Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1203

Last change on this file since 1203 was 1203, checked in by witha, 11 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 55.6 KB

Line
1	#!/bin/ksh
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2012 Leibniz University Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1203 2013-07-10 16:35:03Z witha $
30	#
31	# 1202 2013-07-10 16:22:07Z witha
32	# adjustments for Forwind cluster (lcflow)
33	#
34	# 1199 2013-07-05 14:52:22Z raasch
35	# adjustments for CSC Helsinki (lccrayf)
36	#
37	# use of cluster/express queue enabled (ibmh)
38	# vinessa added (imuk)
39	#
40	# 1103 2013-02-20 02:15:53Z raasch
41	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
42	# instead typeset -L/R),
43	# further adjustments for lckyuh
44	#
45	# 2013-02-10 01:47:43Z raasch
46	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
47	# and for Forwind cluster (lcflow)
48	#
49	# 1094 2013-02-03 01:52:12Z raasch
50	# new option -P for explicit setting of ssh/scp port,
51	# decalpha parts (yonsei) removed
52	#
53	# 2013-02-02 07:06:13Z raasch
54	# adjustments for Kyushu-University computing center (lckyut - tatara)
55	# old changelog messages removed
56	#
57	# 1046 2012-11-09 14:38:45Z maronga
58	# code put under GPL (PALM 3.9)
59	#
60	# 08/07/94 - Siggi - first version finished
61	# 29/06/94 - Siggi - script development started
62	#--------------------------------------------------------------------------------#
63	# subjob - script for automatic generation and submission of batch-job files
64	# for various batch queuing systems
65	#--------------------------------------------------------------------------------#
66
67
68	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
69	delete_dayfile=false
70	email_notification=none
71	group_number=none
72	locat=normal
73	no_default_queue=none
74	no_submit=false
75	job_catalog="~/job_queue"
76	job_name=none
77	local_user=$LOGNAME
78	node_usage=shared
79	numprocs=0
80	punkte="..........................................................."
81	submcom=qsub
82	queue=default
83	remote_host=none
84	remote_user=""
85	verify=true
86
87	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
88	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
89
90
91
92	# FEHLERBEHANDLUNG
93	# BEI EXIT:
94	trap 'if [[ $locat != normal ]]
95	then
96	case $locat in
97	(option) printf "\n --> available optios can be displayed"
98	printf " by typing:"
99	printf "\n \"subjob ?\" \n";;
100	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
101	(*) printf "\n +++ unknown error"
102	printf "\n please inform S. Raasch!\n"
103	esac
104	[[ -f $job_to_send ]] && rm $job_to_send
105	printf "\n\n+++ SUBJOB killed \n\n"
106	fi' exit
107
108
109	# BEI TERMINAL-BREAK:
110	trap '[[ -f $job_to_send ]] && rm $job_to_send
111	printf "\n\n+++ SUBJOB killed \n\n"
112	exit
113	' 2
114
115
116
117
118	# LOKALEN HOSTNAMEN FESTSTELLEN
119	local_host=$(hostname)
120
121
122
123	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
124	# UEBERHAUPT ZULAESSIG IST
125	# Note: One of the entries for "lck" or "lckordi" always has to be
126	# comment out, because the hostname (node*) is same for both machines
127	case $local_host in
128	(atmos) local_addres=172.20.25.35; local_host=lcide;;
129	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
130	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
131	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
132	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
133	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
134	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
135	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
136	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
137	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
138	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
139	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
140	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
141	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
142	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
143	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
144	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
145	(cs*) local_addres=136.172.44.131; local_host=nech;;
146	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
147	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
148	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
149	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
150	(node*) local_addres=165.132.26.61 local_host=lck;;
151	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
152	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
153	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
154	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
155	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
156	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
157	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
158	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
159	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
160	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
161	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
162	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
163	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
164	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
165	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
166	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
167	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
168	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
169	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
170	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
171	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
172	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
173	(meller) local_addres=134.106.74.155; local_host=lcfor;;
174	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
175	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
176	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
177	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
178	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
179	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
180	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
181	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
182	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
183	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
184	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
185	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
186	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
187	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
188	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
189	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
190	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
191	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
192	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
193	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
194	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
195	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
196	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
197	(*) printf "\n +++ \"$local_host\" unknown";
198	printf "\n please inform S. Raasch!";
199	locat=parameter; exit;;
200	esac
201
202
203
204	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
205	remote_host=$local_host
206
207
208
209
210	# PROZEDUROPTIONEN EINLESEN
211	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
212	do
213	case $option in
214	(c) job_catalog=$OPTARG;;
215	(d) delete_dayfile=true;;
216	(D) no_submit=true;;
217	(e) email_notification=$OPTARG;;
218	(g) group_number=$OPTARG;;
219	(h) remote_host=$OPTARG;;
220	(m) memory=$OPTARG;;
221	(n) job_name=$OPTARG;;
222	(N) node_usage=$OPTARG;;
223	(O) threads_per_task=$OPTARG;;
224	(P) scp_port=$OPTARG;;
225	(q) no_default_queue=$OPTARG;;
226	(t) cputime=$OPTARG;;
227	(T) tasks_per_node=$OPTARG;;
228	(u) remote_user=$OPTARG;;
229	(v) verify=false;;
230	(X) numprocs=$OPTARG;;
231	(\?) printf "\n +++ Option $OPTARG unknown \n";
232	locat=option; exit;;
233	esac
234	done
235
236
237	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
238	(( to_shift = $OPTIND - 1 ))
239	shift $to_shift; file_to_send=$1
240
241
242	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
243	if [ "$1" = "?" ]
244	then
245	(printf "\n *** subjob can be called as follows:\n"
246	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
247	printf "\n Description of available options:\n"
248	printf "\n Option Description Default-Value"
249	printf "\n -c job-input- and output-catalog ~/job_queue"
250	printf "\n -d no job-protocol will be created ---"
251	printf "\n -D only the job-file will be created ---"
252	printf "\n -h execution host, available hosts: $remote_host"
253	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
254	printf "\n lckiaps, lctit, nech, necriam, unics"
255	printf "\n -m memory demand per process in MByte ---"
256	printf "\n -n jobname <jobdatei>"
257	printf "\n -O threads per task (for OpenMP usage) 1"
258	printf "\n -P ssh/scp port default port"
259	printf "\n -q job-queue to be used default"
260	printf "\n -t allowed cpu-time in seconds ---"
261	printf "\n -T tasks per node (on parallel hosts) ---"
262	printf "\n -u username on execution host from .netrc"
263	printf "\n -v no prompt for confirmation ---"
264	printf "\n -X # of processors (on parallel hosts) 1"
265	printf "\n "
266	printf "\n The only possible positional parameter is <jobfile>:"
267	printf "\n The complete NQS-job must be provided here."
