Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 206

Last change on this file since 206 was 206, checked in by raasch, 16 years ago
ocean-atmosphere coupling realized with MPI-1, adjustments in mrun, mbuild, subjob for lcxt4
Property svn:keywords set to `Id Rev`
File size: 41.2 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 206 2008-10-13 14:59:11Z raasch $
4
5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
7	# Host zurueckgeschickt werden sollen
8
9
10	# letzte Aenderung:
11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
13	# 06/02/98 - Siggi - berte validiert
14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
15	# moeglich
16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
19	# allgemeine Anpassungen fuer ibm-Rechner
20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
21	# Ruecktransfer des Jobprotokolls fuer diese
22	# Maschine abgeschaltet
23	# 31/10/02 - Siggi - berni validiert
24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
25	# 08/11/02 - Siggi - quanero validiert
26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
27	# 23/01/03 - Siggi - hostname nobel changed to nobela
28	# 06/02/03 - Siggi - gregale validated
29	# 12/02/03 - Siggi - orkan and maestro validated
30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
31	# 12/03/03 - Siggi - nec at DKRZ validated
32	# 13/03/03 - Siggi - new nqs resource variable Memory
33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
34	# node
35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
37	# cxxl added to ibmh
38	# 29/08/03 - Siggi - changes in job queues and communication system on
39	# ibms
40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
41	# 30/10/03 - Siggi - nech is not supported any more
42	# 10/11/03 - Siggi - nech2 renamed to nech
43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
45	# 31/03/04 - Siggi - new option -N for node usage
46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
47	# due to error in ssh installation (otherwise a prompt
48	# for the password appears)
49	# 23/07/04 - Siggi - changes due to the new berni configuration
50	# (federation switch)
51	# 01/09/04 - Gerald new job-classes on hanni
52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
54	# 03/12/04 - Siggi - notification on ibm switched of in case of
55	# delete_dayfile = true, node usage in cdev set to
56	# shared
57	# 16/02/05 - Gerald hababai validated
58	# 29/03/05 - Micha - new job class channi on hanni
59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
60	# on ibms
61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
62	# p_normal
63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
65	# "sn_all", new job class cexp
66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
68	# hurricane (with -X0)
69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
70	# added
71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
72	# 09/02/06 - Siggi - ibmy admitted
73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
74	# 13/04/06 - Siggi - ostria admitted
75	# 18/04/06 - Siggi - new option -O for OpenMP usage
76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
80	# 28/11/06 - Siggi - levanto admitted
81	# 13/02/07 - Siggi - hpmuk releated code removed
82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
84	# 27/07/07 - Siggi - autan admitted
85	# 03/08/07 - Marcus- lcfimm admitted
86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
88	# suggestions
89	# 19/10/07 - Marcus- add new optional argument -g group_number
90	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
91	# 30/10/07 - Marcus- further adjustments for queues on lctit
92	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
93	# Science)
94	# 14/07/08 - Siggi - adjustments for lcsgih
95	# 23/09/08 - Gerald- paesano admitted
96	# 02/10/08 - Siggi - PBS adjustments for lcxt4
97
98
99
100	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
101	delete_dayfile=false
102	group_number=none
103	locat=normal
104	no_default_queue=none
105	no_submit=false
106	job_catalog="~/job_queue"
107	job_name=none
108	local_user=$LOGNAME
109	node_usage=shared
110	numprocs=1
111	punkte="..........................................................."
