Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1202

Last change on this file since 1202 was 1202, checked in by witha, 11 years ago
script adjustments for Forwind cluster (lcflow)
Property svn:keywords set to `Id Rev`
File size: 55.6 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	# adjustments for Forwind cluster (lcflow)
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1202 2013-07-10 16:22:07Z witha $
30	#
31	# 1199 2013-07-05 14:52:22Z raasch
32	# adjustments for CSC Helsinki (lccrayf)
33	#
34	# use of cluster/express queue enabled (ibmh)
35	# vinessa added (imuk)
36	#
37	# 1103 2013-02-20 02:15:53Z raasch
38	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
39	# instead typeset -L/R),
40	# further adjustments for lckyuh
41	#
42	# 2013-02-10 01:47:43Z raasch
43	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
44	# and for Forwind cluster (lcflow)
45	#
46	# 1094 2013-02-03 01:52:12Z raasch
47	# new option -P for explicit setting of ssh/scp port,
48	# decalpha parts (yonsei) removed
49	#
50	# 2013-02-02 07:06:13Z raasch
51	# adjustments for Kyushu-University computing center (lckyut - tatara)
52	# old changelog messages removed
53	#
54	# 1046 2012-11-09 14:38:45Z maronga
55	# code put under GPL (PALM 3.9)
56	#
57	# 08/07/94 - Siggi - first version finished
58	# 29/06/94 - Siggi - script development started
59	#--------------------------------------------------------------------------------#
60	# subjob - script for automatic generation and submission of batch-job files
61	# for various batch queuing systems
62	#--------------------------------------------------------------------------------#
63
64
65	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
66	delete_dayfile=false
67	email_notification=none
68	group_number=none
69	locat=normal
70	no_default_queue=none
71	no_submit=false
72	job_catalog="~/job_queue"
73	job_name=none
74	local_user=$LOGNAME
75	node_usage=shared
76	numprocs=0
77	punkte="..........................................................."
78	submcom=qsub
79	queue=default
80	remote_host=none
81	remote_user=""
82	verify=true
83
84	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
85	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
86
87
88
89	# FEHLERBEHANDLUNG
90	# BEI EXIT:
91	trap 'if [[ $locat != normal ]]
92	then
93	case $locat in
94	(option) printf "\n --> available optios can be displayed"
95	printf " by typing:"
96	printf "\n \"subjob ?\" \n";;
97	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
98	(*) printf "\n +++ unknown error"
99	printf "\n please inform S. Raasch!\n"
100	esac
101	[[ -f $job_to_send ]] && rm $job_to_send
102	printf "\n\n+++ SUBJOB killed \n\n"
103	fi' exit
104
105
106	# BEI TERMINAL-BREAK:
107	trap '[[ -f $job_to_send ]] && rm $job_to_send
108	printf "\n\n+++ SUBJOB killed \n\n"
109	exit
110	' 2
111
112
113
114
115	# LOKALEN HOSTNAMEN FESTSTELLEN
116	local_host=$(hostname)
117
118
119
120	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
121	# UEBERHAUPT ZULAESSIG IST
122	# Note: One of the entries for "lck" or "lckordi" always has to be
123	# comment out, because the hostname (node*) is same for both machines
124	case $local_host in
125	(atmos) local_addres=172.20.25.35; local_host=lcide;;
126	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
127	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
128	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
129	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
130	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
131	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
132	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
133	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
134	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
135	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
136	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
137	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
138	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
139	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
140	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
141	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
142	(cs*) local_addres=136.172.44.131; local_host=nech;;
143	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
144	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
145	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
146	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
147	(node*) local_addres=165.132.26.61 local_host=lck;;
148	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
149	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
150	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
151	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
152	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
153	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
154	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
155	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
156	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
157	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
158	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
159	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
160	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
161	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
162	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
163	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
164	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
165	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
166	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
167	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
168	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
169	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
170	(meller) local_addres=134.106.74.155; local_host=lcfor;;
171	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
172	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
173	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
174	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
175	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
176	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
177	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
178	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
179	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
180	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
181	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
182	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
183	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
184	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
185	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
186	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
187	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
188	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
189	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
190	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
191	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
192	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
193	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
194	(*) printf "\n +++ \"$local_host\" unknown";
195	printf "\n please inform S. Raasch!";
196	locat=parameter; exit;;
197	esac
198
199
200
201	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
202	remote_host=$local_host
203
204
205
206
207	# PROZEDUROPTIONEN EINLESEN
208	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
209	do
210	case $option in
211	(c) job_catalog=$OPTARG;;
212	(d) delete_dayfile=true;;
213	(D) no_submit=true;;
214	(e) email_notification=$OPTARG;;
215	(g) group_number=$OPTARG;;
216	(h) remote_host=$OPTARG;;
217	(m) memory=$OPTARG;;
218	(n) job_name=$OPTARG;;
219	(N) node_usage=$OPTARG;;
220	(O) threads_per_task=$OPTARG;;
221	(P) scp_port=$OPTARG;;
222	(q) no_default_queue=$OPTARG;;
223	(t) cputime=$OPTARG;;
224	(T) tasks_per_node=$OPTARG;;
225	(u) remote_user=$OPTARG;;
226	(v) verify=false;;
227	(X) numprocs=$OPTARG;;
228	(\?) printf "\n +++ Option $OPTARG unknown \n";
229	locat=option; exit;;
230	esac
231	done
232
233
234	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
235	(( to_shift = $OPTIND - 1 ))
236	shift $to_shift; file_to_send=$1
237
238
239	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
240	if [ "$1" = "?" ]
241	then
242	(printf "\n *** subjob can be called as follows:\n"
243	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
244	printf "\n Description of available options:\n"
245	printf "\n Option Description Default-Value"
246	printf "\n -c job-input- and output-catalog ~/job_queue"
247	printf "\n -d no job-protocol will be created ---"
248	printf "\n -D only the job-file will be created ---"
249	printf "\n -h execution host, available hosts: $remote_host"
250	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
251	printf "\n lckiaps, lctit, nech, necriam, unics"
252	printf "\n -m memory demand per process in MByte ---"
253	printf "\n -n jobname <jobdatei>"
254	printf "\n -O threads per task (for OpenMP usage) 1"
255	printf "\n -P ssh/scp port default port"
256	printf "\n -q job-queue to be used default"
257	printf "\n -t allowed cpu-time in seconds ---"
258	printf "\n -T tasks per node (on parallel hosts) ---"
259	printf "\n -u username on execution host from .netrc"
260	printf "\n -v no prompt for confirmation ---"
261	printf "\n -X # of processors (on parallel hosts) 1"
262	printf "\n "
263	printf "\n The only possible positional parameter is <jobfile>:"
264	printf "\n The complete NQS-job must be provided here."
