Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1224

Last change on this file since 1224 was 1224, checked in by raasch, 11 years ago
First adjustments for HLRN-III system (lccrayb). Dummy interface added to avoid compiler warnings.
Property svn:keywords set to `Id Rev`
File size: 56.6 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	# first adjustments for lccrayb
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1224 2013-09-16 07:27:23Z raasch $
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	(ambiel-lx) local_addres=134.106.74.48; local_host=lcfor;;
129	(atmos) local_addres=172.20.25.35; local_host=lcide;;
130	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
131	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
132	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
133	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
134	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
135	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
136	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
137	(b04*) local_addres=133.5.4.33; local_host=lckyuh;;
138	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
139	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
140	(blogin*) local_addres=130.73.233.1; local_host=lccrayb;;
141	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
142	(buran) local_addres=130.75.105.58; local_host=lcmuk;;
143	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
144	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
145	(clogin*) local_addres=86.50.166.21; local_host=lccrayf;;
146	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
147	(cs*) local_addres=136.172.44.131; local_host=nech;;
148	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
149	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
150	(flow01) local_addres=10.141.255.71; local_host=lcflow;;
151	(flow02) local_addres=10.141.255.72; local_host=lcflow;;
152	(node*) local_addres=165.132.26.61 local_host=lck;;
153	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
154	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
155	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
156	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
157	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
158	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
159	(hayaka*) local_addres=133.5.4.33; local_host=lckyuh;;
160	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
161	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
162	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
163	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
164	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
165	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
166	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
167	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
168	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
169	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
170	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
171	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
172	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
173	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
174	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
175	(meller) local_addres=134.106.74.155; local_host=lcfor;;
176	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
177	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
178	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
179	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
180	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
181	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
182	(pcj*) local_addres=172.31.120.1; local_host=lckyut;;
183	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
184	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
185	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
186	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
187	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
188	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
189	(shiokaze-lx) local_addres=134.106.74.123; local_host=lcfor;;
190	(sisu-login*) local_addres=86.50.166.21; local_host=lccrayf;;
191	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
192	(sugoka*) local_addres=172.31.120.1; local_host=lckyut;;
193	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
194	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
195	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
196	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
197	(vinessa) local_addres=130.75.105.112; local_host=lcmuk;;
198	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
199	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
200	(*) printf "\n +++ \"$local_host\" unknown";
201	printf "\n please inform S. Raasch!";
202	locat=parameter; exit;;
203	esac
204
205
206
207	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
208	remote_host=$local_host
209
210
211
212
213	# PROZEDUROPTIONEN EINLESEN
214	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
215	do
216	case $option in
217	(c) job_catalog=$OPTARG;;
218	(d) delete_dayfile=true;;
219	(D) no_submit=true;;
220	(e) email_notification=$OPTARG;;
221	(g) group_number=$OPTARG;;
222	(h) remote_host=$OPTARG;;
223	(m) memory=$OPTARG;;
224	(n) job_name=$OPTARG;;
225	(N) node_usage=$OPTARG;;
226	(O) threads_per_task=$OPTARG;;
227	(P) scp_port=$OPTARG;;
228	(q) no_default_queue=$OPTARG;;
229	(t) cputime=$OPTARG;;
230	(T) tasks_per_node=$OPTARG;;
231	(u) remote_user=$OPTARG;;
232	(v) verify=false;;
233	(X) numprocs=$OPTARG;;
234	(\?) printf "\n +++ Option $OPTARG unknown \n";
235	locat=option; exit;;
236	esac
237	done
238
239
240	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
241	(( to_shift = $OPTIND - 1 ))
242	shift $to_shift; file_to_send=$1
243
244
245	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
246	if [ "$1" = "?" ]
247	then
248	(printf "\n *** subjob can be called as follows:\n"
249	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
250	printf "\n Description of available options:\n"
251	printf "\n Option Description Default-Value"
252	printf "\n -c job-input- and output-catalog ~/job_queue"
253	printf "\n -d no job-protocol will be created ---"
254	printf "\n -D only the job-file will be created ---"
255	printf "\n -h execution host, available hosts: $remote_host"
256	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
257	printf "\n lckiaps, lctit, nech, necriam, unics"
258	printf "\n -m memory demand per process in MByte ---"
259	printf "\n -n jobname <jobdatei>"
260	printf "\n -O threads per task (for OpenMP usage) 1"
261	printf "\n -P ssh/scp port default port"
262	printf "\n -q job-queue to be used default"
263	printf "\n -t allowed cpu-time in seconds ---"
264	printf "\n -T tasks per node (on parallel hosts) ---"
265	printf "\n -u username on execution host from .netrc"
266	printf "\n -v no prompt for confirmation ---"
267	printf "\n -X # of processors (on parallel hosts) 1"
268	printf "\n "
269	printf "\n The only possible positional parameter is <jobfile>:"
270	printf "\n The complete NQS-job must be provided here."
