Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1103

Last change on this file since 1103 was 1103, checked in by raasch, 12 years ago
small bugfixes; mrun and subjob scripts are made bash compatible; further adjustments for lckyuh
Property svn:keywords set to `Id Rev`
File size: 54.0 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	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
26	# instead typeset -L/R),
27	# further adjustments for lckyuh
28	#
29	# Former revisions:
30	# -----------------
31	# $Id: subjob 1103 2013-02-20 02:15:53Z raasch $
32	#
33	# 2013-02-10 01:47:43Z raasch
34	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
35	# and for Forwind cluster (lcflow)
36	#
37	# 1094 2013-02-03 01:52:12Z raasch
38	# new option -P for explicit setting of ssh/scp port,
39	# decalpha parts (yonsei) removed
40	#
41	# 2013-02-02 07:06:13Z raasch
42	# adjustments for Kyushu-University computing center (lckyut - tatara)
43	# old changelog messages removed
44	#
45	# 1046 2012-11-09 14:38:45Z maronga
46	# code put under GPL (PALM 3.9)
47	#
48	# 08/07/94 - Siggi - first version finished
49	# 29/06/94 - Siggi - script development started
50	#--------------------------------------------------------------------------------#
51	# subjob - script for automatic generation and submission of batch-job files
52	# for various batch queuing systems
53	#--------------------------------------------------------------------------------#
54
55
56	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
57	delete_dayfile=false
58	email_notification=none
59	group_number=none
60	locat=normal
61	no_default_queue=none
62	no_submit=false
63	job_catalog="~/job_queue"
64	job_name=none
65	local_user=$LOGNAME
66	node_usage=shared
67	numprocs=0
68	punkte="..........................................................."
69	submcom=qsub
70	queue=default
71	remote_host=none
72	remote_user=""
73	verify=true
74
75	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
76	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
77
78
79
80	# FEHLERBEHANDLUNG
81	# BEI EXIT:
82	trap 'if [[ $locat != normal ]]
83	then
84	case $locat in
85	(option) printf "\n --> available optios can be displayed"
86	printf " by typing:"
87	printf "\n \"subjob ?\" \n";;
88	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
89	(*) printf "\n +++ unknown error"
90	printf "\n please inform S. Raasch!\n"
91	esac
92	[[ -f $job_to_send ]] && rm $job_to_send
93	printf "\n\n+++ SUBJOB killed \n\n"
94	fi' exit
95
96
97	# BEI TERMINAL-BREAK:
98	trap '[[ -f $job_to_send ]] && rm $job_to_send
99	printf "\n\n+++ SUBJOB killed \n\n"
100	exit
101	' 2
102
103
104
105
106	# LOKALEN HOSTNAMEN FESTSTELLEN
107	local_host=$(hostname)
108
109
110
111	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
112	# UEBERHAUPT ZULAESSIG IST
113	# Note: One of the entries for "lck" or "lckordi" always has to be
114	# comment out, because the hostname (node*) is same for both machines
115	case $local_host in
116	(atmos) local_addres=172.20.25.35; local_host=lcide;;
117	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
118	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
119	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
120	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
121	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
122	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
123	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
124	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
125	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
126	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
127	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
128	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
129	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
130	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
131	(cs*) local_addres=136.172.44.131; local_host=nech;;
132	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
133	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
134	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
135	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
136	(node*) local_addres=165.132.26.61 local_host=lck;;
137	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
138	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
139	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
140	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
141	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
142	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
143	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
144	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
145	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
146	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
147	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
148	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
149	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
150	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
151	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
152	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
153	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
154	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
155	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
156	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
157	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
158	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
159	(meller) local_addres=134.106.74.155; local_host=lcfor;;
160	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
161	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
162	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
163	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
164	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
165	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
166	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
167	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
168	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
169	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
170	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
171	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
172	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
173	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
174	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
175	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
176	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
177	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
178	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
179	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
180	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
181	(*) printf "\n +++ \"$local_host\" unknown";
182	printf "\n please inform S. Raasch!";
183	locat=parameter; exit;;
184	esac
185
186
187
188	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
189	remote_host=$local_host
190
191
192
193
194	# PROZEDUROPTIONEN EINLESEN
195	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
196	do
197	case $option in
198	(c) job_catalog=$OPTARG;;
199	(d) delete_dayfile=true;;
200	(D) no_submit=true;;
201	(e) email_notification=$OPTARG;;
202	(g) group_number=$OPTARG;;
203	(h) remote_host=$OPTARG;;
204	(m) memory=$OPTARG;;
205	(n) job_name=$OPTARG;;
206	(N) node_usage=$OPTARG;;
207	(O) threads_per_task=$OPTARG;;
208	(P) scp_port=$OPTARG;;
209	(q) no_default_queue=$OPTARG;;
210	(t) cputime=$OPTARG;;
211	(T) tasks_per_node=$OPTARG;;
212	(u) remote_user=$OPTARG;;
213	(v) verify=false;;
214	(X) numprocs=$OPTARG;;
215	(\?) printf "\n +++ Option $OPTARG unknown \n";
216	locat=option; exit;;
217	esac
218	done
219
220
221	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
222	(( to_shift = $OPTIND - 1 ))
223	shift $to_shift; file_to_send=$1
224
225
226	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
227	if [ "$1" = "?" ]
228	then
229	(printf "\n *** subjob can be called as follows:\n"
230	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
231	printf "\n Description of available options:\n"
232	printf "\n Option Description Default-Value"
233	printf "\n -c job-input- and output-catalog ~/job_queue"
234	printf "\n -d no job-protocol will be created ---"
235	printf "\n -D only the job-file will be created ---"
236	printf "\n -h execution host, available hosts: $remote_host"
237	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
238	printf "\n lckiaps, lctit, nech, necriam, unics"
239	printf "\n -m memory demand per process in MByte ---"
240	printf "\n -n jobname <jobdatei>"
241	printf "\n -O threads per task (for OpenMP usage) 1"
242	printf "\n -P ssh/scp port default port"
243	printf "\n -q job-queue to be used default"
244	printf "\n -t allowed cpu-time in seconds ---"
245	printf "\n -T tasks per node (on parallel hosts) ---"
246	printf "\n -u username on execution host from .netrc"
247	printf "\n -v no prompt for confirmation ---"
248	printf "\n -X # of processors (on parallel hosts) 1"
249	printf "\n "
250	printf "\n The only possible positional parameter is <jobfile>:"
251	printf "\n The complete NQS-job must be provided here."
