Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1264

Last change on this file since 1264 was 1264, checked in by fricke, 11 years ago
Bugfix: Using number of nodes instead of number of processors (lccrayb)
Property svn:keywords set to `Id Rev`
File size: 58.3 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	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1264 2013-12-09 12:46:09Z fricke $
30	#
31	# 1262 2013-12-09 10:57:20Z fricke
32	# further adjustments for lccrayb
33	#
34	# 1260 2013-12-04 12:48:04Z raasch
35	# jaboticaba admitted
36	#
37	# 1255 2013-11-07 14:43:35Z raasch
38	# further adjustments for lccrayb
39	#
40	# 1224 2013-09-16 07:27:23Z raasch
41	# first adjustments for lccrayb
42	#
43	# 1202 2013-07-10 16:22:07Z witha
44	# adjustments for Forwind cluster (lcflow)
45	#
46	# 1199 2013-07-05 14:52:22Z raasch
47	# adjustments for CSC Helsinki (lccrayf)
48	#
49	# use of cluster/express queue enabled (ibmh)
50	# vinessa added (imuk)
51	#
52	# 1103 2013-02-20 02:15:53Z raasch
53	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
54	# instead typeset -L/R),
55	# further adjustments for lckyuh
56	#
57	# 2013-02-10 01:47:43Z raasch
58	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
59	# and for Forwind cluster (lcflow)
60	#
61	# 1094 2013-02-03 01:52:12Z raasch
62	# new option -P for explicit setting of ssh/scp port,
63	# decalpha parts (yonsei) removed
64	#
65	# 2013-02-02 07:06:13Z raasch
66	# adjustments for Kyushu-University computing center (lckyut - tatara)
67	# old changelog messages removed
68	#
69	# 1046 2012-11-09 14:38:45Z maronga
70	# code put under GPL (PALM 3.9)
71	#
72	# 08/07/94 - Siggi - first version finished
73	# 29/06/94 - Siggi - script development started
74	#--------------------------------------------------------------------------------#
75	# subjob - script for automatic generation and submission of batch-job files
76	# for various batch queuing systems
77	#--------------------------------------------------------------------------------#
78
79
80	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
81	delete_dayfile=false
82	email_notification=none
83	group_number=none
84	locat=normal
85	no_default_queue=none
86	no_submit=false
87	job_catalog="~/job_queue"
88	job_name=none
89	local_user=$LOGNAME
90	node_usage=shared
91	numprocs=0
92	punkte="..........................................................."
93	submcom=qsub
94	queue=default
95	remote_host=none
96	remote_user=""
97	verify=true
98
99	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
100	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
101
102
103
104	# FEHLERBEHANDLUNG
105	# BEI EXIT:
106	trap 'if [[ $locat != normal ]]
107	then
108	case $locat in
109	(option) printf "\n --> available optios can be displayed"
110	printf " by typing:"
111	printf "\n \"subjob ?\" \n";;
112	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
113	(*) printf "\n +++ unknown error"
114	printf "\n please inform S. Raasch!\n"
115	esac
116	[[ -f $job_to_send ]] && rm $job_to_send
117	printf "\n\n+++ SUBJOB killed \n\n"
118	fi' exit
119
120
121	# BEI TERMINAL-BREAK:
122	trap '[[ -f $job_to_send ]] && rm $job_to_send
123	printf "\n\n+++ SUBJOB killed \n\n"
124	exit
125	' 2
126
127
128
129
130	# LOKALEN HOSTNAMEN FESTSTELLEN
131	local_host=$(hostname)
132
133
134
135	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
136	# UEBERHAUPT ZULAESSIG IST
137	# Note: One of the entries for "lck" or "lckordi" always has to be
138	# comment out, because the hostname (node*) is same for both machines
139	case $local_host in
140	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
141	(atmos) local_addres=172.20.25.35; local_host=lcide;;
142	(austru) local_addres=130.75.105.128; local_host=lcmuk;;
143	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
144	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
145	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
146	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
147	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
148	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
149	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
150	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
151	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
152	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
153	(blogin\|bxc) local_addres=130.73.233.1; local_host=lccrayb;;
154	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
155	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
156	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
157	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
158	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
159	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
160	(cs*) local_addres=136.172.44.131; local_host=nech;;
161	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
162	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
163	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
164	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
165	(node*) local_addres=165.132.26.61 local_host=lck;;
166	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
167	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
168	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
169	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
170	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
171	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
172	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
173	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
174	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
175	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
176	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
177	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
178	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
179	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
180	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
181	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
182	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