112	submcom=qsub
113	queue=default
114	remote_host=none
115	remote_user=""
116	verify=true
117
118	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
119	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
120	typeset -L20 spalte1
121	typeset -R40 spalte2
122	typeset -L60 spalte3
123
124
125
126	# FEHLERBEHANDLUNG
127	# BEI EXIT:
128	trap 'if [[ $locat != normal ]]
129	then
130	case $locat in
131	(option) printf "\n --> available optios can be displayed"
132	printf " by typing:"
133	printf "\n \"subjob ?\" \n";;
134	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
135	(*) printf "\n +++ unknown error"
136	printf "\n please inform S. Raasch!\n"
137	esac
138	[[ -f $job_to_send ]] && rm $job_to_send
139	printf "\n\n+++ SUBJOB killed \n\n"
140	fi' exit
141
142
143	# BEI TERMINAL-BREAK:
144	trap '[[ -f $job_to_send ]] && rm $job_to_send
145	printf "\n\n+++ SUBJOB killed \n\n"
146	exit
147	' 2
148
149
150
151
152	# LOKALEN HOSTNAMEN FESTSTELLEN
153	local_host=$(hostname)
154
155
156
157	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
158	# UEBERHAUPT ZULAESSIG IST
159	case $local_host in
160	(atmos) local_addres=172.20.25.35; local_host=lcide;;
161	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
162	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
163	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
164	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
165	(b01*) local_addres=130.73.232.102; local_host=lcsgib;;
166	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
167	(cs*) local_addres=136.172.44.131; local_host=nech;;
168	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
169	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
170	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
171	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
172	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
173	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
174	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
175	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
176	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
177	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
178	(h01*) local_addres=130.75.4.102; local_host=lcsgih;;
179	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
180	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
181	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
182	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
183	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
184	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
185	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
186	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
187	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
188	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
189	(service100) local_addres=130.75.4.102; local_host=lcsgih;;
190	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
191	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
192	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
193	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
194	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
195	(*) printf "\n +++ \"$local_host\" unknown";
196	printf "\n please inform S. Raasch!";
197	locat=parameter; exit;;
198	esac
199
200
201
202	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
203	remote_host=$local_host
204
205
206
207
208	# PROZEDUROPTIONEN EINLESEN
209	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
210	do
211	case $option in
212	(c) job_catalog=$OPTARG;;
213	(d) delete_dayfile=true;;
214	(D) no_submit=true;;
215	(g) group_number=$OPTARG;;
216	(h) remote_host=$OPTARG;;
217	(m) memory=$OPTARG;;
218	(n) job_name=$OPTARG;;
219	(N) node_usage=$OPTARG;;
220	(O) threads_per_task=$OPTARG;;
221	(q) no_default_queue=$OPTARG;;
222	(t) cputime=$OPTARG;;
223	(T) tasks_per_node=$OPTARG;;
224	(u) remote_user=$OPTARG;;
225	(v) verify=false;;
226	(X) numprocs=$OPTARG;;
227	(\?) printf "\n +++ Option $OPTARG unknown \n";
228	locat=option; exit;;
229	esac
230	done
231
232
233	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
234	shift OPTIND-1; file_to_send=$1
235
236
237	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
238	if [ "$1" = "?" ]
239	then
240	(printf "\n *** subjob can be called as follows:\n"
241	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
242	printf "\n Description of available options:\n"
243	printf "\n Option Description Default-Value"
244	printf "\n -c job-input- and output-catalog ~/job_queue"
245	printf "\n -d no job-protocol will be created ---"
246	printf "\n -D only the job-file will be created ---"
247	printf "\n -h execution host, available hosts: $remote_host"
248	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
249	printf "\n lctit, nech, neck, unics"
250	printf "\n -m memory demand per process in MByte ---"
251	printf "\n -n jobname <jobdatei>"
252	printf "\n -O threads per task (for OpenMP usage) 1"
253	printf "\n -q job-queue to be used default"
254	printf "\n -t allowed cpu-time in seconds ---"
255	printf "\n -T tasks per node (on parallel hosts) ---"
256	printf "\n -u username on execution host from .netrc"
257	printf "\n -v no prompt for confirmation ---"
258	printf "\n -X # of processors (on parallel hosts) 1"
259	printf "\n "
260	printf "\n The only possible positional parameter is <jobfile>:"
261	printf "\n The complete NQS-job must be provided here."