252	printf "\n <jobfile>=? creates this outline\n\n") \| more
253	exit
254	fi
255
256
257
258	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
259	if [[ "$file_to_send" = "" ]]
260	then
261	printf "\n +++ job-file missing"
262	locat=parameter; exit
263	else
264	if [[ -f $file_to_send ]]
265	then
266	true
267	else
268	printf "\n +++ job-file: "
269	printf "\n $file_to_send"
270	printf "\n does not exist"
271	locat=parameter; exit
272	fi
273	fi
274
275
276
277	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
278	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
279	if [[ $job_name = none ]]
280	then
281	job_name=$file_to_send
282	fi
283	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
284	then
285	printf "\n +++ job-file name: "
286	printf "\n $job_name"
287	printf "\n must not contain \"/\"-characters"
288	locat=parameter; exit
289	fi
290
291
292
293
294	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
295	# ODER WENN HOST NICHT ANGEGEBEN WURDE
296	if [[ $remote_host = none ]]
297	then
298	printf "\n +++ host missing"
299	locat=option; exit
300	else
301	case $remote_host in
302	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
303	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
304	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
305	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
306	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
307	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
308	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
309	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
310	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
311	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
312	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
313	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
314	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
315	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
316	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
317	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
318	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
319	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
320	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
321	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
322	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
323	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
324	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
325	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
326	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
327	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
328	locat=parameter; exit;;
329	esac
330	fi
331
332
333	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
334	if [[ $no_default_queue != none ]]
335	then
336	error=false
337	ndq=$no_default_queue
338	case $remote_host in
339	(ibm) case $ndq in
340	(p690_express\|p690_standard\|p690_long) error=false;;
341	(*) error=true;;
342	esac;;
343	(ibmh) case $ndq in
344	(no_class) error=false;;
345	(*) error=true;;
346	esac;;
347	(ibmkisti) case $ndq in
348	(class.32plus\|class.1-2\|class.2-32) error=false;;
349	(*) error=true;;
350	esac;;
351	(ibmku) case $ndq in
352	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
353	(*) error=true;;
354	esac;;
355	(ibms) case $ndq in
356	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
357	(*) error=true;;
358	esac;;
359	(ibmy) case $ndq in
360	(parallel) error=false;;
361	(*) error=true;;
362	esac;;
363	(lcflow) case $ndq in
364	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
365	(*) error=true;;
366	esac;;
367	(lckiaps) case $ndq in
368	(express\|normal) error=false;;
369	(*) error=true;;
370	esac;;
371	(lckyoto) case $ndq in
372	(eh\|ph) error=false;;
373	(*) error=true;;
374	esac;;
375	(lckyuh) case $ndq in
376	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
377	(*) error=true;;
378	esac;;
379	(lckyut) case $ndq in
380	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
381	(*) error=true;;
382	esac;;
383	(lcsgib\|lcsgih) case $ndq in
384	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
385	(*) error=true;;
386	esac;;
387	(lctit) case $ndq in
388	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
389	(*) error=true;;
390	esac;;
391	(t3eb) case $ndq in
392	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
393	(*) error=true;;
394	esac;;
395	(necriam) case $ndq in
396	(SP\|SS\|P6) error=false;;
397	(*) error=true;;
398	esac;;
399	(t3eh) case $ndq in
400	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
401	(*) error=true;;
402	esac;;
403	(t3ej2\|t3ej5) case $ndq in
404	(low\|normal\|high) error=false;;
405	(*) error=true;;
406	esac;;
407	(t3es) case $ndq in
408	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
409	(*) error=true;;
410	esac;;
411	(unics) case $ndq in
412	(unics\|ht) error=false;;
413	(*) error=true;;
414	esac;;
415	esac
416	if [[ $error = true ]]
417	then
418	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
419	locat=parameter; exit
420	else
421	queue=$no_default_queue
422	fi
423	fi
424
425
426
427	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
