Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1179

Last change on this file since 1179 was 1174, checked in by gryschka, 12 years ago
Bugfix in computing initial profiles for ug, vg, lad, q in case of Atmosphere
Property svn:keywords set to `Id Rev`
File size: 54.1 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2012 Leibniz University Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1174 2013-05-31 10:28:08Z raasch $
30	#
31	# 1103 2013-02-20 02:15:53Z raasch
32	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
33	# instead typeset -L/R),
34	# further adjustments for lckyuh
35	#
36	# 2013-02-10 01:47:43Z raasch
37	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
38	# and for Forwind cluster (lcflow)
39	#
40	# 1094 2013-02-03 01:52:12Z raasch
41	# new option -P for explicit setting of ssh/scp port,
42	# decalpha parts (yonsei) removed
43	#
44	# 2013-02-02 07:06:13Z raasch
45	# adjustments for Kyushu-University computing center (lckyut - tatara)
46	# old changelog messages removed
47	#
48	# 1046 2012-11-09 14:38:45Z maronga
49	# code put under GPL (PALM 3.9)
50	#
51	# 08/07/94 - Siggi - first version finished
52	# 29/06/94 - Siggi - script development started
53	#--------------------------------------------------------------------------------#
54	# subjob - script for automatic generation and submission of batch-job files
55	# for various batch queuing systems
56	#--------------------------------------------------------------------------------#
57
58
59	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
60	delete_dayfile=false
61	email_notification=none
62	group_number=none
63	locat=normal
64	no_default_queue=none
65	no_submit=false
66	job_catalog="~/job_queue"
67	job_name=none
68	local_user=$LOGNAME
69	node_usage=shared
70	numprocs=0
71	punkte="..........................................................."
72	submcom=qsub
73	queue=default
74	remote_host=none
75	remote_user=""
76	verify=true
77
78	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
79	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
80
81
82
83	# FEHLERBEHANDLUNG
84	# BEI EXIT:
85	trap 'if [[ $locat != normal ]]
86	then
87	case $locat in
88	(option) printf "\n --> available optios can be displayed"
89	printf " by typing:"
90	printf "\n \"subjob ?\" \n";;
91	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
92	(*) printf "\n +++ unknown error"
93	printf "\n please inform S. Raasch!\n"
94	esac
95	[[ -f $job_to_send ]] && rm $job_to_send
96	printf "\n\n+++ SUBJOB killed \n\n"
97	fi' exit
98
99
100	# BEI TERMINAL-BREAK:
101	trap '[[ -f $job_to_send ]] && rm $job_to_send
102	printf "\n\n+++ SUBJOB killed \n\n"
103	exit
104	' 2
105
106
107
108
109	# LOKALEN HOSTNAMEN FESTSTELLEN
110	local_host=$(hostname)
111
112
113
114	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
115	# UEBERHAUPT ZULAESSIG IST
116	# Note: One of the entries for "lck" or "lckordi" always has to be
117	# comment out, because the hostname (node*) is same for both machines
118	case $local_host in
119	(atmos) local_addres=172.20.25.35; local_host=lcide;;
120	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
121	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
122	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
123	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
124	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
125	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
126	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
127	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
128	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
129	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
130	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
131	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
132	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
133	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
134	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
135	(cs*) local_addres=136.172.44.131; local_host=nech;;
136	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
137	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
138	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
139	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
140	(node*) local_addres=165.132.26.61 local_host=lck;;
141	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
142	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
143	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
144	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
145	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
146	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
147	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
148	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
149	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
150	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
151	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
152	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
153	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
154	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
155	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
156	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
157	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
158	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
159	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
160	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
161	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
162	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
163	(meller) local_addres=134.106.74.155; local_host=lcfor;;
164	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
165	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
166	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
167	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
168	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
169	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
170	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
171	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
172	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
173	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
174	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
175	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
176	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
177	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
178	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
179	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
180	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
181	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
182	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
183	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
184	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
185	(*) printf "\n +++ \"$local_host\" unknown";
186	printf "\n please inform S. Raasch!";
187	locat=parameter; exit;;
188	esac
189
190
191
192	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
193	remote_host=$local_host
194
195
196
197
198	# PROZEDUROPTIONEN EINLESEN
199	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
200	do
201	case $option in
202	(c) job_catalog=$OPTARG;;
203	(d) delete_dayfile=true;;
204	(D) no_submit=true;;
205	(e) email_notification=$OPTARG;;
206	(g) group_number=$OPTARG;;
207	(h) remote_host=$OPTARG;;
208	(m) memory=$OPTARG;;
209	(n) job_name=$OPTARG;;
210	(N) node_usage=$OPTARG;;
211	(O) threads_per_task=$OPTARG;;
212	(P) scp_port=$OPTARG;;
213	(q) no_default_queue=$OPTARG;;
214	(t) cputime=$OPTARG;;
215	(T) tasks_per_node=$OPTARG;;
216	(u) remote_user=$OPTARG;;
217	(v) verify=false;;
218	(X) numprocs=$OPTARG;;
219	(\?) printf "\n +++ Option $OPTARG unknown \n";
220	locat=option; exit;;
221	esac
222	done
223
224
225	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
226	(( to_shift = $OPTIND - 1 ))
227	shift $to_shift; file_to_send=$1
228
229
230	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
231	if [ "$1" = "?" ]
232	then
233	(printf "\n *** subjob can be called as follows:\n"
234	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
235	printf "\n Description of available options:\n"
236	printf "\n Option Description Default-Value"
237	printf "\n -c job-input- and output-catalog ~/job_queue"
238	printf "\n -d no job-protocol will be created ---"
239	printf "\n -D only the job-file will be created ---"
240	printf "\n -h execution host, available hosts: $remote_host"
241	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
242	printf "\n lckiaps, lctit, nech, necriam, unics"
243	printf "\n -m memory demand per process in MByte ---"
244	printf "\n -n jobname <jobdatei>"
245	printf "\n -O threads per task (for OpenMP usage) 1"
246	printf "\n -P ssh/scp port default port"
247	printf "\n -q job-queue to be used default"
248	printf "\n -t allowed cpu-time in seconds ---"
249	printf "\n -T tasks per node (on parallel hosts) ---"
250	printf "\n -u username on execution host from .netrc"
251	printf "\n -v no prompt for confirmation ---"
252	printf "\n -X # of processors (on parallel hosts) 1"
253	printf "\n "
254	printf "\n The only possible positional parameter is <jobfile>:"
255	printf "\n The complete NQS-job must be provided here."