271	printf "\n <jobfile>=? creates this outline\n\n") \| more
272	exit
273	fi
274
275
276
277	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
278	if [[ "$file_to_send" = "" ]]
279	then
280	printf "\n +++ job-file missing"
281	locat=parameter; exit
282	else
283	if [[ -f $file_to_send ]]
284	then
285	true
286	else
287	printf "\n +++ job-file: "
288	printf "\n $file_to_send"
289	printf "\n does not exist"
290	locat=parameter; exit
291	fi
292	fi
293
294
295
296	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
297	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
298	if [[ $job_name = none ]]
299	then
300	job_name=$file_to_send
301	fi
302	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
303	then
304	printf "\n +++ job-file name: "
305	printf "\n $job_name"
306	printf "\n must not contain \"/\"-characters"
307	locat=parameter; exit
308	fi
309
310
311
312
313	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
314	# ODER WENN HOST NICHT ANGEGEBEN WURDE
315	if [[ $remote_host = none ]]
316	then
317	printf "\n +++ host missing"
318	locat=option; exit
319	else
320	case $remote_host in
321	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
322	(ibmh) queue=cluster; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
323	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
324	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
325	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
327	(lccrayb) queue=testq; remote_addres=130.73.233.1; submcom=/opt/torque/4.2.4.1/bin/qsub;;
328	(lccrayf) queue=small; remote_addres=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
329	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
330	(lcflow) remote_addres=10.140.1.71; submcom=/cm/shared/apps/sge/6.2u5p2/bin/lx26-amd64/qsub;;
331	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
332	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
333	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
334	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
335	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
336	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
337	(lckyuh) remote_addres=133.5.4.33; submcom=/usr/bin/pjsub;;
338	(lckyut) remote_addres=133.5.4.37; submcom=/usr/bin/pjsub;;
339	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
340	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
341	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
342	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
343	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
344	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
345	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
346	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
347	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
348	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
349	locat=parameter; exit;;
350	esac
351	fi
352
353
354	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
355	if [[ $no_default_queue != none ]]
356	then
357	error=false
358	ndq=$no_default_queue
359	case $remote_host in
360	(ibm) case $ndq in
361	(p690_express\|p690_standard\|p690_long) error=false;;
362	(*) error=true;;
363	esac;;
364	(ibmh) case $ndq in
365	(cluster\|express) error=false;;
366	(*) error=true;;
367	esac;;
368	(ibmkisti) case $ndq in
369	(class.32plus\|class.1-2\|class.2-32) error=false;;
370	(*) error=true;;
371	esac;;
372	(ibmku) case $ndq in
373	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
374	(*) error=true;;
375	esac;;
376	(ibms) case $ndq in
377	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
378	(*) error=true;;
379	esac;;
380	(ibmy) case $ndq in
381	(parallel) error=false;;
382	(*) error=true;;
383	esac;;
384	(lccrayb) case $ndq in
385	(mpp1q\|mpp1testq\|testq) error=false;;
386	(*) error=true;;
387	esac;;
388	(lccrayf) case $ndq in
389	(usup\|test*\|small\|large) error=false;;
390	(*) error=true;;
391	esac;;
392	(lcflow) case $ndq in
393	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q) error=false;;
394	(*) error=true;;
395	esac;;
396	(lckiaps) case $ndq in
397	(express\|normal) error=false;;
398	(*) error=true;;
399	esac;;
400	(lckyoto) case $ndq in
401	(eh\|ph) error=false;;
402	(*) error=true;;
403	esac;;
404	(lckyuh) case $ndq in
405	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
406	(*) error=true;;
407	esac;;
408	(lckyut) case $ndq in
409	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
410	(*) error=true;;
411	esac;;
412	(lcsgib\|lcsgih) case $ndq in
413	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
414	(*) error=true;;
415	esac;;
416	(lctit) case $ndq in
417	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
418	(*) error=true;;
419	esac;;
420	(t3eb) case $ndq in
421	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
422	(*) error=true;;
423	esac;;
424	(necriam) case $ndq in
425	(SP\|SS\|P6) error=false;;
426	(*) error=true;;
427	esac;;
428	(t3eh) case $ndq in
429	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
430	(*) error=true;;
431	esac;;
432	(t3ej2\|t3ej5) case $ndq in
433	(low\|normal\|high) error=false;;
434	(*) error=true;;
435	esac;;
436	(t3es) case $ndq in
437	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
438	(*) error=true;;
439	esac;;
440	(unics) case $ndq in
441	(unics\|ht) error=false;;
442	(*) error=true;;
443	esac;;
444	esac
445	if [[ $error = true ]]
446	then
447	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
