Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1119

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