262	printf "\n <jobfile>=? creates this outline\n\n") \| more
263	exit
264	fi
265
266
267
268	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
269	if [[ "$file_to_send" = "" ]]
270	then
271	printf "\n +++ job-file missing"
272	locat=parameter; exit
273	else
274	if [[ -f $file_to_send ]]
275	then
276	true
277	else
278	printf "\n +++ job-file: "
279	printf "\n $file_to_send"
280	printf "\n does not exist"
281	locat=parameter; exit
282	fi
283	fi
284
285
286
287	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
288	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
289	if [[ $job_name = none ]]
290	then
291	job_name=$file_to_send
292	fi
293	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
294	then
295	printf "\n +++ job-file name: "
296	printf "\n $job_name"
297	printf "\n must not contain \"/\"-characters"
298	locat=parameter; exit
299	fi
300
301
302
303
304	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
305	# ODER WENN HOST NICHT ANGEGEBEN WURDE
306	if [[ $remote_host = none ]]
307	then
308	printf "\n +++ host missing"
309	locat=option; exit
310	else
311	case $remote_host in
312	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
313	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
314	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
315	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
316	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
317	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
318	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
319	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
320	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
321	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
322	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
323	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
324	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
325	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
326	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
327	locat=parameter; exit;;
328	esac
329	fi
330
331
332	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
333	if [[ $no_default_queue != none ]]
334	then
335	error=false
336	ndq=$no_default_queue
337	case $remote_host in
338	(ibm) case $ndq in
339	(p690_express\|p690_standard\|p690_long) error=false;;
340	(*) error=true;;
341	esac;;
342	(ibmb) case $ndq in
343	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
344	(*) error=true;;
345	esac;;
346	(ibmh) case $ndq in
347	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
348	(*) error=true;;
349	esac;;
350	(ibms) case $ndq in
351	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
352	(*) error=true;;
353	esac;;
354	(ibmy) case $ndq in
355	(parallel) error=false;;
356	(*) error=true;;
357	esac;;
358	(lcsgib\|lcsgih) case $ndq in
359	(testq\|smallq\|bigq) error=false;;
360	(*) error=true;;
361	esac;;
362	(lctit) case $ndq in
363	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
364	(*) error=true;;
365	esac;;
366	(t3eb) case $ndq in
367	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
368	(*) error=true;;
369	esac;;
370	(t3eh) case $ndq in
371	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
372	(*) error=true;;
373	esac;;
374	(t3ej2\|t3ej5) case $ndq in
375	(low\|normal\|high) error=false;;
376	(*) error=true;;
377	esac;;
378	(t3es) case $ndq in
379	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
380	(*) error=true;;
381	esac;;
382	(unics) case $ndq in
383	(unics\|ht) error=false;;
384	(*) error=true;;
385	esac;;
386	esac
387	if [[ $error = true ]]
388	then
389	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
390	locat=parameter; exit
391	else
392	queue=$no_default_queue
393	fi
394	fi
395
396
397
398	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
399	if [[ $node_usage != shared && $queue = cdev ]]
400	then
401	node_usage=shared
402	fi
403
404
405
406	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
407	# AUFTEILEN
408	done=false
409	while [[ $done = false ]]
410	do
411	if (( $cputime <= 0 ))
412	then
413	printf "\n +++ wrong cpu-time or cpu-time missing"
