Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 366

Last change on this file since 366 was 366, checked in by raasch, 15 years ago
speed optomizations +bugfix in init_ocean
Property svn:keywords set to `Id Rev`
File size: 43.0 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 366 2009-08-25 08:06:27Z 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	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
104	# now run on the data nodes (feature=data)
105
106
107	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
108	delete_dayfile=false
109	email_notification=none
110	group_number=none
111	locat=normal
112	no_default_queue=none
113	no_submit=false
114	job_catalog="~/job_queue"
115	job_name=none
116	local_user=$LOGNAME
117	node_usage=shared
118	numprocs=1
119	punkte="..........................................................."
120	submcom=qsub
121	queue=default
122	remote_host=none
123	remote_user=""
124	verify=true
125
126	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
127	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
128	typeset -L20 spalte1
129	typeset -R40 spalte2
130	typeset -L60 spalte3
131
132
133
134	# FEHLERBEHANDLUNG
135	# BEI EXIT:
136	trap 'if [[ $locat != normal ]]
137	then
138	case $locat in
139	(option) printf "\n --> available optios can be displayed"
140	printf " by typing:"
141	printf "\n \"subjob ?\" \n";;
142	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
143	(*) printf "\n +++ unknown error"
144	printf "\n please inform S. Raasch!\n"
145	esac
146	[[ -f $job_to_send ]] && rm $job_to_send
147	printf "\n\n+++ SUBJOB killed \n\n"
148	fi' exit
149
150
151	# BEI TERMINAL-BREAK:
152	trap '[[ -f $job_to_send ]] && rm $job_to_send
153	printf "\n\n+++ SUBJOB killed \n\n"
154	exit
155	' 2
156
157
158
159
160	# LOKALEN HOSTNAMEN FESTSTELLEN
161	local_host=$(hostname)
162
163
164
165	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
166	# UEBERHAUPT ZULAESSIG IST
167	case $local_host in
168	(atmos) local_addres=172.20.25.35; local_host=lcide;;
169	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
170	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
171	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
172	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
173	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
174	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
175	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
176	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
177	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
178	(cs*) local_addres=136.172.44.131; local_host=nech;;
179	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
180	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
181	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
182	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
183	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
184	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
185	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
186	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
187	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
188	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
189	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
190	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
191	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
192	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
193	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
194	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
195	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
196	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
197	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
198	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
199	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
200	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
201	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
202	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
203	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
204	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
205	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
206	(*) printf "\n +++ \"$local_host\" unknown";
207	printf "\n please inform S. Raasch!";
208	locat=parameter; exit;;
209	esac
210
211
212
213	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
214	remote_host=$local_host
215
216
217
218
219	# PROZEDUROPTIONEN EINLESEN
220	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
221	do
222	case $option in
223	(c) job_catalog=$OPTARG;;
224	(d) delete_dayfile=true;;
225	(D) no_submit=true;;
226	(e) email_notification=$OPTARG;;
227	(g) group_number=$OPTARG;;
228	(h) remote_host=$OPTARG;;
229	(m) memory=$OPTARG;;
230	(n) job_name=$OPTARG;;
231	(N) node_usage=$OPTARG;;
232	(O) threads_per_task=$OPTARG;;
233	(q) no_default_queue=$OPTARG;;
234	(t) cputime=$OPTARG;;
235	(T) tasks_per_node=$OPTARG;;
236	(u) remote_user=$OPTARG;;
237	(v) verify=false;;
238	(X) numprocs=$OPTARG;;
239	(\?) printf "\n +++ Option $OPTARG unknown \n";
240	locat=option; exit;;
241	esac
242	done
243
244
245	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
246	shift OPTIND-1; file_to_send=$1
247
248
249	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
250	if [ "$1" = "?" ]
251	then
252	(printf "\n *** subjob can be called as follows:\n"
253	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
254	printf "\n Description of available options:\n"
255	printf "\n Option Description Default-Value"
256	printf "\n -c job-input- and output-catalog ~/job_queue"
257	printf "\n -d no job-protocol will be created ---"
258	printf "\n -D only the job-file will be created ---"
259	printf "\n -h execution host, available hosts: $remote_host"
260	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
261	printf "\n lctit, nech, necriam, unics"
262	printf "\n -m memory demand per process in MByte ---"
263	printf "\n -n jobname <jobdatei>"
264	printf "\n -O threads per task (for OpenMP usage) 1"
265	printf "\n -q job-queue to be used default"
266	printf "\n -t allowed cpu-time in seconds ---"
267	printf "\n -T tasks per node (on parallel hosts) ---"
268	printf "\n -u username on execution host from .netrc"
269	printf "\n -v no prompt for confirmation ---"
270	printf "\n -X # of processors (on parallel hosts) 1"
271	printf "\n "
272	printf "\n The only possible positional parameter is <jobfile>:"
273	printf "\n The complete NQS-job must be provided here."