183	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
184	(jaboticaba) local_addres=150.163.25.181; local_host=lcbr;;
185	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
186	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
187	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
188	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
189	(meller) local_addres=134.106.74.155; local_host=lcfor;;
190	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
191	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
192	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
193	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
194	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
195	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
196	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
197	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
198	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
199	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
200	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
201	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
202	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
203	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
204	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
205	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
206	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
207	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
208	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
209	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
210	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
211	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
212	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
213	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
214	(*) printf "\n +++ \"$local_host\" unknown";
215	printf "\n please contact the PALM group at IMUK";
216	locat=parameter; exit;;
217	esac
218
219
220
221	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
222	remote_host=$local_host
223
224
225
226
227	# PROZEDUROPTIONEN EINLESEN
228	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
229	do
230	case $option in
231	(c) job_catalog=$OPTARG;;
232	(d) delete_dayfile=true;;
233	(D) no_submit=true;;
234	(e) email_notification=$OPTARG;;
235	(g) group_number=$OPTARG;;
236	(h) remote_host=$OPTARG;;
237	(m) memory=$OPTARG;;
238	(n) job_name=$OPTARG;;
239	(N) node_usage=$OPTARG;;
240	(O) threads_per_task=$OPTARG;;
241	(P) scp_port=$OPTARG;;
242	(q) no_default_queue=$OPTARG;;
243	(t) cputime=$OPTARG;;
244	(T) tasks_per_node=$OPTARG;;
245	(u) remote_user=$OPTARG;;
246	(v) verify=false;;
247	(X) numprocs=$OPTARG;;
248	(\?) printf "\n +++ Option $OPTARG unknown \n";
249	locat=option; exit;;
250	esac
251	done
252
253
254	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
255	(( to_shift = $OPTIND - 1 ))
256	shift $to_shift; file_to_send=$1
257
258
259	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
260	if [ "$1" = "?" ]
261	then
262	(printf "\n *** subjob can be called as follows:\n"
263	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
264	printf "\n Description of available options:\n"
265	printf "\n Option Description Default-Value"
266	printf "\n -c job-input- and output-catalog ~/job_queue"
267	printf "\n -d no job-protocol will be created ---"
268	printf "\n -D only the job-file will be created ---"
269	printf "\n -h execution host, available hosts: $remote_host"
270	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
271	printf "\n lckiaps, lctit, nech, necriam, unics"
272	printf "\n -m memory demand per process in MByte ---"
273	printf "\n -n jobname <jobdatei>"
274	printf "\n -O threads per task (for OpenMP usage) 1"
275	printf "\n -P ssh/scp port default port"
276	printf "\n -q job-queue to be used default"
277	printf "\n -t allowed cpu-time in seconds ---"
278	printf "\n -T tasks per node (on parallel hosts) ---"
279	printf "\n -u username on execution host from .netrc"
280	printf "\n -v no prompt for confirmation ---"
281	printf "\n -X # of processors (on parallel hosts) 1"
282	printf "\n "
283	printf "\n The only possible positional parameter is <jobfile>:"
284	printf "\n The complete NQS-job must be provided here."
285	printf "\n <jobfile>=? creates this outline\n\n") \| more
286	exit
287	fi
288
289
290
291	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
292	if [[ "$file_to_send" = "" ]]
293	then
294	printf "\n +++ job-file missing"
295	locat=parameter; exit
296	else
297	if [[ -f $file_to_send ]]
298	then
299	true
300	else
301	printf "\n +++ job-file: "
302	printf "\n $file_to_send"
303	printf "\n does not exist"
304	locat=parameter; exit
305	fi
306	fi
307
308
309
310	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
311	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
312	if [[ $job_name = none ]]
313	then
314	job_name=$file_to_send
315	fi
316	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
317	then
318	printf "\n +++ job-file name: "
319	printf "\n $job_name"
320	printf "\n must not contain \"/\"-characters"
321	locat=parameter; exit
322	fi
323
324
325
326
327	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
328	# ODER WENN HOST NICHT ANGEGEBEN WURDE
329	if [[ $remote_host = none ]]
330	then
331	printf "\n +++ host missing"
332	locat=option; exit
333	else
334	case $remote_host in
335	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
336	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
337	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
338	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
339	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
340	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
341	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/moab/7.2.6/bin/msub;;
342	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
343	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
344	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
345	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
346	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