428	# AUFTEILEN
429	done=false
430	while [[ $done = false ]]
431	do
432	if (( $cputime <= 0 ))
433	then
434	printf "\n +++ wrong cpu-time or cpu-time missing"
435	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
436	printf "\n >>> "
437	read cputime 1>/dev/null 2>&1
438	else
439	done=true
440	fi
441	done
442	if [[ $remote_host = nech ]]
443	then
444	if (( tasks_per_node != 0 ))
445	then
446	(( cputime = cputime * tasks_per_node ))
447	elif [[ $numprocs != 0 ]]
448	then
449	(( cputime = cputime * numprocs ))
450	fi
451	fi
452	(( stunden = cputime / 3600 ))
453	(( resttime = cputime - stunden * 3600 ))
454	(( minuten = resttime / 60 ))
455	(( sekunden = resttime - minuten * 60 ))
456	timestring=${stunden}:${minuten}:${sekunden}
457
458
459
460	# PRUEFEN DER KERNSPEICHERANFORDERUNG
461	done=false
462	while [[ $done = false ]]
463	do
464	if (( memory <= 0 ))
465	then
466	printf "\n +++ wrong memory demand or memory demand missing"
467	printf "\n >>> Please type memory in MByte per process as INTEGER:"
468	printf "\n >>> "
469	read memory 1>/dev/null 2>&1
470	else
471	done=true
472	fi
473	done
474
475	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
476	then
477	if (( tasks_per_node != 0 ))
478	then
479	(( Memory = memory * tasks_per_node / 1000 ))
480	elif [[ $numprocs != 0 ]]
481	then
482	(( Memory = memory * numprocs / 1000 ))
483	else
484	(( Memory = memory / 1000 ))
485	fi
486	elif [[ $remote_host = lctit ]]
487	then
488	(( Memory = memory * tasks_per_node / 1000 ))
489	fi
490
491
492	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
493	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
494	then
495	(( memory = memory * threads_per_task ))
496	fi
497
498
499	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
500	if (( tasks_per_node != 0 ))
501	then
502	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
503	fi
504
505
506	# Calculate number of processes per node
507	(( processes_per_node = tasks_per_node * threads_per_task ))
508
509
510	# Calculate number of MPI tasks
511	(( mpi_tasks = numprocs / threads_per_task ))
512
513
514	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
515	if [[ "$scp_port" != "" ]]
516	then
517	PORTOPT="-P $scp_port"
518	SSH_PORTOPT="-p $scp_port"
519	fi
520
521
522	# HEADER-AUSGABE
523	if [[ $verify = true ]]
524	then
525	printf "\n\n"
526	printf "#--------------------------------------------------------------# \n"
527	spalte1=SUBJOB;spalte2=$(date)
528	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
529	printf "\| \| \n"
530	printf "\| values of parameters/options: \| \n"
531	spalte1=$(echo local_host$punkte \| cut -c-20)
532	spalte2=$punkte$local_host
533	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
534	spalte1=$(echo remote_host$punkte \| cut -c-20)
535	spalte2=$punkte$remote_host
536	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
537	spalte1=$(echo queue$punkte \| cut -c-20)
538	spalte2=$punkte$queue
539	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
540	spalte1=$(echo memory$punkte \| cut -c-20)
541	spalte2="$punkte$memory mb"
542	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
543	spalte1=$(echo cputime$punkte \| cut -c-20)
544	spalte2="$punkte$cputime sec"
545	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
546	spalte1=$(echo job_name$punkte \| cut -c-20)
547	spalte2="$punkte$job_name"
548	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
549	printf "#--------------------------------------------------------------# \n\n"
550
551
552	# KONTROLLABFRAGE, OB ALLES O.K.
553	antwort="dummy"
554	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
555	do
556	read antwort?" >>> continue (y/n) ? "
557	done
558	if [[ $antwort = n \|\| $antwort = N ]]
559	then
560	locat=verify; exit
561	fi
562	printf "\n"
563	fi
564
565
566
567	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
568	kennung=$RANDOM
569	job_on_remhost=${job_name}_${kennung}_$local_host
570	job_to_send=job_to_send_$kennung
571	if [[ $delete_dayfile = false ]]
572	then
573	remote_dayfile=${local_host}_${job_name}_result_$kennung
574	local_dayfile=${remote_host}_${job_name}
575	else
576	remote_dayfile=/dev/null
577	fi
578
579
580
581	# Generate the batch job scripts (qsub/msub/LoadLeveler)
582	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
583	then
584
585	# General LoadLeveler settings
586	execute_in_shell="#!/bin/ksh"
587	use_shell="# @ shell = /bin/ksh"
588	consumable_memory="ConsumableMemory($memory mb)"
589	class="# @ class = $queue"
590	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
591	network_to_use="# @ network.mpi = sn_all,shared,us"
592	data_limit="# @ data_limit = 1.76gb"
593	image_size="# @ image_size = 50"
594	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
595
596	if [[ $email_notification = none ]]
597	then
598	notify_user=""
599	else
600	notify_user="# @ notify_user = $email_notification"
601	if [[ $delete_dayfile = true ]]
602	then
603	notification='# @ notification = never'
604	fi
605	fi
606
607	if [[ $remote_host = ibmh ]]
608	then
609	data_limit=""
610	network_to_use=""