274	printf "\n <jobfile>=? creates this outline\n\n") \| more
275	exit
276	fi
277
278
279
280	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
281	if [[ "$file_to_send" = "" ]]
282	then
283	printf "\n +++ job-file missing"
284	locat=parameter; exit
285	else
286	if [[ -f $file_to_send ]]
287	then
288	true
289	else
290	printf "\n +++ job-file: "
291	printf "\n $file_to_send"
292	printf "\n does not exist"
293	locat=parameter; exit
294	fi
295	fi
296
297
298
299	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
300	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
301	if [[ $job_name = none ]]
302	then
303	job_name=$file_to_send
304	fi
305	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
306	then
307	printf "\n +++ job-file name: "
308	printf "\n $job_name"
309	printf "\n must not contain \"/\"-characters"
310	locat=parameter; exit
311	fi
312
313
314
315
316	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
317	# ODER WENN HOST NICHT ANGEGEBEN WURDE
318	if [[ $remote_host = none ]]
319	then
320	printf "\n +++ host missing"
321	locat=option; exit
322	else
323	case $remote_host in
324	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
325	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
327	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
328	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
329	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
330	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
331	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
332	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
333	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
334	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
335	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
336	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
337	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
338	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
339	locat=parameter; exit;;
340	esac
341	fi
342
343
344	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
345	if [[ $no_default_queue != none ]]
346	then
347	error=false
348	ndq=$no_default_queue
349	case $remote_host in
350	(ibm) case $ndq in
351	(p690_express\|p690_standard\|p690_long) error=false;;
352	(*) error=true;;
353	esac;;
354	(ibmb) case $ndq in
355	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
356	(*) error=true;;
357	esac;;
358	(ibmh) case $ndq in
359	(no_class) error=false;;
360	(*) error=true;;
361	esac;;
362	(ibms) case $ndq in
363	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
364	(*) error=true;;
365	esac;;
366	(ibmy) case $ndq in
367	(parallel) error=false;;
368	(*) error=true;;
369	esac;;
370	(lcsgib\|lcsgih) case $ndq in
371	(testq\|serialq\|smallq\|bigq\|workq) error=false;;
372	(*) error=true;;
373	esac;;
374	(lctit) case $ndq in
375	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
376	(*) error=true;;
377	esac;;
378	(t3eb) case $ndq in
379	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
380	(*) error=true;;
381	esac;;
382	(necriam) case $ndq in
383	(SP\|SS\|P6) error=false;;
384	(*) error=true;;
385	esac;;
386	(t3eh) case $ndq in
387	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
388	(*) error=true;;
389	esac;;
390	(t3ej2\|t3ej5) case $ndq in
391	(low\|normal\|high) error=false;;
392	(*) error=true;;
393	esac;;
394	(t3es) case $ndq in
395	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
396	(*) error=true;;
397	esac;;
398	(unics) case $ndq in
399	(unics\|ht) error=false;;
400	(*) error=true;;
401	esac;;
402	esac
403	if [[ $error = true ]]
404	then
405	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
406	locat=parameter; exit
407	else
408	queue=$no_default_queue
409	fi
410	fi
411
412
413
414	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
415	if [[ $node_usage != shared && $queue = cdev ]]
416	then
417	node_usage=shared
418	fi
419
420
421
422	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
423	# AUFTEILEN
424	done=false
425	while [[ $done = false ]]
426	do
427	if (( $cputime <= 0 ))
428	then
429	printf "\n +++ wrong cpu-time or cpu-time missing"
430	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
431	printf "\n >>> "
432	read cputime 1>/dev/null 2>&1
433	else