265	printf "\n <jobfile>=? creates this outline\n\n") \| more
266	exit
267	fi
268
269
270
271	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
272	if [[ "$file_to_send" = "" ]]
273	then
274	printf "\n +++ job-file missing"
275	locat=parameter; exit
276	else
277	if [[ -f $file_to_send ]]
278	then
279	true
280	else
281	printf "\n +++ job-file: "
282	printf "\n $file_to_send"
283	printf "\n does not exist"
284	locat=parameter; exit
285	fi
286	fi
287
288
289
290	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
291	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
292	if [[ $job_name = none ]]
293	then
294	job_name=$file_to_send
295	fi
296	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
297	then
298	printf "\n +++ job-file name: "
299	printf "\n $job_name"
300	printf "\n must not contain \"/\"-characters"
301	locat=parameter; exit
302	fi
303
304
305
306
307	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
308	# ODER WENN HOST NICHT ANGEGEBEN WURDE
309	if [[ $remote_host = none ]]
310	then
311	printf "\n +++ host missing"
312	locat=option; exit
313	else
314	case $remote_host in
315	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
316	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
317	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
318	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
319	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
320	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
321	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
322	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
323	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
324	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
325	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
326	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
327	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
328	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
329	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
330	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
331	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
332	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
333	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
334	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
335	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
336	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
337	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
338	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
339	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
340	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
341	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
342	locat=parameter; exit;;
343	esac
344	fi
345
346
347	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
348	if [[ $no_default_queue != none ]]
349	then
350	error=false
351	ndq=$no_default_queue
352	case $remote_host in
353	(ibm) case $ndq in
354	(p690_express\|p690_standard\|p690_long) error=false;;
355	(*) error=true;;
356	esac;;
357	(ibmh) case $ndq in
358	(cluster\|express) error=false;;
359	(*) error=true;;
360	esac;;
361	(ibmkisti) case $ndq in
362	(class.32plus\|class.1-2\|class.2-32) error=false;;
363	(*) error=true;;
364	esac;;
365	(ibmku) case $ndq in
366	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
367	(*) error=true;;
368	esac;;
369	(ibms) case $ndq in
370	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
371	(*) error=true;;
372	esac;;
373	(ibmy) case $ndq in
374	(parallel) error=false;;
375	(*) error=true;;
376	esac;;
377	(lccrayf) case $ndq in
378	(usup\|test*\|small\|large) error=false;;
379	(*) error=true;;
380	esac;;
381	(lcflow) case $ndq in
382	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
383	(*) error=true;;
384	esac;;
385	(lckiaps) case $ndq in
386	(express\|normal) error=false;;
387	(*) error=true;;
388	esac;;
389	(lckyoto) case $ndq in
390	(eh\|ph) error=false;;
391	(*) error=true;;
392	esac;;
393	(lckyuh) case $ndq in
394	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
395	(*) error=true;;
396	esac;;
397	(lckyut) case $ndq in
398	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
399	(*) error=true;;
400	esac;;
401	(lcsgib\|lcsgih) case $ndq in
402	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
403	(*) error=true;;
404	esac;;
405	(lctit) case $ndq in
406	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
407	(*) error=true;;
408	esac;;
409	(t3eb) case $ndq in
410	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
411	(*) error=true;;
412	esac;;
413	(necriam) case $ndq in
414	(SP\|SS\|P6) error=false;;
415	(*) error=true;;
416	esac;;
417	(t3eh) case $ndq in
418	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
419	(*) error=true;;
420	esac;;
421	(t3ej2\|t3ej5) case $ndq in
422	(low\|normal\|high) error=false;;
423	(*) error=true;;
424	esac;;
425	(t3es) case $ndq in
426	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
427	(*) error=true;;
428	esac;;
429	(unics) case $ndq in
430	(unics\|ht) error=false;;
431	(*) error=true;;
432	esac;;
433	esac
434	if [[ $error = true ]]
435	then
436	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
437	locat=parameter; exit
438	else
439	queue=$no_default_queue
440	fi
441	fi
442
443
444