347	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
348	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
349	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
350	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
351	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
352	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
353	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
354	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
355	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
356	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
357	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
358	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
359	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
360	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
361	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
362	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
363	locat=parameter; exit;;
364	esac
365	fi
366
367
368	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
369	if [[ $no_default_queue != none ]]
370	then
371	error=false
372	ndq=$no_default_queue
373	case $remote_host in
374	(ibm) case $ndq in
375	(p690_express\|p690_standard\|p690_long) error=false;;
376	(*) error=true;;
377	esac;;
378	(ibmh) case $ndq in
379	(cluster\|express) error=false;;
380	(*) error=true;;
381	esac;;
382	(ibmkisti) case $ndq in
383	(class.32plus\|class.1-2\|class.2-32) error=false;;
384	(*) error=true;;
385	esac;;
386	(ibmku) case $ndq in
387	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
388	(*) error=true;;
389	esac;;
390	(ibms) case $ndq in
391	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
392	(*) error=true;;
393	esac;;
394	(ibmy) case $ndq in
395	(parallel) error=false;;
396	(*) error=true;;
397	esac;;
398	(lccrayb) case $ndq in
399	(mpp1q\|mpp1testq\|testq) error=false;;
400	(*) error=true;;
401	esac;;
402	(lccrayf) case $ndq in
403	(usup\|test*\|small\|large) error=false;;
404	(*) error=true;;
405	esac;;
406	(lcflow) case $ndq in
407	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
408	(*) error=true;;
409	esac;;
410	(lckiaps) case $ndq in
411	(express\|normal) error=false;;
412	(*) error=true;;
413	esac;;
414	(lckyoto) case $ndq in
415	(eh\|ph) error=false;;
416	(*) error=true;;
417	esac;;
418	(lckyuh) case $ndq in
419	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
420	(*) error=true;;
421	esac;;
422	(lckyut) case $ndq in
423	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
424	(*) error=true;;
425	esac;;
426	(lcsgib\|lcsgih) case $ndq in
427	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
428	(*) error=true;;
429	esac;;
430	(lctit) case $ndq in
431	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
432	(*) error=true;;
433	esac;;
434	(t3eb) case $ndq in
435	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
436	(*) error=true;;
437	esac;;
438	(necriam) case $ndq in
439	(SP\|SS\|P6) error=false;;
440	(*) error=true;;
441	esac;;
442	(t3eh) case $ndq in
443	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
444	(*) error=true;;
445	esac;;
446	(t3ej2\|t3ej5) case $ndq in
447	(low\|normal\|high) error=false;;
448	(*) error=true;;
449	esac;;
450	(t3es) case $ndq in
451	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
452	(*) error=true;;
453	esac;;
454	(unics) case $ndq in
455	(unics\|ht) error=false;;
456	(*) error=true;;
457	esac;;
458	esac
459	if [[ $error = true ]]
460	then
461	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
462	locat=parameter; exit
463	else
464	queue=$no_default_queue
465	fi
466	fi
467
468
469
470	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
471	# AUFTEILEN
472	done=false
473	while [[ $done = false ]]
474	do
475	if (( $cputime <= 0 ))
476	then
477	printf "\n +++ wrong cpu-time or cpu-time missing"
478	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
479	printf "\n >>> "
480	read cputime 1>/dev/null 2>&1
481	else
482	done=true
483	fi
484	done
485	if [[ $remote_host = nech ]]
486	then
487	if (( tasks_per_node != 0 ))
488	then
489	(( cputime = cputime * tasks_per_node ))
490	elif [[ $numprocs != 0 ]]
491	then
492	(( cputime = cputime * numprocs ))
493	fi
494	fi
495	(( stunden = cputime / 3600 ))
496	(( resttime = cputime - stunden * 3600 ))
497	(( minuten = resttime / 60 ))
498	(( sekunden = resttime - minuten * 60 ))
499	timestring=${stunden}:${minuten}:${sekunden}
500
501
502
503	# PRUEFEN DER KERNSPEICHERANFORDERUNG
504	done=false
505	while [[ $done = false ]]
506	do
507	if (( memory <= 0 ))
508	then
509	printf "\n +++ wrong memory demand or memory demand missing"
510	printf "\n >>> Please type memory in MByte per process as INTEGER:"
511	printf "\n >>> "
512	read memory 1>/dev/null 2>&1
513	else
514	done=true
515	fi
516	done
517
518	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
519	then
520	if (( tasks_per_node != 0 ))
521	then
522	(( Memory = memory * tasks_per_node / 1000 ))
523	elif [[ $numprocs != 0 ]]
524	then
525	(( Memory = memory * numprocs / 1000 ))
526	else
527	(( Memory = memory / 1000 ))
528	fi
529	elif [[ $remote_host = lctit ]]
530	then
531	(( Memory = memory * tasks_per_node / 1000 ))
532	fi
533
534
535	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
536	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
537	then
538	(( memory = memory * threads_per_task ))
539	fi
540
541
542	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
543	if (( tasks_per_node != 0 ))
544	then
545	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
546	fi
547
548
549	# Calculate number of processes per node
550	(( processes_per_node = tasks_per_node * threads_per_task ))
551
552
553	# Calculate number of MPI tasks
554	(( mpi_tasks = numprocs / threads_per_task ))
555
556
557	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
558	if [[ "$scp_port" != "" ]]
559	then
560	PORTOPT="-P $scp_port"
561	SSH_PORTOPT="-p $scp_port"
562	fi
563
564
565	# HEADER-AUSGABE
566	if [[ $verify = true ]]
567	then
568	printf "\n\n"
569	printf "#--------------------------------------------------------------# \n"