448	locat=parameter; exit
449	else
450	queue=$no_default_queue
451	fi
452	fi
453
454
455
456	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
457	# AUFTEILEN
458	done=false
459	while [[ $done = false ]]
460	do
461	if (( $cputime <= 0 ))
462	then
463	printf "\n +++ wrong cpu-time or cpu-time missing"
464	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
465	printf "\n >>> "
466	read cputime 1>/dev/null 2>&1
467	else
468	done=true
469	fi
470	done
471	if [[ $remote_host = nech ]]
472	then
473	if (( tasks_per_node != 0 ))
474	then
475	(( cputime = cputime * tasks_per_node ))
476	elif [[ $numprocs != 0 ]]
477	then
478	(( cputime = cputime * numprocs ))
479	fi
480	fi
481	(( stunden = cputime / 3600 ))
482	(( resttime = cputime - stunden * 3600 ))
483	(( minuten = resttime / 60 ))
484	(( sekunden = resttime - minuten * 60 ))
485	timestring=${stunden}:${minuten}:${sekunden}
486
487
488
489	# PRUEFEN DER KERNSPEICHERANFORDERUNG
490	done=false
491	while [[ $done = false ]]
492	do
493	if (( memory <= 0 ))
494	then
495	printf "\n +++ wrong memory demand or memory demand missing"
496	printf "\n >>> Please type memory in MByte per process as INTEGER:"
497	printf "\n >>> "
498	read memory 1>/dev/null 2>&1
499	else
500	done=true
501	fi
502	done
503
504	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
505	then
506	if (( tasks_per_node != 0 ))
507	then
508	(( Memory = memory * tasks_per_node / 1000 ))
509	elif [[ $numprocs != 0 ]]
510	then
511	(( Memory = memory * numprocs / 1000 ))
512	else
513	(( Memory = memory / 1000 ))
514	fi
515	elif [[ $remote_host = lctit ]]
516	then
517	(( Memory = memory * tasks_per_node / 1000 ))
518	fi
519
520
521	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
522	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
523	then
524	(( memory = memory * threads_per_task ))
525	fi
526
527
528	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
529	if (( tasks_per_node != 0 ))
530	then
531	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
532	fi
533
534
535	# Calculate number of processes per node
536	(( processes_per_node = tasks_per_node * threads_per_task ))
537
538
539	# Calculate number of MPI tasks
540	(( mpi_tasks = numprocs / threads_per_task ))
541
542
543	# Set port number option for calls of ssh/scp, subjob and batch_scp scripts
544	if [[ "$scp_port" != "" ]]
545	then
546	PORTOPT="-P $scp_port"
547	SSH_PORTOPT="-p $scp_port"
548	fi
549
550
551	# HEADER-AUSGABE
552	if [[ $verify = true ]]
553	then
554	printf "\n\n"
555	printf "#--------------------------------------------------------------# \n"
556	spalte1=SUBJOB;spalte2=$(date)
557	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
558	printf "\| \| \n"
559	printf "\| values of parameters/options: \| \n"
560	spalte1=$(echo local_host$punkte \| cut -c-20)
561	spalte2=$punkte$local_host
562	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
563	spalte1=$(echo remote_host$punkte \| cut -c-20)
564	spalte2=$punkte$remote_host
565	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
566	spalte1=$(echo queue$punkte \| cut -c-20)
567	spalte2=$punkte$queue
568	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
569	spalte1=$(echo memory$punkte \| cut -c-20)
570	spalte2="$punkte$memory mb"
571	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
572	spalte1=$(echo cputime$punkte \| cut -c-20)
573	spalte2="$punkte$cputime sec"
574	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
575	spalte1=$(echo job_name$punkte \| cut -c-20)
576	spalte2="$punkte$job_name"
577	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
578	printf "#--------------------------------------------------------------# \n\n"
579
580
581	# KONTROLLABFRAGE, OB ALLES O.K.
582	antwort="dummy"
583	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
584	do
585	read antwort?" >>> continue (y/n) ? "
586	done
587	if [[ $antwort = n \|\| $antwort = N ]]
588	then
589	locat=verify; exit
590	fi
591	printf "\n"
592	fi
593
594
595
596	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
597	kennung=$RANDOM
598	job_on_remhost=${job_name}_${kennung}_$local_host
599	job_to_send=job_to_send_$kennung
600	if [[ $delete_dayfile = false ]]
601	then
602	remote_dayfile=${local_host}_${job_name}_result_$kennung
603	local_dayfile=${remote_host}_${job_name}
604	else
605	remote_dayfile=/dev/null
606	fi
607
608
609
610	# Generate the batch job scripts (qsub/msub/LoadLeveler)
611	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
612	then
613
614	# General LoadLeveler settings
615	execute_in_shell="#!/bin/ksh"
616	use_shell="# @ shell = /bin/ksh"
617	consumable_memory="ConsumableMemory($memory mb)"
618	class="# @ class = $queue"
619	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
620	network_to_use="# @ network.mpi = sn_all,shared,us"
621	data_limit="# @ data_limit = 1.76gb"
622	image_size="# @ image_size = 50"
623	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
624
625	if [[ $email_notification = none ]]
626	then
627	notify_user=""
628	else
629	notify_user="# @ notify_user = $email_notification"
630	if [[ $delete_dayfile = true ]]
631	then
632	notification='# @ notification = never'
633	fi
634	fi
635
636	if [[ $remote_host = ibmh ]]
637	then
638	data_limit=""
639	network_to_use=""
640	class="# @ class = $queue"
641	environment=""
642	rset="# @ rset = RSET_MCM_AFFINITY"
643	task_affinity="# @ task_affinity = core(1)"
644	elif [[ $remote_host = ibmkisti ]]