445	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
446	# AUFTEILEN
447	done=false
448	while [[ $done = false ]]
449	do
450	if (( $cputime <= 0 ))
451	then
452	printf "\n +++ wrong cpu-time or cpu-time missing"
453	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
454	printf "\n >>> "
455	read cputime 1>/dev/null 2>&1
456	else
457	done=true
458	fi
459	done
460	if [[ $remote_host = nech ]]
461	then
462	if (( tasks_per_node != 0 ))
463	then
464	(( cputime = cputime * tasks_per_node ))
465	elif [[ $numprocs != 0 ]]
466	then
467	(( cputime = cputime * numprocs ))
468	fi
469	fi
470	(( stunden = cputime / 3600 ))
471	(( resttime = cputime - stunden * 3600 ))
472	(( minuten = resttime / 60 ))
473	(( sekunden = resttime - minuten * 60 ))
474	timestring=${stunden}:${minuten}:${sekunden}
475
476
477
478	# PRUEFEN DER KERNSPEICHERANFORDERUNG
479	done=false
480	while [[ $done = false ]]
481	do
482	if (( memory <= 0 ))
483	then
484	printf "\n +++ wrong memory demand or memory demand missing"
485	printf "\n >>> Please type memory in MByte per process as INTEGER:"
486	printf "\n >>> "
487	read memory 1>/dev/null 2>&1
488	else
489	done=true
490	fi
491	done
492
493	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
494	then
495	if (( tasks_per_node != 0 ))
496	then
497	(( Memory = memory * tasks_per_node / 1000 ))
498	elif [[ $numprocs != 0 ]]
499	then
500	(( Memory = memory * numprocs / 1000 ))
501	else
502	(( Memory = memory / 1000 ))
503	fi
504	elif [[ $remote_host = lctit ]]
505	then
506	(( Memory = memory * tasks_per_node / 1000 ))
507	fi
508
509
510	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
511	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
512	then
513	(( memory = memory * threads_per_task ))
514	fi
515
516
517	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
518	if (( tasks_per_node != 0 ))
519	then
520	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
521	fi
522
523
524	# Calculate number of processes per node
525	(( processes_per_node = tasks_per_node * threads_per_task ))
526
527
528	# Calculate number of MPI tasks
529	(( mpi_tasks = numprocs / threads_per_task ))
530
531
532	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
533	if [[ "$scp_port" != "" ]]
534	then
535	PORTOPT="-P $scp_port"
536	SSH_PORTOPT="-p $scp_port"
537	fi
538
539
540	# HEADER-AUSGABE
541	if [[ $verify = true ]]
542	then
543	printf "\n\n"
544	printf "#--------------------------------------------------------------# \n"
545	spalte1=SUBJOB;spalte2=$(date)
546	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
547	printf "\| \| \n"
548	printf "\| values of parameters/options: \| \n"
549	spalte1=$(echo local_host$punkte \| cut -c-20)
550	spalte2=$punkte$local_host
551	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
552	spalte1=$(echo remote_host$punkte \| cut -c-20)
553	spalte2=$punkte$remote_host
554	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
555	spalte1=$(echo queue$punkte \| cut -c-20)
556	spalte2=$punkte$queue
557	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
558	spalte1=$(echo memory$punkte \| cut -c-20)
559	spalte2="$punkte$memory mb"
560	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
561	spalte1=$(echo cputime$punkte \| cut -c-20)
562	spalte2="$punkte$cputime sec"
563	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
564	spalte1=$(echo job_name$punkte \| cut -c-20)
565	spalte2="$punkte$job_name"
566	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
567	printf "#--------------------------------------------------------------# \n\n"
568
569
570	# KONTROLLABFRAGE, OB ALLES O.K.
571	antwort="dummy"
572	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
573	do
574	read antwort?" >>> continue (y/n) ? "
575	done
576	if [[ $antwort = n \|\| $antwort = N ]]
577	then
578	locat=verify; exit
579	fi
580	printf "\n"
581	fi
582
583
584
585	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
586	kennung=$RANDOM
587	job_on_remhost=${job_name}_${kennung}_$local_host
588	job_to_send=job_to_send_$kennung
589	if [[ $delete_dayfile = false ]]
590	then
591	remote_dayfile=${local_host}_${job_name}_result_$kennung
592	local_dayfile=${remote_host}_${job_name}
593	else
594	remote_dayfile=/dev/null
595	fi
596
597
598
599	# Generate the batch job scripts (qsub/msub/LoadLeveler)
600	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
601	then
602
603	# General LoadLeveler settings
604	execute_in_shell="#!/bin/ksh"
605	use_shell="# @ shell = /bin/ksh"
606	consumable_memory="ConsumableMemory($memory mb)"
607	class="# @ class = $queue"
608	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
609	network_to_use="# @ network.mpi = sn_all,shared,us"
610	data_limit="# @ data_limit = 1.76gb"
611	image_size="# @ image_size = 50"
612	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
613
614	if [[ $email_notification = none ]]
615	then
616	notify_user=""
617	else
618	notify_user="# @ notify_user = $email_notification"
619	if [[ $delete_dayfile = true ]]
620	then
621	notification='# @ notification = never'
622	fi
623	fi
624
625	if [[ $remote_host = ibmh ]]
626	then
627	data_limit=""
628	network_to_use=""
629	class="# @ class = $queue"
630	environment=""
631	rset="# @ rset = RSET_MCM_AFFINITY"
632	task_affinity="# @ task_affinity = core(1)"