414	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
415	printf "\n >>> "
416	read cputime 1>/dev/null 2>&1
417	else
418	done=true
419	fi
420	done
421	if [[ $remote_host = nech ]]
422	then
423	if (( tasks_per_node != 0 ))
424	then
425	(( cputime = cputime * tasks_per_node ))
426	elif [[ $numprocs != 0 ]]
427	then
428	(( cputime = cputime * numprocs ))
429	fi
430	fi
431	(( stunden = cputime / 3600 ))
432	(( resttime = cputime - stunden * 3600 ))
433	(( minuten = resttime / 60 ))
434	(( sekunden = resttime - minuten * 60 ))
435	timestring=${stunden}:${minuten}:${sekunden}
436
437
438
439	# PRUEFEN DER KERNSPEICHERANFORDERUNG
440	done=false
441	while [[ $done = false ]]
442	do
443	if (( memory <= 0 ))
444	then
445	printf "\n +++ wrong memory demand or memory demand missing"
446	printf "\n >>> Please type memory in MByte per process as INTEGER:"
447	printf "\n >>> "
448	read memory 1>/dev/null 2>&1
449	else
450	done=true
451	fi
452	done
453
454	if [[ $remote_host = nech \|\| $remote_host = neck ]]
455	then
456	if (( tasks_per_node != 0 ))
457	then
458	(( Memory = memory * tasks_per_node / 1000 ))
459	elif [[ $numprocs != 0 ]]
460	then
461	(( Memory = memory * numprocs / 1000 ))
462	else
463	(( Memory = memory / 1000 ))
464	fi
465	fi
466
467
468	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
469	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
470	then
471	(( memory = memory * threads_per_task ))
472	fi
473
474
475	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
476	if (( tasks_per_node != 0 ))
477	then
478	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
479	fi
480
481
482
483	# HEADER-AUSGABE
484	if [[ $verify = true ]]
485	then
486	printf "\n\n"
487	printf "#--------------------------------------------------------------# \n"
488	spalte1=SUBJOB;spalte2=$(date)
489	printf "\| $spalte1$spalte2 \| \n"
490	printf "\| \| \n"
491	printf "\| values of parameters/options: \| \n"
492	spalte1=local_host$punkte; spalte2=$punkte$local_host
493	printf "\| $spalte1$spalte2 \| \n"
494	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
495	printf "\| $spalte1$spalte2 \| \n"
496	spalte1=queue$punkte; spalte2=$punkte$queue
497	printf "\| $spalte1$spalte2 \| \n"
498	spalte1=memory$punkte; spalte2="$punkte$memory mb"
499	printf "\| $spalte1$spalte2 \| \n"
500	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
501	printf "\| $spalte1$spalte2 \| \n"
502	spalte1=job_name$punkte; spalte2="$punkte$job_name"
503	printf "\| $spalte1$spalte2 \| \n"
504	printf "#--------------------------------------------------------------# \n\n"
505
506
507	# KONTROLLABFRAGE, OB ALLES O.K.
508	antwort="dummy"
509	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
510	do
511	read antwort?" >>> continue (y/n) ? "
512	done
513	if [[ $antwort = n \|\| $antwort = N ]]
514	then
515	locat=verify; exit
516	fi
517	printf "\n"
518	fi
519
520
521
522	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
523	kennung=$RANDOM
524	job_on_remhost=${job_name}_${kennung}_$local_host
525	job_to_send=job_to_send_$kennung
526	if [[ $delete_dayfile = false ]]
527	then
528	remote_dayfile=${local_host}_${job_name}_result_$kennung
529	local_dayfile=${remote_host}_${job_name}
530	else
531	remote_dayfile=/dev/null
532	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
533	then
534	notification='# @ notification = never'
535	fi
536	fi
537
538
539
540	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
541	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
542	then
543
544	if [[ $remote_host = ibmy ]]
545	then
546	consumable_memory=""
547	else
548	consumable_memory="ConsumableMemory($memory mb)"
549	fi
550	if [[ $queue = cdev ]]
551	then
552	data_limit="# @ data_limit = 1.76gb"
553	network_to_use="# @ network.mpi = sn_all,shared,ip"
554	else
555	if [[ $remote_host = ibms ]]
556	then
557	network_to_use="# @ network.mpi = csss,shared,us"
558	elif [[ $remote_host = ibmy ]]
559	then
560	network_to_use=""
561	else
562	network_to_use="# @ network.mpi = sn_all,shared,us"
563	data_limit="# @ data_limit = 1.76gb"
564	fi
565	fi
566