570	spalte1=SUBJOB;spalte2=$(date)
571	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
572	printf "\| \| \n"
573	printf "\| values of parameters/options: \| \n"
574	spalte1=$(echo local_host$punkte \| cut -c-20)
575	spalte2=$punkte$local_host
576	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
577	spalte1=$(echo remote_host$punkte \| cut -c-20)
578	spalte2=$punkte$remote_host
579	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
580	spalte1=$(echo queue$punkte \| cut -c-20)
581	spalte2=$punkte$queue
582	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
583	spalte1=$(echo memory$punkte \| cut -c-20)
584	spalte2="$punkte$memory mb"
585	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
586	spalte1=$(echo cputime$punkte \| cut -c-20)
587	spalte2="$punkte$cputime sec"
588	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
589	spalte1=$(echo job_name$punkte \| cut -c-20)
590	spalte2="$punkte$job_name"
591	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
592	printf "#--------------------------------------------------------------# \n\n"
593
594
595	# KONTROLLABFRAGE, OB ALLES O.K.
596	antwort="dummy"
597	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
598	do
599	read antwort?" >>> continue (y/n) ? "
600	done
601	if [[ $antwort = n \|\| $antwort = N ]]
602	then
603	locat=verify; exit
604	fi
605	printf "\n"
606	fi
607
608
609
610	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
611	kennung=$RANDOM
612	job_on_remhost=${job_name}_${kennung}_$local_host
613	job_to_send=job_to_send_$kennung
614	if [[ $delete_dayfile = false ]]
615	then
616	remote_dayfile=${local_host}_${job_name}_result_$kennung
617	local_dayfile=${remote_host}_${job_name}
618	else
619	remote_dayfile=/dev/null
620	fi
621
622
623
624	# Generate the batch job scripts (qsub/msub/LoadLeveler)
625	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
626	then
627
628	# General LoadLeveler settings
629	execute_in_shell="#!/bin/ksh"
630	use_shell="# @ shell = /bin/ksh"
631	consumable_memory="ConsumableMemory($memory mb)"
632	class="# @ class = $queue"
633	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
634	network_to_use="# @ network.mpi = sn_all,shared,us"
635	data_limit="# @ data_limit = 1.76gb"
636	image_size="# @ image_size = 50"
637	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
638
639	if [[ $email_notification = none ]]
640	then
641	notify_user=""
642	else
643	notify_user="# @ notify_user = $email_notification"
644	if [[ $delete_dayfile = true ]]
645	then
646	notification='# @ notification = never'
647	fi
648	fi
649
650	if [[ $remote_host = ibmh ]]
651	then
652	data_limit=""
653	network_to_use=""
654	class="# @ class = $queue"
655	environment=""
656	rset="# @ rset = RSET_MCM_AFFINITY"
657	task_affinity="# @ task_affinity = core(1)"
658	elif [[ $remote_host = ibmkisti ]]
659	then
660	network_to_use="# @ network.MPI = sn_all,shared,US"
661	wall_clock_limit="# @ wall_clock_limit = $timestring"
662	if [[ $threads_per_task = 1 ]]
663	then
664	rset="# @ rset = RSET_MCM_AFFINITY"
665	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
666	fi
667	environment=""
668	use_shell=""
669	data_limit=""
670	image_size=""
671	elif [[ $remote_host = ibmku ]]
672	then
673	execute_in_shell="#!/usr/bin/ksh"
674	use_shell="# @ shell = /usr/bin/ksh"
675	consumable_memory=""
676	environment=""
677	network_to_use="# @ network.mpi = sn_all,shared,us"
678	data_limit=""
679	image_size=""
680	elif [[ $remote_host = ibms ]]
681	then
682	network_to_use="# @ network.mpi = csss,shared,us"
683	elif [[ $remote_host = ibmy ]]
684	then
685	consumable_memory=""
686	network_to_use=""
687	fi
688
689	cat > $job_to_send << %%END%%
690	$execute_in_shell
691	$use_shell
692
693	# @ job_type = parallel
694	# @ job_name = $job_name
695	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
696	# @ output = $remote_dayfile
697	# @ error = $remote_dayfile
698	$wall_clock_limit
699	$image_size
700	$class
701	$environment
702	$network_to_use
703	$data_limit
704	$rset
705	$mcm_affinity_options
706	$task_affinity
707	$notification
708	$notify_user
709
710	%%END%%
711
712	if (( nodes > 0 ))
713	then
714
715	if [[ $remote_host != ibmkisti ]]
716	then
717
718	cat >> $job_to_send << %%END%%
719	# @ node = $nodes
720	# @ tasks_per_node = $processes_per_node
721	# @ node_usage = $node_usage
722	# @ queue
723
724	%%END%%
725
726	else
727
728	cat >> $job_to_send << %%END%%
729	# @ total_tasks = $mpi_tasks
730	# @ blocking = unlimited
731	# @ queue
732
733	%%END%%
734
735	fi
736
737	else
738
739	if [[ $remote_host != ibmy ]]
740	then
741
742	cat >> $job_to_send << %%END%%
743	# @ blocking = unlimited
744	# @ total_tasks = $numprocs
745	# @ node_usage = $node_usage
746	# @ queue
747
748	%%END%%
749
750	else
751
752	cat >> $job_to_send << %%END%%
753	# @ node = 1
754	# @ total_tasks = $numprocs
755	# @ queue
756
757	%%END%%
758
759	fi
760
761	fi
762
763	# workaround because of silly job filter on ibmkisti
764	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
765	then
766	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
767	fi
768
769	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
770	then
771
772	cat > $job_to_send << %%END%%
773	#!/bin/ksh
774
775	# @ job_type = serial
776	# @ node_usage = $node_usage
777	# @ job_name = palm
778	# @ wall_clock_limit = ${timestring},$timestring
779	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
780	# @ output = $remote_dayfile
781	# @ error = $remote_dayfile
782	$class
783	$notification
784
785	# @ queue
786
787	%%END%%
788
789	elif [[ $remote_host = lccrayb ]]
790	then
791
792	if [[ $numprocs != 0 ]]
793	then
794	cat > $job_to_send << %%END%%
795	#!/bin/bash
796	#PBS -N $job_name
797	#PBS -l walltime=$timestring
798	#PBS -l nodes=$nodes:ppn=$processes_per_node
799	#PBS -o $remote_dayfile
800	#PBS -j oe
801	#PBS -q $queue
802
803	$module_calls
804
805	%%END%%
806
807	else
808
809	continue
810
811	fi
812
813	elif [[ $remote_host = lccrayf ]]
814	then
815
816	if [[ $numprocs != 0 ]]
817	then
818	cat > $job_to_send << %%END%%