256	printf "\n <jobfile>=? creates this outline\n\n") \| more
257	exit
258	fi
259
260
261
262	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
263	if [[ "$file_to_send" = "" ]]
264	then
265	printf "\n +++ job-file missing"
266	locat=parameter; exit
267	else
268	if [[ -f $file_to_send ]]
269	then
270	true
271	else
272	printf "\n +++ job-file: "
273	printf "\n $file_to_send"
274	printf "\n does not exist"
275	locat=parameter; exit
276	fi
277	fi
278
279
280
281	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
282	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
283	if [[ $job_name = none ]]
284	then
285	job_name=$file_to_send
286	fi
287	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
288	then
289	printf "\n +++ job-file name: "
290	printf "\n $job_name"
291	printf "\n must not contain \"/\"-characters"
292	locat=parameter; exit
293	fi
294
295
296
297
298	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
299	# ODER WENN HOST NICHT ANGEGEBEN WURDE
300	if [[ $remote_host = none ]]
301	then
302	printf "\n +++ host missing"
303	locat=option; exit
304	else
305	case $remote_host in
306	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
307	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
308	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
309	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
310	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
311	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
312	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
313	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
314	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
315	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
316	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
317	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
318	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
319	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
320	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
321	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
322	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
323	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
324	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
325	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
326	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
327	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
328	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
329	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
330	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
331	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
332	locat=parameter; exit;;
333	esac
334	fi
335
336
337	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
338	if [[ $no_default_queue != none ]]
339	then
340	error=false
341	ndq=$no_default_queue
342	case $remote_host in
343	(ibm) case $ndq in
344	(p690_express\|p690_standard\|p690_long) error=false;;
345	(*) error=true;;
346	esac;;
347	(ibmh) case $ndq in
348	(no_class) error=false;;
349	(*) error=true;;
350	esac;;
351	(ibmkisti) case $ndq in
352	(class.32plus\|class.1-2\|class.2-32) error=false;;
353	(*) error=true;;
354	esac;;
355	(ibmku) case $ndq in
356	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
357	(*) error=true;;
358	esac;;
359	(ibms) case $ndq in
360	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
361	(*) error=true;;
362	esac;;
363	(ibmy) case $ndq in
364	(parallel) error=false;;
365	(*) error=true;;
366	esac;;
367	(lcflow) case $ndq in
368	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
369	(*) error=true;;
370	esac;;
371	(lckiaps) case $ndq in
372	(express\|normal) error=false;;
373	(*) error=true;;
374	esac;;
375	(lckyoto) case $ndq in
376	(eh\|ph) error=false;;
377	(*) error=true;;
378	esac;;
379	(lckyuh) case $ndq in
380	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
381	(*) error=true;;
382	esac;;
383	(lckyut) case $ndq in
384	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
385	(*) error=true;;
386	esac;;
387	(lcsgib\|lcsgih) case $ndq in
388	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
389	(*) error=true;;
390	esac;;
391	(lctit) case $ndq in
392	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
393	(*) error=true;;
394	esac;;
395	(t3eb) case $ndq in
396	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
397	(*) error=true;;
398	esac;;
399	(necriam) case $ndq in
400	(SP\|SS\|P6) error=false;;
401	(*) error=true;;
402	esac;;
403	(t3eh) case $ndq in
404	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
405	(*) error=true;;
406	esac;;
407	(t3ej2\|t3ej5) case $ndq in
408	(low\|normal\|high) error=false;;
409	(*) error=true;;
410	esac;;
411	(t3es) case $ndq in
412	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
413	(*) error=true;;
414	esac;;
415	(unics) case $ndq in
416	(unics\|ht) error=false;;
417	(*) error=true;;
418	esac;;
419	esac
420	if [[ $error = true ]]
421	then
422	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
423	locat=parameter; exit
424	else
425	queue=$no_default_queue
426	fi
427	fi
428
429
430
431	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
