Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 344

Last change on this file since 344 was 343, checked in by maronga, 16 years ago
adjustments for lcxt4 and ibmy, allow user 2d xy cross section output at z=nzb+1
Property svn:keywords set to `Id Rev`
File size: 42.5 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 343 2009-06-24 12:59:09Z maronga $
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
103
104	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
105	delete_dayfile=false
106	group_number=none
107	locat=normal
108	no_default_queue=none
109	no_submit=false
110	job_catalog="~/job_queue"
111	job_name=none
112	local_user=$LOGNAME
113	node_usage=shared
114	numprocs=1
115	punkte="..........................................................."
116	submcom=qsub
117	queue=default
118	remote_host=none
119	remote_user=""
120	verify=true
121
122	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
123	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
124	typeset -L20 spalte1
125	typeset -R40 spalte2
126	typeset -L60 spalte3
127
128
129
130	# FEHLERBEHANDLUNG
131	# BEI EXIT:
132	trap 'if [[ $locat != normal ]]
133	then
134	case $locat in
135	(option) printf "\n --> available optios can be displayed"
136	printf " by typing:"
137	printf "\n \"subjob ?\" \n";;
138	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
139	(*) printf "\n +++ unknown error"
140	printf "\n please inform S. Raasch!\n"
141	esac
142	[[ -f $job_to_send ]] && rm $job_to_send
143	printf "\n\n+++ SUBJOB killed \n\n"
144	fi' exit
145
146
147	# BEI TERMINAL-BREAK:
148	trap '[[ -f $job_to_send ]] && rm $job_to_send
149	printf "\n\n+++ SUBJOB killed \n\n"
150	exit
151	' 2
152
153
154
155
156	# LOKALEN HOSTNAMEN FESTSTELLEN
157	local_host=$(hostname)
158
159
160
161	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
162	# UEBERHAUPT ZULAESSIG IST
163	case $local_host in
164	(atmos) local_addres=172.20.25.35; local_host=lcide;;
165	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
166	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
167	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
168	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
169	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
170	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
171	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
172	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
173	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
174	(cs*) local_addres=136.172.44.131; local_host=nech;;
175	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
176	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
177	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
178	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
179	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
180	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
181	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
182	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
183	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
184	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
185	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
186	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
187	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
188	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
189	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
190	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
191	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
192	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
193	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
194	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
195	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
196	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
197	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
198	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
199	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
200	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
201	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
202	(*) printf "\n +++ \"$local_host\" unknown";
203	printf "\n please inform S. Raasch!";
204	locat=parameter; exit;;
205	esac
206
207
208
209	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
210	remote_host=$local_host
211
212
213
214
215	# PROZEDUROPTIONEN EINLESEN
216	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
217	do
218	case $option in
219	(c) job_catalog=$OPTARG;;
220	(d) delete_dayfile=true;;
221	(D) no_submit=true;;
222	(g) group_number=$OPTARG;;
223	(h) remote_host=$OPTARG;;
224	(m) memory=$OPTARG;;
225	(n) job_name=$OPTARG;;
226	(N) node_usage=$OPTARG;;
227	(O) threads_per_task=$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	shift OPTIND-1; file_to_send=$1
242
243
244	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
245	if [ "$1" = "?" ]
246	then
247	(printf "\n *** subjob can be called as follows:\n"
248	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
249	printf "\n Description of available options:\n"
250	printf "\n Option Description Default-Value"
251	printf "\n -c job-input- and output-catalog ~/job_queue"
252	printf "\n -d no job-protocol will be created ---"
253	printf "\n -D only the job-file will be created ---"
254	printf "\n -h execution host, available hosts: $remote_host"
255	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
256	printf "\n lctit, nech, necriam, unics"
257	printf "\n -m memory demand per process in MByte ---"
258	printf "\n -n jobname <jobdatei>"
259	printf "\n -O threads per task (for OpenMP usage) 1"
260	printf "\n -q job-queue to be used default"
261	printf "\n -t allowed cpu-time in seconds ---"
262	printf "\n -T tasks per node (on parallel hosts) ---"
263	printf "\n -u username on execution host from .netrc"
264	printf "\n -v no prompt for confirmation ---"
265	printf "\n -X # of processors (on parallel hosts) 1"
266	printf "\n "
267	printf "\n The only possible positional parameter is <jobfile>:"
268	printf "\n The complete NQS-job must be provided here."