434	done=true
435	fi
436	done
437	if [[ $remote_host = nech ]]
438	then
439	if (( tasks_per_node != 0 ))
440	then
441	(( cputime = cputime * tasks_per_node ))
442	elif [[ $numprocs != 0 ]]
443	then
444	(( cputime = cputime * numprocs ))
445	fi
446	fi
447	(( stunden = cputime / 3600 ))
448	(( resttime = cputime - stunden * 3600 ))
449	(( minuten = resttime / 60 ))
450	(( sekunden = resttime - minuten * 60 ))
451	timestring=${stunden}:${minuten}:${sekunden}
452
453
454
455	# PRUEFEN DER KERNSPEICHERANFORDERUNG
456	done=false
457	while [[ $done = false ]]
458	do
459	if (( memory <= 0 ))
460	then
461	printf "\n +++ wrong memory demand or memory demand missing"
462	printf "\n >>> Please type memory in MByte per process as INTEGER:"
463	printf "\n >>> "
464	read memory 1>/dev/null 2>&1
465	else
466	done=true
467	fi
468	done
469
470	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
471	then
472	if (( tasks_per_node != 0 ))
473	then
474	(( Memory = memory * tasks_per_node / 1000 ))
475	elif [[ $numprocs != 0 ]]
476	then
477	(( Memory = memory * numprocs / 1000 ))
478	else
479	(( Memory = memory / 1000 ))
480	fi
481	fi
482
483
484	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
485	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
486	then
487	(( memory = memory * threads_per_task ))
488	fi
489
490
491	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
492	if (( tasks_per_node != 0 ))
493	then
494	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
495	fi
496
497
498
499	# HEADER-AUSGABE
500	if [[ $verify = true ]]
501	then
502	printf "\n\n"
503	printf "#--------------------------------------------------------------# \n"
504	spalte1=SUBJOB;spalte2=$(date)
505	printf "\| $spalte1$spalte2 \| \n"
506	printf "\| \| \n"
507	printf "\| values of parameters/options: \| \n"
508	spalte1=local_host$punkte; spalte2=$punkte$local_host
509	printf "\| $spalte1$spalte2 \| \n"
510	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
511	printf "\| $spalte1$spalte2 \| \n"
512	spalte1=queue$punkte; spalte2=$punkte$queue
513	printf "\| $spalte1$spalte2 \| \n"
514	spalte1=memory$punkte; spalte2="$punkte$memory mb"
515	printf "\| $spalte1$spalte2 \| \n"
516	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
517	printf "\| $spalte1$spalte2 \| \n"
518	spalte1=job_name$punkte; spalte2="$punkte$job_name"
519	printf "\| $spalte1$spalte2 \| \n"
520	printf "#--------------------------------------------------------------# \n\n"
521
522
523	# KONTROLLABFRAGE, OB ALLES O.K.
524	antwort="dummy"
525	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
526	do
527	read antwort?" >>> continue (y/n) ? "
528	done
529	if [[ $antwort = n \|\| $antwort = N ]]
530	then
531	locat=verify; exit
532	fi
533	printf "\n"
534	fi
535
536
537
538	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
539	kennung=$RANDOM
540	job_on_remhost=${job_name}_${kennung}_$local_host
541	job_to_send=job_to_send_$kennung
542	if [[ $delete_dayfile = false ]]
543	then
544	remote_dayfile=${local_host}_${job_name}_result_$kennung
545	local_dayfile=${remote_host}_${job_name}
546	else
547	remote_dayfile=/dev/null
548	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
549	then
550	notification='# @ notification = never'
551	fi
552	fi
553
554
555
556	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
557	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
558	then
559
560	if [[ $remote_host = ibmy ]]
561	then
562	consumable_memory=""
563	else
564	consumable_memory="ConsumableMemory($memory mb)"
565	fi
566
567	if [[ $remote_host = ibmh ]]
568	then
569	data_limit=""
570	network_to_use=""
571	class=""
572	environment=""
573	else
574	class="# @ class = $queue"
575	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
576	if [[ $queue = cdev ]]
577	then
578	data_limit="# @ data_limit = 1.76gb"
579	network_to_use="# @ network.mpi = sn_all,shared,ip"
580	else
581	if [[ $remote_host = ibms ]]
582	then
583	network_to_use="# @ network.mpi = csss,shared,us"
584	elif [[ $remote_host = ibmy ]]
585	then
586	network_to_use=""
587	else
588	network_to_use="# @ network.mpi = sn_all,shared,us"
589	data_limit="# @ data_limit = 1.76gb"
590	fi
591	fi
592	fi
593
594	cat > $job_to_send << %%END%%