633	elif [[ $remote_host = ibmkisti ]]
634	then
635	network_to_use="# @ network.MPI = sn_all,shared,US"
636	wall_clock_limit="# @ wall_clock_limit = $timestring"
637	if [[ $threads_per_task = 1 ]]
638	then
639	rset="# @ rset = RSET_MCM_AFFINITY"
640	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
641	fi
642	environment=""
643	use_shell=""
644	data_limit=""
645	image_size=""
646	elif [[ $remote_host = ibmku ]]
647	then
648	execute_in_shell="#!/usr/bin/ksh"
649	use_shell="# @ shell = /usr/bin/ksh"
650	consumable_memory=""
651	environment=""
652	network_to_use="# @ network.mpi = sn_all,shared,us"
653	data_limit=""
654	image_size=""
655	elif [[ $remote_host = ibms ]]
656	then
657	network_to_use="# @ network.mpi = csss,shared,us"
658	elif [[ $remote_host = ibmy ]]
659	then
660	consumable_memory=""
661	network_to_use=""
662	fi
663
664	cat > $job_to_send << %%END%%
665	$execute_in_shell
666	$use_shell
667
668	# @ job_type = parallel
669	# @ job_name = $job_name
670	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
671	# @ output = $remote_dayfile
672	# @ error = $remote_dayfile
673	$wall_clock_limit
674	$image_size
675	$class
676	$environment
677	$network_to_use
678	$data_limit
679	$rset
680	$mcm_affinity_options
681	$task_affinity
682	$notification
683	$notify_user
684
685	%%END%%
686
687	if (( nodes > 0 ))
688	then
689
690	if [[ $remote_host != ibmkisti ]]
691	then
692
693	cat >> $job_to_send << %%END%%
694	# @ node = $nodes
695	# @ tasks_per_node = $processes_per_node
696	# @ node_usage = $node_usage
697	# @ queue
698
699	%%END%%
700
701	else
702
703	cat >> $job_to_send << %%END%%
704	# @ total_tasks = $mpi_tasks
705	# @ blocking = unlimited
706	# @ queue
707
708	%%END%%
709
710	fi
711
712	else
713
714	if [[ $remote_host != ibmy ]]
715	then
716
717	cat >> $job_to_send << %%END%%
718	# @ blocking = unlimited
719	# @ total_tasks = $numprocs
720	# @ node_usage = $node_usage
721	# @ queue
722
723	%%END%%
724
725	else
726
727	cat >> $job_to_send << %%END%%
728	# @ node = 1
729	# @ total_tasks = $numprocs
730	# @ queue
731
732	%%END%%
733
734	fi
735
736	fi
737
738	# workaround because of silly job filter on ibmkisti
739	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
740	then
741	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
742	fi
743
744	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
745	then
746
747	cat > $job_to_send << %%END%%
748	#!/bin/ksh
749
750	# @ job_type = serial
751	# @ node_usage = $node_usage
752	# @ job_name = palm
753	# @ wall_clock_limit = ${timestring},$timestring
754	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
755	# @ output = $remote_dayfile
756	# @ error = $remote_dayfile
757	$class
758	$notification
759
760	# @ queue
761
762	%%END%%
763
764	elif [[ $remote_host = lccrayf ]]
765	then
766
767	if [[ $numprocs != 0 ]]
768	then
769	cat > $job_to_send << %%END%%
770	#!/bin/bash -l
771	#SBATCH -J $job_name
772	#SBATCH -t $timestring
773	#SBATCH -N $nodes
774	#SBATCH --ntasks-per-node=$processes_per_node
775	#SBATCH -p $queue
776	#SBATCH -o $remote_dayfile
777	#SBATCH -e $remote_dayfile
778
779	$init_cmds
780	$module_calls
781
782	%%END%%
783
784	else
785	cat > $job_to_send << %%END%%
786	#!/bin/bash -l
787	#SBATCH -J $job_name
788	#SBATCH -t $timestring
789	#SBATCH -l ncpus=1
790	#SBATCH -l pmem=${memory}mb
791	#SBATCH -m abe
792	#SBATCH -o $remote_dayfile
793	#SBATCH -e $remote_dayfile
794
795	$init_cmds
796	$module_calls
797
798	%%END%%
799
800	fi
801
802	elif [[ $remote_host = lcfimm ]]
803	then
804
805	if [[ $numprocs != 0 ]]
806	then
807	cat > $job_to_send << %%END%%
808	#!/bin/ksh
809	#PBS -N $job_name
810	#PBS -A $project_account
811	#PBS -l walltime=$timestring
812	#PBS -l nodes=${nodes}:ppn=$processes_per_node
813	#PBS -l pmem=${memory}mb
814	#PBS -m abe
815	#PBS -o $remote_dayfile
816	#PBS -j oe
817	mpd &
818
819	%%END%%
820
821	else
822	cat > $job_to_send << %%END%%
823	#!/bin/ksh
824	#PBS -N $job_name
825	#PBS -A $project_account
826	#PBS -l walltime=$timestring
827	#PBS -l ncpus=1
828	#PBS -l pmem=${memory}mb
829	#PBS -m abe
830	#PBS -o $remote_dayfile
831	#PBS -j oe
832
833	%%END%%
834
835	fi
836
837	elif [[ $remote_host = lcflow ]]
838	then
839	if [ $memory -gt 1800 ]; then
840	use_himem=""
841	else
842	use_himem="#"
843	fi
844	if [[ $numprocs != 0 ]]
845	then
846	pe_set="#$ -pe impi41 $numprocs"
847	else
848	pe_set="#$ -pe impi41 1"
849	fi
850	if [[ $queue = default ]]
851	then
852	queue_set=""
853	else
854	queue_set="#$ -q $queue"
855	fi
856	[[ "$disc_space" = "" ]] && disc_space=50
857
858	cat > $job_to_send << %%END%%
859	#!/bin/bash
860	#$ -S /bin/bash
861	#$ -N $job_name
862	#$ -cwd
863	#$ -l h_rt=$timestring
864	#$ -l h_vmem=${memory}M
865	#$ -o $remote_dayfile
866	#$ -j y
867	$pe_set
868	#$ -R y
869	${use_himem}#$ -l highmem=true
870	#$ -l h_fsize=${disc_space}G
871	$queue_set
872
873	%%END%%
874
875	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
876	then
877
878	if [[ $numprocs != 0 ]]
879	then
880	cat > $job_to_send << %%END%%
881	#!/bin/ksh
882	#PBS -N $job_name
883	#PBS -l walltime=$timestring
884	#PBS -l ncpus=$numprocs
885	#PBS -l pmem=${memory}mb
886	#PBS -o $remote_dayfile
887	#PBS -l nodes=$nodes:ppn=${processes_per_node}
888	#PBS -j oe
889
890	mpd &