819	#!/bin/bash -l
820	#SBATCH -J $job_name
821	#SBATCH -t $timestring
822	#SBATCH -N $nodes
823	#SBATCH --ntasks-per-node=$processes_per_node
824	#SBATCH -p $queue
825	#SBATCH -o $remote_dayfile
826	#SBATCH -e $remote_dayfile
827
828	$init_cmds
829	$module_calls
830
831	%%END%%
832
833	else
834	cat > $job_to_send << %%END%%
835	#!/bin/bash -l
836	#SBATCH -J $job_name
837	#SBATCH -t $timestring
838	#SBATCH -l ncpus=1
839	#SBATCH -l pmem=${memory}mb
840	#SBATCH -m abe
841	#SBATCH -o $remote_dayfile
842	#SBATCH -e $remote_dayfile
843
844	$init_cmds
845	$module_calls
846
847	%%END%%
848
849	fi
850
851	elif [[ $remote_host = lcfimm ]]
852	then
853
854	if [[ $numprocs != 0 ]]
855	then
856	cat > $job_to_send << %%END%%
857	#!/bin/ksh
858	#PBS -N $job_name
859	#PBS -A $project_account
860	#PBS -l walltime=$timestring
861	#PBS -l nodes=${nodes}:ppn=$processes_per_node
862	#PBS -l pmem=${memory}mb
863	#PBS -m abe
864	#PBS -o $remote_dayfile
865	#PBS -j oe
866	mpd &
867
868	%%END%%
869
870	else
871	cat > $job_to_send << %%END%%
872	#!/bin/ksh
873	#PBS -N $job_name
874	#PBS -A $project_account
875	#PBS -l walltime=$timestring
876	#PBS -l ncpus=1
877	#PBS -l pmem=${memory}mb
878	#PBS -m abe
879	#PBS -o $remote_dayfile
880	#PBS -j oe
881
882	%%END%%
883
884	fi
885
886	elif [[ $remote_host = lcflow ]]
887	then
888	if [ $memory -gt 1800 ]; then
889	use_himem=""
890	else
891	use_himem="#"
892	fi
893	if [[ $numprocs != 0 ]]
894	then
895	pe_set="#$ -pe impi41 $numprocs"
896	else
897	pe_set="#$ -pe impi41 1"
898	fi
899	if [[ $queue = default ]]
900	then
901	queue_set=""
902	else
903	queue_set="#$ -q $queue"
904	fi
905	[[ "$disc_space" = "" ]] && disc_space=50
906
907	cat > $job_to_send << %%END%%
908	#!/bin/bash
909	#$ -S /bin/bash
910	#$ -N $job_name
911	#$ -cwd
912	#$ -l h_rt=$timestring
913	#$ -l h_vmem=${memory}M
914	#$ -o $remote_dayfile
915	#$ -j y
916	$pe_set
917	#$ -R y
918	${use_himem}#$ -l highmem=true
919	#$ -l h_fsize=${disc_space}G
920	$queue_set
921
922	%%END%%
923
924	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
925	then
926
927	if [[ $numprocs != 0 ]]
928	then
929	cat > $job_to_send << %%END%%
930	#!/bin/ksh
931	#PBS -N $job_name
932	#PBS -l walltime=$timestring
933	#PBS -l ncpus=$numprocs
934	#PBS -l pmem=${memory}mb
935	#PBS -o $remote_dayfile
936	#PBS -l nodes=$nodes:ppn=${processes_per_node}
937	#PBS -j oe
938
939	mpd &
940
941	%%END%%
942
943	else
944	cat > $job_to_send << %%END%%
945	#!/bin/ksh
946	#PBS -N $job_name
947	#PBS -l walltime=$timestring
948	#PBS -l ncpus=1
949	#PBS -l pmem=${memory}mb
950	#PBS -o $remote_dayfile
951	#PBS -j oe
952
953	%%END%%
954
955	fi
956
957	elif [[ $remote_host = lckiaps ]]
958	then
959
960	if [[ $numprocs != 0 ]]
961	then
962	cat > $job_to_send << %%END%%
963	#!/bin/ksh
964	#PBS -N $job_name
965	#PBS -l walltime=$timestring
966	#PBS -l select=1:ncpus=$numprocs
967	#PBS -l pmem=${memory}mb
968	#PBS -q $queue
969	#PBS -o $remote_dayfile
970	#PBS -j oe
971	#PBS -V
972
973	%%END%%
974
975	else
976	cat > $job_to_send << %%END%%
977	#!/bin/ksh
978	#PBS -N $job_name
979	#PBS -l walltime=$timestring
980	#PBS -l ncpus=1
981	#PBS -l pmem=${memory}mb
982	#PBS -o $remote_dayfile
983	#PBS -j oe
984
985	%%END%%
986
987	fi
988
989	elif [[ $remote_host = lcyon ]]
990	then
991
992	if [[ $numprocs != 0 ]]
993	then
994	cat > $job_to_send << %%END%%
995	#!/bin/ksh
996	#PBS -N $job_name
997	#PBS -l walltime=$timestring
998	#PBS -l ncpus=$numprocs
999	#PBS -l pmem=${memory}mb
1000	#PBS -o $remote_dayfile
1001	#PBS -j oe
1002
1003	%%END%%
1004
1005	else
1006	cat > $job_to_send << %%END%%
1007	#!/bin/ksh
1008	#PBS -N $job_name
1009	#PBS -l walltime=$timestring
1010	#PBS -l ncpus=1
1011	#PBS -l pmem=${memory}mb
1012	#PBS -o $remote_dayfile
1013	#PBS -j oe
1014
1015	%%END%%
1016
1017	fi
1018
1019	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1020	then
1021
1022	if [[ "$sgi_feature" != "" ]]
1023	then
1024	feature_directive="#PBS -l feature=$sgi_feature"
1025	else
1026	feature_directive=""
1027	fi
1028
1029	if [[ $queue = dataq \|\| $queue = permq ]]
1030	then
1031	feature_directive="#PBS -l feature=data"
1032	fi
1033
1034	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1035	then
1036	queue_directive="#PBS -q $queue"
1037	else
1038	queue_directive=""
1039	fi
1040
1041	if [[ $email_notification = none ]]
1042	then
1043	email_directive=""
1044	else
1045	email_directive="#PBS -M $email_notification"
1046	fi
1047
1048	if [[ $numprocs != 0 ]]
1049	then
1050	cat > $job_to_send << %%END%%
1051	#!/bin/bash
1052	#PBS -N $job_name
1053	#PBS -l walltime=$timestring
1054	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1055	#PBS -l naccesspolicy=$node_usage
1056	#PBS -o $remote_dayfile
1057	#PBS -j oe
1058	$feature_directive
1059	$queue_directive
1060	$email_directive
1061
1062	eval \`/sw/swdist/bin/modulesinit\`
1063	#. /usr/share/modules/init/bash
1064	$init_cmds
1065	$module_calls
1066
1067	echo ld_library_path=\$LD_LIBRARY_PATH
1068
1069	%%END%%
1070
1071	else
1072	cat > $job_to_send << %%END%%
1073	#PBS -S /bin/bash
1074	#PBS -N $job_name
1075	#PBS -l walltime=$timestring
1076	#PBS -l ncpus=1
1077	#PBS -o $remote_dayfile
1078	#PBS -j oe
1079	$queue_directive
1080	$email_directive
1081
1082	eval \`/sw/swdist/bin/modulesinit\`
1083	#. /usr/share/modules/init/bash
1084	$init_cmds
1085	$module_calls
1086
1087	%%END%%
1088
1089	fi
1090
1091	elif [[ $remote_host = lcxe6 ]]
1092	then
1093
1094	if [[ $numprocs != 0 ]]
1095	then
1096	cat > $job_to_send << %%END%%
1097	#!/bin/ksh
1098	#PBS -S /bin/ksh
1099	#PBS -N $job_name
1100	#PBS -A $project_account
1101	#PBS -j oe
1102	#PBS -l walltime=$timestring
1103	#PBS -l mppwidth=${numprocs}
1104	#PBS -l mppnppn=${processes_per_node}
1105	#PBS -m abe
1106	#PBS -o $remote_dayfile
1107	$email_directive
1108
1109	$init_cmds
1110	$module_calls
1111
1112	%%END%%
1113
1114	else
1115	cat > $job_to_send << %%END%%
1116	#!/bin/ksh
1117	#PBS -S /bin/ksh
1118	#PBS -N $job_name
1119	#PBS -A $project_account
1120	#PBS -j oe
1121	#PBS -l walltime=$timestring
1122	#PBS -l ncpus=1
1123	#PBS -l pmem=${memory}mb