645	then
646	network_to_use="# @ network.MPI = sn_all,shared,US"
647	wall_clock_limit="# @ wall_clock_limit = $timestring"
648	if [[ $threads_per_task = 1 ]]
649	then
650	rset="# @ rset = RSET_MCM_AFFINITY"
651	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
652	fi
653	environment=""
654	use_shell=""
655	data_limit=""
656	image_size=""
657	elif [[ $remote_host = ibmku ]]
658	then
659	execute_in_shell="#!/usr/bin/ksh"
660	use_shell="# @ shell = /usr/bin/ksh"
661	consumable_memory=""
662	environment=""
663	network_to_use="# @ network.mpi = sn_all,shared,us"
664	data_limit=""
665	image_size=""
666	elif [[ $remote_host = ibms ]]
667	then
668	network_to_use="# @ network.mpi = csss,shared,us"
669	elif [[ $remote_host = ibmy ]]
670	then
671	consumable_memory=""
672	network_to_use=""
673	fi
674
675	cat > $job_to_send << %%END%%
676	$execute_in_shell
677	$use_shell
678
679	# @ job_type = parallel
680	# @ job_name = $job_name
681	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
682	# @ output = $remote_dayfile
683	# @ error = $remote_dayfile
684	$wall_clock_limit
685	$image_size
686	$class
687	$environment
688	$network_to_use
689	$data_limit
690	$rset
691	$mcm_affinity_options
692	$task_affinity
693	$notification
694	$notify_user
695
696	%%END%%
697
698	if (( nodes > 0 ))
699	then
700
701	if [[ $remote_host != ibmkisti ]]
702	then
703
704	cat >> $job_to_send << %%END%%
705	# @ node = $nodes
706	# @ tasks_per_node = $processes_per_node
707	# @ node_usage = $node_usage
708	# @ queue
709
710	%%END%%
711
712	else
713
714	cat >> $job_to_send << %%END%%
715	# @ total_tasks = $mpi_tasks
716	# @ blocking = unlimited
717	# @ queue
718
719	%%END%%
720
721	fi
722
723	else
724
725	if [[ $remote_host != ibmy ]]
726	then
727
728	cat >> $job_to_send << %%END%%
729	# @ blocking = unlimited
730	# @ total_tasks = $numprocs
731	# @ node_usage = $node_usage
732	# @ queue
733
734	%%END%%
735
736	else
737
738	cat >> $job_to_send << %%END%%
739	# @ node = 1
740	# @ total_tasks = $numprocs
741	# @ queue
742
743	%%END%%
744
745	fi
746
747	fi
748
749	# workaround because of silly job filter on ibmkisti
750	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
751	then
752	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
753	fi
754
755	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
756	then
757
758	cat > $job_to_send << %%END%%
759	#!/bin/ksh
760
761	# @ job_type = serial
762	# @ node_usage = $node_usage
763	# @ job_name = palm
764	# @ wall_clock_limit = ${timestring},$timestring
765	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
766	# @ output = $remote_dayfile
767	# @ error = $remote_dayfile
768	$class
769	$notification
770
771	# @ queue
772
773	%%END%%
774
775	elif [[ $remote_host = lccrayb ]]
776	then
777
778	if [[ $numprocs != 0 ]]
779	then
780	cat > $job_to_send << %%END%%
781	#!/bin/bash --login
782	#PBS -N $job_name
783	#PBS -l walltime=$timestring
784	#PBS -l mppwidth=$numprocs
785	#PBS -l mppnppn=$processes_per_node
786	#PBS -o $remote_dayfile
787	#PBS -j oe
788	#PBS -q $queue
789
790	$module_calls
791	export ATP_ENABLED=0
792	export MPICH_GNI_BTE_MULTI_CHANNEL=disabled
793
794	%%END%%
795
796	else
797
798	continue
799
800	fi
801
802	elif [[ $remote_host = lccrayf ]]
803	then
804
805	if [[ $numprocs != 0 ]]
806	then
807	cat > $job_to_send << %%END%%
808	#!/bin/bash -l
809	#SBATCH -J $job_name
810	#SBATCH -t $timestring
811	#SBATCH -N $nodes
812	#SBATCH --ntasks-per-node=$processes_per_node
813	#SBATCH -p $queue
814	#SBATCH -o $remote_dayfile
815	#SBATCH -e $remote_dayfile
816
817	$init_cmds
818	$module_calls
819
820	%%END%%
821
822	else
823	cat > $job_to_send << %%END%%
824	#!/bin/bash -l
825	#SBATCH -J $job_name
826	#SBATCH -t $timestring
827	#SBATCH -l ncpus=1
828	#SBATCH -l pmem=${memory}mb
829	#SBATCH -m abe
830	#SBATCH -o $remote_dayfile
831	#SBATCH -e $remote_dayfile
832
833	$init_cmds
834	$module_calls
835
836	%%END%%
837
838	fi
839
840	elif [[ $remote_host = lcfimm ]]
841	then
842
843	if [[ $numprocs != 0 ]]
844	then
845	cat > $job_to_send << %%END%%
846	#!/bin/ksh
847	#PBS -N $job_name
848	#PBS -A $project_account
849	#PBS -l walltime=$timestring
850	#PBS -l nodes=${nodes}:ppn=$processes_per_node
851	#PBS -l pmem=${memory}mb
852	#PBS -m abe
853	#PBS -o $remote_dayfile
854	#PBS -j oe
855	mpd &
856
857	%%END%%
858
859	else
860	cat > $job_to_send << %%END%%
861	#!/bin/ksh
862	#PBS -N $job_name
863	#PBS -A $project_account
864	#PBS -l walltime=$timestring
865	#PBS -l ncpus=1
866	#PBS -l pmem=${memory}mb
867	#PBS -m abe
868	#PBS -o $remote_dayfile
869	#PBS -j oe
870
871	%%END%%
872
873	fi
874
875	elif [[ $remote_host = lcflow ]]
876	then
877	if [ $memory -gt 1800 ]; then
878	use_himem=""
879	else
880	use_himem="#"
881	fi
882	if [[ $numprocs != 0 ]]
883	then
884	pe_set="#$ -pe impi41 $numprocs"
885	else
886	pe_set="#$ -pe impi41 1"
887	fi
888	if [[ $queue = default ]]
889	then
890	queue_set=""
891	else
892	queue_set="#$ -q $queue"
893	fi
894	[[ "$disc_space" = "" ]] && disc_space=50
895
896	cat > $job_to_send << %%END%%
897	#!/bin/bash
898	#$ -S /bin/bash
899	#$ -N $job_name
900	#$ -cwd
901	#$ -l h_rt=$timestring
902	#$ -l h_vmem=${memory}M
903	#$ -o $remote_dayfile
904	#$ -j y
905	$pe_set
906	#$ -R y
907	${use_himem}#$ -l highmem=true
908	#$ -l h_fsize=${disc_space}G
909	$queue_set
910
911	%%END%%
912
913	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
914	then
915
916	if [[ $numprocs != 0 ]]