891
892	%%END%%
893
894	else
895	cat > $job_to_send << %%END%%
896	#!/bin/ksh
897	#PBS -N $job_name
898	#PBS -l walltime=$timestring
899	#PBS -l ncpus=1
900	#PBS -l pmem=${memory}mb
901	#PBS -o $remote_dayfile
902	#PBS -j oe
903
904	%%END%%
905
906	fi
907
908	elif [[ $remote_host = lckiaps ]]
909	then
910
911	if [[ $numprocs != 0 ]]
912	then
913	cat > $job_to_send << %%END%%
914	#!/bin/ksh
915	#PBS -N $job_name
916	#PBS -l walltime=$timestring
917	#PBS -l select=1:ncpus=$numprocs
918	#PBS -l pmem=${memory}mb
919	#PBS -q $queue
920	#PBS -o $remote_dayfile
921	#PBS -j oe
922	#PBS -V
923
924	%%END%%
925
926	else
927	cat > $job_to_send << %%END%%
928	#!/bin/ksh
929	#PBS -N $job_name
930	#PBS -l walltime=$timestring
931	#PBS -l ncpus=1
932	#PBS -l pmem=${memory}mb
933	#PBS -o $remote_dayfile
934	#PBS -j oe
935
936	%%END%%
937
938	fi
939
940	elif [[ $remote_host = lcyon ]]
941	then
942
943	if [[ $numprocs != 0 ]]
944	then
945	cat > $job_to_send << %%END%%
946	#!/bin/ksh
947	#PBS -N $job_name
948	#PBS -l walltime=$timestring
949	#PBS -l ncpus=$numprocs
950	#PBS -l pmem=${memory}mb
951	#PBS -o $remote_dayfile
952	#PBS -j oe
953
954	%%END%%
955
956	else
957	cat > $job_to_send << %%END%%
958	#!/bin/ksh
959	#PBS -N $job_name
960	#PBS -l walltime=$timestring
961	#PBS -l ncpus=1
962	#PBS -l pmem=${memory}mb
963	#PBS -o $remote_dayfile
964	#PBS -j oe
965
966	%%END%%
967
968	fi
969
970	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
971	then
972
973	if [[ "$sgi_feature" != "" ]]
974	then
975	feature_directive="#PBS -l feature=$sgi_feature"
976	else
977	feature_directive=""
978	fi
979
980	if [[ $queue = dataq \|\| $queue = permq ]]
981	then
982	feature_directive="#PBS -l feature=data"
983	fi
984
985	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
986	then
987	queue_directive="#PBS -q $queue"
988	else
989	queue_directive=""
990	fi
991
992	if [[ $email_notification = none ]]
993	then
994	email_directive=""
995	else
996	email_directive="#PBS -M $email_notification"
997	fi
998
999	if [[ $numprocs != 0 ]]
1000	then
1001	cat > $job_to_send << %%END%%
1002	#!/bin/bash
1003	#PBS -N $job_name
1004	#PBS -l walltime=$timestring
1005	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1006	#PBS -l naccesspolicy=$node_usage
1007	#PBS -o $remote_dayfile
1008	#PBS -j oe
1009	$feature_directive
1010	$queue_directive
1011	$email_directive
1012
1013	eval \`/sw/swdist/bin/modulesinit\`
1014	#. /usr/share/modules/init/bash
1015	$init_cmds
1016	$module_calls
1017
1018	echo ld_library_path=\$LD_LIBRARY_PATH
1019
1020	%%END%%
1021
1022	else
1023	cat > $job_to_send << %%END%%
1024	#PBS -S /bin/bash
1025	#PBS -N $job_name
1026	#PBS -l walltime=$timestring
1027	#PBS -l ncpus=1
1028	#PBS -o $remote_dayfile
1029	#PBS -j oe
1030	$queue_directive
1031	$email_directive
1032
1033	eval \`/sw/swdist/bin/modulesinit\`
1034	#. /usr/share/modules/init/bash
1035	$init_cmds
1036	$module_calls
1037
1038	%%END%%
1039
1040	fi
1041
1042	elif [[ $remote_host = lcxe6 ]]
1043	then
1044
1045	if [[ $numprocs != 0 ]]
1046	then
1047	cat > $job_to_send << %%END%%
1048	#!/bin/ksh
1049	#PBS -S /bin/ksh
1050	#PBS -N $job_name
1051	#PBS -A $project_account
1052	#PBS -j oe
1053	#PBS -l walltime=$timestring
1054	#PBS -l mppwidth=${numprocs}
1055	#PBS -l mppnppn=${processes_per_node}
1056	#PBS -m abe
1057	#PBS -o $remote_dayfile
1058	$email_directive
1059
1060	$init_cmds
1061	$module_calls
1062
1063	%%END%%
1064
1065	else
1066	cat > $job_to_send << %%END%%
1067	#!/bin/ksh
1068	#PBS -S /bin/ksh
1069	#PBS -N $job_name
1070	#PBS -A $project_account
1071	#PBS -j oe
1072	#PBS -l walltime=$timestring
1073	#PBS -l ncpus=1
1074	#PBS -l pmem=${memory}mb
1075	#PBS -m abe
1076	$email_directive
1077	#PBS -o $remote_dayfile
1078
1079	$init_cmds
1080	$module_calls
1081
1082	%%END%%
1083
1084	fi
1085
1086	elif [[ $remote_host = lckyoto ]]
1087	then
1088
1089	cat > $job_to_send << %%END%%
1090	#!/bin/ksh
1091	# @\$-o $remote_dayfile
1092	# @\$-eo -oi
1093	# @\$-lP 16
1094	# @\$-lp 1
1095	# @\$-lm 28gb -llm unlimited -ls unlimited
1096	# @\$-q $queue
1097	# @\$-Pvn abs_pack
1098	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1099	#. /thin/local/etc/setprofile/intel-11.0.sh
1100	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1101	. ~/.myprofile
1102	#. /home2/t/t51254/palm/current_version/myprofile
1103	#. /thin/apps/pgi/mpi.sh
1104	#
1105	env
1106	#
1107	set -x
1108
1109	%%END%%
1110
1111	elif [[ $remote_host = lcxt5m ]]
1112	then
1113
1114	if [[ $numprocs != 0 ]]
1115	then
1116	cat > $job_to_send << %%END%%
1117	#!/bin/ksh
1118	#PBS -S /bin/ksh
1119	#PBS -N $job_name
1120	#PBS -j oe
1121	#PBS -l walltime=$timestring
1122	#PBS -l mppwidth=${numprocs}
1123	#PBS -l mppnppn=${processes_per_node}
1124	#PBS -m abe
1125	#PBS -o $remote_dayfile
1126
1127	$init_cmds
1128	$module_calls
1129
1130	%%END%%
1131
1132	else
1133	cat > $job_to_send << %%END%%
1134	#!/bin/ksh
1135	#PBS -S /bin/ksh
1136	#PBS -N $job_name
1137	#PBS -j oe
1138	#PBS -l walltime=$timestring
1139	#PBS -l ncpus=1
1140	#PBS -l pmem=${memory}mb
1141	#PBS -m abe
1142	#PBS -o $remote_dayfile
1143
1144	$init_cmds
1145	$module_calls
1146
1147	%%END%%
1148
1149	fi
1150
1151	elif [[ $remote_host = lckyuh ]]
1152	then
1153	cat > $job_to_send << %%END%%