269	printf "\n <jobfile>=? creates this outline\n\n") \| more
270	exit
271	fi
272
273
274
275	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
276	if [[ "$file_to_send" = "" ]]
277	then
278	printf "\n +++ job-file missing"
279	locat=parameter; exit
280	else
281	if [[ -f $file_to_send ]]
282	then
283	true
284	else
285	printf "\n +++ job-file: "
286	printf "\n $file_to_send"
287	printf "\n does not exist"
288	locat=parameter; exit
289	fi
290	fi
291
292
293
294	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
295	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
296	if [[ $job_name = none ]]
297	then
298	job_name=$file_to_send
299	fi
300	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
301	then
302	printf "\n +++ job-file name: "
303	printf "\n $job_name"
304	printf "\n must not contain \"/\"-characters"
305	locat=parameter; exit
306	fi
307
308
309
310
311	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
312	# ODER WENN HOST NICHT ANGEGEBEN WURDE
313	if [[ $remote_host = none ]]
314	then
315	printf "\n +++ host missing"
316	locat=option; exit
317	else
318	case $remote_host in
319	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
320	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
321	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
322	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
323	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
324	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
325	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
326	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
327	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
328	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
329	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
330	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
331	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
332	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
333	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
334	locat=parameter; exit;;
335	esac
336	fi
337
338
339	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
340	if [[ $no_default_queue != none ]]
341	then
342	error=false
343	ndq=$no_default_queue
344	case $remote_host in
345	(ibm) case $ndq in
346	(p690_express\|p690_standard\|p690_long) error=false;;
347	(*) error=true;;
348	esac;;
349	(ibmb) case $ndq in
350	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
351	(*) error=true;;
352	esac;;
353	(ibmh) case $ndq in
354	(no_class) error=false;;
355	(*) error=true;;
356	esac;;
357	(ibms) case $ndq in
358	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
359	(*) error=true;;
360	esac;;
361	(ibmy) case $ndq in
362	(parallel) error=false;;
363	(*) error=true;;
364	esac;;
365	(lcsgib\|lcsgih) case $ndq in
366	(testq\|serialq\|smallq\|bigq) error=false;;
367	(*) error=true;;
368	esac;;
369	(lctit) case $ndq in
370	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
371	(*) error=true;;
372	esac;;
373	(t3eb) case $ndq in
374	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
375	(*) error=true;;
376	esac;;
377	(necriam) case $ndq in
378	(SP\|SS\|P6) error=false;;
379	(*) error=true;;
380	esac;;
381	(t3eh) case $ndq in
382	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
383	(*) error=true;;
384	esac;;
385	(t3ej2\|t3ej5) case $ndq in
386	(low\|normal\|high) error=false;;
387	(*) error=true;;
388	esac;;
389	(t3es) case $ndq in
390	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
391	(*) error=true;;
392	esac;;
393	(unics) case $ndq in
394	(unics\|ht) error=false;;
395	(*) error=true;;
396	esac;;
397	esac
398	if [[ $error = true ]]
399	then
400	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
401	locat=parameter; exit
402	else
403	queue=$no_default_queue
404	fi
405	fi
406
407
408
409	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
410	if [[ $node_usage != shared && $queue = cdev ]]
411	then
412	node_usage=shared
413	fi
414
415
416
417	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
418	# AUFTEILEN
419	done=false
420	while [[ $done = false ]]
421	do
422	if (( $cputime <= 0 ))
423	then
424	printf "\n +++ wrong cpu-time or cpu-time missing"