917	then
918	cat > $job_to_send << %%END%%
919	#!/bin/ksh
920	#PBS -N $job_name
921	#PBS -l walltime=$timestring
922	#PBS -l ncpus=$numprocs
923	#PBS -l pmem=${memory}mb
924	#PBS -o $remote_dayfile
925	#PBS -l nodes=$nodes:ppn=${processes_per_node}
926	#PBS -j oe
927
928	mpd &
929
930	%%END%%
931
932	else
933	cat > $job_to_send << %%END%%
934	#!/bin/ksh
935	#PBS -N $job_name
936	#PBS -l walltime=$timestring
937	#PBS -l ncpus=1
938	#PBS -l pmem=${memory}mb
939	#PBS -o $remote_dayfile
940	#PBS -j oe
941
942	%%END%%
943
944	fi
945
946	elif [[ $remote_host = lckiaps ]]
947	then
948
949	if [[ $numprocs != 0 ]]
950	then
951	cat > $job_to_send << %%END%%
952	#!/bin/ksh
953	#PBS -N $job_name
954	#PBS -l walltime=$timestring
955	#PBS -l select=1:ncpus=$numprocs
956	#PBS -l pmem=${memory}mb
957	#PBS -q $queue
958	#PBS -o $remote_dayfile
959	#PBS -j oe
960	#PBS -V
961
962	%%END%%
963
964	else
965	cat > $job_to_send << %%END%%
966	#!/bin/ksh
967	#PBS -N $job_name
968	#PBS -l walltime=$timestring
969	#PBS -l ncpus=1
970	#PBS -l pmem=${memory}mb
971	#PBS -o $remote_dayfile
972	#PBS -j oe
973
974	%%END%%
975
976	fi
977
978	elif [[ $remote_host = lcyon ]]
979	then
980
981	if [[ $numprocs != 0 ]]
982	then
983	cat > $job_to_send << %%END%%
984	#!/bin/ksh
985	#PBS -N $job_name
986	#PBS -l walltime=$timestring
987	#PBS -l ncpus=$numprocs
988	#PBS -l pmem=${memory}mb
989	#PBS -o $remote_dayfile
990	#PBS -j oe
991
992	%%END%%
993
994	else
995	cat > $job_to_send << %%END%%
996	#!/bin/ksh
997	#PBS -N $job_name
998	#PBS -l walltime=$timestring
999	#PBS -l ncpus=1
1000	#PBS -l pmem=${memory}mb
1001	#PBS -o $remote_dayfile
1002	#PBS -j oe
1003
1004	%%END%%
1005
1006	fi
1007
1008	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1009	then
1010
1011	if [[ "$sgi_feature" != "" ]]
1012	then
1013	feature_directive="#PBS -l feature=$sgi_feature"
1014	else
1015	feature_directive=""
1016	fi
1017
1018	if [[ $queue = dataq \|\| $queue = permq ]]
1019	then
1020	feature_directive="#PBS -l feature=data"
1021	fi
1022
1023	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
1024	then
1025	queue_directive="#PBS -q $queue"
1026	else
1027	queue_directive=""
1028	fi
1029
1030	if [[ $email_notification = none ]]
1031	then
1032	email_directive=""
1033	else
1034	email_directive="#PBS -M $email_notification"
1035	fi
1036
1037	if [[ $numprocs != 0 ]]
1038	then
1039	cat > $job_to_send << %%END%%
1040	#!/bin/bash
1041	#PBS -N $job_name
1042	#PBS -l walltime=$timestring
1043	#PBS -l nodes=$nodes:ppn=${processes_per_node}
1044	#PBS -l naccesspolicy=$node_usage
1045	#PBS -o $remote_dayfile
1046	#PBS -j oe
1047	$feature_directive
1048	$queue_directive
1049	$email_directive
1050
1051	eval \`/sw/swdist/bin/modulesinit\`
1052	#. /usr/share/modules/init/bash
1053	$init_cmds
1054	$module_calls
1055
1056	echo ld_library_path=\$LD_LIBRARY_PATH
1057
1058	%%END%%
1059
1060	else
1061	cat > $job_to_send << %%END%%
1062	#PBS -S /bin/bash
1063	#PBS -N $job_name
1064	#PBS -l walltime=$timestring
1065	#PBS -l ncpus=1
1066	#PBS -o $remote_dayfile
1067	#PBS -j oe
1068	$queue_directive
1069	$email_directive
1070
1071	eval \`/sw/swdist/bin/modulesinit\`
1072	#. /usr/share/modules/init/bash
1073	$init_cmds
1074	$module_calls
1075
1076	%%END%%
1077
1078	fi
1079
1080	elif [[ $remote_host = lcxe6 ]]
1081	then
1082
1083	if [[ $numprocs != 0 ]]
1084	then
1085	cat > $job_to_send << %%END%%
1086	#!/bin/ksh
1087	#PBS -S /bin/ksh
1088	#PBS -N $job_name
1089	#PBS -A $project_account
1090	#PBS -j oe
1091	#PBS -l walltime=$timestring
1092	#PBS -l mppwidth=${numprocs}
1093	#PBS -l mppnppn=${processes_per_node}
1094	#PBS -m abe
1095	#PBS -o $remote_dayfile
1096	$email_directive
1097
1098	$init_cmds
1099	$module_calls
1100
1101	%%END%%
1102
1103	else
1104	cat > $job_to_send << %%END%%
1105	#!/bin/ksh
1106	#PBS -S /bin/ksh
1107	#PBS -N $job_name
1108	#PBS -A $project_account
1109	#PBS -j oe
1110	#PBS -l walltime=$timestring
1111	#PBS -l ncpus=1
1112	#PBS -l pmem=${memory}mb
1113	#PBS -m abe
1114	$email_directive
1115	#PBS -o $remote_dayfile
1116
1117	$init_cmds
1118	$module_calls
1119
1120	%%END%%
1121
1122	fi
1123
1124	elif [[ $remote_host = lckyoto ]]
1125	then
1126
1127	cat > $job_to_send << %%END%%
1128	#!/bin/ksh
1129	# @\$-o $remote_dayfile
1130	# @\$-eo -oi
1131	# @\$-lP 16
1132	# @\$-lp 1
1133	# @\$-lm 28gb -llm unlimited -ls unlimited
1134	# @\$-q $queue
1135	# @\$-Pvn abs_pack
1136	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1137	#. /thin/local/etc/setprofile/intel-11.0.sh
1138	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1139	. ~/.myprofile
1140	#. /home2/t/t51254/palm/current_version/myprofile
1141	#. /thin/apps/pgi/mpi.sh
1142	#
1143	env
1144	#
1145	set -x
1146
1147	%%END%%
1148
1149	elif [[ $remote_host = lcxt5m ]]
1150	then
1151
1152	if [[ $numprocs != 0 ]]
1153	then
1154	cat > $job_to_send << %%END%%
1155	#!/bin/ksh
1156	#PBS -S /bin/ksh
1157	#PBS -N $job_name
1158	#PBS -j oe
1159	#PBS -l walltime=$timestring
1160	#PBS -l mppwidth=${numprocs}
1161	#PBS -l mppnppn=${processes_per_node}
1162	#PBS -m abe
1163	#PBS -o $remote_dayfile
1164
1165	$init_cmds
1166	$module_calls
1167
1168	%%END%%
1169
1170	else
1171	cat > $job_to_send << %%END%%
1172	#!/bin/ksh
1173	#PBS -S /bin/ksh
1174	#PBS -N $job_name
1175	#PBS -j oe
1176	#PBS -l walltime=$timestring
1177	#PBS -l ncpus=1
1178	#PBS -l pmem=${memory}mb
1179	#PBS -m abe
1180	#PBS -o $remote_dayfile
1181
1182	$init_cmds