611	class=""
612	environment=""
613	rset="# @ rset = RSET_MCM_AFFINITY"
614	task_affinity="# @ task_affinity = core(1)"
615	elif [[ $remote_host = ibmkisti ]]
616	then
617	network_to_use="# @ network.MPI = sn_all,shared,US"
618	wall_clock_limit="# @ wall_clock_limit = $timestring"
619	if [[ $threads_per_task = 1 ]]
620	then
621	rset="# @ rset = RSET_MCM_AFFINITY"
622	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
623	fi
624	environment=""
625	use_shell=""
626	data_limit=""
627	image_size=""
628	elif [[ $remote_host = ibmku ]]
629	then
630	execute_in_shell="#!/usr/bin/ksh"
631	use_shell="# @ shell = /usr/bin/ksh"
632	consumable_memory=""
633	environment=""
634	network_to_use="# @ network.mpi = sn_all,shared,us"
635	data_limit=""
636	image_size=""
637	elif [[ $remote_host = ibms ]]
638	then
639	network_to_use="# @ network.mpi = csss,shared,us"
640	elif [[ $remote_host = ibmy ]]
641	then
642	consumable_memory=""
643	network_to_use=""
644	fi
645
646	cat > $job_to_send << %%END%%
647	$execute_in_shell
648	$use_shell
649
650	# @ job_type = parallel
651	# @ job_name = $job_name
652	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
653	# @ output = $remote_dayfile
654	# @ error = $remote_dayfile
655	$wall_clock_limit
656	$image_size
657	$class
658	$environment
659	$network_to_use
660	$data_limit
661	$rset
662	$mcm_affinity_options
663	$task_affinity
664	$notification
665	$notify_user
666
667	%%END%%
668
669	if (( nodes > 0 ))
670	then
671
672	if [[ $remote_host != ibmkisti ]]
673	then
674
675	cat >> $job_to_send << %%END%%
676	# @ node = $nodes
677	# @ tasks_per_node = $processes_per_node
678	# @ node_usage = $node_usage
679	# @ queue
680
681	%%END%%
682
683	else
684
685	cat >> $job_to_send << %%END%%
686	# @ total_tasks = $mpi_tasks
687	# @ blocking = unlimited
688	# @ queue
689
690	%%END%%
691
692	fi
693
694	else
695
696	if [[ $remote_host != ibmy ]]
697	then
698
699	cat >> $job_to_send << %%END%%
700	# @ blocking = unlimited
701	# @ total_tasks = $numprocs
702	# @ node_usage = $node_usage
703	# @ queue
704
705	%%END%%
706
707	else
708
709	cat >> $job_to_send << %%END%%
710	# @ node = 1
711	# @ total_tasks = $numprocs
712	# @ queue
713
714	%%END%%
715
716	fi
717
718	fi
719
720	# workaround because of silly job filter on ibmkisti
721	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
722	then
723	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
724	fi
725
726	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
727	then
728
729	cat > $job_to_send << %%END%%
730	#!/bin/ksh
731
732	# @ job_type = serial
733	# @ node_usage = $node_usage
734	# @ job_name = palm
735	# @ wall_clock_limit = ${timestring},$timestring
736	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
737	# @ output = $remote_dayfile
738	# @ error = $remote_dayfile
739	$class
740	$notification
741
742	# @ queue
743
744	%%END%%
745
746	elif [[ $remote_host = lcfimm ]]
747	then
748
749	if [[ $numprocs != 0 ]]
750	then
751	cat > $job_to_send << %%END%%
752	#!/bin/ksh
753	#PBS -N $job_name
754	#PBS -A $project_account
755	#PBS -l walltime=$timestring
756	#PBS -l nodes=${nodes}:ppn=$processes_per_node
757	#PBS -l pmem=${memory}mb
758	#PBS -m abe
759	#PBS -o $remote_dayfile
760	#PBS -j oe
761	mpd &
762
763	%%END%%
764
765	else
766	cat > $job_to_send << %%END%%
767	#!/bin/ksh
768	#PBS -N $job_name
769	#PBS -A $project_account
770	#PBS -l walltime=$timestring
771	#PBS -l ncpus=1
772	#PBS -l pmem=${memory}mb
773	#PBS -m abe
774	#PBS -o $remote_dayfile
775	#PBS -j oe
776
777	%%END%%
778
779	fi
780
781	elif [[ $remote_host = lcflow ]]
782	then
783	if [ $memory -gt 1900 ]; then
784	use_himem=""
785	else
786	use_himem="#"
787	fi
788	if [[ $numprocs != 0 ]]
789	then
790	pe_set="#$ -pe impi $numprocs"
791	else
792	pe_set="#$ -pe impi 1"
793	fi
794	if [[ $queue = default ]]
795	then
796	queue_set=""
797	else
798	queue_set="#$ -q $queue"
799	fi
800	[[ "$disc_space" = "" ]] && disc_space=50
801
802	cat > $job_to_send << %%END%%
803	#!/bin/bash
804	#$ -S /bin/bash
805	#$ -N $job_name
806	#$ -cwd
807	#$ -l h_rt=$timestring
808	#$ -l h_vmem=${memory}M
809	#$ -o $remote_dayfile
810	#$ -j y
811	$pe_set
812	#$ -R y
813	${use_himem}#$ -l highmem=true
814	#$ -l h_fsize=${disc_space}G
815	$queue_set
816
817	%%END%%
818
819	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
820	then
821
822	if [[ $numprocs != 0 ]]
823	then
824	cat > $job_to_send << %%END%%
825	#!/bin/ksh
826	#PBS -N $job_name
827	#PBS -l walltime=$timestring
828	#PBS -l ncpus=$numprocs
829	#PBS -l pmem=${memory}mb
830	#PBS -o $remote_dayfile
831	#PBS -l nodes=$nodes:ppn=${processes_per_node}
832	#PBS -j oe
833
834	mpd &
835
836	%%END%%
837
838	else
839	cat > $job_to_send << %%END%%
840	#!/bin/ksh
841	#PBS -N $job_name
842	#PBS -l walltime=$timestring
843	#PBS -l ncpus=1
844	#PBS -l pmem=${memory}mb
845	#PBS -o $remote_dayfile
846	#PBS -j oe
847
848	%%END%%
849
850	fi
851
852	elif [[ $remote_host = lckiaps ]]
853	then
854
855	if [[ $numprocs != 0 ]]
856	then
857	cat > $job_to_send << %%END%%
858	#!/bin/ksh
859	#PBS -N $job_name
860	#PBS -l walltime=$timestring