268	printf "\n <jobfile>=? creates this outline\n\n") \| more
269	exit
270	fi
271
272
273
274	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
275	if [[ "$file_to_send" = "" ]]
276	then
277	printf "\n +++ job-file missing"
278	locat=parameter; exit
279	else
280	if [[ -f $file_to_send ]]
281	then
282	true
283	else
284	printf "\n +++ job-file: "
285	printf "\n $file_to_send"
286	printf "\n does not exist"
287	locat=parameter; exit
288	fi
289	fi
290
291
292
293	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
294	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
295	if [[ $job_name = none ]]
296	then
297	job_name=$file_to_send
298	fi
299	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
300	then
301	printf "\n +++ job-file name: "
302	printf "\n $job_name"
303	printf "\n must not contain \"/\"-characters"
304	locat=parameter; exit
305	fi
306
307
308
309
310	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
311	# ODER WENN HOST NICHT ANGEGEBEN WURDE
312	if [[ $remote_host = none ]]
313	then
314	printf "\n +++ host missing"
315	locat=option; exit
316	else
317	case $remote_host in
318	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
319	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
320	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
321	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
322	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
323	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
324	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
325	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
326	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
327	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
328	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
329	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
330	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
331	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
332	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
333	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
334	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
335	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
336	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
337	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
338	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
339	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
340	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
341	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
342	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
343	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
344	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
345	locat=parameter; exit;;
346	esac
347	fi
348
349
350	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
351	if [[ $no_default_queue != none ]]
352	then
353	error=false
354	ndq=$no_default_queue
355	case $remote_host in
356	(ibm) case $ndq in
357	(p690_express\|p690_standard\|p690_long) error=false;;
358	(*) error=true;;
359	esac;;
360	(ibmh) case $ndq in
361	(cluster\|express) error=false;;
362	(*) error=true;;
363	esac;;
364	(ibmkisti) case $ndq in
365	(class.32plus\|class.1-2\|class.2-32) error=false;;
366	(*) error=true;;
367	esac;;
368	(ibmku) case $ndq in
369	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
370	(*) error=true;;
371	esac;;
372	(ibms) case $ndq in
373	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
374	(*) error=true;;
375	esac;;
376	(ibmy) case $ndq in
377	(parallel) error=false;;
378	(*) error=true;;
379	esac;;
380	(lccrayf) case $ndq in
381	(usup\|test*\|small\|large) error=false;;
382	(*) error=true;;
383	esac;;
384	(lcflow) case $ndq in
385	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
386	(*) error=true;;
387	esac;;
388	(lckiaps) case $ndq in
389	(express\|normal) error=false;;
390	(*) error=true;;
391	esac;;
392	(lckyoto) case $ndq in
393	(eh\|ph) error=false;;
394	(*) error=true;;
395	esac;;
396	(lckyuh) case $ndq in
397	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
398	(*) error=true;;
399	esac;;
400	(lckyut) case $ndq in
401	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
402	(*) error=true;;
403	esac;;
404	(lcsgib\|lcsgih) case $ndq in
405	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
406	(*) error=true;;
407	esac;;
408	(lctit) case $ndq in
409	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
410	(*) error=true;;
411	esac;;
412	(t3eb) case $ndq in
413	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
414	(*) error=true;;
415	esac;;
416	(necriam) case $ndq in
417	(SP\|SS\|P6) error=false;;
418	(*) error=true;;
419	esac;;
420	(t3eh) case $ndq in
421	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
422	(*) error=true;;
423	esac;;
424	(t3ej2\|t3ej5) case $ndq in
425	(low\|normal\|high) error=false;;
426	(*) error=true;;
427	esac;;
428	(t3es) case $ndq in
429	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
430	(*) error=true;;
431	esac;;
432	(unics) case $ndq in
433	(unics\|ht) error=false;;
434	(*) error=true;;
435	esac;;
436	esac
437	if [[ $error = true ]]
438	then
439	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
440	locat=parameter; exit
441	else
442	queue=$no_default_queue
443	fi
444	fi
445
446
447