1183	$module_calls
1184
1185	%%END%%
1186
1187	fi
1188
1189	elif [[ $remote_host = lckyuh ]]
1190	then
1191	cat > $job_to_send << %%END%%
1192	#!/bin/bash
1193	#PJM -L "rscgrp=$queue"
1194	#PJM -L "node=$nodes"
1195	#PJM --mpi "proc=$numprocs"
1196	#PJM -L "elapse=$timestring"
1197	#PJM -o $remote_dayfile
1198	#PJM -j
1199	#PJM -X
1200	#PJM --no-stging
1201
1202	export LANG=en_US.UTF-8
1203	%%END%%
1204
1205	elif [[ $remote_host = lckyut ]]
1206	then
1207	cat > $job_to_send << %%END%%
1208	#!/bin/bash
1209	#PJM -L "rscgrp=$queue"
1210	#PJM -L "vnode=$numprocs"
1211	#PJM -L "vnode-core=1"
1212	#PJM -L "elapse=$timestring"
1213	#PJM --mpi proc=$numprocs
1214	#PJM -o $remote_dayfile
1215	#PJM -j
1216	#PJM -X
1217	#PJM --no-stging
1218
1219	export LANG=en_US.UTF-8
1220	%%END%%
1221
1222	elif [[ $remote_host = nech ]]
1223	then
1224
1225	if (( nodes > 1 ))
1226	then
1227	cat > $job_to_send << %%END%%
1228	#!/bin/ksh
1229	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1230	#PBS -l ${qsubmem}=${Memory}gb
1231	#PBS -b $nodes
1232	#PBS -o $remote_dayfile
1233	#PBS -N palm
1234	#PBS -j o
1235	#PBS -T mpisx
1236
1237	%%END%%
1238
1239	elif [[ $numprocs != 0 ]]
1240	then
1241	cat > $job_to_send << %%END%%
1242	#!/bin/ksh
1243	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1244	#PBS -l ${qsubmem}=${Memory}gb
1245	#PBS -o $remote_dayfile
1246	#PBS -N palm
1247	#PBS -j o
1248
1249	%%END%%
1250
1251	else
1252	cat > $job_to_send << %%END%%
1253	#!/bin/ksh
1254	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1255	#PBS -o $remote_dayfile
1256	#PBS -j o
1257
1258	%%END%%
1259
1260	fi
1261
1262	elif [[ $remote_host = necriam ]]
1263	then
1264
1265	if (( nodes > 1 ))
1266	then
1267	cat > $job_to_send << %%END%%
1268	#!/bin/ksh
1269	#PBS -b $nodes
1270	#PBS -o $remote_dayfile
1271	#PBS -N $job_name
1272	#PBS -j o
1273	#PBS -v MPIPROGINV=YES
1274
1275	%%END%%
1276
1277	elif [[ $numprocs != 0 ]]
1278	then
1279	cat > $job_to_send << %%END%%
1280	#!/bin/ksh
1281	#PBS -q ${queue}
1282	#PBS -o $remote_dayfile
1283	#PBS -N $job_name
1284	#PBS -j o
1285	#PBS -v MPIPROGINV=YES
1286
1287	%%END%%
1288
1289	else
1290	cat > $job_to_send << %%END%%
1291	#!/bin/ksh
1292	#PBS -o $remote_dayfile
1293	#PBS -j o
1294
1295	%%END%%
1296
1297	fi
1298
1299	elif [[ $remote_host = lctit ]]
1300	then
1301	cat > $job_to_send << %%END%%
1302	#!/bin/ksh
1303	$init_cmds
1304	$module_calls
1305
1306	%%END%%
1307
1308	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1309	if [[ $tasks_per_node != $processes_per_node ]]
1310	then
1311	submcom="$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 $queue "
1312	else
1313	submcom="$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 "
1314	fi
1315
1316	else
1317
1318	cat > $job_to_send << %%END%%
1319	# @\$-q ${queue}
1320	# @\$-l${qsubtime} $timestring
1321	# @\$-l${qsubmem} ${memory}mb
1322	# @\$-o $remote_dayfile
1323	# @\$-eo
1324
1325	%%END%%
1326
1327	fi
1328
1329
1330	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1331	# BEI EXIT VERANLASSEN
1332	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1333	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1334	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1335	then
1336	echo "set +vx" >> $job_to_send
1337	echo "trap '" >> $job_to_send
1338	echo "set +vx" >> $job_to_send
1339	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* ]]
1340	then
1341	if [[ $remote_host = ibmh ]]
1342	then
1343	return_queue=c1
1344	elif [[ $remote_host = ibmkisti ]]
1345	then
1346	return_queue=class.1-2
1347	elif [[ $remote_host = ibmku ]]
1348	then
1349	return_queue=sdbg2
1350	elif [[ $remote_host = ibms ]]
1351	then
1352	return_queue=p_normal
1353	elif [[ $remote_host = ibmy ]]
1354	then
1355	return_queue=serial
1356	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1357	then
1358	return_queue=serialq
1359	elif [[ $remote_host = necriam ]]
1360	then
1361	return_queue=SP
1362	elif [[ $remote_host = lckiaps ]]
1363	then
1364	return_queue=express
1365	elif [[ $remote_host = lckyuh ]]
1366	then
1367	return_queue=cx-single
1368	elif [[ $remote_host = lckyut ]]
1369	then
1370	return_queue=cx-single
1371	else
1372	return_queue=unknown
1373	fi
1374
1375	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1376	then
1377
1378	if [[ $remote_host = ibmku ]]
1379	then
1380	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1381	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1382	else
1383	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1384	fi
1385	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1386	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1387	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1388	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1389	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1390	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1391	if [[ $host != "ibmh" ]]
1392	then
1393	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1394	fi
1395	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1396	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1397
1398	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1399	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1400
1401	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1402	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
1403	if [[ $remote_host = ibmku ]]
1404	then
1405	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1406	fi