861	#PBS -l select=1:ncpus=$numprocs
862	#PBS -l pmem=${memory}mb
863	#PBS -q $queue
864	#PBS -o $remote_dayfile
865	#PBS -j oe
866	#PBS -V
867
868	%%END%%
869
870	else
871	cat > $job_to_send << %%END%%
872	#!/bin/ksh
873	#PBS -N $job_name
874	#PBS -l walltime=$timestring
875	#PBS -l ncpus=1
876	#PBS -l pmem=${memory}mb
877	#PBS -o $remote_dayfile
878	#PBS -j oe
879
880	%%END%%
881
882	fi
883
884	elif [[ $remote_host = lcyon ]]
885	then
886
887	if [[ $numprocs != 0 ]]
888	then
889	cat > $job_to_send << %%END%%
890	#!/bin/ksh
891	#PBS -N $job_name
892	#PBS -l walltime=$timestring
893	#PBS -l ncpus=$numprocs
894	#PBS -l pmem=${memory}mb
895	#PBS -o $remote_dayfile
896	#PBS -j oe
897
898	%%END%%
899
900	else
901	cat > $job_to_send << %%END%%
902	#!/bin/ksh
903	#PBS -N $job_name
904	#PBS -l walltime=$timestring
905	#PBS -l ncpus=1
906	#PBS -l pmem=${memory}mb
907	#PBS -o $remote_dayfile
908	#PBS -j oe
909
910	%%END%%
911
912	fi
913
914	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
915	then
916
917	if [[ "$sgi_feature" != "" ]]
918	then
919	feature_directive="#PBS -l feature=$sgi_feature"
920	else
921	feature_directive=""
922	fi
923
924	if [[ $queue = dataq \|\| $queue = permq ]]
925	then
926	feature_directive="#PBS -l feature=data"
927	fi
928
929	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
930	then
931	queue_directive="#PBS -q $queue"
932	else
933	queue_directive=""
934	fi
935
936	if [[ $email_notification = none ]]
937	then
938	email_directive=""
939	else
940	email_directive="#PBS -M $email_notification"
941	fi
942
943	if [[ $numprocs != 0 ]]
944	then
945	cat > $job_to_send << %%END%%
946	#!/bin/bash
947	#PBS -N $job_name
948	#PBS -l walltime=$timestring
949	#PBS -l nodes=$nodes:ppn=${processes_per_node}
950	#PBS -l naccesspolicy=$node_usage
951	#PBS -o $remote_dayfile
952	#PBS -j oe
953	$feature_directive
954	$queue_directive
955	$email_directive
956
957	eval \`/sw/swdist/bin/modulesinit\`
958	#. /usr/share/modules/init/bash
959	$init_cmds
960	$module_calls
961
962	echo ld_library_path=\$LD_LIBRARY_PATH
963
964	%%END%%
965
966	else
967	cat > $job_to_send << %%END%%
968	#PBS -S /bin/bash
969	#PBS -N $job_name
970	#PBS -l walltime=$timestring
971	#PBS -l ncpus=1
972	#PBS -o $remote_dayfile
973	#PBS -j oe
974	$queue_directive
975	$email_directive
976
977	eval \`/sw/swdist/bin/modulesinit\`
978	#. /usr/share/modules/init/bash
979	$init_cmds
980	$module_calls
981
982	%%END%%
983
984	fi
985
986	elif [[ $remote_host = lcxe6 ]]
987	then
988
989	if [[ $numprocs != 0 ]]
990	then
991	cat > $job_to_send << %%END%%
992	#!/bin/ksh
993	#PBS -S /bin/ksh
994	#PBS -N $job_name
995	#PBS -A $project_account
996	#PBS -j oe
997	#PBS -l walltime=$timestring
998	#PBS -l mppwidth=${numprocs}
999	#PBS -l mppnppn=${processes_per_node}
1000	#PBS -m abe
1001	#PBS -o $remote_dayfile
1002	$email_directive
1003
1004	$init_cmds
1005	$module_calls
1006
1007	%%END%%
1008
1009	else
1010	cat > $job_to_send << %%END%%
1011	#!/bin/ksh
1012	#PBS -S /bin/ksh
1013	#PBS -N $job_name
1014	#PBS -A $project_account
1015	#PBS -j oe
1016	#PBS -l walltime=$timestring
1017	#PBS -l ncpus=1
1018	#PBS -l pmem=${memory}mb
1019	#PBS -m abe
1020	$email_directive
1021	#PBS -o $remote_dayfile
1022
1023	$init_cmds
1024	$module_calls
1025
1026	%%END%%
1027
1028	fi
1029
1030	elif [[ $remote_host = lckyoto ]]
1031	then
1032
1033	cat > $job_to_send << %%END%%
1034	#!/bin/ksh
1035	# @\$-o $remote_dayfile
1036	# @\$-eo -oi
1037	# @\$-lP 16
1038	# @\$-lp 1
1039	# @\$-lm 28gb -llm unlimited -ls unlimited
1040	# @\$-q $queue
1041	# @\$-Pvn abs_pack
1042	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1043	#. /thin/local/etc/setprofile/intel-11.0.sh
1044	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1045	. ~/.myprofile
1046	#. /home2/t/t51254/palm/current_version/myprofile
1047	#. /thin/apps/pgi/mpi.sh
1048	#
1049	env
1050	#
1051	set -x
1052
1053	%%END%%
1054
1055	elif [[ $remote_host = lcxt5m ]]
1056	then
1057
1058	if [[ $numprocs != 0 ]]
1059	then
1060	cat > $job_to_send << %%END%%
1061	#!/bin/ksh
1062	#PBS -S /bin/ksh
1063	#PBS -N $job_name
1064	#PBS -j oe
1065	#PBS -l walltime=$timestring
1066	#PBS -l mppwidth=${numprocs}
1067	#PBS -l mppnppn=${processes_per_node}
1068	#PBS -m abe
1069	#PBS -o $remote_dayfile
1070
1071	$init_cmds
1072	$module_calls
1073
1074	%%END%%
1075
1076	else
1077	cat > $job_to_send << %%END%%
1078	#!/bin/ksh
1079	#PBS -S /bin/ksh
1080	#PBS -N $job_name
1081	#PBS -j oe
1082	#PBS -l walltime=$timestring
1083	#PBS -l ncpus=1
1084	#PBS -l pmem=${memory}mb
1085	#PBS -m abe
1086	#PBS -o $remote_dayfile
1087
1088	$init_cmds
1089	$module_calls
1090
1091	%%END%%
1092
1093	fi
1094
1095	elif [[ $remote_host = lckyuh ]]
1096	then
1097	cat > $job_to_send << %%END%%
1098	#!/bin/bash
1099	#PJM -L "rscgrp=$queue"
1100	#PJM -L "node=$nodes"
1101	#PJM --mpi "proc=$numprocs"
1102	#PJM -L "elapse=$timestring"
1103	#PJM -o $remote_dayfile
1104	#PJM -j
1105	#PJM -X
1106	#PJM --no-stging
1107
1108	export LANG=en_US.UTF-8
1109	%%END%%
1110
1111	elif [[ $remote_host = lckyut ]]
1112	then
1113	cat > $job_to_send << %%END%%
1114	#!/bin/bash