595	#!/bin/ksh
596	# @ shell = /bin/ksh
597
598	# @ job_type = parallel
599	# @ job_name = palm
600	# @ wall_clock_limit = ${timestring},$timestring
601	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
602	# @ output = $remote_dayfile
603	# @ error = $remote_dayfile
604	# @ image_size = 50
605	$class
606	$environment
607	$network_to_use
608	$data_limit
609	$notification
610
611	%%END%%
612
613	if (( nodes > 0 ))
614	then
615
616	cat >> $job_to_send << %%END%%
617	# @ node = $nodes
618	# @ tasks_per_node = $tasks_per_node
619	# @ node_usage = $node_usage
620	# @ queue
621
622	%%END%%
623
624	else
625
626	if [[ $remote_host != ibmy ]]
627	then
628
629	cat >> $job_to_send << %%END%%
630	# @ blocking = unlimited
631	# @ total_tasks = $numprocs
632	# @ node_usage = $node_usage
633	# @ queue
634
635	%%END%%
636
637	else
638
639	cat >> $job_to_send << %%END%%
640	# @ node = 1
641	# @ total_tasks = $numprocs
642	# @ queue
643
644	%%END%%
645
646	fi
647
648	fi
649
650	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
651	then
652
653	cat > $job_to_send << %%END%%
654	#!/bin/ksh
655
656	# @ job_type = serial
657	# @ node_usage = $node_usage
658	# @ job_name = palm
659	# @ wall_clock_limit = ${timestring},$timestring
660	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
661	# @ output = $remote_dayfile
662	# @ error = $remote_dayfile
663	$class
664	$notification
665
666	# @ queue
667
668	%%END%%
669
670	elif [[ $remote_host = lcfimm ]]
671	then
672
673	if [[ $numprocs != 0 ]]
674	then
675	cat > $job_to_send << %%END%%
676	#!/bin/ksh
677	#PBS -N $job_name
678	#PBS -A geofysisk
679	#PBS -l walltime=$timestring
680	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
681	#PBS -l pmem=${memory}mb
682	#PBS -m abe
683	#PBS -M maronga@muk.uni-hannover.de
684	#PBS -o $remote_dayfile
685	#PBS -j oe
686	mpd &
687
688	%%END%%
689
690	else
691	cat > $job_to_send << %%END%%
692	#!/bin/ksh
693	#PBS -N $job_name
694	#PBS -A geofysisk
695	#PBS -l walltime=$timestring
696	#PBS -l ncpus=1
697	#PBS -l pmem=${memory}mb
698	#PBS -m abe
699	#PBS -M maronga@muk.uni-hannover.de
700	#PBS -o $remote_dayfile
701	#PBS -j oe
702
703	%%END%%
704
705	fi
706
707	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
708	then
709
710	if [[ $queue = testq \|\| $queue = workq ]]
711	then
712	queue_directive="#PBS -q $queue"
713	else
714	queue_directive=""
715	fi
716
717	if [[ $email_notification = none ]]
718	then
719	email_directive=""
720	else
721	email_directive="#PBS -M $email_notification"
722	fi
723
724	if [[ $numprocs != 0 ]]
725	then
726	cat > $job_to_send << %%END%%
727	#!/bin/bash
728	#PBS -N $job_name
729	#PBS -l walltime=$timestring
730	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
731	#PBS -l feature=$sgi_feature
732	#PBS -l naccesspolicy=$node_usage
733	#PBS -o $remote_dayfile
734	#PBS -j oe
735	$queue_directive
736	$email_directive
737
738	. /usr/share/modules/init/bash
739	module load ifort/11.0.069
740	module load $mpilib
741	# module load mvapich2/1.2rc1-intel
742	# module load mvapich2
743	module load netcdf
744	#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
745	echo ld_library_path=\$LD_LIBRARY_PATH
746
747	%%END%%
748
749	else
750	cat > $job_to_send << %%END%%
751	#PBS -S /bin/bash
752	#PBS -N $job_name
753	#PBS -l walltime=$timestring
754	#PBS -l ncpus=1
755	#PBS -o $remote_dayfile
756	#PBS -e $remote_dayfile
757
758	%%END%%
759
760	fi
761
762	elif [[ $remote_host = lcxt4 ]]
763	then
764
765	if [[ $numprocs != 0 ]]
766	then
767	cat > $job_to_send << %%END%%
768	#!/bin/ksh
769	#PBS -S /bin/ksh
770	#PBS -N $job_name
771	#PBS -A geofysisk
772	#PBS -j oe
773	#PBS -l walltime=$timestring
774	#PBS -l mppwidth=${numprocs}
775	#PBS -l mppnppn=${tasks_per_node}
776	#PBS -m abe
777	#PBS -M maronga@muk.uni-hannover.de
778	#PBS -o $remote_dayfile
779
780	. /opt/modules/default/init/ksh
781	module load modules/3.1.6
782	module load PrgEnv-pgi/2.1.56HD
783	module load Base-opts/2.1.56HD
784	module load moab/5.2.5-s13441
785	module load torque/2.3.4-snap.200809221601
786
787	%%END%%
788
789	else
790	cat > $job_to_send << %%END%%
791	#!/bin/ksh
792	#PBS -S /bin/ksh
793	#PBS -N $job_name
794	#PBS -A geofysisk
795	#PBS -j oe
796	#PBS -l walltime=$timestring