448	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
449	# AUFTEILEN
450	done=false
451	while [[ $done = false ]]
452	do
453	if (( $cputime <= 0 ))
454	then
455	printf "\n +++ wrong cpu-time or cpu-time missing"
456	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
457	printf "\n >>> "
458	read cputime 1>/dev/null 2>&1
459	else
460	done=true
461	fi
462	done
463	if [[ $remote_host = nech ]]
464	then
465	if (( tasks_per_node != 0 ))
466	then
467	(( cputime = cputime * tasks_per_node ))
468	elif [[ $numprocs != 0 ]]
469	then
470	(( cputime = cputime * numprocs ))
471	fi
472	fi
473	(( stunden = cputime / 3600 ))
474	(( resttime = cputime - stunden * 3600 ))
475	(( minuten = resttime / 60 ))
476	(( sekunden = resttime - minuten * 60 ))
477	timestring=${stunden}:${minuten}:${sekunden}
478
479
480
481	# PRUEFEN DER KERNSPEICHERANFORDERUNG
482	done=false
483	while [[ $done = false ]]
484	do
485	if (( memory <= 0 ))
486	then
487	printf "\n +++ wrong memory demand or memory demand missing"
488	printf "\n >>> Please type memory in MByte per process as INTEGER:"
489	printf "\n >>> "
490	read memory 1>/dev/null 2>&1
491	else
492	done=true
493	fi
494	done
495
496	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
497	then
498	if (( tasks_per_node != 0 ))
499	then
500	(( Memory = memory * tasks_per_node / 1000 ))
501	elif [[ $numprocs != 0 ]]
502	then
503	(( Memory = memory * numprocs / 1000 ))
504	else
505	(( Memory = memory / 1000 ))
506	fi
507	elif [[ $remote_host = lctit ]]
508	then
509	(( Memory = memory * tasks_per_node / 1000 ))
510	fi
511
512
513	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
514	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
515	then
516	(( memory = memory * threads_per_task ))
517	fi
518
519
520	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
521	if (( tasks_per_node != 0 ))
522	then
523	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
524	fi
525
526
527	# Calculate number of processes per node
528	(( processes_per_node = tasks_per_node * threads_per_task ))
529
530
531	# Calculate number of MPI tasks
532	(( mpi_tasks = numprocs / threads_per_task ))
533
534
535	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
536	if [[ "$scp_port" != "" ]]
537	then
538	PORTOPT="-P $scp_port"
539	SSH_PORTOPT="-p $scp_port"
540	fi
541
542
543	# HEADER-AUSGABE
544	if [[ $verify = true ]]
545	then
546	printf "\n\n"
547	printf "#--------------------------------------------------------------# \n"
548	spalte1=SUBJOB;spalte2=$(date)
549	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
550	printf "\| \| \n"
551	printf "\| values of parameters/options: \| \n"
552	spalte1=$(echo local_host$punkte \| cut -c-20)
553	spalte2=$punkte$local_host
554	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
555	spalte1=$(echo remote_host$punkte \| cut -c-20)
556	spalte2=$punkte$remote_host
557	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
558	spalte1=$(echo queue$punkte \| cut -c-20)
559	spalte2=$punkte$queue
560	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
561	spalte1=$(echo memory$punkte \| cut -c-20)
562	spalte2="$punkte$memory mb"
563	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
564	spalte1=$(echo cputime$punkte \| cut -c-20)
565	spalte2="$punkte$cputime sec"
566	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
567	spalte1=$(echo job_name$punkte \| cut -c-20)
568	spalte2="$punkte$job_name"
569	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
570	printf "#--------------------------------------------------------------# \n\n"
571
572
573	# KONTROLLABFRAGE, OB ALLES O.K.
574	antwort="dummy"
575	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
576	do
577	read antwort?" >>> continue (y/n) ? "
578	done
579	if [[ $antwort = n \|\| $antwort = N ]]
580	then
581	locat=verify; exit
582	fi
583	printf "\n"
584	fi
585
586
587
588	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
589	kennung=$RANDOM
590	job_on_remhost=${job_name}_${kennung}_$local_host
591	job_to_send=job_to_send_$kennung
592	if [[ $delete_dayfile = false ]]
593	then
594	remote_dayfile=${local_host}_${job_name}_result_$kennung
595	local_dayfile=${remote_host}_${job_name}
596	else
597	remote_dayfile=/dev/null
598	fi
599
600
601
602	# Generate the batch job scripts (qsub/msub/LoadLeveler)
603	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
604	then
605
606	# General LoadLeveler settings
607	execute_in_shell="#!/bin/ksh"
608	use_shell="# @ shell = /bin/ksh"
609	consumable_memory="ConsumableMemory($memory mb)"
610	class="# @ class = $queue"
611	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
612	network_to_use="# @ network.mpi = sn_all,shared,us"
613	data_limit="# @ data_limit = 1.76gb"
614	image_size="# @ image_size = 50"
615	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
616
617	if [[ $email_notification = none ]]
618	then
619	notify_user=""
620	else
621	notify_user="# @ notify_user = $email_notification"
622	if [[ $delete_dayfile = true ]]
623	then
624	notification='# @ notification = never'
625	fi
626	fi
627
628	if [[ $remote_host = ibmh ]]
629	then
630	data_limit=""
631	network_to_use=""
632	class="# @ class = $queue"
633	environment=""
634	rset="# @ rset = RSET_MCM_AFFINITY"
635	task_affinity="# @ task_affinity = core(1)"