1115	#PJM -L "rscgrp=$queue"
1116	#PJM -L "vnode=$numprocs"
1117	#PJM -L "vnode-core=1"
1118	#PJM -L "elapse=$timestring"
1119	#PJM --mpi proc=$numprocs
1120	#PJM -o $remote_dayfile
1121	#PJM -j
1122	#PJM -X
1123	#PJM --no-stging
1124
1125	export LANG=en_US.UTF-8
1126	%%END%%
1127
1128	elif [[ $remote_host = nech ]]
1129	then
1130
1131	if (( nodes > 1 ))
1132	then
1133	cat > $job_to_send << %%END%%
1134	#!/bin/ksh
1135	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1136	#PBS -l ${qsubmem}=${Memory}gb
1137	#PBS -b $nodes
1138	#PBS -o $remote_dayfile
1139	#PBS -N palm
1140	#PBS -j o
1141	#PBS -T mpisx
1142
1143	%%END%%
1144
1145	elif [[ $numprocs != 0 ]]
1146	then
1147	cat > $job_to_send << %%END%%
1148	#!/bin/ksh
1149	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1150	#PBS -l ${qsubmem}=${Memory}gb
1151	#PBS -o $remote_dayfile
1152	#PBS -N palm
1153	#PBS -j o
1154
1155	%%END%%
1156
1157	else
1158	cat > $job_to_send << %%END%%
1159	#!/bin/ksh
1160	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1161	#PBS -o $remote_dayfile
1162	#PBS -j o
1163
1164	%%END%%
1165
1166	fi
1167
1168	elif [[ $remote_host = necriam ]]
1169	then
1170
1171	if (( nodes > 1 ))
1172	then
1173	cat > $job_to_send << %%END%%
1174	#!/bin/ksh
1175	#PBS -b $nodes
1176	#PBS -o $remote_dayfile
1177	#PBS -N $job_name
1178	#PBS -j o
1179	#PBS -v MPIPROGINV=YES
1180
1181	%%END%%
1182
1183	elif [[ $numprocs != 0 ]]
1184	then
1185	cat > $job_to_send << %%END%%
1186	#!/bin/ksh
1187	#PBS -q ${queue}
1188	#PBS -o $remote_dayfile
1189	#PBS -N $job_name
1190	#PBS -j o
1191	#PBS -v MPIPROGINV=YES
1192
1193	%%END%%
1194
1195	else
1196	cat > $job_to_send << %%END%%
1197	#!/bin/ksh
1198	#PBS -o $remote_dayfile
1199	#PBS -j o
1200
1201	%%END%%
1202
1203	fi
1204
1205	elif [[ $remote_host = lctit ]]
1206	then
1207	cat > $job_to_send << %%END%%
1208	#!/bin/ksh
1209	$init_cmds
1210	$module_calls
1211
1212	%%END%%
1213
1214	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1215	if [[ $tasks_per_node != $processes_per_node ]]
1216	then
1217	submcom="$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 $queue "
1218	else
1219	submcom="$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 "
1220	fi
1221
1222	else
1223
1224	cat > $job_to_send << %%END%%
1225	# @\$-q ${queue}
1226	# @\$-l${qsubtime} $timestring
1227	# @\$-l${qsubmem} ${memory}mb
1228	# @\$-o $remote_dayfile
1229	# @\$-eo
1230
1231	%%END%%
1232
1233	fi
1234
1235
1236	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1237	# BEI EXIT VERANLASSEN
1238	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1239	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1240	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1241	then
1242	echo "set +vx" >> $job_to_send
1243	echo "trap '" >> $job_to_send
1244	echo "set +vx" >> $job_to_send
1245	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* ]]
1246	then
1247	if [[ $remote_host = ibmh ]]
1248	then
1249	return_queue=c1
1250	elif [[ $remote_host = ibmkisti ]]
1251	then
1252	return_queue=class.1-2
1253	elif [[ $remote_host = ibmku ]]
1254	then
1255	return_queue=sdbg2
1256	elif [[ $remote_host = ibms ]]
1257	then
1258	return_queue=p_normal
1259	elif [[ $remote_host = ibmy ]]
1260	then
1261	return_queue=serial
1262	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1263	then
1264	return_queue=serialq
1265	elif [[ $remote_host = necriam ]]
1266	then
1267	return_queue=SP
1268	elif [[ $remote_host = lckiaps ]]
1269	then
1270	return_queue=express
1271	elif [[ $remote_host = lckyuh ]]
1272	then
1273	return_queue=cx-single
1274	elif [[ $remote_host = lckyut ]]
1275	then
1276	return_queue=cx-single
1277	else
1278	return_queue=unknown
1279	fi
1280
1281	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1282	then
1283
1284	if [[ $remote_host = ibmku ]]
1285	then
1286	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1287	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1288	else
1289	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1290	fi
1291	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1292	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1293	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1294	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1295	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1296	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1297	if [[ $host != "ibmh" ]]
1298	then
1299	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1300	fi
1301	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1302	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1303
1304	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1305	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1306
1307	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1308	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
1309	if [[ $remote_host = ibmku ]]
1310	then
1311	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1312	fi
1313	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1314
1315	elif [[ $remote_host = nech ]]
1316	then