567	cat > $job_to_send << %%END%%
568	#!/bin/ksh
569
570	# @ job_type = parallel
571	# @ job_name = palm
572	# @ wall_clock_limit = ${timestring},$timestring
573	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
574	# @ output = $remote_dayfile
575	# @ error = $remote_dayfile
576	# @ image_size = 50
577	# @ class = $queue
578	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
579	$network_to_use
580	$data_limit
581	$notification
582
583	%%END%%
584
585	if (( nodes > 0 ))
586	then
587
588	cat >> $job_to_send << %%END%%
589	# @ node = $nodes
590	# @ tasks_per_node = $tasks_per_node
591	# @ node_usage = $node_usage
592	# @ queue
593
594	%%END%%
595
596	else
597
598	if [[ $remote_host != ibmy ]]
599	then
600
601	cat >> $job_to_send << %%END%%
602	# @ blocking = unlimited
603	# @ total_tasks = $numprocs
604	# @ node_usage = $node_usage
605	# @ queue
606
607	%%END%%
608
609	else
610
611	cat >> $job_to_send << %%END%%
612	# @ node = 1
613	# @ total_tasks = $numprocs
614	# @ queue
615
616	%%END%%
617
618	fi
619
620	fi
621
622	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
623	then
624
625	cat > $job_to_send << %%END%%
626	#!/bin/ksh
627
628	# @ job_type = serial
629	# @ node_usage = $node_usage
630	# @ job_name = palm
631	# @ wall_clock_limit = ${timestring},$timestring
632	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
633	# @ output = $remote_dayfile
634	# @ error = $remote_dayfile
635	# @ class = $queue
636	$notification
637
638	# @ queue
639
640	%%END%%
641
642	elif [[ $remote_host = lcfimm ]]
643	then
644
645	if [[ $numprocs != 0 ]]
646	then
647	cat > $job_to_send << %%END%%
648	#!/bin/ksh
649	#PBS -N $job_name
650	#PBS -A geofysisk
651	#PBS -l walltime=$timestring
652	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
653	#PBS -l pmem=${memory}mb
654	#PBS -m abe
655	#PBS -M bjorn.maronga@student.uib.no
656	#PBS -o $remote_dayfile
657	#PBS -j oe
658	mpd &
659
660	%%END%%
661
662	else
663	cat > $job_to_send << %%END%%
664	#!/bin/ksh
665	#PBS -N $job_name
666	#PBS -A geofysisk
667	#PBS -l walltime=$timestring
668	#PBS -l ncpus=1
669	#PBS -l pmem=${memory}mb
670	#PBS -m abe
671	#PBS -M bjorn.maronga@student.uib.no
672	#PBS -o $remote_dayfile
673	#PBS -j oe
674
675	%%END%%
676
677	fi
678
679	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
680	then
681
682	if [[ $queue = testq ]]
683	then
684	queue_directive="#PBS -q $queue"
685	else
686	queue_directive=""
687	fi
688
689	if [[ $numprocs != 0 ]]
690	then
691	cat > $job_to_send << %%END%%
692	#!/bin/bash
693	#PBS -N $job_name
694	#PBS -l walltime=$timestring
695	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
696	#PBS -l feature=$sgi_feature
697	#PBS -l naccesspolicy=$node_usage
698	#PBS -o $remote_dayfile
699	#PBS -j oe
700	$queue_directive
701
702	. /usr/share/modules/init/bash
703	module load mpt
704	# module load mvapich2/1.2rc1-intel
705	# module load mvapich2
706	module load netcdf
707	export LD_LIBRARY_PATH=/sw/intel/cce/10.1.015/lib:/sw/intel/fce/10.1.015/lib:/sw/dataformats/netcdf/3.6.2/lib:\$LD_LIBRARY_PATH
708	echo ld_library_path=\$LD_LIBRARY_PATH
709
710	%%END%%
711
712	else
713	cat > $job_to_send << %%END%%
714	#PBS -S /bin/bash
715	#PBS -N $job_name
716	#PBS -l walltime=$timestring
717	#PBS -l ncpus=1
718	#PBS -o $remote_dayfile
719	#PBS -e $remote_dayfile
720
721	%%END%%
722
723	fi
724
725	elif [[ $remote_host = lcxt4 ]]
726	then
727
728	if [[ $numprocs != 0 ]]
729	then
730	cat > $job_to_send << %%END%%
731	#!/bin/ksh
732	#PBS -S /bin/ksh
733	#PBS -N $job_name
734	#PBS -A geofysisk
735	#PBS -j oe
736	#PBS -l walltime=$timestring
737	#PBS -l mppwidth=${numprocs}
738	#PBS -l mppnppn=${tasks_per_node}
739	#PBS -m abe
740	#PBS -M bjorn.maronga@student.uib.no
741	#PBS -o $remote_dayfile
742
743	%%END%%
744
745	else
746	cat > $job_to_send << %%END%%
747	#!/bin/ksh
748	#PBS -S /bin/ksh
749	#PBS -N $job_name
750	#PBS -A geofysisk
751	#PBS -j oe
752	#PBS -l walltime=$timestring
753	#PBS -l ncpus=1
754	#PBS -l pmem=${memory}mb
755	#PBS -m abe
756	#PBS -M bjorn.maronga@student.uib.no
757	#PBS -o $remote_dayfile
758
759	%%END%%
760
761	fi