432	# AUFTEILEN
433	done=false
434	while [[ $done = false ]]
435	do
436	if (( $cputime <= 0 ))
437	then
438	printf "\n +++ wrong cpu-time or cpu-time missing"
439	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
440	printf "\n >>> "
441	read cputime 1>/dev/null 2>&1
442	else
443	done=true
444	fi
445	done
446	if [[ $remote_host = nech ]]
447	then
448	if (( tasks_per_node != 0 ))
449	then
450	(( cputime = cputime * tasks_per_node ))
451	elif [[ $numprocs != 0 ]]
452	then
453	(( cputime = cputime * numprocs ))
454	fi
455	fi
456	(( stunden = cputime / 3600 ))
457	(( resttime = cputime - stunden * 3600 ))
458	(( minuten = resttime / 60 ))
459	(( sekunden = resttime - minuten * 60 ))
460	timestring=${stunden}:${minuten}:${sekunden}
461
462
463
464	# PRUEFEN DER KERNSPEICHERANFORDERUNG
465	done=false
466	while [[ $done = false ]]
467	do
468	if (( memory <= 0 ))
469	then
470	printf "\n +++ wrong memory demand or memory demand missing"
471	printf "\n >>> Please type memory in MByte per process as INTEGER:"
472	printf "\n >>> "
473	read memory 1>/dev/null 2>&1
474	else
475	done=true
476	fi
477	done
478
479	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
480	then
481	if (( tasks_per_node != 0 ))
482	then
483	(( Memory = memory * tasks_per_node / 1000 ))
484	elif [[ $numprocs != 0 ]]
485	then
486	(( Memory = memory * numprocs / 1000 ))
487	else
488	(( Memory = memory / 1000 ))
489	fi
490	elif [[ $remote_host = lctit ]]
491	then
492	(( Memory = memory * tasks_per_node / 1000 ))
493	fi
494
495
496	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
497	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
498	then
499	(( memory = memory * threads_per_task ))
500	fi
501
502
503	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
504	if (( tasks_per_node != 0 ))
505	then
506	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
507	fi
508
509
510	# Calculate number of processes per node
511	(( processes_per_node = tasks_per_node * threads_per_task ))
512
513
514	# Calculate number of MPI tasks
515	(( mpi_tasks = numprocs / threads_per_task ))
516
517
518	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
519	if [[ "$scp_port" != "" ]]
520	then
521	PORTOPT="-P $scp_port"
522	SSH_PORTOPT="-p $scp_port"
523	fi
524
525
526	# HEADER-AUSGABE
527	if [[ $verify = true ]]
528	then
529	printf "\n\n"
530	printf "#--------------------------------------------------------------# \n"
531	spalte1=SUBJOB;spalte2=$(date)
532	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
533	printf "\| \| \n"
534	printf "\| values of parameters/options: \| \n"
535	spalte1=$(echo local_host$punkte \| cut -c-20)
536	spalte2=$punkte$local_host
537	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
538	spalte1=$(echo remote_host$punkte \| cut -c-20)
539	spalte2=$punkte$remote_host
540	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
541	spalte1=$(echo queue$punkte \| cut -c-20)
542	spalte2=$punkte$queue
543	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
544	spalte1=$(echo memory$punkte \| cut -c-20)
545	spalte2="$punkte$memory mb"
546	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
547	spalte1=$(echo cputime$punkte \| cut -c-20)
548	spalte2="$punkte$cputime sec"
549	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
550	spalte1=$(echo job_name$punkte \| cut -c-20)
551	spalte2="$punkte$job_name"
552	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
553	printf "#--------------------------------------------------------------# \n\n"
554
555
556	# KONTROLLABFRAGE, OB ALLES O.K.
557	antwort="dummy"
558	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
559	do
560	read antwort?" >>> continue (y/n) ? "
561	done
562	if [[ $antwort = n \|\| $antwort = N ]]
563	then
564	locat=verify; exit
565	fi
566	printf "\n"
567	fi
568
569
570
571	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
572	kennung=$RANDOM
573	job_on_remhost=${job_name}_${kennung}_$local_host
574	job_to_send=job_to_send_$kennung
575	if [[ $delete_dayfile = false ]]
576	then
577	remote_dayfile=${local_host}_${job_name}_result_$kennung
578	local_dayfile=${remote_host}_${job_name}
579	else
580	remote_dayfile=/dev/null
581	fi
582
583
584
585	# Generate the batch job scripts (qsub/msub/LoadLeveler)
586	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
587	then
588
589	# General LoadLeveler settings
590	execute_in_shell="#!/bin/ksh"
591	use_shell="# @ shell = /bin/ksh"
592	consumable_memory="ConsumableMemory($memory mb)"
593	class="# @ class = $queue"
594	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
595	network_to_use="# @ network.mpi = sn_all,shared,us"
596	data_limit="# @ data_limit = 1.76gb"
597	image_size="# @ image_size = 50"
598	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
599
600	if [[ $email_notification = none ]]
601	then
602	notify_user=""
603	else
604	notify_user="# @ notify_user = $email_notification"
605	if [[ $delete_dayfile = true ]]
606	then
607	notification='# @ notification = never'
608	fi
609	fi
610
611	if [[ $remote_host = ibmh ]]
612	then
613	data_limit=""
614	network_to_use=""
