echo "exit" >> $job_to_send
1595	fi
1596
1597	# remove job file
1598	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1599	then
1600	echo " " >> $job_to_send
1601	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1602	fi
1603
1604
1605
1606	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1607	if [[ -z $remote_user ]]
1608	then
1609	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1610	then
1611	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1612	fi
1613	fi
1614
1615
1616
1617	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1618	if [[ $no_submit = false ]]
1619	then
1620	if [[ $remote_host != $local_host ]]
1621	then
1622	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1623	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1624	then
1625	job_catalog_save=$job_catalog
1626	job_catalog=job_queue
1627	elif [[ $remote_host = nech ]]
1628	then
1629	job_catalog_save=$job_catalog
1630	job_catalog=/hpf/b/${remote_user}/job_queue
1631	fi
1632	if [[ $remote_host = nech ]]
1633	then
1634	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1635	# TRANSFERIERT WERDEN
1636	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1637	else
1638	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1639	fi
1640	if [[ $? = 1 ]]
1641	then
1642	locat=scp; exit
1643	fi
1644	if [[ $remote_host = ibms ]]
1645	then
1646	job_catalog=$job_catalog_save
1647	fi
1648	[[ $verify = true ]] && printf "\n >>> finished\n"
1649	else
1650	eval job_catalog=$job_catalog
1651	cp $job_to_send ${job_catalog}/$job_on_remhost
1652	fi
1653
1654
1655
1656	# NQS- BZW. LOADLEVELER-JOB STARTEN
1657	if [[ $remote_host != $local_host ]]
1658	then
1659	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1660
1661	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1662	then
1663	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1664	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1665	elif [[ $remote_host = ibmku ]]
1666	then
1667	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1668	elif [[ $remote_host = lcflow ]]
1669	then
1670	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1671	else
1672	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1673	fi
1674
1675	[[ $verify = true ]] && printf " >>> o.k.\n"
1676	else
1677	cd $job_catalog
1678	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1679	then
1680	eval $submcom $job_on_remhost
1681	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1682	then
1683	chmod u+x $job_on_remhost
1684	eval $submcom $job_on_remhost
1685	elif [[ $local_host = nech ]]
1686	then
1687	if [[ $queue = default ]]
1688	then
1689	eval $submcom $job_on_remhost
1690	else
1691	eval $submcom -q $queue $job_on_remhost
1692	fi
1693	else
1694	qsub $job_on_remhost
1695	fi
1696
1697	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1698	# only at the end of the job.
1699	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1700	then
1701	rm $job_on_remhost
1702	fi
1703	cd - > /dev/null
1704	fi
1705	fi
1706
1707
1708
1709	# ABSCHLUSSARBEITEN
1710
1711	if [[ $no_submit = false ]]
1712	then
1713	rm -f $job_to_send
1714	fi
1715	[[ $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 |