636	elif [[ $remote_host = ibmkisti ]]
637	then
638	network_to_use="# @ network.MPI = sn_all,shared,US"
639	wall_clock_limit="# @ wall_clock_limit = $timestring"
640	if [[ $threads_per_task = 1 ]]
641	then
642	rset="# @ rset = RSET_MCM_AFFINITY"
643	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
644	fi
645	environment=""
646	use_shell=""
647	data_limit=""
648	image_size=""
649	elif [[ $remote_host = ibmku ]]
650	then
651	execute_in_shell="#!/usr/bin/ksh"
652	use_shell="# @ shell = /usr/bin/ksh"
653	consumable_memory=""
654	environment=""
655	network_to_use="# @ network.mpi = sn_all,shared,us"
656	data_limit=""
657	image_size=""
658	elif [[ $remote_host = ibms ]]
659	then
660	network_to_use="# @ network.mpi = csss,shared,us"
661	elif [[ $remote_host = ibmy ]]
662	then
663	consumable_memory=""
664	network_to_use=""
665	fi
666
667	cat > $job_to_send << %%END%%
668	$execute_in_shell
669	$use_shell
670
671	# @ job_type = parallel
672	# @ job_name = $job_name
673	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
674	# @ output = $remote_dayfile
675	# @ error = $remote_dayfile
676	$wall_clock_limit
677	$image_size
678	$class
679	$environment
680	$network_to_use
681	$data_limit
682	$rset
683	$mcm_affinity_options
684	$task_affinity
685	$notification
686	$notify_user
687
688	%%END%%
689
690	if (( nodes > 0 ))
691	then
692
693	if [[ $remote_host != ibmkisti ]]
694	then
695
696	cat >> $job_to_send << %%END%%
697	# @ node = $nodes
698	# @ tasks_per_node = $processes_per_node
699	# @ node_usage = $node_usage
700	# @ queue
701
702	%%END%%
703
704	else
705
706	cat >> $job_to_send << %%END%%
707	# @ total_tasks = $mpi_tasks
708	# @ blocking = unlimited
709	# @ queue
710
711	%%END%%
712
713	fi
714
715	else
716
717	if [[ $remote_host != ibmy ]]
718	then
719
720	cat >> $job_to_send << %%END%%
721	# @ blocking = unlimited
722	# @ total_tasks = $numprocs
723	# @ node_usage = $node_usage
724	# @ queue
725
726	%%END%%
727
728	else
729
730	cat >> $job_to_send << %%END%%
731	# @ node = 1
732	# @ total_tasks = $numprocs
733	# @ queue
734
735	%%END%%
736
737	fi
738
739	fi
740
741	# workaround because of silly job filter on ibmkisti
742	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
743	then
744	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
745	fi
746
747	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
748	then
749
750	cat > $job_to_send << %%END%%
751	#!/bin/ksh
752
753	# @ job_type = serial
754	# @ node_usage = $node_usage
755	# @ job_name = palm
756	# @ wall_clock_limit = ${timestring},$timestring
757	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
758	# @ output = $remote_dayfile
759	# @ error = $remote_dayfile
760	$class
761	$notification
762
763	# @ queue
764
765	%%END%%
766
767	elif [[ $remote_host = lccrayf ]]
768	then
769
770	if [[ $numprocs != 0 ]]
771	then
772	cat > $job_to_send << %%END%%
773	#!/bin/bash -l
774	#SBATCH -J $job_name
775	#SBATCH -t $timestring
776	#SBATCH -N $nodes
777	#SBATCH --ntasks-per-node=$processes_per_node
778	#SBATCH -p $queue
779	#SBATCH -o $remote_dayfile
780	#SBATCH -e $remote_dayfile
781
782	$init_cmds
783	$module_calls
784
785	%%END%%
786
787	else
788	cat > $job_to_send << %%END%%
789	#!/bin/bash -l
790	#SBATCH -J $job_name
791	#SBATCH -t $timestring
792	#SBATCH -l ncpus=1
793	#SBATCH -l pmem=${memory}mb
794	#SBATCH -m abe
795	#SBATCH -o $remote_dayfile
796	#SBATCH -e $remote_dayfile
797
798	$init_cmds
799	$module_calls
800
801	%%END%%
802
803	fi
804
805	elif [[ $remote_host = lcfimm ]]
806	then
807
808	if [[ $numprocs != 0 ]]
809	then
810	cat > $job_to_send << %%END%%
811	#!/bin/ksh
812	#PBS -N $job_name
813	#PBS -A $project_account
814	#PBS -l walltime=$timestring
815	#PBS -l nodes=${nodes}:ppn=$processes_per_node
816	#PBS -l pmem=${memory}mb
817	#PBS -m abe
818	#PBS -o $remote_dayfile
819	#PBS -j oe
820	mpd &
821
822	%%END%%
823
824	else
825	cat > $job_to_send << %%END%%
826	#!/bin/ksh
827	#PBS -N $job_name
828	#PBS -A $project_account
829	#PBS -l walltime=$timestring
830	#PBS -l ncpus=1
831	#PBS -l pmem=${memory}mb
832	#PBS -m abe
833	#PBS -o $remote_dayfile
834	#PBS -j oe
835
836	%%END%%
837
838	fi
839
840	elif [[ $remote_host = lcflow ]]
841	then
842	if [ $memory -gt 1800 ]; then
843	use_himem=""
844	else
845	use_himem="#"
846	fi
847	if [[ $numprocs != 0 ]]
848	then
849	pe_set="#$ -pe impi41 $numprocs"
850	else
851	pe_set="#$ -pe impi41 1"
852	fi
853	if [[ $queue = default ]]
854	then
855	queue_set=""
856	else
857	queue_set="#$ -q $queue"
858	fi
859	[[ "$disc_space" = "" ]] && disc_space=50
860
861	cat > $job_to_send << %%END%%
862	#!/bin/bash
863	#$ -S /bin/bash
864	#$ -N $job_name
865	#$ -cwd
866	#$ -l h_rt=$timestring
867	#$ -l h_vmem=${memory}M
868	#$ -o $remote_dayfile
869	#$ -j y
870	$pe_set
871	#$ -R y
872	${use_himem}#$ -l highmem=true
873	#$ -l h_fsize=${disc_space}G
874	$queue_set
875
876	%%END%%
877
878	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
879	then
880
881	if [[ $numprocs != 0 ]]
882	then
883	cat > $job_to_send << %%END%%
884	#!/bin/ksh
885	#PBS -N $job_name
886	#PBS -l walltime=$timestring
887	#PBS -l ncpus=$numprocs
888	#PBS -l pmem=${memory}mb
889	#PBS -o $remote_dayfile
890	#PBS -l nodes=$nodes:ppn=${processes_per_node}
891	#PBS -j oe
892
893	mpd &
