Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 364

Last change on this file since 364 was 352, checked in by raasch, 15 years ago
+option -e in subjob
Property svn:keywords set to `Id Rev`
File size: 42.8 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 352 2009-07-08 16:26:02Z letzel $
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
104
105	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
106	delete_dayfile=false
107	email_notification=none
108	group_number=none
109	locat=normal
110	no_default_queue=none
111	no_submit=false
112	job_catalog="~/job_queue"
113	job_name=none
114	local_user=$LOGNAME
115	node_usage=shared
116	numprocs=1
117	punkte="..........................................................."
118	submcom=qsub
119	queue=default
120	remote_host=none
121	remote_user=""
122	verify=true
123
124	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
125	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
126	typeset -L20 spalte1
127	typeset -R40 spalte2
128	typeset -L60 spalte3
129
130
131
132	# FEHLERBEHANDLUNG
133	# BEI EXIT:
134	trap 'if [[ $locat != normal ]]
135	then
136	case $locat in
137	(option) printf "\n --> available optios can be displayed"
138	printf " by typing:"
139	printf "\n \"subjob ?\" \n";;
140	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
141	(*) printf "\n +++ unknown error"
142	printf "\n please inform S. Raasch!\n"
143	esac
144	[[ -f $job_to_send ]] && rm $job_to_send
145	printf "\n\n+++ SUBJOB killed \n\n"
146	fi' exit
147
148
149	# BEI TERMINAL-BREAK:
150	trap '[[ -f $job_to_send ]] && rm $job_to_send
151	printf "\n\n+++ SUBJOB killed \n\n"
152	exit
153	' 2
154
155
156
157
158	# LOKALEN HOSTNAMEN FESTSTELLEN
159	local_host=$(hostname)
160
161
162
163	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
164	# UEBERHAUPT ZULAESSIG IST
165	case $local_host in
166	(atmos) local_addres=172.20.25.35; local_host=lcide;;
167	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
168	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
169	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
170	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
171	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
172	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
173	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
174	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
175	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
176	(cs*) local_addres=136.172.44.131; local_host=nech;;
177	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
178	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
179	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
180	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
181	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
182	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
183	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
184	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
185	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
186	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
187	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
188	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
189	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
190	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
191	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
192	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
193	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
194	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
195	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
196	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
197	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
198	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
199	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
200	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
201	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
202	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
203	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
204	(*) printf "\n +++ \"$local_host\" unknown";
205	printf "\n please inform S. Raasch!";
206	locat=parameter; exit;;
207	esac
208
209
210
211	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
212	remote_host=$local_host
213
214
215
216
217	# PROZEDUROPTIONEN EINLESEN
218	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
219	do
220	case $option in
221	(c) job_catalog=$OPTARG;;
222	(d) delete_dayfile=true;;
223	(D) no_submit=true;;
224	(e) email_notification=$OPTARG;;
225	(g) group_number=$OPTARG;;
226	(h) remote_host=$OPTARG;;
227	(m) memory=$OPTARG;;
228	(n) job_name=$OPTARG;;
229	(N) node_usage=$OPTARG;;
230	(O) threads_per_task=$OPTARG;;
231	(q) no_default_queue=$OPTARG;;
232	(t) cputime=$OPTARG;;
233	(T) tasks_per_node=$OPTARG;;
234	(u) remote_user=$OPTARG;;
235	(v) verify=false;;
236	(X) numprocs=$OPTARG;;
237	(\?) printf "\n +++ Option $OPTARG unknown \n";
238	locat=option; exit;;
239	esac
240	done
241
242
243	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
244	shift OPTIND-1; file_to_send=$1
245
246
247	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
248	if [ "$1" = "?" ]
249	then
250	(printf "\n *** subjob can be called as follows:\n"
251	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
252	printf "\n Description of available options:\n"
253	printf "\n Option Description Default-Value"
254	printf "\n -c job-input- and output-catalog ~/job_queue"
255	printf "\n -d no job-protocol will be created ---"
256	printf "\n -D only the job-file will be created ---"
257	printf "\n -h execution host, available hosts: $remote_host"
258	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
259	printf "\n lctit, nech, necriam, unics"
260	printf "\n -m memory demand per process in MByte ---"
261	printf "\n -n jobname <jobdatei>"
262	printf "\n -O threads per task (for OpenMP usage) 1"
263	printf "\n -q job-queue to be used default"
264	printf "\n -t allowed cpu-time in seconds ---"
265	printf "\n -T tasks per node (on parallel hosts) ---"
266	printf "\n -u username on execution host from .netrc"
267	printf "\n -v no prompt for confirmation ---"
268	printf "\n -X # of processors (on parallel hosts) 1"
269	printf "\n "
270	printf "\n The only possible positional parameter is <jobfile>:"
271	printf "\n The complete NQS-job must be provided here."