1317	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1318	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1319	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1320	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1321	echo "#PBS -j o" >> $job_to_send
1322	echo " " >> $job_to_send
1323	echo "set -x" >> $job_to_send
1324	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1325	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1326	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1327	echo "%%END%%" >> $job_to_send
1328
1329	elif [[ $remote_host = necriam ]]
1330	then
1331	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1332	echo "#PBS -q $return_queue" >> $job_to_send
1333	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1334	echo "#PBS -j o" >> $job_to_send
1335	echo " " >> $job_to_send
1336	echo "set -x" >> $job_to_send
1337	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1338	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1339	echo "%%END%%" >> $job_to_send
1340
1341	elif [[ $remote_host = lckyuh ]]
1342	then
1343	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1344	echo "#!/bin/bash" >> $job_to_send
1345	echo "#PJM -L \"node=1\"" >> $job_to_send
1346	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1347	echo "#PJM --no-stging" >> $job_to_send
1348	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1349	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1350	echo "#PJM -j" >> $job_to_send
1351	echo " " >> $job_to_send
1352	echo "export LANG=en_US.UTF-8" >> $job_to_send
1353	echo "set -x" >> $job_to_send
1354	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1355	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1356	echo "%%END%%" >> $job_to_send
1357
1358	elif [[ $remote_host = lckyut ]]
1359	then
1360	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1361	echo "#!/bin/bash" >> $job_to_send
1362	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1363	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1364	echo "#PJM --no-stging" >> $job_to_send
1365	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1366	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1367	echo "#PJM -j" >> $job_to_send
1368	echo " " >> $job_to_send
1369	echo "export LANG=en_US.UTF-8" >> $job_to_send
1370	echo "set -x" >> $job_to_send
1371	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1372	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1373	echo "%%END%%" >> $job_to_send
1374
1375	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1376	then
1377	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1378	echo "#!/bin/bash" >> $job_to_send
1379	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1380	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1381	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1382	echo "#PBS -l feature=data" >> $job_to_send
1383	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1384	echo "#PBS -j oe" >> $job_to_send
1385	echo " " >> $job_to_send
1386	echo ". /usr/share/modules/init/bash" >> $job_to_send
1387	echo "set -x" >> $job_to_send
1388	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1389	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1390	echo "%%END%%" >> $job_to_send
1391
1392	elif [[ $remote_host = lcflow ]]
1393	then
1394	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1395	echo "#!/bin/bash" >> $job_to_send
1396	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1397	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1398	echo "SGEPREFIX -cwd" >> $job_to_send
1399	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1400	echo "SGEPREFIX -l h_vmem=100M" >> $job_to_send
1401	echo "SGEPREFIX -j y" >> $job_to_send
1402	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1403	echo " " >> $job_to_send
1404	echo "set -x" >> $job_to_send
1405	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1406	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1407	echo "" >> $job_to_send
1408	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1409	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1410	echo "rm -f scpjob.${kennung}" >> $job_to_send
1411	echo "%%END%%" >> $job_to_send
1412	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1413	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1414
1415	else
1416
1417	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1418	echo "# @\\\$-q $return_queue" >> $job_to_send
1419	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1420	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1421	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1422	then
1423	echo "# @\$-l mpp_p=0" >> $job_to_send
1424	fi
1425	echo '# @\$-lF 10mb' >> $job_to_send
1426	# echo '# @\$-o /dev/null' >> $job_to_send
1427	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1428	echo '# @\\\$-eo' >> $job_to_send
1429	echo " " >> $job_to_send
1430	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1431	then
1432	echo "set +vx" >> $job_to_send
1433	echo ". .profile" >> $job_to_send
1434	fi
1435	echo "set -x" >> $job_to_send
1436	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