894
895	%%END%%
896
897	else
898	cat > $job_to_send << %%END%%
899	#!/bin/ksh
900	#PBS -N $job_name
901	#PBS -l walltime=$timestring
902	#PBS -l ncpus=1
903	#PBS -l pmem=${memory}mb
904	#PBS -o $remote_dayfile
905	#PBS -j oe
906
907	%%END%%
908
909	fi
910
911	elif [[ $remote_host = lckiaps ]]
912	then
913
914	if [[ $numprocs != 0 ]]
915	then
916	cat > $job_to_send << %%END%%
917	#!/bin/ksh
918	#PBS -N $job_name
919	#PBS -l walltime=$timestring
920	#PBS -l select=1:ncpus=$numprocs
921	#PBS -l pmem=${memory}mb
922	#PBS -q $queue
923	#PBS -o $remote_dayfile
924	#PBS -j oe
925	#PBS -V
926
927	%%END%%
928
929	else
930	cat > $job_to_send << %%END%%
931	#!/bin/ksh
932	#PBS -N $job_name
933	#PBS -l walltime=$timestring
934	#PBS -l ncpus=1
935	#PBS -l pmem=${memory}mb
936	#PBS -o $remote_dayfile
937	#PBS -j oe
938
939	%%END%%
940
941	fi
942
943	elif [[ $remote_host = lcyon ]]
944	then
945
946	if [[ $numprocs != 0 ]]
947	then
948	cat > $job_to_send << %%END%%
949	#!/bin/ksh
950	#PBS -N $job_name
951	#PBS -l walltime=$timestring
952	#PBS -l ncpus=$numprocs
953	#PBS -l pmem=${memory}mb
954	#PBS -o $remote_dayfile
955	#PBS -j oe
956
957	%%END%%
958
959	else
960	cat > $job_to_send << %%END%%
961	#!/bin/ksh
962	#PBS -N $job_name
963	#PBS -l walltime=$timestring
964	#PBS -l ncpus=1
965	#PBS -l pmem=${memory}mb
966	#PBS -o $remote_dayfile
967	#PBS -j oe
968
969	%%END%%
970
971	fi
972
973	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
974	then
975
976	if [[ "$sgi_feature" != "" ]]
977	then
978	feature_directive="#PBS -l feature=$sgi_feature"
979	else
980	feature_directive=""
981	fi
982
983	if [[ $queue = dataq \|\| $queue = permq ]]
984	then
985	feature_directive="#PBS -l feature=data"
986	fi
987
988	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
989	then
990	queue_directive="#PBS -q $queue"
991	else
992	queue_directive=""
993	fi
994
995	if [[ $email_notification = none ]]
996	then
997	email_directive=""
998	else
999	email_directive="#PBS -M $email_notification"
1000	fi
1001
1002	if [[ $numprocs != 0 ]]
1003	then
1004	cat > $job_to_send << %%END%%
1005	#!/bin/bash
1006	#PBS -N $job_name
1007	#PBS -l walltime=$timestring
1008	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1009	#PBS -l naccesspolicy=$node_usage
1010	#PBS -o $remote_dayfile
1011	#PBS -j oe
1012	$feature_directive
1013	$queue_directive
1014	$email_directive
1015
1016	eval \`/sw/swdist/bin/modulesinit\`
1017	#. /usr/share/modules/init/bash
1018	$init_cmds
1019	$module_calls
1020
1021	echo ld_library_path=\$LD_LIBRARY_PATH
1022
1023	%%END%%
1024
1025	else
1026	cat > $job_to_send << %%END%%
1027	#PBS -S /bin/bash
1028	#PBS -N $job_name
1029	#PBS -l walltime=$timestring
1030	#PBS -l ncpus=1
1031	#PBS -o $remote_dayfile
1032	#PBS -j oe
1033	$queue_directive
1034	$email_directive
1035
1036	eval \`/sw/swdist/bin/modulesinit\`
1037	#. /usr/share/modules/init/bash
1038	$init_cmds
1039	$module_calls
1040
1041	%%END%%
1042
1043	fi
1044
1045	elif [[ $remote_host = lcxe6 ]]
1046	then
1047
1048	if [[ $numprocs != 0 ]]
1049	then
1050	cat > $job_to_send << %%END%%
1051	#!/bin/ksh
1052	#PBS -S /bin/ksh
1053	#PBS -N $job_name
1054	#PBS -A $project_account
1055	#PBS -j oe
1056	#PBS -l walltime=$timestring
1057	#PBS -l mppwidth=${numprocs}
1058	#PBS -l mppnppn=${processes_per_node}
1059	#PBS -m abe
1060	#PBS -o $remote_dayfile
1061	$email_directive
1062
1063	$init_cmds
1064	$module_calls
1065
1066	%%END%%
1067
1068	else
1069	cat > $job_to_send << %%END%%
1070	#!/bin/ksh
1071	#PBS -S /bin/ksh
1072	#PBS -N $job_name
1073	#PBS -A $project_account
1074	#PBS -j oe
1075	#PBS -l walltime=$timestring
1076	#PBS -l ncpus=1
1077	#PBS -l pmem=${memory}mb
1078	#PBS -m abe
1079	$email_directive
1080	#PBS -o $remote_dayfile
1081
1082	$init_cmds
1083	$module_calls
1084
1085	%%END%%
1086
1087	fi
1088
1089	elif [[ $remote_host = lckyoto ]]
1090	then
1091
1092	cat > $job_to_send << %%END%%
1093	#!/bin/ksh
1094	# @\$-o $remote_dayfile
1095	# @\$-eo -oi
1096	# @\$-lP 16
1097	# @\$-lp 1
1098	# @\$-lm 28gb -llm unlimited -ls unlimited
1099	# @\$-q $queue
1100	# @\$-Pvn abs_pack
1101	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1102	#. /thin/local/etc/setprofile/intel-11.0.sh
1103	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1104	. ~/.myprofile
1105	#. /home2/t/t51254/palm/current_version/myprofile
1106	#. /thin/apps/pgi/mpi.sh
1107	#
1108	env
1109	#
1110	set -x
1111
1112	%%END%%
1113
1114	elif [[ $remote_host = lcxt5m ]]
1115	then
1116
1117	if [[ $numprocs != 0 ]]
1118	then
1119	cat > $job_to_send << %%END%%
1120	#!/bin/ksh
1121	#PBS -S /bin/ksh
1122	#PBS -N $job_name
1123	#PBS -j oe
1124	#PBS -l walltime=$timestring
1125	#PBS -l mppwidth=${numprocs}
1126	#PBS -l mppnppn=${processes_per_node}
1127	#PBS -m abe
1128	#PBS -o $remote_dayfile
1129
1130	$init_cmds
1131	$module_calls
1132
1133	%%END%%
1134
1135	else
1136	cat > $job_to_send << %%END%%
1137	#!/bin/ksh
1138	#PBS -S /bin/ksh
1139	#PBS -N $job_name
1140	#PBS -j oe
1141	#PBS -l walltime=$timestring
1142	#PBS -l ncpus=1
1143	#PBS -l pmem=${memory}mb
1144	#PBS -m abe
1145	#PBS -o $remote_dayfile
1146
1147	$init_cmds
1148	$module_calls
1149
1150	%%END%%
1151
1152	fi
1153
1154	elif [[ $remote_host = lckyuh ]]
1155	then
1156	cat > $job_to_send << %%END%%