615	class=""
616	environment=""
617	rset="# @ rset = RSET_MCM_AFFINITY"
618	task_affinity="# @ task_affinity = core(1)"
619	elif [[ $remote_host = ibmkisti ]]
620	then
621	network_to_use="# @ network.MPI = sn_all,shared,US"
622	wall_clock_limit="# @ wall_clock_limit = $timestring"
623	if [[ $threads_per_task = 1 ]]
624	then
625	rset="# @ rset = RSET_MCM_AFFINITY"
626	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
627	fi
628	environment=""
629	use_shell=""
630	data_limit=""
631	image_size=""
632	elif [[ $remote_host = ibmku ]]
633	then
634	execute_in_shell="#!/usr/bin/ksh"
635	use_shell="# @ shell = /usr/bin/ksh"
636	consumable_memory=""
637	environment=""
638	network_to_use="# @ network.mpi = sn_all,shared,us"
639	data_limit=""
640	image_size=""
641	elif [[ $remote_host = ibms ]]
642	then
643	network_to_use="# @ network.mpi = csss,shared,us"
644	elif [[ $remote_host = ibmy ]]
645	then
646	consumable_memory=""
647	network_to_use=""
648	fi
649
650	cat > $job_to_send << %%END%%
651	$execute_in_shell
652	$use_shell
653
654	# @ job_type = parallel
655	# @ job_name = $job_name
656	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
657	# @ output = $remote_dayfile
658	# @ error = $remote_dayfile
659	$wall_clock_limit
660	$image_size
661	$class
662	$environment
663	$network_to_use
664	$data_limit
665	$rset
666	$mcm_affinity_options
667	$task_affinity
668	$notification
669	$notify_user
670
671	%%END%%
672
673	if (( nodes > 0 ))
674	then
675
676	if [[ $remote_host != ibmkisti ]]
677	then
678
679	cat >> $job_to_send << %%END%%
680	# @ node = $nodes
681	# @ tasks_per_node = $processes_per_node
682	# @ node_usage = $node_usage
683	# @ queue
684
685	%%END%%
686
687	else
688
689	cat >> $job_to_send << %%END%%
690	# @ total_tasks = $mpi_tasks
691	# @ blocking = unlimited
692	# @ queue
693
694	%%END%%
695
696	fi
697
698	else
699
700	if [[ $remote_host != ibmy ]]
701	then
702
703	cat >> $job_to_send << %%END%%
704	# @ blocking = unlimited
705	# @ total_tasks = $numprocs
706	# @ node_usage = $node_usage
707	# @ queue
708
709	%%END%%
710
711	else
712
713	cat >> $job_to_send << %%END%%
714	# @ node = 1
715	# @ total_tasks = $numprocs
716	# @ queue
717
718	%%END%%
719
720	fi
721
722	fi
723
724	# workaround because of silly job filter on ibmkisti
725	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
726	then
727	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
728	fi
729
730	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
731	then
732
733	cat > $job_to_send << %%END%%
734	#!/bin/ksh
735
736	# @ job_type = serial
737	# @ node_usage = $node_usage
738	# @ job_name = palm
739	# @ wall_clock_limit = ${timestring},$timestring
740	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
741	# @ output = $remote_dayfile
742	# @ error = $remote_dayfile
743	$class
744	$notification
745
746	# @ queue
747
748	%%END%%
749
750	elif [[ $remote_host = lcfimm ]]
751	then
752
753	if [[ $numprocs != 0 ]]
754	then
755	cat > $job_to_send << %%END%%
756	#!/bin/ksh
757	#PBS -N $job_name
758	#PBS -A $project_account
759	#PBS -l walltime=$timestring
760	#PBS -l nodes=${nodes}:ppn=$processes_per_node
761	#PBS -l pmem=${memory}mb
762	#PBS -m abe
763	#PBS -o $remote_dayfile
764	#PBS -j oe
765	mpd &
766
767	%%END%%
768
769	else
770	cat > $job_to_send << %%END%%
771	#!/bin/ksh
772	#PBS -N $job_name
773	#PBS -A $project_account
774	#PBS -l walltime=$timestring
775	#PBS -l ncpus=1
776	#PBS -l pmem=${memory}mb
777	#PBS -m abe
778	#PBS -o $remote_dayfile
779	#PBS -j oe
780
781	%%END%%
782
783	fi
784
785	elif [[ $remote_host = lcflow ]]
786	then
787	if [ $memory -gt 1900 ]; then
788	use_himem=""
789	else
790	use_himem="#"
791	fi
792	if [[ $numprocs != 0 ]]
793	then
794	pe_set="#$ -pe impi $numprocs"
795	else
796	pe_set="#$ -pe impi 1"
797	fi
798	if [[ $queue = default ]]
799	then
800	queue_set=""
801	else
802	queue_set="#$ -q $queue"
803	fi
804	[[ "$disc_space" = "" ]] && disc_space=50
805
806	cat > $job_to_send << %%END%%
807	#!/bin/bash
808	#$ -S /bin/bash
809	#$ -N $job_name
810	#$ -cwd
811	#$ -l h_rt=$timestring
812	#$ -l h_vmem=${memory}M
813	#$ -o $remote_dayfile
814	#$ -j y
815	$pe_set
816	#$ -R y
817	${use_himem}#$ -l highmem=true
818	#$ -l h_fsize=${disc_space}G
819	$queue_set
820
821	%%END%%
822
823	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
824	then
825
826	if [[ $numprocs != 0 ]]
827	then
828	cat > $job_to_send << %%END%%
829	#!/bin/ksh
830	#PBS -N $job_name
831	#PBS -l walltime=$timestring
832	#PBS -l ncpus=$numprocs
833	#PBS -l pmem=${memory}mb
834	#PBS -o $remote_dayfile
835	#PBS -l nodes=$nodes:ppn=${processes_per_node}
836	#PBS -j oe
837
838	mpd &
839
840	%%END%%
841
842	else
843	cat > $job_to_send << %%END%%
844	#!/bin/ksh
845	#PBS -N $job_name
846	#PBS -l walltime=$timestring
847	#PBS -l ncpus=1
848	#PBS -l pmem=${memory}mb
849	#PBS -o $remote_dayfile
850	#PBS -j oe
851
852	%%END%%
853
854	fi
855
856	elif [[ $remote_host = lckiaps ]]
857	then
858
859	if [[ $numprocs != 0 ]]
860	then
861	cat > $job_to_send << %%END%%
862	#!/bin/ksh
863	#PBS -N $job_name
864	#PBS -l walltime=$timestring