272	printf "\n <jobfile>=? creates this outline\n\n") \| more
273	exit
274	fi
275
276
277
278	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
279	if [[ "$file_to_send" = "" ]]
280	then
281	printf "\n +++ job-file missing"
282	locat=parameter; exit
283	else
284	if [[ -f $file_to_send ]]
285	then
286	true
287	else
288	printf "\n +++ job-file: "
289	printf "\n $file_to_send"
290	printf "\n does not exist"
291	locat=parameter; exit
292	fi
293	fi
294
295
296
297	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
298	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
299	if [[ $job_name = none ]]
300	then
301	job_name=$file_to_send
302	fi
303	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
304	then
305	printf "\n +++ job-file name: "
306	printf "\n $job_name"
307	printf "\n must not contain \"/\"-characters"
308	locat=parameter; exit
309	fi
310
311
312
313
314	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
315	# ODER WENN HOST NICHT ANGEGEBEN WURDE
316	if [[ $remote_host = none ]]
317	then
318	printf "\n +++ host missing"
319	locat=option; exit
320	else
321	case $remote_host in
322	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
323	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
324	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
325	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
326	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
327	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
328	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
329	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
330	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
331	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
332	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
333	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
334	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
335	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
336	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
337	locat=parameter; exit;;
338	esac
339	fi
340
341
342	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
343	if [[ $no_default_queue != none ]]
344	then
345	error=false
346	ndq=$no_default_queue
347	case $remote_host in
348	(ibm) case $ndq in
349	(p690_express\|p690_standard\|p690_long) error=false;;
350	(*) error=true;;
351	esac;;
352	(ibmb) case $ndq in
353	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
354	(*) error=true;;
355	esac;;
356	(ibmh) case $ndq in
357	(no_class) error=false;;
358	(*) error=true;;
359	esac;;
360	(ibms) case $ndq in
361	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
362	(*) error=true;;
363	esac;;
364	(ibmy) case $ndq in
365	(parallel) error=false;;
366	(*) error=true;;
367	esac;;
368	(lcsgib\|lcsgih) case $ndq in
369	(testq\|serialq\|smallq\|bigq) error=false;;
370	(*) error=true;;
371	esac;;
372	(lctit) case $ndq in
373	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
374	(*) error=true;;
375	esac;;
376	(t3eb) case $ndq in
377	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
378	(*) error=true;;
379	esac;;
380	(necriam) case $ndq in
381	(SP\|SS\|P6) error=false;;
382	(*) error=true;;
383	esac;;
384	(t3eh) case $ndq in
385	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
386	(*) error=true;;
387	esac;;
388	(t3ej2\|t3ej5) case $ndq in
389	(low\|normal\|high) error=false;;
390	(*) error=true;;
391	esac;;
392	(t3es) case $ndq in
393	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
394	(*) error=true;;
395	esac;;
396	(unics) case $ndq in
397	(unics\|ht) error=false;;
398	(*) error=true;;
399	esac;;
400	esac
401	if [[ $error = true ]]
402	then
403	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
404	locat=parameter; exit
405	else
406	queue=$no_default_queue
407	fi
408	fi
409
410
411
412	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
413	if [[ $node_usage != shared && $queue = cdev ]]
414	then
415	node_usage=shared
416	fi
417
418
419
420	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
421	# AUFTEILEN
422	done=false
423	while [[ $done = false ]]
424	do
425	if (( $cputime <= 0 ))
426	then