1154	#!/bin/bash
1155	#PJM -L "rscgrp=$queue"
1156	#PJM -L "node=$nodes"
1157	#PJM --mpi "proc=$numprocs"
1158	#PJM -L "elapse=$timestring"
1159	#PJM -o $remote_dayfile
1160	#PJM -j
1161	#PJM -X
1162	#PJM --no-stging
1163
1164	export LANG=en_US.UTF-8
1165	%%END%%
1166
1167	elif [[ $remote_host = lckyut ]]
1168	then
1169	cat > $job_to_send << %%END%%
1170	#!/bin/bash
1171	#PJM -L "rscgrp=$queue"
1172	#PJM -L "vnode=$numprocs"
1173	#PJM -L "vnode-core=1"
1174	#PJM -L "elapse=$timestring"
1175	#PJM --mpi proc=$numprocs
1176	#PJM -o $remote_dayfile
1177	#PJM -j
1178	#PJM -X
1179	#PJM --no-stging
1180
1181	export LANG=en_US.UTF-8
1182	%%END%%
1183
1184	elif [[ $remote_host = nech ]]
1185	then
1186
1187	if (( nodes > 1 ))
1188	then
1189	cat > $job_to_send << %%END%%
1190	#!/bin/ksh
1191	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1192	#PBS -l ${qsubmem}=${Memory}gb
1193	#PBS -b $nodes
1194	#PBS -o $remote_dayfile
1195	#PBS -N palm
1196	#PBS -j o
1197	#PBS -T mpisx
1198
1199	%%END%%
1200
1201	elif [[ $numprocs != 0 ]]
1202	then
1203	cat > $job_to_send << %%END%%
1204	#!/bin/ksh
1205	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1206	#PBS -l ${qsubmem}=${Memory}gb
1207	#PBS -o $remote_dayfile
1208	#PBS -N palm
1209	#PBS -j o
1210
1211	%%END%%
1212
1213	else
1214	cat > $job_to_send << %%END%%
1215	#!/bin/ksh
1216	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1217	#PBS -o $remote_dayfile
1218	#PBS -j o
1219
1220	%%END%%
1221
1222	fi
1223
1224	elif [[ $remote_host = necriam ]]
1225	then
1226
1227	if (( nodes > 1 ))
1228	then
1229	cat > $job_to_send << %%END%%
1230	#!/bin/ksh
1231	#PBS -b $nodes
1232	#PBS -o $remote_dayfile
1233	#PBS -N $job_name
1234	#PBS -j o
1235	#PBS -v MPIPROGINV=YES
1236
1237	%%END%%
1238
1239	elif [[ $numprocs != 0 ]]
1240	then
1241	cat > $job_to_send << %%END%%
1242	#!/bin/ksh
1243	#PBS -q ${queue}
1244	#PBS -o $remote_dayfile
1245	#PBS -N $job_name
1246	#PBS -j o
1247	#PBS -v MPIPROGINV=YES
1248
1249	%%END%%
1250
1251	else
1252	cat > $job_to_send << %%END%%
1253	#!/bin/ksh
1254	#PBS -o $remote_dayfile
1255	#PBS -j o
1256
1257	%%END%%
1258
1259	fi
1260
1261	elif [[ $remote_host = lctit ]]
1262	then
1263	cat > $job_to_send << %%END%%
1264	#!/bin/ksh
1265	$init_cmds
1266	$module_calls
1267
1268	%%END%%
1269
1270	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1271	if [[ $tasks_per_node != $processes_per_node ]]
1272	then
1273	submcom="$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 $queue "
1274	else
1275	submcom="$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 "
1276	fi
1277
1278	else
1279
1280	cat > $job_to_send << %%END%%
1281	# @\$-q ${queue}
1282	# @\$-l${qsubtime} $timestring
1283	# @\$-l${qsubmem} ${memory}mb
1284	# @\$-o $remote_dayfile
1285	# @\$-eo
1286
1287	%%END%%
1288
1289	fi
1290
1291
1292	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1293	# BEI EXIT VERANLASSEN
1294	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1295	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1296	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1297	then
1298	echo "set +vx" >> $job_to_send
1299	echo "trap '" >> $job_to_send
1300	echo "set +vx" >> $job_to_send
1301	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* ]]
1302	then
1303	if [[ $remote_host = ibmh ]]
1304	then
1305	return_queue=c1
1306	elif [[ $remote_host = ibmkisti ]]
1307	then
1308	return_queue=class.1-2
1309	elif [[ $remote_host = ibmku ]]
1310	then
1311	return_queue=sdbg2
1312	elif [[ $remote_host = ibms ]]
1313	then
1314	return_queue=p_normal
1315	elif [[ $remote_host = ibmy ]]
1316	then
1317	return_queue=serial
1318	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1319	then
1320	return_queue=serialq
1321	elif [[ $remote_host = necriam ]]
1322	then
1323	return_queue=SP
1324	elif [[ $remote_host = lckiaps ]]
1325	then
1326	return_queue=express
1327	elif [[ $remote_host = lckyuh ]]
1328	then
1329	return_queue=cx-single
1330	elif [[ $remote_host = lckyut ]]
1331	then
1332	return_queue=cx-single
1333	else
1334	return_queue=unknown
1335	fi
1336
1337	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1338	then
1339
1340	if [[ $remote_host = ibmku ]]
1341	then
1342	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1343	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1344	else
1345	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1346	fi
1347	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1348	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1349	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1350	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1351	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1352	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1353	if [[ $host != "ibmh" ]]
1354	then
1355	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1356	fi
1357	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1358	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1359
1360	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1361	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1362
1363	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1364	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
1365	if [[ $remote_host = ibmku ]]
1366	then