797	#PBS -l ncpus=1
798	#PBS -l pmem=${memory}mb
799	#PBS -m abe
800	#PBS -M maronga@muk.uni-hannover.de
801	#PBS -o $remote_dayfile
802
803	. /opt/modules/default/init/ksh
804	module load modules/3.1.6
805	module load PrgEnv-pgi/2.1.56HD
806	module load Base-opts/2.1.56HD
807	module load moab/5.2.5-s13441
808	module load torque/2.3.4-snap.200809221601
809
810	%%END%%
811
812	fi
813
814	elif [[ $remote_host = nech ]]
815	then
816
817	if (( nodes > 1 ))
818	then
819	cat > $job_to_send << %%END%%
820	#!/bin/ksh
821	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
822	#PBS -l ${qsubmem}=${Memory}gb
823	#PBS -b $nodes
824	#PBS -o $remote_dayfile
825	#PBS -N palm
826	#PBS -j o
827	#PBS -T mpisx
828
829	%%END%%
830
831	elif [[ $numprocs != 0 ]]
832	then
833	cat > $job_to_send << %%END%%
834	#!/bin/ksh
835	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
836	#PBS -l ${qsubmem}=${Memory}gb
837	#PBS -o $remote_dayfile
838	#PBS -N palm
839	#PBS -j o
840
841	%%END%%
842
843	else
844	cat > $job_to_send << %%END%%
845	#!/bin/ksh
846	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
847	#PBS -o $remote_dayfile
848	#PBS -j o
849
850	%%END%%
851
852	fi
853
854	elif [[ $remote_host = necriam ]]
855	then
856
857	if (( nodes > 1 ))
858	then
859	cat > $job_to_send << %%END%%
860	#!/bin/ksh
861	#PBS -b $nodes
862	#PBS -o $remote_dayfile
863	#PBS -N $job_name
864	#PBS -j o
865	#PBS -v MPIPROGINV=YES
866
867	%%END%%
868
869	elif [[ $numprocs != 0 ]]
870	then
871	cat > $job_to_send << %%END%%
872	#!/bin/ksh
873	#PBS -q ${queue}
874	#PBS -o $remote_dayfile
875	#PBS -N $job_name
876	#PBS -j o
877	#PBS -v MPIPROGINV=YES
878
879	%%END%%
880
881	else
882	cat > $job_to_send << %%END%%
883	#!/bin/ksh
884	#PBS -o $remote_dayfile
885	#PBS -j o
886
887	%%END%%
888
889	fi
890
891	elif [[ $remote_host = lctit ]]
892	then
893	cat > $job_to_send << %%END%%
894	#!/bin/bash
895	#$ -S /bin/bash
896	cd $job_catalog
897	export PATH=$PALM_BIN:\$PATH
898	echo \$PATH
899
900	%%END%%
901
902	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
903	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
904	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
905
906	else
907
908	cat > $job_to_send << %%END%%
909	# @\$-q ${queue}
910	# @\$-l${qsubtime} $timestring
911	# @\$-l${qsubmem} ${memory}mb
912	# @\$-o $remote_dayfile
913	# @\$-eo
914
915	%%END%%
916
917	fi
918
919
920	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
921	# BEI EXIT VERANLASSEN
922	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
923	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
924	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
925	then
926	echo "set +vx" >> $job_to_send
927	echo "trap '" >> $job_to_send
928	echo "set +vx" >> $job_to_send
929	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 ]]
930	then
931	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
932	then
933	return_queue=c1
934	elif [[ $remote_host = ibms ]]
935	then
936	return_queue=p_normal
937	elif [[ $remote_host = ibmy ]]
938	then
939	return_queue=serial
940	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
941	then
942	return_queue=serialq
943	elif [[ $remote_host = lctit ]]
944	then
945	return_queue=sla3
946	elif [[ $remote_host = necriam ]]
947	then
948	return_queue=SP
949	else
950	return_queue=unknown
951	fi
952
953	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
954	then
955
956	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
957	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
958	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
959	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
960	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
961	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
962	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
963	if [[ $host != "ibmh" ]]
964	then
965	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
966	fi
967	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
968	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
969
970	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
971	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
972
973	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