425	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
426	printf "\n >>> "
427	read cputime 1>/dev/null 2>&1
428	else
429	done=true
430	fi
431	done
432	if [[ $remote_host = nech ]]
433	then
434	if (( tasks_per_node != 0 ))
435	then
436	(( cputime = cputime * tasks_per_node ))
437	elif [[ $numprocs != 0 ]]
438	then
439	(( cputime = cputime * numprocs ))
440	fi
441	fi
442	(( stunden = cputime / 3600 ))
443	(( resttime = cputime - stunden * 3600 ))
444	(( minuten = resttime / 60 ))
445	(( sekunden = resttime - minuten * 60 ))
446	timestring=${stunden}:${minuten}:${sekunden}
447
448
449
450	# PRUEFEN DER KERNSPEICHERANFORDERUNG
451	done=false
452	while [[ $done = false ]]
453	do
454	if (( memory <= 0 ))
455	then
456	printf "\n +++ wrong memory demand or memory demand missing"
457	printf "\n >>> Please type memory in MByte per process as INTEGER:"
458	printf "\n >>> "
459	read memory 1>/dev/null 2>&1
460	else
461	done=true
462	fi
463	done
464
465	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
466	then
467	if (( tasks_per_node != 0 ))
468	then
469	(( Memory = memory * tasks_per_node / 1000 ))
470	elif [[ $numprocs != 0 ]]
471	then
472	(( Memory = memory * numprocs / 1000 ))
473	else
474	(( Memory = memory / 1000 ))
475	fi
476	fi
477
478
479	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
480	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
481	then
482	(( memory = memory * threads_per_task ))
483	fi
484
485
486	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
487	if (( tasks_per_node != 0 ))
488	then
489	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
490	fi
491
492
493
494	# HEADER-AUSGABE
495	if [[ $verify = true ]]
496	then
497	printf "\n\n"
498	printf "#--------------------------------------------------------------# \n"
499	spalte1=SUBJOB;spalte2=$(date)
500	printf "\| $spalte1$spalte2 \| \n"
501	printf "\| \| \n"
502	printf "\| values of parameters/options: \| \n"
503	spalte1=local_host$punkte; spalte2=$punkte$local_host
504	printf "\| $spalte1$spalte2 \| \n"
505	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
506	printf "\| $spalte1$spalte2 \| \n"
507	spalte1=queue$punkte; spalte2=$punkte$queue
508	printf "\| $spalte1$spalte2 \| \n"
509	spalte1=memory$punkte; spalte2="$punkte$memory mb"
510	printf "\| $spalte1$spalte2 \| \n"
511	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
512	printf "\| $spalte1$spalte2 \| \n"
513	spalte1=job_name$punkte; spalte2="$punkte$job_name"
514	printf "\| $spalte1$spalte2 \| \n"
515	printf "#--------------------------------------------------------------# \n\n"
516
517
518	# KONTROLLABFRAGE, OB ALLES O.K.
519	antwort="dummy"
520	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
521	do
522	read antwort?" >>> continue (y/n) ? "
523	done
524	if [[ $antwort = n \|\| $antwort = N ]]
525	then
526	locat=verify; exit
527	fi
528	printf "\n"
529	fi
530
531
532
533	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
534	kennung=$RANDOM
535	job_on_remhost=${job_name}_${kennung}_$local_host
536	job_to_send=job_to_send_$kennung
537	if [[ $delete_dayfile = false ]]
538	then
539	remote_dayfile=${local_host}_${job_name}_result_$kennung
540	local_dayfile=${remote_host}_${job_name}
541	else
542	remote_dayfile=/dev/null
543	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
544	then
545	notification='# @ notification = never'
546	fi
547	fi
548
549
550
551	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
552	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
553	then
554
555	if [[ $remote_host = ibmy ]]
556	then
557	consumable_memory=""
558	else
559	consumable_memory="ConsumableMemory($memory mb)"
560	fi
561
562	if [[ $remote_host = ibmh ]]
563	then
564	data_limit=""
565	network_to_use=""
566	class=""
567	environment=""
568	else
569	class="# @ class = $queue"
570	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
571	if [[ $queue = cdev ]]
572	then
573	data_limit="# @ data_limit = 1.76gb"
574	network_to_use="# @ network.mpi = sn_all,shared,ip"
575	else
576	if [[ $remote_host = ibms ]]
577	then
578	network_to_use="# @ network.mpi = csss,shared,us"
579	elif [[ $remote_host = ibmy ]]
580	then
581	network_to_use=""
582	else