1367	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1368	fi
1369	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1370
1371	elif [[ $remote_host = nech ]]
1372	then
1373	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1374	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1375	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1376	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1377	echo "#PBS -j o" >> $job_to_send
1378	echo " " >> $job_to_send
1379	echo "set -x" >> $job_to_send
1380	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1381	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1382	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1383	echo "%%END%%" >> $job_to_send
1384
1385	elif [[ $remote_host = necriam ]]
1386	then
1387	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1388	echo "#PBS -q $return_queue" >> $job_to_send
1389	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1390	echo "#PBS -j o" >> $job_to_send
1391	echo " " >> $job_to_send
1392	echo "set -x" >> $job_to_send
1393	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1394	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1395	echo "%%END%%" >> $job_to_send
1396
1397	elif [[ $remote_host = lckyuh ]]
1398	then
1399	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1400	echo "#!/bin/bash" >> $job_to_send
1401	echo "#PJM -L \"node=1\"" >> $job_to_send
1402	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1403	echo "#PJM --no-stging" >> $job_to_send
1404	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1405	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1406	echo "#PJM -j" >> $job_to_send
1407	echo " " >> $job_to_send
1408	echo "export LANG=en_US.UTF-8" >> $job_to_send
1409	echo "set -x" >> $job_to_send
1410	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1411	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1412	echo "%%END%%" >> $job_to_send
1413
1414	elif [[ $remote_host = lckyut ]]
1415	then
1416	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1417	echo "#!/bin/bash" >> $job_to_send
1418	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1419	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1420	echo "#PJM --no-stging" >> $job_to_send
1421	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1422	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1423	echo "#PJM -j" >> $job_to_send
1424	echo " " >> $job_to_send
1425	echo "export LANG=en_US.UTF-8" >> $job_to_send
1426	echo "set -x" >> $job_to_send
1427	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1428	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1429	echo "%%END%%" >> $job_to_send
1430
1431	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1432	then
1433	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1434	echo "#!/bin/bash" >> $job_to_send
1435	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1436	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1437	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1438	echo "#PBS -l feature=data" >> $job_to_send
1439	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1440	echo "#PBS -j oe" >> $job_to_send
1441	echo " " >> $job_to_send
1442	echo ". /usr/share/modules/init/bash" >> $job_to_send
1443	echo "set -x" >> $job_to_send
1444	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1445	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1446	echo "%%END%%" >> $job_to_send
1447
1448	elif [[ $remote_host = lcflow ]]
1449	then
1450	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1451	echo "#!/bin/bash" >> $job_to_send
1452	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1453	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1454	echo "SGEPREFIX -cwd" >> $job_to_send
1455	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1456	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1457	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1458	echo "SGEPREFIX -j y" >> $job_to_send
1459	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1460	echo " " >> $job_to_send
1461	echo "set -x" >> $job_to_send
1462	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1463	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1464	echo "" >> $job_to_send
1465	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1466	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1467	echo "rm -f scpjob.${kennung}" >> $job_to_send
1468	echo "%%END%%" >> $job_to_send
1469	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1470	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1471
1472	else
1473
1474	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1475	echo "# @\\\$-q $return_queue" >> $job_to_send
1476	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1477	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1478	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1479	then
1480	echo "# @\$-l mpp_p=0" >> $job_to_send
1481	fi
1482	echo '# @\$-lF 10mb' >> $job_to_send
1483	# echo '# @\$-o /dev/null' >> $job_to_send
1484	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1485	echo '# @\\\$-eo' >> $job_to_send
1486	echo " " >> $job_to_send
1487	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1488	then
1489	echo "set +vx" >> $job_to_send
1490	echo ". .profile" >> $job_to_send
1491	fi