1124	#PBS -m abe
1125	$email_directive
1126	#PBS -o $remote_dayfile
1127
1128	$init_cmds
1129	$module_calls
1130
1131	%%END%%
1132
1133	fi
1134
1135	elif [[ $remote_host = lckyoto ]]
1136	then
1137
1138	cat > $job_to_send << %%END%%
1139	#!/bin/ksh
1140	# @\$-o $remote_dayfile
1141	# @\$-eo -oi
1142	# @\$-lP 16
1143	# @\$-lp 1
1144	# @\$-lm 28gb -llm unlimited -ls unlimited
1145	# @\$-q $queue
1146	# @\$-Pvn abs_pack
1147	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1148	#. /thin/local/etc/setprofile/intel-11.0.sh
1149	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1150	. ~/.myprofile
1151	#. /home2/t/t51254/palm/current_version/myprofile
1152	#. /thin/apps/pgi/mpi.sh
1153	#
1154	env
1155	#
1156	set -x
1157
1158	%%END%%
1159
1160	elif [[ $remote_host = lcxt5m ]]
1161	then
1162
1163	if [[ $numprocs != 0 ]]
1164	then
1165	cat > $job_to_send << %%END%%
1166	#!/bin/ksh
1167	#PBS -S /bin/ksh
1168	#PBS -N $job_name
1169	#PBS -j oe
1170	#PBS -l walltime=$timestring
1171	#PBS -l mppwidth=${numprocs}
1172	#PBS -l mppnppn=${processes_per_node}
1173	#PBS -m abe
1174	#PBS -o $remote_dayfile
1175
1176	$init_cmds
1177	$module_calls
1178
1179	%%END%%
1180
1181	else
1182	cat > $job_to_send << %%END%%
1183	#!/bin/ksh
1184	#PBS -S /bin/ksh
1185	#PBS -N $job_name
1186	#PBS -j oe
1187	#PBS -l walltime=$timestring
1188	#PBS -l ncpus=1
1189	#PBS -l pmem=${memory}mb
1190	#PBS -m abe
1191	#PBS -o $remote_dayfile
1192
1193	$init_cmds
1194	$module_calls
1195
1196	%%END%%
1197
1198	fi
1199
1200	elif [[ $remote_host = lckyuh ]]
1201	then
1202	cat > $job_to_send << %%END%%
1203	#!/bin/bash
1204	#PJM -L "rscgrp=$queue"
1205	#PJM -L "node=$nodes"
1206	#PJM --mpi "proc=$numprocs"
1207	#PJM -L "elapse=$timestring"
1208	#PJM -o $remote_dayfile
1209	#PJM -j
1210	#PJM -X
1211	#PJM --no-stging
1212
1213	export LANG=en_US.UTF-8
1214	%%END%%
1215
1216	elif [[ $remote_host = lckyut ]]
1217	then
1218	cat > $job_to_send << %%END%%
1219	#!/bin/bash
1220	#PJM -L "rscgrp=$queue"
1221	#PJM -L "vnode=$numprocs"
1222	#PJM -L "vnode-core=1"
1223	#PJM -L "elapse=$timestring"
1224	#PJM --mpi proc=$numprocs
1225	#PJM -o $remote_dayfile
1226	#PJM -j
1227	#PJM -X
1228	#PJM --no-stging
1229
1230	export LANG=en_US.UTF-8
1231	%%END%%
1232
1233	elif [[ $remote_host = nech ]]
1234	then
1235
1236	if (( nodes > 1 ))
1237	then
1238	cat > $job_to_send << %%END%%
1239	#!/bin/ksh
1240	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1241	#PBS -l ${qsubmem}=${Memory}gb
1242	#PBS -b $nodes
1243	#PBS -o $remote_dayfile
1244	#PBS -N palm
1245	#PBS -j o
1246	#PBS -T mpisx
1247
1248	%%END%%
1249
1250	elif [[ $numprocs != 0 ]]
1251	then
1252	cat > $job_to_send << %%END%%
1253	#!/bin/ksh
1254	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1255	#PBS -l ${qsubmem}=${Memory}gb
1256	#PBS -o $remote_dayfile
1257	#PBS -N palm
1258	#PBS -j o
1259
1260	%%END%%
1261
1262	else
1263	cat > $job_to_send << %%END%%
1264	#!/bin/ksh
1265	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1266	#PBS -o $remote_dayfile
1267	#PBS -j o
1268
1269	%%END%%
1270
1271	fi
1272
1273	elif [[ $remote_host = necriam ]]
1274	then
1275
1276	if (( nodes > 1 ))
1277	then
1278	cat > $job_to_send << %%END%%
1279	#!/bin/ksh
1280	#PBS -b $nodes
1281	#PBS -o $remote_dayfile
1282	#PBS -N $job_name
1283	#PBS -j o
1284	#PBS -v MPIPROGINV=YES
1285
1286	%%END%%
1287
1288	elif [[ $numprocs != 0 ]]
1289	then
1290	cat > $job_to_send << %%END%%
1291	#!/bin/ksh
1292	#PBS -q ${queue}
1293	#PBS -o $remote_dayfile
1294	#PBS -N $job_name
1295	#PBS -j o
1296	#PBS -v MPIPROGINV=YES
1297
1298	%%END%%
1299
1300	else
1301	cat > $job_to_send << %%END%%
1302	#!/bin/ksh
1303	#PBS -o $remote_dayfile
1304	#PBS -j o
1305
1306	%%END%%
1307
1308	fi
1309
1310	elif [[ $remote_host = lctit ]]
1311	then
1312	cat > $job_to_send << %%END%%
1313	#!/bin/ksh
1314	$init_cmds
1315	$module_calls
1316
1317	%%END%%
1318
1319	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1320	if [[ $tasks_per_node != $processes_per_node ]]
1321	then
1322	submcom="$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 $queue "
1323	else
1324	submcom="$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 "
1325	fi
1326
1327	else
1328
1329	cat > $job_to_send << %%END%%
1330	# @\$-q ${queue}
1331	# @\$-l${qsubtime} $timestring
1332	# @\$-l${qsubmem} ${memory}mb
1333	# @\$-o $remote_dayfile
1334	# @\$-eo
1335
1336	%%END%%
1337
1338	fi
1339
1340
1341	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1342	# BEI EXIT VERANLASSEN
1343	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1344	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1345	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1346	then
1347	echo "set +vx" >> $job_to_send
1348	echo "trap '" >> $job_to_send
1349	echo "set +vx" >> $job_to_send
1350	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lccrayb \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* ]]
1351	then
1352	if [[ $remote_host = ibmh ]]
1353	then
1354	return_queue=c1
1355	elif [[ $remote_host = ibmkisti ]]
1356	then
1357	return_queue=class.1-2
1358	elif [[ $remote_host = ibmku ]]
1359	then
1360	return_queue=sdbg2
1361	elif [[ $remote_host = ibms ]]
1362	then
1363	return_queue=p_normal
1364	elif [[ $remote_host = ibmy ]]
1365	then
1366	return_queue=serial
1367	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1368	then
1369	return_queue=serialq
1370	elif [[ $remote_host = necriam ]]
1371	then
1372	return_queue=SP
1373	elif [[ $remote_host = lccrayb ]]
1374	then
1375	return_queue=dataq
1376	elif [[ $remote_host = lckiaps ]]
1377	then
1378	return_queue=express
1379	elif [[ $remote_host = lckyuh ]]
1380	then
1381	return_queue=cx-single
1382	elif [[ $remote_host = lckyut ]]
1383	then
1384	return_queue=cx-single
1385	else
1386	return_queue=unknown
1387	fi
1388
1389	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1390	then
1391
1392	if [[ $remote_host = ibmku ]]
1393	then
1394	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1395	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1396	else
1397	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1398	fi