762
763	elif [[ $remote_host = nech ]]
764	then
765
766	if (( nodes > 1 ))
767	then
768	cat > $job_to_send << %%END%%
769	#!/bin/ksh
770	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
771	#PBS -l ${qsubmem}=${Memory}gb
772	#PBS -b $nodes
773	#PBS -o $remote_dayfile
774	#PBS -N palm
775	#PBS -j o
776	#PBS -T mpisx
777
778	%%END%%
779
780	elif [[ $numprocs != 0 ]]
781	then
782	cat > $job_to_send << %%END%%
783	#!/bin/ksh
784	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
785	#PBS -l ${qsubmem}=${Memory}gb
786	#PBS -o $remote_dayfile
787	#PBS -N palm
788	#PBS -j o
789
790	%%END%%
791
792	else
793	cat > $job_to_send << %%END%%
794	#!/bin/ksh
795	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
796	#PBS -o $remote_dayfile
797	#PBS -j o
798
799	%%END%%
800
801	fi
802
803	elif [[ $remote_host = neck ]]
804	then
805
806	if (( nodes > 1 ))
807	then
808	cat > $job_to_send << %%END%%
809	#!/bin/ksh
810	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
811	#PBS -l ${qsubmem}=${Memory}gb
812	#PBS -b $nodes
813	#PBS -o $remote_dayfile
814	#PBS -N palm
815	#PBS -j o
816	#PBS -T mpisx
817
818	%%END%%
819
820	elif [[ $numprocs != 0 ]]
821	then
822	cat > $job_to_send << %%END%%
823	#!/bin/ksh
824	#PBS -q ${queue}
825	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
826	#PBS -l ${qsubmem}=${Memory}gb
827	#PBS -o $remote_dayfile
828	#PBS -N palm
829	#PBS -j o
830	#PBS -T mpisx
831
832	%%END%%
833
834	else
835	cat > $job_to_send << %%END%%
836	#!/bin/ksh
837	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
838	#PBS -o $remote_dayfile
839	#PBS -j o
840
841	%%END%%
842
843	fi
844
845	elif [[ $remote_host = lctit ]]
846	then
847	cat > $job_to_send << %%END%%
848	#!/bin/bash
849	#$ -S /bin/bash
850	cd $job_catalog
851	export PATH=$PALM_BIN:\$PATH
852	echo \$PATH
853
854	%%END%%
855
856	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
857	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
858	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
859
860	else
861
862	cat > $job_to_send << %%END%%
863	# @\$-q ${queue}
864	# @\$-l${qsubtime} $timestring
865	# @\$-l${qsubmem} ${memory}mb
866	# @\$-o $remote_dayfile
867	# @\$-eo
868
869	%%END%%
870
871	fi
872
873
874	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
875	# BEI EXIT VERANLASSEN
876	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
877	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
878	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
879	then
880	echo "set +vx" >> $job_to_send
881	echo "trap '" >> $job_to_send
882	echo "set +vx" >> $job_to_send
883	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
884	then
885	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
886	then
887	return_queue=c1
888	elif [[ $remote_host = ibms ]]
889	then
890	return_queue=p_normal
891	elif [[ $remote_host = ibmy ]]
892	then
893	return_queue=serial
894	elif [[ $remote_host = lctit ]]
895	then
896	return_queue=sla3
897	elif [[ $remote_host = neck ]]
898	then
899	return_queue=S
900	else
901	return_queue=unknown
902	fi
903
904	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
905	then
906
907	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
908	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
909	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
910	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
911	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
912	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
913	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
914	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
915	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
916	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
917
918	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
919	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
920
921	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
922	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
923	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
924
925	elif [[ $remote_host = nech ]]
926	then