1157	#!/bin/bash
1158	#PJM -L "rscgrp=$queue"
1159	#PJM -L "node=$nodes"
1160	#PJM --mpi "proc=$numprocs"
1161	#PJM -L "elapse=$timestring"
1162	#PJM -o $remote_dayfile
1163	#PJM -j
1164	#PJM -X
1165	#PJM --no-stging
1166
1167	export LANG=en_US.UTF-8
1168	%%END%%
1169
1170	elif [[ $remote_host = lckyut ]]
1171	then
1172	cat > $job_to_send << %%END%%
1173	#!/bin/bash
1174	#PJM -L "rscgrp=$queue"
1175	#PJM -L "vnode=$numprocs"
1176	#PJM -L "vnode-core=1"
1177	#PJM -L "elapse=$timestring"
1178	#PJM --mpi proc=$numprocs
1179	#PJM -o $remote_dayfile
1180	#PJM -j
1181	#PJM -X
1182	#PJM --no-stging
1183
1184	export LANG=en_US.UTF-8
1185	%%END%%
1186
1187	elif [[ $remote_host = nech ]]
1188	then
1189
1190	if (( nodes > 1 ))
1191	then
1192	cat > $job_to_send << %%END%%
1193	#!/bin/ksh
1194	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1195	#PBS -l ${qsubmem}=${Memory}gb
1196	#PBS -b $nodes
1197	#PBS -o $remote_dayfile
1198	#PBS -N palm
1199	#PBS -j o
1200	#PBS -T mpisx
1201
1202	%%END%%
1203
1204	elif [[ $numprocs != 0 ]]
1205	then
1206	cat > $job_to_send << %%END%%
1207	#!/bin/ksh
1208	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1209	#PBS -l ${qsubmem}=${Memory}gb
1210	#PBS -o $remote_dayfile
1211	#PBS -N palm
1212	#PBS -j o
1213
1214	%%END%%
1215
1216	else
1217	cat > $job_to_send << %%END%%
1218	#!/bin/ksh
1219	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1220	#PBS -o $remote_dayfile
1221	#PBS -j o
1222
1223	%%END%%
1224
1225	fi
1226
1227	elif [[ $remote_host = necriam ]]
1228	then
1229
1230	if (( nodes > 1 ))
1231	then
1232	cat > $job_to_send << %%END%%
1233	#!/bin/ksh
1234	#PBS -b $nodes
1235	#PBS -o $remote_dayfile
1236	#PBS -N $job_name
1237	#PBS -j o
1238	#PBS -v MPIPROGINV=YES
1239
1240	%%END%%
1241
1242	elif [[ $numprocs != 0 ]]
1243	then
1244	cat > $job_to_send << %%END%%
1245	#!/bin/ksh
1246	#PBS -q ${queue}
1247	#PBS -o $remote_dayfile
1248	#PBS -N $job_name
1249	#PBS -j o
1250	#PBS -v MPIPROGINV=YES
1251
1252	%%END%%
1253
1254	else
1255	cat > $job_to_send << %%END%%
1256	#!/bin/ksh
1257	#PBS -o $remote_dayfile
1258	#PBS -j o
1259
1260	%%END%%
1261
1262	fi
1263
1264	elif [[ $remote_host = lctit ]]
1265	then
1266	cat > $job_to_send << %%END%%
1267	#!/bin/ksh
1268	$init_cmds
1269	$module_calls
1270
1271	%%END%%
1272
1273	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1274	if [[ $tasks_per_node != $processes_per_node ]]
1275	then
1276	submcom="$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 $queue "
1277	else
1278	submcom="$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 "
1279	fi
1280
1281	else
1282
1283	cat > $job_to_send << %%END%%
1284	# @\$-q ${queue}
1285	# @\$-l${qsubtime} $timestring
1286	# @\$-l${qsubmem} ${memory}mb
1287	# @\$-o $remote_dayfile
1288	# @\$-eo
1289
1290	%%END%%
1291
1292	fi
1293
1294
1295	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1296	# BEI EXIT VERANLASSEN
1297	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1298	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1299	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1300	then
1301	echo "set +vx" >> $job_to_send
1302	echo "trap '" >> $job_to_send
1303	echo "set +vx" >> $job_to_send
1304	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* ]]
1305	then
1306	if [[ $remote_host = ibmh ]]
1307	then
1308	return_queue=c1
1309	elif [[ $remote_host = ibmkisti ]]
1310	then
1311	return_queue=class.1-2
1312	elif [[ $remote_host = ibmku ]]
1313	then
1314	return_queue=sdbg2
1315	elif [[ $remote_host = ibms ]]
1316	then
1317	return_queue=p_normal
1318	elif [[ $remote_host = ibmy ]]
1319	then
1320	return_queue=serial
1321	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1322	then
1323	return_queue=serialq
1324	elif [[ $remote_host = necriam ]]
1325	then
1326	return_queue=SP
1327	elif [[ $remote_host = lckiaps ]]
1328	then
1329	return_queue=express
1330	elif [[ $remote_host = lckyuh ]]
1331	then
1332	return_queue=cx-single
1333	elif [[ $remote_host = lckyut ]]
1334	then
1335	return_queue=cx-single
1336	else
1337	return_queue=unknown
1338	fi
1339
1340	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1341	then
1342
1343	if [[ $remote_host = ibmku ]]
1344	then
1345	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1346	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1347	else
1348	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1349	fi
1350	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1351	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1352	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1353	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1354	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1355	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1356	if [[ $host != "ibmh" ]]
1357	then
1358	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1359	fi
1360	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1361	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1362
1363	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1364	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1365
1366	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1367	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
1368	if [[ $remote_host = ibmku ]]
1369	then