865	#PBS -l select=1:ncpus=$numprocs
866	#PBS -l pmem=${memory}mb
867	#PBS -q $queue
868	#PBS -o $remote_dayfile
869	#PBS -j oe
870	#PBS -V
871
872	%%END%%
873
874	else
875	cat > $job_to_send << %%END%%
876	#!/bin/ksh
877	#PBS -N $job_name
878	#PBS -l walltime=$timestring
879	#PBS -l ncpus=1
880	#PBS -l pmem=${memory}mb
881	#PBS -o $remote_dayfile
882	#PBS -j oe
883
884	%%END%%
885
886	fi
887
888	elif [[ $remote_host = lcyon ]]
889	then
890
891	if [[ $numprocs != 0 ]]
892	then
893	cat > $job_to_send << %%END%%
894	#!/bin/ksh
895	#PBS -N $job_name
896	#PBS -l walltime=$timestring
897	#PBS -l ncpus=$numprocs
898	#PBS -l pmem=${memory}mb
899	#PBS -o $remote_dayfile
900	#PBS -j oe
901
902	%%END%%
903
904	else
905	cat > $job_to_send << %%END%%
906	#!/bin/ksh
907	#PBS -N $job_name
908	#PBS -l walltime=$timestring
909	#PBS -l ncpus=1
910	#PBS -l pmem=${memory}mb
911	#PBS -o $remote_dayfile
912	#PBS -j oe
913
914	%%END%%
915
916	fi
917
918	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
919	then
920
921	if [[ "$sgi_feature" != "" ]]
922	then
923	feature_directive="#PBS -l feature=$sgi_feature"
924	else
925	feature_directive=""
926	fi
927
928	if [[ $queue = dataq \|\| $queue = permq ]]
929	then
930	feature_directive="#PBS -l feature=data"
931	fi
932
933	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
934	then
935	queue_directive="#PBS -q $queue"
936	else
937	queue_directive=""
938	fi
939
940	if [[ $email_notification = none ]]
941	then
942	email_directive=""
943	else
944	email_directive="#PBS -M $email_notification"
945	fi
946
947	if [[ $numprocs != 0 ]]
948	then
949	cat > $job_to_send << %%END%%
950	#!/bin/bash
951	#PBS -N $job_name
952	#PBS -l walltime=$timestring
953	#PBS -l nodes=$nodes:ppn=${processes_per_node}
954	#PBS -l naccesspolicy=$node_usage
955	#PBS -o $remote_dayfile
956	#PBS -j oe
957	$feature_directive
958	$queue_directive
959	$email_directive
960
961	eval \`/sw/swdist/bin/modulesinit\`
962	#. /usr/share/modules/init/bash
963	$init_cmds
964	$module_calls
965
966	echo ld_library_path=\$LD_LIBRARY_PATH
967
968	%%END%%
969
970	else
971	cat > $job_to_send << %%END%%
972	#PBS -S /bin/bash
973	#PBS -N $job_name
974	#PBS -l walltime=$timestring
975	#PBS -l ncpus=1
976	#PBS -o $remote_dayfile
977	#PBS -j oe
978	$queue_directive
979	$email_directive
980
981	eval \`/sw/swdist/bin/modulesinit\`
982	#. /usr/share/modules/init/bash
983	$init_cmds
984	$module_calls
985
986	%%END%%
987
988	fi
989
990	elif [[ $remote_host = lcxe6 ]]
991	then
992
993	if [[ $numprocs != 0 ]]
994	then
995	cat > $job_to_send << %%END%%
996	#!/bin/ksh
997	#PBS -S /bin/ksh
998	#PBS -N $job_name
999	#PBS -A $project_account
1000	#PBS -j oe
1001	#PBS -l walltime=$timestring
1002	#PBS -l mppwidth=${numprocs}
1003	#PBS -l mppnppn=${processes_per_node}
1004	#PBS -m abe
1005	#PBS -o $remote_dayfile
1006	$email_directive
1007
1008	$init_cmds
1009	$module_calls
1010
1011	%%END%%
1012
1013	else
1014	cat > $job_to_send << %%END%%
1015	#!/bin/ksh
1016	#PBS -S /bin/ksh
1017	#PBS -N $job_name
1018	#PBS -A $project_account
1019	#PBS -j oe
1020	#PBS -l walltime=$timestring
1021	#PBS -l ncpus=1
1022	#PBS -l pmem=${memory}mb
1023	#PBS -m abe
1024	$email_directive
1025	#PBS -o $remote_dayfile
1026
1027	$init_cmds
1028	$module_calls
1029
1030	%%END%%
1031
1032	fi
1033
1034	elif [[ $remote_host = lckyoto ]]
1035	then
1036
1037	cat > $job_to_send << %%END%%
1038	#!/bin/ksh
1039	# @\$-o $remote_dayfile
1040	# @\$-eo -oi
1041	# @\$-lP 16
1042	# @\$-lp 1
1043	# @\$-lm 28gb -llm unlimited -ls unlimited
1044	# @\$-q $queue
1045	# @\$-Pvn abs_pack
1046	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1047	#. /thin/local/etc/setprofile/intel-11.0.sh
1048	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1049	. ~/.myprofile
1050	#. /home2/t/t51254/palm/current_version/myprofile
1051	#. /thin/apps/pgi/mpi.sh
1052	#
1053	env
1054	#
1055	set -x
1056
1057	%%END%%
1058
1059	elif [[ $remote_host = lcxt5m ]]
1060	then
1061
1062	if [[ $numprocs != 0 ]]
1063	then
1064	cat > $job_to_send << %%END%%
1065	#!/bin/ksh
1066	#PBS -S /bin/ksh
1067	#PBS -N $job_name
1068	#PBS -j oe
1069	#PBS -l walltime=$timestring
1070	#PBS -l mppwidth=${numprocs}
1071	#PBS -l mppnppn=${processes_per_node}
1072	#PBS -m abe
1073	#PBS -o $remote_dayfile
1074
1075	$init_cmds
1076	$module_calls
1077
1078	%%END%%
1079
1080	else
1081	cat > $job_to_send << %%END%%
1082	#!/bin/ksh
1083	#PBS -S /bin/ksh
1084	#PBS -N $job_name
1085	#PBS -j oe
1086	#PBS -l walltime=$timestring
1087	#PBS -l ncpus=1
1088	#PBS -l pmem=${memory}mb
1089	#PBS -m abe
1090	#PBS -o $remote_dayfile
1091
1092	$init_cmds
1093	$module_calls
1094
1095	%%END%%
1096
1097	fi
1098
1099	elif [[ $remote_host = lckyuh ]]
1100	then
1101	cat > $job_to_send << %%END%%
1102	#!/bin/bash
1103	#PJM -L "rscgrp=$queue"
1104	#PJM -L "node=$nodes"
1105	#PJM --mpi "proc=$numprocs"
1106	#PJM -L "elapse=$timestring"
1107	#PJM -o $remote_dayfile
1108	#PJM -j
1109	#PJM -X
1110	#PJM --no-stging
1111
1112	export LANG=en_US.UTF-8
1113	%%END%%
1114
1115	elif [[ $remote_host = lckyut ]]
1116	then
1117	cat > $job_to_send << %%END%%
1118	#!/bin/bash