583	network_to_use="# @ network.mpi = sn_all,shared,us"
584	data_limit="# @ data_limit = 1.76gb"
585	fi
586	fi
587	fi
588
589	cat > $job_to_send << %%END%%
590	#!/bin/ksh
591	# @ shell = /bin/ksh
592
593	# @ job_type = parallel
594	# @ job_name = palm
595	# @ wall_clock_limit = ${timestring},$timestring
596	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
597	# @ output = $remote_dayfile
598	# @ error = $remote_dayfile
599	# @ image_size = 50
600	$class
601	$environment
602	$network_to_use
603	$data_limit
604	$notification
605
606	%%END%%
607
608	if (( nodes > 0 ))
609	then
610
611	cat >> $job_to_send << %%END%%
612	# @ node = $nodes
613	# @ tasks_per_node = $tasks_per_node
614	# @ node_usage = $node_usage
615	# @ queue
616
617	%%END%%
618
619	else
620
621	if [[ $remote_host != ibmy ]]
622	then
623
624	cat >> $job_to_send << %%END%%
625	# @ blocking = unlimited
626	# @ total_tasks = $numprocs
627	# @ node_usage = $node_usage
628	# @ queue
629
630	%%END%%
631
632	else
633
634	cat >> $job_to_send << %%END%%
635	# @ node = 1
636	# @ total_tasks = $numprocs
637	# @ queue
638
639	%%END%%
640
641	fi
642
643	fi
644
645	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
646	then
647
648	cat > $job_to_send << %%END%%
649	#!/bin/ksh
650
651	# @ job_type = serial
652	# @ node_usage = $node_usage
653	# @ job_name = palm
654	# @ wall_clock_limit = ${timestring},$timestring
655	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
656	# @ output = $remote_dayfile
657	# @ error = $remote_dayfile
658	$class
659	$notification
660
661	# @ queue
662
663	%%END%%
664
665	elif [[ $remote_host = lcfimm ]]
666	then
667
668	if [[ $numprocs != 0 ]]
669	then
670	cat > $job_to_send << %%END%%
671	#!/bin/ksh
672	#PBS -N $job_name
673	#PBS -A geofysisk
674	#PBS -l walltime=$timestring
675	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
676	#PBS -l pmem=${memory}mb
677	#PBS -m abe
678	#PBS -M maronga@muk.uni-hannover.de
679	#PBS -o $remote_dayfile
680	#PBS -j oe
681	mpd &
682
683	%%END%%
684
685	else
686	cat > $job_to_send << %%END%%
687	#!/bin/ksh
688	#PBS -N $job_name
689	#PBS -A geofysisk
690	#PBS -l walltime=$timestring
691	#PBS -l ncpus=1
692	#PBS -l pmem=${memory}mb
693	#PBS -m abe
694	#PBS -M maronga@muk.uni-hannover.de
695	#PBS -o $remote_dayfile
696	#PBS -j oe
697
698	%%END%%
699
700	fi
701
702	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
703	then
704
705	if [[ $queue = testq ]]
706	then
707	queue_directive="#PBS -q $queue"
708	else
709	queue_directive=""
710	fi
711
712	if [[ $numprocs != 0 ]]
713	then
714	cat > $job_to_send << %%END%%
715	#!/bin/bash
716	#PBS -N $job_name
717	#PBS -l walltime=$timestring
718	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
719	#PBS -l feature=$sgi_feature
720	#PBS -l naccesspolicy=$node_usage
721	#PBS -o $remote_dayfile
722	#PBS -j oe
723	$queue_directive
724
725	. /usr/share/modules/init/bash
726	module load $mpilib
727	# module load mvapich2/1.2rc1-intel
728	# module load mvapich2
729	module load netcdf
730	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
731	echo ld_library_path=\$LD_LIBRARY_PATH
732
733	%%END%%
734
735	else
736	cat > $job_to_send << %%END%%
737	#PBS -S /bin/bash
738	#PBS -N $job_name
739	#PBS -l walltime=$timestring
740	#PBS -l ncpus=1
741	#PBS -o $remote_dayfile
742	#PBS -e $remote_dayfile
743
744	%%END%%
745
746	fi
747
748	elif [[ $remote_host = lcxt4 ]]
749	then
750
751	if [[ $numprocs != 0 ]]
752	then
753	cat > $job_to_send << %%END%%
754	#!/bin/ksh
755	#PBS -S /bin/ksh
756	#PBS -N $job_name
757	#PBS -A geofysisk
758	#PBS -j oe
759	#PBS -l walltime=$timestring
760	#PBS -l mppwidth=${numprocs}
761	#PBS -l mppnppn=${tasks_per_node}
762	#PBS -m abe
763	#PBS -M maronga@muk.uni-hannover.de
764	#PBS -o $remote_dayfile
765
766	. /opt/modules/default/init/ksh
767	module load modules/3.1.6
768	module load PrgEnv-pgi/2.1.56HD
769	module load Base-opts/2.1.56HD
770	module load moab/5.2.5-s13441
771	module load torque/2.3.4-snap.200809221601
772
773	%%END%%
774
775	else
776	cat > $job_to_send << %%END%%
777	#!/bin/ksh
778	#PBS -S /bin/ksh
779	#PBS -N $job_name
780	#PBS -A geofysisk
781	#PBS -j oe
782	#PBS -l walltime=$timestring
783	#PBS -l ncpus=1