1370	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1371	fi
1372	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1373
1374	elif [[ $remote_host = nech ]]
1375	then
1376	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1377	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1378	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1379	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1380	echo "#PBS -j o" >> $job_to_send
1381	echo " " >> $job_to_send
1382	echo "set -x" >> $job_to_send
1383	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1384	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1385	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1386	echo "%%END%%" >> $job_to_send
1387
1388	elif [[ $remote_host = necriam ]]
1389	then
1390	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1391	echo "#PBS -q $return_queue" >> $job_to_send
1392	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1393	echo "#PBS -j o" >> $job_to_send
1394	echo " " >> $job_to_send
1395	echo "set -x" >> $job_to_send
1396	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1397	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1398	echo "%%END%%" >> $job_to_send
1399
1400	elif [[ $remote_host = lckyuh ]]
1401	then
1402	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1403	echo "#!/bin/bash" >> $job_to_send
1404	echo "#PJM -L \"node=1\"" >> $job_to_send
1405	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1406	echo "#PJM --no-stging" >> $job_to_send
1407	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1408	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1409	echo "#PJM -j" >> $job_to_send
1410	echo " " >> $job_to_send
1411	echo "export LANG=en_US.UTF-8" >> $job_to_send
1412	echo "set -x" >> $job_to_send
1413	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1414	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1415	echo "%%END%%" >> $job_to_send
1416
1417	elif [[ $remote_host = lckyut ]]
1418	then
1419	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1420	echo "#!/bin/bash" >> $job_to_send
1421	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1422	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1423	echo "#PJM --no-stging" >> $job_to_send
1424	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1425	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1426	echo "#PJM -j" >> $job_to_send
1427	echo " " >> $job_to_send
1428	echo "export LANG=en_US.UTF-8" >> $job_to_send
1429	echo "set -x" >> $job_to_send
1430	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1431	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1432	echo "%%END%%" >> $job_to_send
1433
1434	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1435	then
1436	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1437	echo "#!/bin/bash" >> $job_to_send
1438	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1439	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1440	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1441	echo "#PBS -l feature=data" >> $job_to_send
1442	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1443	echo "#PBS -j oe" >> $job_to_send
1444	echo " " >> $job_to_send
1445	echo ". /usr/share/modules/init/bash" >> $job_to_send
1446	echo "set -x" >> $job_to_send
1447	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1448	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1449	echo "%%END%%" >> $job_to_send
1450
1451	elif [[ $remote_host = lcflow ]]
1452	then
1453	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1454	echo "#!/bin/bash" >> $job_to_send
1455	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1456	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1457	echo "SGEPREFIX -cwd" >> $job_to_send
1458	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1459	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1460	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1461	echo "SGEPREFIX -j y" >> $job_to_send
1462	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1463	echo " " >> $job_to_send
1464	echo "set -x" >> $job_to_send
1465	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1466	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1467	echo "" >> $job_to_send
1468	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1469	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1470	echo "rm -f scpjob.${kennung}" >> $job_to_send
1471	echo "%%END%%" >> $job_to_send
1472	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1473	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1474
1475	else
1476
1477	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1478	echo "# @\\\$-q $return_queue" >> $job_to_send
1479	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1480	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1481	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1482	then
1483	echo "# @\$-l mpp_p=0" >> $job_to_send
1484	fi
1485	echo '# @\$-lF 10mb' >> $job_to_send
1486	# echo '# @\$-o /dev/null' >> $job_to_send
1487	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1488	echo '# @\\\$-eo' >> $job_to_send
1489	echo " " >> $job_to_send
1490	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1491	then
1492	echo "set +vx" >> $job_to_send
1493	echo ". .profile" >> $job_to_send
1494	fi