1399	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1400	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1401	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1402	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1403	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1404	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1405	if [[ $host != "ibmh" ]]
1406	then
1407	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1408	fi
1409	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1410	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1411
1412	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1413	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1414
1415	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1416	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
1417	if [[ $remote_host = ibmku ]]
1418	then
1419	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1420	fi
1421	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1422
1423	elif [[ $remote_host = nech ]]
1424	then
1425	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1426	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1427	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1428	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1429	echo "#PBS -j o" >> $job_to_send
1430	echo " " >> $job_to_send
1431	echo "set -x" >> $job_to_send
1432	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1433	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1434	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1435	echo "%%END%%" >> $job_to_send
1436
1437	elif [[ $remote_host = necriam ]]
1438	then
1439	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1440	echo "#PBS -q $return_queue" >> $job_to_send
1441	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1442	echo "#PBS -j o" >> $job_to_send
1443	echo " " >> $job_to_send
1444	echo "set -x" >> $job_to_send
1445	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1446	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1447	echo "%%END%%" >> $job_to_send
1448
1449	elif [[ $remote_host = lckyuh ]]
1450	then
1451	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1452	echo "#!/bin/bash" >> $job_to_send
1453	echo "#PJM -L \"node=1\"" >> $job_to_send
1454	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1455	echo "#PJM --no-stging" >> $job_to_send
1456	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1457	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1458	echo "#PJM -j" >> $job_to_send
1459	echo " " >> $job_to_send
1460	echo "export LANG=en_US.UTF-8" >> $job_to_send
1461	echo "set -x" >> $job_to_send
1462	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1463	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1464	echo "%%END%%" >> $job_to_send
1465
1466	elif [[ $remote_host = lckyut ]]
1467	then
1468	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1469	echo "#!/bin/bash" >> $job_to_send
1470	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1471	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1472	echo "#PJM --no-stging" >> $job_to_send
1473	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1474	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1475	echo "#PJM -j" >> $job_to_send
1476	echo " " >> $job_to_send
1477	echo "export LANG=en_US.UTF-8" >> $job_to_send
1478	echo "set -x" >> $job_to_send
1479	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1480	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1481	echo "%%END%%" >> $job_to_send
1482
1483	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1484	then
1485	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1486	echo "#!/bin/bash" >> $job_to_send
1487	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1488	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1489	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1490	echo "#PBS -l feature=data" >> $job_to_send
1491	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1492	echo "#PBS -j oe" >> $job_to_send
1493	echo " " >> $job_to_send
1494	echo ". /usr/share/modules/init/bash" >> $job_to_send
1495	echo "set -x" >> $job_to_send
1496	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1497	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1498	echo "%%END%%" >> $job_to_send
1499
1500	elif [[ $remote_host = lccrayb ]]
1501	then
1502	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1503	echo "#!/bin/bash" >> $job_to_send
1504	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1505	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1506	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1507	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1508	echo "#PBS -j oe" >> $job_to_send
1509	echo " " >> $job_to_send
1510	echo "set -x" >> $job_to_send
1511	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1512	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1513	echo "%%END%%" >> $job_to_send
1514
1515	elif [[ $remote_host = lcflow ]]
1516	then
1517	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1518	echo "#!/bin/bash" >> $job_to_send
1519	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1520	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1521	echo "SGEPREFIX -cwd" >> $job_to_send
1522	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1523	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1524	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1525	echo "SGEPREFIX -j y" >> $job_to_send
1526	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1527	echo " " >> $job_to_send
1528	echo "set -x" >> $job_to_send
1529	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1530	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1531	echo "" >> $job_to_send
1532	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1533	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1534	echo "rm -f scpjob.${kennung}" >> $job_to_send
1535	echo "%%END%%" >> $job_to_send
1536	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1537	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1538