784	#PBS -l pmem=${memory}mb
785	#PBS -m abe
786	#PBS -M maronga@muk.uni-hannover.de
787	#PBS -o $remote_dayfile
788
789	. /opt/modules/default/init/ksh
790	module load modules/3.1.6
791	module load PrgEnv-pgi/2.1.56HD
792	module load Base-opts/2.1.56HD
793	module load moab/5.2.5-s13441
794	module load torque/2.3.4-snap.200809221601
795
796	%%END%%
797
798	fi
799
800	elif [[ $remote_host = nech ]]
801	then
802
803	if (( nodes > 1 ))
804	then
805	cat > $job_to_send << %%END%%
806	#!/bin/ksh
807	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
808	#PBS -l ${qsubmem}=${Memory}gb
809	#PBS -b $nodes
810	#PBS -o $remote_dayfile
811	#PBS -N palm
812	#PBS -j o
813	#PBS -T mpisx
814
815	%%END%%
816
817	elif [[ $numprocs != 0 ]]
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 -o $remote_dayfile
824	#PBS -N palm
825	#PBS -j o
826
827	%%END%%
828
829	else
830	cat > $job_to_send << %%END%%
831	#!/bin/ksh
832	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
833	#PBS -o $remote_dayfile
834	#PBS -j o
835
836	%%END%%
837
838	fi
839
840	elif [[ $remote_host = necriam ]]
841	then
842
843	if (( nodes > 1 ))
844	then
845	cat > $job_to_send << %%END%%
846	#!/bin/ksh
847	#PBS -b $nodes
848	#PBS -o $remote_dayfile
849	#PBS -N $job_name
850	#PBS -j o
851	#PBS -v MPIPROGINV=YES
852
853	%%END%%
854
855	elif [[ $numprocs != 0 ]]
856	then
857	cat > $job_to_send << %%END%%
858	#!/bin/ksh
859	#PBS -q ${queue}
860	#PBS -o $remote_dayfile
861	#PBS -N $job_name
862	#PBS -j o
863	#PBS -v MPIPROGINV=YES
864
865	%%END%%
866
867	else
868	cat > $job_to_send << %%END%%
869	#!/bin/ksh
870	#PBS -o $remote_dayfile
871	#PBS -j o
872
873	%%END%%
874
875	fi
876
877	elif [[ $remote_host = lctit ]]
878	then
879	cat > $job_to_send << %%END%%
880	#!/bin/bash
881	#$ -S /bin/bash
882	cd $job_catalog
883	export PATH=$PALM_BIN:\$PATH
884	echo \$PATH
885
886	%%END%%
887
888	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
889	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
890	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
891
892	else
893
894	cat > $job_to_send << %%END%%
895	# @\$-q ${queue}
896	# @\$-l${qsubtime} $timestring
897	# @\$-l${qsubmem} ${memory}mb
898	# @\$-o $remote_dayfile
899	# @\$-eo
900
901	%%END%%
902
903	fi
904
905
906	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
907	# BEI EXIT VERANLASSEN
908	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
909	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
910	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
911	then
912	echo "set +vx" >> $job_to_send
913	echo "trap '" >> $job_to_send
914	echo "set +vx" >> $job_to_send
915	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 ]]
916	then
917	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
918	then
919	return_queue=c1
920	elif [[ $remote_host = ibms ]]
921	then
922	return_queue=p_normal
923	elif [[ $remote_host = ibmy ]]
924	then
925	return_queue=serial
926	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
927	then
928	return_queue=serialq
929	elif [[ $remote_host = lctit ]]
930	then
931	return_queue=sla3
932	elif [[ $remote_host = necriam ]]
933	then
934	return_queue=SP
935	else
936	return_queue=unknown
937	fi
938
939	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
940	then
941
942	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
943	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
944	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
945	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
946	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
947	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
948	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
949	if [[ $host != "ibmh" ]]
950	then
951	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
952	fi
953	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
954	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
955
956	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
957	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
958
959	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
960	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
961	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