927	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
928	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
929	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
930	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
931	echo "#PBS -j o" >> $job_to_send
932	echo " " >> $job_to_send
933	echo "set -x" >> $job_to_send
934	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
935	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
936	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
937	echo "%%END%%" >> $job_to_send
938
939	elif [[ $remote_host = neck ]]
940	then
941	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
942	echo "#PBS -q $return_queue" >> $job_to_send
943	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
944	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
945	echo "#PBS -j o" >> $job_to_send
946	echo " " >> $job_to_send
947	echo "set -x" >> $job_to_send
948	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
949	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
950	echo "%%END%%" >> $job_to_send
951
952	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
953	then
954	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
955	echo "#!/bin/bash" >> $job_to_send
956	echo "#PBS -N job_protocol_transfer" >> $job_to_send
957	echo "#PBS -l walltime=00:02:00" >> $job_to_send
958	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
959	echo "#PBS -l feature=xe" >> $job_to_send
960	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
961	echo "#PBS -j oe" >> $job_to_send
962	echo " " >> $job_to_send
963	echo ". /usr/share/modules/init/bash" >> $job_to_send
964	echo "set -x" >> $job_to_send
965	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
966	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
967	echo "%%END%%" >> $job_to_send
968
969	elif [[ $remote_host = lctit ]]
970	then
971	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
972	echo "set -x" >> $job_to_send
973	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
974	echo "cd $job_catalog" >> $job_to_send
975	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
976	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
977	echo "%%END%%" >> $job_to_send
978
979	else
980
981	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
982	echo "# @\\\$-q $return_queue" >> $job_to_send
983	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
984	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
985	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
986	then
987	echo "# @\$-l mpp_p=0" >> $job_to_send
988	fi
989	echo '# @\$-lF 10mb' >> $job_to_send
990	# echo '# @\$-o /dev/null' >> $job_to_send
991	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
992	echo '# @\\\$-eo' >> $job_to_send
993	echo " " >> $job_to_send
994	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
995	then
996	echo "set +vx" >> $job_to_send
997	echo ". .profile" >> $job_to_send
998	fi
999	echo "set -x" >> $job_to_send
1000	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1001	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1002	echo "%%END%%" >> $job_to_send
1003	fi
1004
1005	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1006	then
1007	echo "llsubmit scpjob.$kennung" >> $job_to_send
1008	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1009	then
1010	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1011	echo "msub scpjob.$kennung" >> $job_to_send
1012	elif [[ $remote_host = lctit ]]
1013	then
1014	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1015	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1016	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1017	echo "sleep 10" >> $job_to_send
1018	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1019	then
1020	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1021	elif [[ $remote_host = t3es ]]
1022	then
1023	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1024	else
1025	echo "qsub scpjob.$kennung" >> $job_to_send
1026	fi
1027	echo "rm scpjob.$kennung" >> $job_to_send
1028	if [[ $remote_host = nech ]]
1029	then
1030	echo "cd -" >> $job_to_send