974	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
975	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
976
977	elif [[ $remote_host = nech ]]
978	then
979	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
980	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
981	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
982	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
983	echo "#PBS -j o" >> $job_to_send
984	echo " " >> $job_to_send
985	echo "set -x" >> $job_to_send
986	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
987	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
988	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
989	echo "%%END%%" >> $job_to_send
990
991	elif [[ $remote_host = necriam ]]
992	then
993	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
994	echo "#PBS -q $return_queue" >> $job_to_send
995	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
996	echo "#PBS -j o" >> $job_to_send
997	echo " " >> $job_to_send
998	echo "set -x" >> $job_to_send
999	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1000	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1001	echo "%%END%%" >> $job_to_send
1002
1003	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1004	then
1005	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1006	echo "#!/bin/bash" >> $job_to_send
1007	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1008	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1009	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1010	echo "#PBS -l feature=data" >> $job_to_send
1011	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1012	echo "#PBS -j oe" >> $job_to_send
1013	echo " " >> $job_to_send
1014	echo ". /usr/share/modules/init/bash" >> $job_to_send
1015	echo "set -x" >> $job_to_send
1016	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1017	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1018	echo "%%END%%" >> $job_to_send
1019
1020	elif [[ $remote_host = lctit ]]
1021	then
1022	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1023	echo "set -x" >> $job_to_send
1024	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1025	echo "cd $job_catalog" >> $job_to_send
1026	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1027	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1028	echo "%%END%%" >> $job_to_send
1029
1030	else
1031
1032	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1033	echo "# @\\\$-q $return_queue" >> $job_to_send
1034	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1035	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1036	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1037	then
1038	echo "# @\$-l mpp_p=0" >> $job_to_send
1039	fi
1040	echo '# @\$-lF 10mb' >> $job_to_send
1041	# echo '# @\$-o /dev/null' >> $job_to_send
1042	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1043	echo '# @\\\$-eo' >> $job_to_send
1044	echo " " >> $job_to_send
1045	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1046	then
1047	echo "set +vx" >> $job_to_send
1048	echo ". .profile" >> $job_to_send
1049	fi
1050	echo "set -x" >> $job_to_send
1051	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1052	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1053	echo "%%END%%" >> $job_to_send
1054	fi
1055
1056	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1057	then
1058	echo "llsubmit scpjob.$kennung" >> $job_to_send
1059	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1060	then
1061	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1062	echo "msub scpjob.$kennung" >> $job_to_send
1063	elif [[ $remote_host = lctit ]]
1064	then
1065	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1066	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1067	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1068	echo "sleep 10" >> $job_to_send
1069	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1070	then
1071	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1072	elif [[ $remote_host = t3es ]]
1073	then
1074	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1075	else
1076	echo "qsub scpjob.$kennung" >> $job_to_send
1077	fi
1078	echo "rm scpjob.$kennung" >> $job_to_send
1079	if [[ $remote_host = nech ]]
1080	then