962
963	elif [[ $remote_host = nech ]]
964	then
965	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
966	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
967	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
968	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
969	echo "#PBS -j o" >> $job_to_send
970	echo " " >> $job_to_send
971	echo "set -x" >> $job_to_send
972	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
973	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
974	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
975	echo "%%END%%" >> $job_to_send
976
977	elif [[ $remote_host = necriam ]]
978	then
979	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
980	echo "#PBS -q $return_queue" >> $job_to_send
981	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
982	echo "#PBS -j o" >> $job_to_send
983	echo " " >> $job_to_send
984	echo "set -x" >> $job_to_send
985	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
986	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
987	echo "%%END%%" >> $job_to_send
988
989	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
990	then
991	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
992	echo "#!/bin/bash" >> $job_to_send
993	echo "#PBS -N job_protocol_transfer" >> $job_to_send
994	echo "#PBS -l walltime=00:30:00" >> $job_to_send
995	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
996	echo "#PBS -l feature=xe" >> $job_to_send
997	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
998	echo "#PBS -j oe" >> $job_to_send
999	echo " " >> $job_to_send
1000	echo ". /usr/share/modules/init/bash" >> $job_to_send
1001	echo "set -x" >> $job_to_send
1002	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1003	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1004	echo "%%END%%" >> $job_to_send
1005
1006	elif [[ $remote_host = lctit ]]
1007	then
1008	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1009	echo "set -x" >> $job_to_send
1010	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1011	echo "cd $job_catalog" >> $job_to_send
1012	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1013	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1014	echo "%%END%%" >> $job_to_send
1015
1016	else
1017
1018	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1019	echo "# @\\\$-q $return_queue" >> $job_to_send
1020	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1021	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1022	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1023	then
1024	echo "# @\$-l mpp_p=0" >> $job_to_send
1025	fi
1026	echo '# @\$-lF 10mb' >> $job_to_send
1027	# echo '# @\$-o /dev/null' >> $job_to_send
1028	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1029	echo '# @\\\$-eo' >> $job_to_send
1030	echo " " >> $job_to_send
1031	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1032	then
1033	echo "set +vx" >> $job_to_send
1034	echo ". .profile" >> $job_to_send
1035	fi
1036	echo "set -x" >> $job_to_send
1037	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1038	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1039	echo "%%END%%" >> $job_to_send
1040	fi
1041
1042	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1043	then
1044	echo "llsubmit scpjob.$kennung" >> $job_to_send
1045	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1046	then
1047	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1048	echo "msub scpjob.$kennung" >> $job_to_send
1049	elif [[ $remote_host = lctit ]]
1050	then
1051	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1052	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1053	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1054	echo "sleep 10" >> $job_to_send
1055	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1056	then
1057	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1058	elif [[ $remote_host = t3es ]]
1059	then
1060	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1061	else
1062	echo "qsub scpjob.$kennung" >> $job_to_send
1063	fi
1064	echo "rm scpjob.$kennung" >> $job_to_send
1065	if [[ $remote_host = nech ]]
1066	then
1067	echo "cd -" >> $job_to_send
1068	fi
1069	else