1119	#PJM -L "rscgrp=$queue"
1120	#PJM -L "vnode=$numprocs"
1121	#PJM -L "vnode-core=1"
1122	#PJM -L "elapse=$timestring"
1123	#PJM --mpi proc=$numprocs
1124	#PJM -o $remote_dayfile
1125	#PJM -j
1126	#PJM -X
1127	#PJM --no-stging
1128
1129	export LANG=en_US.UTF-8
1130	%%END%%
1131
1132	elif [[ $remote_host = nech ]]
1133	then
1134
1135	if (( nodes > 1 ))
1136	then
1137	cat > $job_to_send << %%END%%
1138	#!/bin/ksh
1139	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1140	#PBS -l ${qsubmem}=${Memory}gb
1141	#PBS -b $nodes
1142	#PBS -o $remote_dayfile
1143	#PBS -N palm
1144	#PBS -j o
1145	#PBS -T mpisx
1146
1147	%%END%%
1148
1149	elif [[ $numprocs != 0 ]]
1150	then
1151	cat > $job_to_send << %%END%%
1152	#!/bin/ksh
1153	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1154	#PBS -l ${qsubmem}=${Memory}gb
1155	#PBS -o $remote_dayfile
1156	#PBS -N palm
1157	#PBS -j o
1158
1159	%%END%%
1160
1161	else
1162	cat > $job_to_send << %%END%%
1163	#!/bin/ksh
1164	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1165	#PBS -o $remote_dayfile
1166	#PBS -j o
1167
1168	%%END%%
1169
1170	fi
1171
1172	elif [[ $remote_host = necriam ]]
1173	then
1174
1175	if (( nodes > 1 ))
1176	then
1177	cat > $job_to_send << %%END%%
1178	#!/bin/ksh
1179	#PBS -b $nodes
1180	#PBS -o $remote_dayfile
1181	#PBS -N $job_name
1182	#PBS -j o
1183	#PBS -v MPIPROGINV=YES
1184
1185	%%END%%
1186
1187	elif [[ $numprocs != 0 ]]
1188	then
1189	cat > $job_to_send << %%END%%
1190	#!/bin/ksh
1191	#PBS -q ${queue}
1192	#PBS -o $remote_dayfile
1193	#PBS -N $job_name
1194	#PBS -j o
1195	#PBS -v MPIPROGINV=YES
1196
1197	%%END%%
1198
1199	else
1200	cat > $job_to_send << %%END%%
1201	#!/bin/ksh
1202	#PBS -o $remote_dayfile
1203	#PBS -j o
1204
1205	%%END%%
1206
1207	fi
1208
1209	elif [[ $remote_host = lctit ]]
1210	then
1211	cat > $job_to_send << %%END%%
1212	#!/bin/ksh
1213	$init_cmds
1214	$module_calls
1215
1216	%%END%%
1217
1218	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1219	if [[ $tasks_per_node != $processes_per_node ]]
1220	then
1221	submcom="$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 $queue "
1222	else
1223	submcom="$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 "
1224	fi
1225
1226	else
1227
1228	cat > $job_to_send << %%END%%
1229	# @\$-q ${queue}
1230	# @\$-l${qsubtime} $timestring
1231	# @\$-l${qsubmem} ${memory}mb
1232	# @\$-o $remote_dayfile
1233	# @\$-eo
1234
1235	%%END%%
1236
1237	fi
1238
1239
1240	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1241	# BEI EXIT VERANLASSEN
1242	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1243	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1244	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1245	then
1246	echo "set +vx" >> $job_to_send
1247	echo "trap '" >> $job_to_send
1248	echo "set +vx" >> $job_to_send
1249	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* ]]
1250	then
1251	if [[ $remote_host = ibmh ]]
1252	then
1253	return_queue=c1
1254	elif [[ $remote_host = ibmkisti ]]
1255	then
1256	return_queue=class.1-2
1257	elif [[ $remote_host = ibmku ]]
1258	then
1259	return_queue=sdbg2
1260	elif [[ $remote_host = ibms ]]
1261	then
1262	return_queue=p_normal
1263	elif [[ $remote_host = ibmy ]]
1264	then
1265	return_queue=serial
1266	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1267	then
1268	return_queue=serialq
1269	elif [[ $remote_host = necriam ]]
1270	then
1271	return_queue=SP
1272	elif [[ $remote_host = lckiaps ]]
1273	then
1274	return_queue=express
1275	elif [[ $remote_host = lckyuh ]]
1276	then
1277	return_queue=cx-single
1278	elif [[ $remote_host = lckyut ]]
1279	then
1280	return_queue=cx-single
1281	else
1282	return_queue=unknown
1283	fi
1284
1285	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1286	then
1287
1288	if [[ $remote_host = ibmku ]]
1289	then
1290	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1291	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1292	else
1293	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1294	fi
1295	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1296	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1297	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1298	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1299	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1300	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1301	if [[ $host != "ibmh" ]]
1302	then
1303	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1304	fi
1305	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1306	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1307
1308	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1309	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1310
1311	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1312	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
1313	if [[ $remote_host = ibmku ]]
1314	then
1315	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1316	fi
1317	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1318
1319	elif [[ $remote_host = nech ]]
1320	then