1081	echo "cd -" >> $job_to_send
1082	fi
1083	else
1084	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1085	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1086	fi
1087	echo "set -x" >> $job_to_send
1088	echo " ' exit" >> $job_to_send
1089	echo "set -x" >> $job_to_send
1090	fi
1091
1092
1093
1094	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1095	cat $file_to_send >> $job_to_send
1096
1097	if [[ $remote_host = ibm ]]
1098	then
1099	echo " " >> $job_to_send
1100	echo "exit" >> $job_to_send
1101	fi
1102	if [[ $remote_host = lctit ]]
1103	then
1104	echo " " >> $job_to_send
1105	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1106	fi
1107
1108
1109
1110	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1111	if [[ -z $remote_user ]]
1112	then
1113	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1114	then
1115	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1116	fi
1117	fi
1118
1119
1120
1121	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1122	if [[ $no_submit = false ]]
1123	then
1124	if [[ $remote_host != $local_host ]]
1125	then
1126	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1127	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1128	then
1129	job_catalog_save=$job_catalog
1130	job_catalog=job_queue
1131	elif [[ $remote_host = nech ]]
1132	then
1133	job_catalog_save=$job_catalog
1134	job_catalog=/hpf/b/${remote_user}/job_queue
1135	fi
1136	if [[ $local_host = decalpha ]]
1137	then
1138	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1139	# PASSWORT AB
1140	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1141	elif [[ $remote_host = nech ]]
1142	then
1143	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1144	# TRANSFERIERT WERDEN
1145	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1146	else
1147	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1148	fi
1149	if [[ $? = 1 ]]
1150	then
1151	locat=scp; exit
1152	fi
1153	if [[ $remote_host = ibms ]]
1154	then
1155	job_catalog=$job_catalog_save
1156	fi
1157	[[ $verify = true ]] && printf "\n >>> finished\n"
1158	else
1159	eval job_catalog=$job_catalog
1160	cp $job_to_send ${job_catalog}/$job_on_remhost
1161	fi
1162
1163
1164
1165	# NQS- BZW. LOADLEVELER-JOB STARTEN
1166	if [[ $remote_host != $local_host ]]
1167	then
1168	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1169	if [[ $remote_host != lctit ]]
1170	then
1171	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1172	else
1173	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1174	# MIT SSH, DESHALB AUFRUF PER PIPE
1175	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1176	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1177	echo "first try" > send_protocol
1178	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1179	do
1180	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1181	then
1182	printf "\n trying to submit job to TIT cluster..."
1183	else
1184	printf "\n +++ failed ... trying to submit again ..."
1185	fi
1186	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1187	cat send_protocol
1188	done
1189	sleep 10
1190	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1191	echo "$submcom $job_on_remhost"
1192	rm send_protocol
1193	fi
1194
1195	[[ $verify = true ]] && printf " >>> o.k.\n"
1196	else
1197	cd $job_catalog
1198	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1199	then
1200	eval $submcom $job_on_remhost
1201	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1202	then
1203	eval $submcom $job_on_remhost
1204	echo "$submcom $job_on_remhost"
1205	chmod u+x $job_on_remhost
1206	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1207	then
1208	eval $submcom $job_on_remhost
1209	elif [[ $local_host = nech ]]
1210	then
1211	if [[ $queue = default ]]
1212	then
1213	eval $submcom $job_on_remhost
1214	else
1215	eval $submcom -q $queue $job_on_remhost
1216	fi
1217	else
1218	qsub $job_on_remhost
1219	fi
1220	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1221	[[ $local_host != lctit ]] && rm $job_on_remhost
1222	cd - > /dev/null
1223	fi
1224	fi
1225
1226
1227
1228	# ABSCHLUSSARBEITEN
1229
1230	if [[ $no_submit = false ]]
1231	then
1232	rm $job_to_send
1233	fi
1234	[[ $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 |