1407	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1408
1409	elif [[ $remote_host = nech ]]
1410	then
1411	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1412	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1413	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1414	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1415	echo "#PBS -j o" >> $job_to_send
1416	echo " " >> $job_to_send
1417	echo "set -x" >> $job_to_send
1418	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1419	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1420	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1421	echo "%%END%%" >> $job_to_send
1422
1423	elif [[ $remote_host = necriam ]]
1424	then
1425	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1426	echo "#PBS -q $return_queue" >> $job_to_send
1427	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1428	echo "#PBS -j o" >> $job_to_send
1429	echo " " >> $job_to_send
1430	echo "set -x" >> $job_to_send
1431	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1432	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1433	echo "%%END%%" >> $job_to_send
1434
1435	elif [[ $remote_host = lckyuh ]]
1436	then
1437	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1438	echo "#!/bin/bash" >> $job_to_send
1439	echo "#PJM -L \"node=1\"" >> $job_to_send
1440	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1441	echo "#PJM --no-stging" >> $job_to_send
1442	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1443	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1444	echo "#PJM -j" >> $job_to_send
1445	echo " " >> $job_to_send
1446	echo "export LANG=en_US.UTF-8" >> $job_to_send
1447	echo "set -x" >> $job_to_send
1448	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1449	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1450	echo "%%END%%" >> $job_to_send
1451
1452	elif [[ $remote_host = lckyut ]]
1453	then
1454	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1455	echo "#!/bin/bash" >> $job_to_send
1456	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1457	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1458	echo "#PJM --no-stging" >> $job_to_send
1459	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1460	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1461	echo "#PJM -j" >> $job_to_send
1462	echo " " >> $job_to_send
1463	echo "export LANG=en_US.UTF-8" >> $job_to_send
1464	echo "set -x" >> $job_to_send
1465	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1466	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1467	echo "%%END%%" >> $job_to_send
1468
1469	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1470	then
1471	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1472	echo "#!/bin/bash" >> $job_to_send
1473	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1474	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1475	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1476	echo "#PBS -l feature=data" >> $job_to_send
1477	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1478	echo "#PBS -j oe" >> $job_to_send
1479	echo " " >> $job_to_send
1480	echo ". /usr/share/modules/init/bash" >> $job_to_send
1481	echo "set -x" >> $job_to_send
1482	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1483	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1484	echo "%%END%%" >> $job_to_send
1485
1486	elif [[ $remote_host = lcflow ]]
1487	then
1488	echo "cat > scpjob.${kennung}.tmp << %%END%%" >> $job_to_send
1489	echo "#!/bin/bash" >> $job_to_send
1490	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1491	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1492	echo "SGEPREFIX -cwd" >> $job_to_send
1493	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1494	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1495	echo "SGEPREFIX -l excl_flow=false" >> $job_to_send
1496	echo "SGEPREFIX -j y" >> $job_to_send
1497	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$kennung" >> $job_to_send
1498	echo " " >> $job_to_send
1499	echo "set -x" >> $job_to_send
1500	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1501	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1502	echo "" >> $job_to_send
1503	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1504	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1505	echo "rm -f scpjob.${kennung}" >> $job_to_send
1506	echo "%%END%%" >> $job_to_send
1507	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${kennung}.tmp > scpjob.${kennung}" >> $job_to_send
1508	echo "rm -f scpjob.${kennung}.tmp" >> $job_to_send
1509
1510	else
1511
1512	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1513	echo "# @\\\$-q $return_queue" >> $job_to_send
1514	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1515	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1516	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1517	then
1518	echo "# @\$-l mpp_p=0" >> $job_to_send
1519	fi
1520	echo '# @\$-lF 10mb' >> $job_to_send
1521	# echo '# @\$-o /dev/null' >> $job_to_send
1522	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1523	echo '# @\\\$-eo' >> $job_to_send
1524	echo " " >> $job_to_send
1525	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1526	then
1527	echo "set +vx" >> $job_to_send
1528	echo ". .profile" >> $job_to_send
1529	fi
1530	echo "set -x" >> $job_to_send
1531	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