1539	else
1540
1541	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1542	echo "# @\\\$-q $return_queue" >> $job_to_send
1543	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1544	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1545	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1546	then
1547	echo "# @\$-l mpp_p=0" >> $job_to_send
1548	fi
1549	echo '# @\$-lF 10mb' >> $job_to_send
1550	# echo '# @\$-o /dev/null' >> $job_to_send
1551	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1552	echo '# @\\\$-eo' >> $job_to_send
1553	echo " " >> $job_to_send
1554	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1555	then
1556	echo "set +vx" >> $job_to_send
1557	echo ". .profile" >> $job_to_send
1558	fi
1559	echo "set -x" >> $job_to_send
1560	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
1561	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1562	echo "%%END%%" >> $job_to_send
1563	fi
1564
1565	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1566	then
1567	echo "llsubmit scpjob.$kennung" >> $job_to_send
1568	elif [[ $remote_host = lccrayb ]]
1569	then
1570	echo "qsub -q $return_queue scpjob.$kennung" >> $job_to_send
1571	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1572	then
1573	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1574	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1575	echo "msub scpjob.$kennung" >> $job_to_send
1576	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1577	then
1578	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1579	elif [[ $remote_host = t3es ]]
1580	then
1581	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1582	elif [[ $remote_host = lckiaps ]]
1583	then
1584	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1585	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1586	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1587	elif [[ $remote_host = lckyu* ]]
1588	then
1589	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1590	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1591	elif [[ $remote_host = lcflow ]]
1592	then
1593	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1594	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1595	else
1596	echo "$submcom scpjob.$kennung" >> $job_to_send
1597	fi
1598	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1599	then
1600	echo "rm scpjob.$kennung" >> $job_to_send
1601	fi
1602	if [[ $remote_host = nech ]]
1603	then
1604	echo "cd -" >> $job_to_send
1605	fi
1606	else
1607	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1608	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1609	fi
1610	echo "set -x" >> $job_to_send
1611	echo " ' exit" >> $job_to_send
1612	echo "set -x" >> $job_to_send
1613	fi
1614
1615
1616
1617	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1618	cat $file_to_send >> $job_to_send
1619
1620	if [[ $remote_host = ibm ]]
1621	then
1622	echo " " >> $job_to_send
1623	echo "exit" >> $job_to_send
1624	fi
1625
1626	# remove job file
1627	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1628	then
1629	echo " " >> $job_to_send
1630	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1631	fi
1632
1633
1634
1635	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1636	if [[ -z $remote_user ]]
1637	then
1638	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1639	then
1640	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1641	fi
1642	fi
1643
1644
1645
1646	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1647	if [[ $no_submit = false ]]
1648	then
1649	if [[ $remote_host != $local_host ]]
1650	then
1651	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1652	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1653	then
1654	job_catalog_save=$job_catalog
1655	job_catalog=job_queue
1656	elif [[ $remote_host = nech ]]
1657	then
1658	job_catalog_save=$job_catalog
1659	job_catalog=/hpf/b/${remote_user}/job_queue
1660	fi
1661	if [[ $remote_host = nech ]]
1662	then
1663	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1664	# TRANSFERIERT WERDEN
1665	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1666	else
1667	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1668	fi
1669	if [[ $? = 1 ]]
1670	then
1671	locat=scp; exit
1672	fi
1673	if [[ $remote_host = ibms ]]
1674	then
1675	job_catalog=$job_catalog_save
1676	fi
1677	[[ $verify = true ]] && printf "\n >>> finished\n"
1678	else
1679	eval job_catalog=$job_catalog
1680	cp $job_to_send ${job_catalog}/$job_on_remhost
1681	fi
1682
1683
1684
1685	# NQS- BZW. LOADLEVELER-JOB STARTEN
1686	if [[ $remote_host != $local_host ]]
1687	then
1688	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1689
1690	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1691	then
1692	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1693	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1694	elif [[ $remote_host = ibmku ]]
1695	then
1696	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1697	elif [[ $remote_host = lcflow ]]
1698	then
1699	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1700	else
1701	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1702	fi
1703
1704	[[ $verify = true ]] && printf " >>> o.k.\n"
1705	else
1706	cd $job_catalog
1707	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1708	then
1709	eval $submcom $job_on_remhost
1710	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1711	then
1712	chmod u+x $job_on_remhost
1713	eval $submcom $job_on_remhost
1714	elif [[ $local_host = nech ]]
1715	then
1716	if [[ $queue = default ]]
1717	then
1718	eval $submcom $job_on_remhost
1719	else
1720	eval $submcom -q $queue $job_on_remhost
1721	fi
1722	else
1723	qsub $job_on_remhost
1724	fi
1725
1726	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1727	# only at the end of the job.
1728	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1729	then
1730	rm $job_on_remhost
1731	fi
1732	cd - > /dev/null
1733	fi
1734	fi
1735
1736
1737
1738	# ABSCHLUSSARBEITEN
1739
1740	if [[ $no_submit = false ]]
1741	then
1742	rm -f $job_to_send
1743	fi
1744	[[ $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 |