1321	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1322	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1323	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1324	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1325	echo "#PBS -j o" >> $job_to_send
1326	echo " " >> $job_to_send
1327	echo "set -x" >> $job_to_send
1328	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1329	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1330	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1331	echo "%%END%%" >> $job_to_send
1332
1333	elif [[ $remote_host = necriam ]]
1334	then
1335	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1336	echo "#PBS -q $return_queue" >> $job_to_send
1337	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1338	echo "#PBS -j o" >> $job_to_send
1339	echo " " >> $job_to_send
1340	echo "set -x" >> $job_to_send
1341	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1342	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1343	echo "%%END%%" >> $job_to_send
1344
1345	elif [[ $remote_host = lckyuh ]]
1346	then
1347	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1348	echo "#!/bin/bash" >> $job_to_send
1349	echo "#PJM -L \"node=1\"" >> $job_to_send
1350	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1351	echo "#PJM --no-stging" >> $job_to_send
1352	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1353	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1354	echo "#PJM -j" >> $job_to_send
1355	echo " " >> $job_to_send
1356	echo "export LANG=en_US.UTF-8" >> $job_to_send
1357	echo "set -x" >> $job_to_send
1358	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1359	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1360	echo "%%END%%" >> $job_to_send
1361
1362	elif [[ $remote_host = lckyut ]]
1363	then
1364	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1365	echo "#!/bin/bash" >> $job_to_send
1366	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1367	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1368	echo "#PJM --no-stging" >> $job_to_send
1369	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1370	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1371	echo "#PJM -j" >> $job_to_send
1372	echo " " >> $job_to_send
1373	echo "export LANG=en_US.UTF-8" >> $job_to_send
1374	echo "set -x" >> $job_to_send
1375	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1376	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1377	echo "%%END%%" >> $job_to_send
1378
1379	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1380	then
1381	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1382	echo "#!/bin/bash" >> $job_to_send
1383	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1384	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1385	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1386	echo "#PBS -l feature=data" >> $job_to_send
1387	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1388	echo "#PBS -j oe" >> $job_to_send
1389	echo " " >> $job_to_send
1390	echo ". /usr/share/modules/init/bash" >> $job_to_send
1391	echo "set -x" >> $job_to_send
1392	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1393	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1394	echo "%%END%%" >> $job_to_send
1395
1396	elif [[ $remote_host = lcflow ]]
1397	then
1398	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1399	echo "#!/bin/bash" >> $job_to_send
1400	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1401	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1402	echo "SGEPREFIX -cwd" >> $job_to_send
1403	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1404	echo "SGEPREFIX -l h_vmem=100M" >> $job_to_send
1405	echo "SGEPREFIX -j y" >> $job_to_send
1406	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1407	echo " " >> $job_to_send
1408	echo "set -x" >> $job_to_send
1409	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1410	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1411	echo "" >> $job_to_send
1412	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1413	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1414	echo "rm -f scpjob.${kennung}" >> $job_to_send
1415	echo "%%END%%" >> $job_to_send
1416	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1417	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1418
1419	else
1420
1421	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1422	echo "# @\\\$-q $return_queue" >> $job_to_send
1423	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1424	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1425	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1426	then
1427	echo "# @\$-l mpp_p=0" >> $job_to_send
1428	fi
1429	echo '# @\$-lF 10mb' >> $job_to_send
1430	# echo '# @\$-o /dev/null' >> $job_to_send
1431	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1432	echo '# @\\\$-eo' >> $job_to_send
1433	echo " " >> $job_to_send
1434	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1435	then
1436	echo "set +vx" >> $job_to_send
1437	echo ". .profile" >> $job_to_send
1438	fi
1439	echo "set -x" >> $job_to_send
1440	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