1532	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1533	echo "%%END%%" >> $job_to_send
1534	fi
1535
1536	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1537	then
1538	echo "llsubmit scpjob.$kennung" >> $job_to_send
1539	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1540	then
1541	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1542	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1543	echo "msub scpjob.$kennung" >> $job_to_send
1544	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1545	then
1546	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1547	elif [[ $remote_host = t3es ]]
1548	then
1549	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1550	elif [[ $remote_host = lckiaps ]]
1551	then
1552	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1553	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1554	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1555	elif [[ $remote_host = lckyu* ]]
1556	then
1557	echo "scp $PORTOPT scpjob.$kennung ${remote_username}@${remote_addres}:job_queue" >> $job_to_send
1558	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_addres} \"cd job_queue; $submcom scpjob.$kennung; rm scpjob.$kennung\" " >> $job_to_send
1559	elif [[ $remote_host = lcflow ]]
1560	then
1561	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1562	echo "/usr/bin/ssh ${remote_username}@${remote_addres} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$kennung\" " >> $job_to_send
1563	else
1564	echo "$submcom scpjob.$kennung" >> $job_to_send
1565	fi
1566	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1567	then
1568	echo "rm scpjob.$kennung" >> $job_to_send
1569	fi
1570	if [[ $remote_host = nech ]]
1571	then
1572	echo "cd -" >> $job_to_send
1573	fi
1574	else
1575	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1576	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1577	fi
1578	echo "set -x" >> $job_to_send
1579	echo " ' exit" >> $job_to_send
1580	echo "set -x" >> $job_to_send
1581	fi
1582
1583
1584
1585	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1586	cat $file_to_send >> $job_to_send
1587
1588	if [[ $remote_host = ibm ]]
1589	then
1590	echo " " >> $job_to_send
1591	echo "exit" >> $job_to_send
1592	fi
1593
1594	# remove job file
1595	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1596	then
1597	echo " " >> $job_to_send
1598	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1599	fi
1600
1601
1602
1603	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1604	if [[ -z $remote_user ]]
1605	then
1606	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1607	then
1608	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1609	fi
1610	fi
1611
1612
1613
1614	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1615	if [[ $no_submit = false ]]
1616	then
1617	if [[ $remote_host != $local_host ]]
1618	then
1619	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1620	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1621	then
1622	job_catalog_save=$job_catalog
1623	job_catalog=job_queue
1624	elif [[ $remote_host = nech ]]
1625	then
1626	job_catalog_save=$job_catalog
1627	job_catalog=/hpf/b/${remote_user}/job_queue
1628	fi
1629	if [[ $remote_host = nech ]]
1630	then
1631	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1632	# TRANSFERIERT WERDEN
1633	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1634	else
1635	scp $PORTOPT $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1636	fi
1637	if [[ $? = 1 ]]
1638	then
1639	locat=scp; exit
1640	fi
1641	if [[ $remote_host = ibms ]]
1642	then
1643	job_catalog=$job_catalog_save
1644	fi
1645	[[ $verify = true ]] && printf "\n >>> finished\n"
1646	else
1647	eval job_catalog=$job_catalog
1648	cp $job_to_send ${job_catalog}/$job_on_remhost
1649	fi
1650
1651
1652
1653	# NQS- BZW. LOADLEVELER-JOB STARTEN
1654	if [[ $remote_host != $local_host ]]
1655	then
1656	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1657
1658	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1659	then
1660	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1661	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1662	elif [[ $remote_host = ibmku ]]
1663	then
1664	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1665	elif [[ $remote_host = lcflow ]]
1666	then
1667	/usr/bin/ssh $SSH_PORTOPT $remote_addres -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1668	else
1669	ssh $SSH_PORTOPT $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1670	fi
1671
1672	[[ $verify = true ]] && printf " >>> o.k.\n"
1673	else
1674	cd $job_catalog
1675	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1676	then
1677	eval $submcom $job_on_remhost
1678	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1679	then
1680	chmod u+x $job_on_remhost
1681	eval $submcom $job_on_remhost
1682	elif [[ $local_host = nech ]]
1683	then
1684	if [[ $queue = default ]]
1685	then
1686	eval $submcom $job_on_remhost
1687	else
1688	eval $submcom -q $queue $job_on_remhost
1689	fi
1690	else
1691	qsub $job_on_remhost
1692	fi
1693
1694	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1695	# only at the end of the job.
1696	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1697	then
1698	rm $job_on_remhost
1699	fi
1700	cd - > /dev/null
1701	fi
1702	fi
1703
1704
1705
1706	# ABSCHLUSSARBEITEN
1707
1708	if [[ $no_submit = false ]]
1709	then
1710	rm -f $job_to_send
1711	fi
1712	[[ $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 |