1070	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1071	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1072	fi
1073	echo "set -x" >> $job_to_send
1074	echo " ' exit" >> $job_to_send
1075	echo "set -x" >> $job_to_send
1076	fi
1077
1078
1079
1080	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1081	cat $file_to_send >> $job_to_send
1082
1083	if [[ $remote_host = ibm ]]
1084	then
1085	echo " " >> $job_to_send
1086	echo "exit" >> $job_to_send
1087	fi
1088	if [[ $remote_host = lctit ]]
1089	then
1090	echo " " >> $job_to_send
1091	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1092	fi
1093
1094
1095
1096	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1097	if [[ -z $remote_user ]]
1098	then
1099	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1100	then
1101	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1102	fi
1103	fi
1104
1105
1106
1107	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1108	if [[ $no_submit = false ]]
1109	then
1110	if [[ $remote_host != $local_host ]]
1111	then
1112	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1113	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1114	then
1115	job_catalog_save=$job_catalog
1116	job_catalog=job_queue
1117	elif [[ $remote_host = nech ]]
1118	then
1119	job_catalog_save=$job_catalog
1120	job_catalog=/hpf/b/${remote_user}/job_queue
1121	fi
1122	if [[ $local_host = decalpha ]]
1123	then
1124	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1125	# PASSWORT AB
1126	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1127	elif [[ $remote_host = nech ]]
1128	then
1129	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1130	# TRANSFERIERT WERDEN
1131	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1132	else
1133	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1134	fi
1135	if [[ $? = 1 ]]
1136	then
1137	locat=scp; exit
1138	fi
1139	if [[ $remote_host = ibms ]]
1140	then
1141	job_catalog=$job_catalog_save
1142	fi
1143	[[ $verify = true ]] && printf "\n >>> finished\n"
1144	else
1145	eval job_catalog=$job_catalog
1146	cp $job_to_send ${job_catalog}/$job_on_remhost
1147	fi
1148
1149
1150
1151	# NQS- BZW. LOADLEVELER-JOB STARTEN
1152	if [[ $remote_host != $local_host ]]
1153	then
1154	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1155	if [[ $remote_host != lctit ]]
1156	then
1157	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1158	else
1159	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1160	# MIT SSH, DESHALB AUFRUF PER PIPE
1161	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1162	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1163	echo "first try" > send_protocol
1164	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1165	do
1166	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1167	then
1168	printf "\n trying to submit job to TIT cluster..."
1169	else
1170	printf "\n +++ failed ... trying to submit again ..."
1171	fi
1172	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1173	cat send_protocol
1174	done
1175	sleep 10
1176	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1177	echo "$submcom $job_on_remhost"
1178	rm send_protocol
1179	fi
1180
1181	[[ $verify = true ]] && printf " >>> o.k.\n"
1182	else
1183	cd $job_catalog
1184	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1185	then
1186	eval $submcom $job_on_remhost
1187	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1188	then
1189	eval $submcom $job_on_remhost
1190	echo "$submcom $job_on_remhost"
1191	chmod u+x $job_on_remhost
1192	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1193	then
1194	eval $submcom $job_on_remhost
1195	elif [[ $local_host = nech ]]
1196	then
1197	if [[ $queue = default ]]
1198	then
1199	eval $submcom $job_on_remhost
1200	else
1201	eval $submcom -q $queue $job_on_remhost
1202	fi
1203	else
1204	qsub $job_on_remhost
1205	fi
1206	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1207	[[ $local_host != lctit ]] && rm $job_on_remhost
1208	cd - > /dev/null
1209	fi
1210	fi
1211
1212
1213
1214	# ABSCHLUSSARBEITEN
1215
1216	if [[ $no_submit = false ]]
1217	then
1218	rm $job_to_send
1219	fi
1220	[[ $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 |