1495	echo "set -x" >> $job_to_send
1496	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1497	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1498	echo "%%END%%" >> $job_to_send
1499	fi
1500
1501	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1502	then
1503	echo "llsubmit scpjob.$kennung" >> $job_to_send
1504	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1505	then
1506	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1507	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1508	echo "msub scpjob.$kennung" >> $job_to_send
1509	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1510	then
1511	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1512	elif [[ $remote_host = t3es ]]
1513	then
1514	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1515	elif [[ $remote_host = lckiaps ]]
1516	then
1517	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1518	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1519	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1520	elif [[ $remote_host = lckyu* ]]
1521	then
1522	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1523	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1524	elif [[ $remote_host = lcflow ]]
1525	then
1526	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1527	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1528	else
1529	echo "$submcom scpjob.$kennung" >> $job_to_send
1530	fi
1531	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1532	then
1533	echo "rm scpjob.$kennung" >> $job_to_send
1534	fi
1535	if [[ $remote_host = nech ]]
1536	then
1537	echo "cd -" >> $job_to_send
1538	fi
1539	else
1540	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1541	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1542	fi
1543	echo "set -x" >> $job_to_send
1544	echo " ' exit" >> $job_to_send
1545	echo "set -x" >> $job_to_send
1546	fi
1547
1548
1549
1550	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1551	cat $file_to_send >> $job_to_send
1552
1553	if [[ $remote_host = ibm ]]
1554	then
1555	echo " " >> $job_to_send
1556	echo "exit" >> $job_to_send
1557	fi
1558
1559	# remove job file
1560	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1561	then
1562	echo " " >> $job_to_send
1563	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1564	fi
1565
1566
1567
1568	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1569	if [[ -z $remote_user ]]
1570	then
1571	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1572	then
1573	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1574	fi
1575	fi
1576
1577
1578
1579	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1580	if [[ $no_submit = false ]]
1581	then
1582	if [[ $remote_host != $local_host ]]
1583	then
1584	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1585	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1586	then
1587	job_catalog_save=$job_catalog
1588	job_catalog=job_queue
1589	elif [[ $remote_host = nech ]]
1590	then
1591	job_catalog_save=$job_catalog
1592	job_catalog=/hpf/b/${remote_user}/job_queue
1593	fi
1594	if [[ $remote_host = nech ]]
1595	then
1596	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1597	# TRANSFERIERT WERDEN
1598	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1599	else
1600	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1601	fi
1602	if [[ $? = 1 ]]
1603	then
1604	locat=scp; exit
1605	fi
1606	if [[ $remote_host = ibms ]]
1607	then
1608	job_catalog=$job_catalog_save
1609	fi
1610	[[ $verify = true ]] && printf "\n >>> finished\n"
1611	else
1612	eval job_catalog=$job_catalog
1613	cp $job_to_send ${job_catalog}/$job_on_remhost
1614	fi
1615
1616
1617
1618	# NQS- BZW. LOADLEVELER-JOB STARTEN
1619	if [[ $remote_host != $local_host ]]
1620	then
1621	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1622
1623	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1624	then
1625	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1626	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1627	elif [[ $remote_host = ibmku ]]
1628	then
1629	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1630	elif [[ $remote_host = lcflow ]]
1631	then
1632	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1633	else
1634	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1635	fi
1636
1637	[[ $verify = true ]] && printf " >>> o.k.\n"
1638	else
1639	cd $job_catalog
1640	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $local_host = lccrayf ]]
1641	then
1642	eval $submcom $job_on_remhost
1643	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1644	then
1645	chmod u+x $job_on_remhost
1646	eval $submcom $job_on_remhost
1647	elif [[ $local_host = nech ]]
1648	then
1649	if [[ $queue = default ]]
1650	then
1651	eval $submcom $job_on_remhost
1652	else
1653	eval $submcom -q $queue $job_on_remhost
1654	fi
1655	else
1656	qsub $job_on_remhost
1657	fi
1658
1659	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1660	# only at the end of the job.
1661	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1662	then
1663	rm $job_on_remhost
1664	fi
1665	cd - > /dev/null
1666	fi
1667	fi
1668
1669
1670
1671	# ABSCHLUSSARBEITEN
1672
1673	if [[ $no_submit = false ]]
1674	then
1675	rm -f $job_to_send
1676	fi
1677	[[ $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 |