1492	echo "set -x" >> $job_to_send
1493	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
1494	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1495	echo "%%END%%" >> $job_to_send
1496	fi
1497
1498	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1499	then
1500	echo "llsubmit scpjob.$kennung" >> $job_to_send
1501	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1502	then
1503	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1504	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1505	echo "msub scpjob.$kennung" >> $job_to_send
1506	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1507	then
1508	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1509	elif [[ $remote_host = t3es ]]
1510	then
1511	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1512	elif [[ $remote_host = lckiaps ]]
1513	then
1514	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1515	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1516	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1517	elif [[ $remote_host = lckyu* ]]
1518	then
1519	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1520	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1521	elif [[ $remote_host = lcflow ]]
1522	then
1523	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1524	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1525	else
1526	echo "$submcom scpjob.$kennung" >> $job_to_send
1527	fi
1528	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1529	then
1530	echo "rm scpjob.$kennung" >> $job_to_send
1531	fi
1532	if [[ $remote_host = nech ]]
1533	then
1534	echo "cd -" >> $job_to_send
1535	fi
1536	else
1537	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1538	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1539	fi
1540	echo "set -x" >> $job_to_send
1541	echo " ' exit" >> $job_to_send
1542	echo "set -x" >> $job_to_send
1543	fi
1544
1545
1546
1547	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1548	cat $file_to_send >> $job_to_send
1549
1550	if [[ $remote_host = ibm ]]
1551	then
1552	echo " " >> $job_to_send
1553	echo "exit" >> $job_to_send
1554	fi
1555
1556	# remove job file
1557	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1558	then
1559	echo " " >> $job_to_send
1560	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1561	fi
1562
1563
1564
1565	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1566	if [[ -z $remote_user ]]
1567	then
1568	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1569	then
1570	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1571	fi
1572	fi
1573
1574
1575
1576	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1577	if [[ $no_submit = false ]]
1578	then
1579	if [[ $remote_host != $local_host ]]
1580	then
1581	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1582	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1583	then
1584	job_catalog_save=$job_catalog
1585	job_catalog=job_queue
1586	elif [[ $remote_host = nech ]]
1587	then
1588	job_catalog_save=$job_catalog
1589	job_catalog=/hpf/b/${remote_user}/job_queue
1590	fi
1591	if [[ $remote_host = nech ]]
1592	then
1593	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1594	# TRANSFERIERT WERDEN
1595	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1596	else
1597	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1598	fi
1599	if [[ $? = 1 ]]
1600	then
1601	locat=scp; exit
1602	fi
1603	if [[ $remote_host = ibms ]]
1604	then
1605	job_catalog=$job_catalog_save
1606	fi
1607	[[ $verify = true ]] && printf "\n >>> finished\n"
1608	else
1609	eval job_catalog=$job_catalog
1610	cp $job_to_send ${job_catalog}/$job_on_remhost
1611	fi
1612
1613
1614
1615	# NQS- BZW. LOADLEVELER-JOB STARTEN
1616	if [[ $remote_host != $local_host ]]
1617	then
1618	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1619
1620	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1621	then
1622	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1623	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1624	elif [[ $remote_host = ibmku ]]
1625	then
1626	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1627	elif [[ $remote_host = lcflow ]]
1628	then
1629	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1630	else
1631	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1632	fi
1633
1634	[[ $verify = true ]] && printf " >>> o.k.\n"
1635	else
1636	cd $job_catalog
1637	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $local_host = lccrayf ]]
1638	then
1639	eval $submcom $job_on_remhost
1640	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1641	then
1642	chmod u+x $job_on_remhost
1643	eval $submcom $job_on_remhost
1644	elif [[ $local_host = nech ]]
1645	then
1646	if [[ $queue = default ]]
1647	then
1648	eval $submcom $job_on_remhost
1649	else
1650	eval $submcom -q $queue $job_on_remhost
1651	fi
1652	else
1653	qsub $job_on_remhost
1654	fi
1655
1656	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1657	# only at the end of the job.
1658	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1659	then
1660	rm $job_on_remhost
1661	fi
1662	cd - > /dev/null
1663	fi
1664	fi
1665
1666
1667
1668	# ABSCHLUSSARBEITEN
1669
1670	if [[ $no_submit = false ]]
1671	then
1672	rm -f $job_to_send
1673	fi
1674	[[ $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 |