1441	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1442	echo "%%END%%" >> $job_to_send
1443	fi
1444
1445	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1446	then
1447	echo "llsubmit scpjob.$kennung" >> $job_to_send
1448	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1449	then
1450	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1451	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1452	echo "msub scpjob.$kennung" >> $job_to_send
1453	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1454	then
1455	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1456	elif [[ $remote_host = t3es ]]
1457	then
1458	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1459	elif [[ $remote_host = lckiaps ]]
1460	then
1461	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1462	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1463	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1464	elif [[ $remote_host = lckyu* ]]
1465	then
1466	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1467	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1468	elif [[ $remote_host = lcflow ]]
1469	then
1470	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1471	echo "ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1472	else
1473	echo "$submcom scpjob.$kennung" >> $job_to_send
1474	fi
1475	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1476	then
1477	echo "rm scpjob.$kennung" >> $job_to_send
1478	fi
1479	if [[ $remote_host = nech ]]
1480	then
1481	echo "cd -" >> $job_to_send
1482	fi
1483	else
1484	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1485	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1486	fi
1487	echo "set -x" >> $job_to_send
1488	echo " ' exit" >> $job_to_send
1489	echo "set -x" >> $job_to_send
1490	fi
1491
1492
1493
1494	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1495	cat $file_to_send >> $job_to_send
1496
1497	if [[ $remote_host = ibm ]]
1498	then
1499	echo " " >> $job_to_send
1500	echo "exit" >> $job_to_send
1501	fi
1502
1503	# remove job file
1504	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1505	then
1506	echo " " >> $job_to_send
1507	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1508	fi
1509
1510
1511
1512	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1513	if [[ -z $remote_user ]]
1514	then
1515	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1516	then
1517	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1518	fi
1519	fi
1520
1521
1522
1523	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1524	if [[ $no_submit = false ]]
1525	then
1526	if [[ $remote_host != $local_host ]]
1527	then
1528	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1529	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1530	then
1531	job_catalog_save=$job_catalog
1532	job_catalog=job_queue
1533	elif [[ $remote_host = nech ]]
1534	then
1535	job_catalog_save=$job_catalog
1536	job_catalog=/hpf/b/${remote_user}/job_queue
1537	fi
1538	if [[ $remote_host = nech ]]
1539	then
1540	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1541	# TRANSFERIERT WERDEN
1542	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1543	else
1544	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1545	fi
1546	if [[ $? = 1 ]]
1547	then
1548	locat=scp; exit
1549	fi
1550	if [[ $remote_host = ibms ]]
1551	then
1552	job_catalog=$job_catalog_save
1553	fi
1554	[[ $verify = true ]] && printf "\n >>> finished\n"
1555	else
1556	eval job_catalog=$job_catalog
1557	cp $job_to_send ${job_catalog}/$job_on_remhost
1558	fi
1559
1560
1561
1562	# NQS- BZW. LOADLEVELER-JOB STARTEN
1563	if [[ $remote_host != $local_host ]]
1564	then
1565	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1566
1567	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1568	then
1569	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1570	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1571	elif [[ $remote_host = ibmku ]]
1572	then
1573	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1574	elif [[ $remote_host = lcflow ]]
1575	then
1576	ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1577	else
1578	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1579	fi
1580
1581	[[ $verify = true ]] && printf " >>> o.k.\n"
1582	else
1583	cd $job_catalog
1584	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1585	then
1586	eval $submcom $job_on_remhost
1587	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1588	then
1589	chmod u+x $job_on_remhost
1590	eval $submcom $job_on_remhost
1591	elif [[ $local_host = nech ]]
1592	then
1593	if [[ $queue = default ]]
1594	then
1595	eval $submcom $job_on_remhost
1596	else
1597	eval $submcom -q $queue $job_on_remhost
1598	fi
1599	else
1600	qsub $job_on_remhost
1601	fi
1602
1603	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1604	# only at the end of the job.
1605	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1606	then
1607	rm $job_on_remhost
1608	fi
1609	cd - > /dev/null
1610	fi
1611	fi
1612
1613
1614
1615	# ABSCHLUSSARBEITEN
1616
1617	if [[ $no_submit = false ]]
1618	then
1619	rm -f $job_to_send
1620	fi
1621	[[ $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 |