1031	fi
1032	else
1033	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1034	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1035	fi
1036	echo "set -x" >> $job_to_send
1037	echo " ' exit" >> $job_to_send
1038	echo "set -x" >> $job_to_send
1039	fi
1040
1041
1042
1043	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1044	cat $file_to_send >> $job_to_send
1045
1046	if [[ $remote_host = ibm ]]
1047	then
1048	echo " " >> $job_to_send
1049	echo "exit" >> $job_to_send
1050	fi
1051	if [[ $remote_host = lctit ]]
1052	then
1053	echo " " >> $job_to_send
1054	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1055	fi
1056
1057
1058
1059	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1060	if [[ -z $remote_user ]]
1061	then
1062	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1063	then
1064	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1065	fi
1066	fi
1067
1068
1069
1070	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1071	if [[ $no_submit = false ]]
1072	then
1073	if [[ $remote_host != $local_host ]]
1074	then
1075	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1076	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1077	then
1078	job_catalog_save=$job_catalog
1079	job_catalog=job_queue
1080	elif [[ $remote_host = nech ]]
1081	then
1082	job_catalog_save=$job_catalog
1083	job_catalog=/hpf/b/${remote_user}/job_queue
1084	fi
1085	if [[ $local_host = decalpha ]]
1086	then
1087	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1088	# PASSWORT AB
1089	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1090	elif [[ $remote_host = nech ]]
1091	then
1092	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1093	# TRANSFERIERT WERDEN
1094	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1095	else
1096	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1097	fi
1098	if [[ $? = 1 ]]
1099	then
1100	locat=scp; exit
1101	fi
1102	if [[ $remote_host = ibms ]]
1103	then
1104	job_catalog=$job_catalog_save
1105	fi
1106	[[ $verify = true ]] && printf "\n >>> finished\n"
1107	else
1108	eval job_catalog=$job_catalog
1109	cp $job_to_send ${job_catalog}/$job_on_remhost
1110	fi
1111
1112
1113
1114	# NQS- BZW. LOADLEVELER-JOB STARTEN
1115	if [[ $remote_host != $local_host ]]
1116	then
1117	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1118	if [[ $remote_host != lctit ]]
1119	then
1120	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1121	else
1122	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1123	# MIT SSH, DESHALB AUFRUF PER PIPE
1124	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1125	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1126	echo "first try" > send_protocol
1127	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1128	do
1129	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1130	then
1131	printf "\n trying to submit job to TIT cluster..."
1132	else
1133	printf "\n +++ failed ... trying to submit again ..."
1134	fi
1135	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1136	cat send_protocol
1137	done
1138	sleep 10
1139	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1140	echo "$submcom $job_on_remhost"
1141	rm send_protocol
1142	fi
1143
1144	[[ $verify = true ]] && printf " >>> o.k.\n"
1145	else
1146	cd $job_catalog
1147	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1148	then
1149	eval $submcom $job_on_remhost
1150	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1151	then
1152	eval $submcom $job_on_remhost
1153	echo "$submcom $job_on_remhost"
1154	chmod u+x $job_on_remhost
1155	elif [[ $local_host = nech ]]
1156	then
1157	if [[ $queue = default ]]
1158	then
1159	eval $submcom $job_on_remhost
1160	else
1161	eval $submcom -q $queue $job_on_remhost
1162	fi
1163	else
1164	qsub $job_on_remhost
1165	fi
1166	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1167	[[ $local_host != lctit ]] && rm $job_on_remhost
1168	cd - > /dev/null
1169	fi
1170	fi
1171
1172
1173
1174	# ABSCHLUSSARBEITEN
1175
1176	if [[ $no_submit = false ]]
1177	then
1178	rm $job_to_send
1179	fi
1180	[[ $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 |