1437	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1438	echo "%%END%%" >> $job_to_send
1439	fi
1440
1441	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1442	then
1443	echo "llsubmit scpjob.$kennung" >> $job_to_send
1444	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1445	then
1446	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1447	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1448	echo "msub scpjob.$kennung" >> $job_to_send
1449	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1450	then
1451	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1452	elif [[ $remote_host = t3es ]]
1453	then
1454	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1455	elif [[ $remote_host = lckiaps ]]
1456	then
1457	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1458	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1459	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1460	elif [[ $remote_host = lckyu* ]]
1461	then
1462	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1463	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1464	elif [[ $remote_host = lcflow ]]
1465	then
1466	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1467	echo "ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1468	else
1469	echo "$submcom scpjob.$kennung" >> $job_to_send
1470	fi
1471	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1472	then
1473	echo "rm scpjob.$kennung" >> $job_to_send
1474	fi
1475	if [[ $remote_host = nech ]]
1476	then
1477	echo "cd -" >> $job_to_send
1478	fi
1479	else
1480	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1481	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1482	fi
1483	echo "set -x" >> $job_to_send
1484	echo " ' exit" >> $job_to_send
1485	echo "set -x" >> $job_to_send
1486	fi
1487
1488
1489
1490	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1491	cat $file_to_send >> $job_to_send
1492
1493	if [[ $remote_host = ibm ]]
1494	then
1495	echo " " >> $job_to_send
1496	echo "exit" >> $job_to_send
1497	fi
1498
1499	# remove job file
1500	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1501	then
1502	echo " " >> $job_to_send
1503	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1504	fi
1505
1506
1507
1508	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1509	if [[ -z $remote_user ]]
1510	then
1511	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1512	then
1513	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1514	fi
1515	fi
1516
1517
1518
1519	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1520	if [[ $no_submit = false ]]
1521	then
1522	if [[ $remote_host != $local_host ]]
1523	then
1524	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1525	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1526	then
1527	job_catalog_save=$job_catalog
1528	job_catalog=job_queue
1529	elif [[ $remote_host = nech ]]
1530	then
1531	job_catalog_save=$job_catalog
1532	job_catalog=/hpf/b/${remote_user}/job_queue
1533	fi
1534	if [[ $remote_host = nech ]]
1535	then
1536	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1537	# TRANSFERIERT WERDEN
1538	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1539	else
1540	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1541	fi
1542	if [[ $? = 1 ]]
1543	then
1544	locat=scp; exit
1545	fi
1546	if [[ $remote_host = ibms ]]
1547	then
1548	job_catalog=$job_catalog_save
1549	fi
1550	[[ $verify = true ]] && printf "\n >>> finished\n"
1551	else
1552	eval job_catalog=$job_catalog
1553	cp $job_to_send ${job_catalog}/$job_on_remhost
1554	fi
1555
1556
1557
1558	# NQS- BZW. LOADLEVELER-JOB STARTEN
1559	if [[ $remote_host != $local_host ]]
1560	then
1561	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1562
1563	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1564	then
1565	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1566	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1567	elif [[ $remote_host = ibmku ]]
1568	then
1569	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1570	elif [[ $remote_host = lcflow ]]
1571	then
1572	ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1573	else
1574	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1575	fi
1576
1577	[[ $verify = true ]] && printf " >>> o.k.\n"
1578	else
1579	cd $job_catalog
1580	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1581	then
1582	eval $submcom $job_on_remhost
1583	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1584	then
1585	chmod u+x $job_on_remhost
1586	eval $submcom $job_on_remhost
1587	elif [[ $local_host = nech ]]
1588	then
1589	if [[ $queue = default ]]
1590	then
1591	eval $submcom $job_on_remhost
1592	else
1593	eval $submcom -q $queue $job_on_remhost
1594	fi
1595	else
1596	qsub $job_on_remhost
1597	fi
1598
1599	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1600	# only at the end of the job.
1601	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1602	then
1603	rm $job_on_remhost
1604	fi
1605	cd - > /dev/null
1606	fi
1607	fi
1608
1609
1610
1611	# ABSCHLUSSARBEITEN
1612
1613	if [[ $no_submit = false ]]
1614	then
1615	rm -f $job_to_send
1616	fi
1617	[[ $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 |