427	printf "\n +++ wrong cpu-time or cpu-time missing"
428	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
429	printf "\n >>> "
430	read cputime 1>/dev/null 2>&1
431	else
432	done=true
433	fi
434	done
435	if [[ $remote_host = nech ]]
436	then
437	if (( tasks_per_node != 0 ))
438	then
439	(( cputime = cputime * tasks_per_node ))
440	elif [[ $numprocs != 0 ]]
441	then
442	(( cputime = cputime * numprocs ))
443	fi
444	fi
445	(( stunden = cputime / 3600 ))
446	(( resttime = cputime - stunden * 3600 ))
447	(( minuten = resttime / 60 ))
448	(( sekunden = resttime - minuten * 60 ))
449	timestring=${stunden}:${minuten}:${sekunden}
450
451
452
453	# PRUEFEN DER KERNSPEICHERANFORDERUNG
454	done=false
455	while [[ $done = false ]]
456	do
457	if (( memory <= 0 ))
458	then
459	printf "\n +++ wrong memory demand or memory demand missing"
460	printf "\n >>> Please type memory in MByte per process as INTEGER:"
461	printf "\n >>> "
462	read memory 1>/dev/null 2>&1
463	else
464	done=true
465	fi
466	done
467
468	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
469	then
470	if (( tasks_per_node != 0 ))
471	then
472	(( Memory = memory * tasks_per_node / 1000 ))
473	elif [[ $numprocs != 0 ]]
474	then
475	(( Memory = memory * numprocs / 1000 ))
476	else
477	(( Memory = memory / 1000 ))
478	fi
479	fi
480
481
482	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
483	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
484	then
485	(( memory = memory * threads_per_task ))
486	fi
487
488
489	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
490	if (( tasks_per_node != 0 ))
491	then
492	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
493	fi
494
495
496
497	# HEADER-AUSGABE
498	if [[ $verify = true ]]
499	then
500	printf "\n\n"
501	printf "#--------------------------------------------------------------# \n"
502	spalte1=SUBJOB;spalte2=$(date)
503	printf "\| $spalte1$spalte2 \| \n"
504	printf "\| \| \n"
505	printf "\| values of parameters/options: \| \n"
506	spalte1=local_host$punkte; spalte2=$punkte$local_host
507	printf "\| $spalte1$spalte2 \| \n"
508	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
509	printf "\| $spalte1$spalte2 \| \n"
510	spalte1=queue$punkte; spalte2=$punkte$queue
511	printf "\| $spalte1$spalte2 \| \n"
512	spalte1=memory$punkte; spalte2="$punkte$memory mb"
513	printf "\| $spalte1$spalte2 \| \n"
514	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
515	printf "\| $spalte1$spalte2 \| \n"
516	spalte1=job_name$punkte; spalte2="$punkte$job_name"
517	printf "\| $spalte1$spalte2 \| \n"
518	printf "#--------------------------------------------------------------# \n\n"
519
520
521	# KONTROLLABFRAGE, OB ALLES O.K.
522	antwort="dummy"
523	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
524	do
525	read antwort?" >>> continue (y/n) ? "
526	done
527	if [[ $antwort = n \|\| $antwort = N ]]
528	then
529	locat=verify; exit
530	fi
531	printf "\n"
532	fi
533
534
535
536	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
537	kennung=$RANDOM
538	job_on_remhost=${job_name}_${kennung}_$local_host
539	job_to_send=job_to_send_$kennung
540	if [[ $delete_dayfile = false ]]
541	then
542	remote_dayfile=${local_host}_${job_name}_result_$kennung
543	local_dayfile=${remote_host}_${job_name}
544	else
545	remote_dayfile=/dev/null
546	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
547	then
548	notification='# @ notification = never'
549	fi
550	fi
551
552
553
554	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
555	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
556	then
557
558	if [[ $remote_host = ibmy ]]
559	then
560	consumable_memory=""
561	else
562	consumable_memory="ConsumableMemory($memory mb)"
563	fi
564
565	if [[ $remote_host = ibmh ]]
566	then
567	data_limit=""
568	network_to_use=""
569	class=""
570	environment=""
571	else
572	class="# @ class = $queue"
573	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
574	if [[ $queue = cdev ]]
575	then
576	data_limit="# @ data_limit = 1.76gb"
577	network_to_use="# @ network.mpi = sn_all,shared,ip"
578	else
579	if [[ $remote_host = ibms ]]
580	then
581	network_to_use="# @ network.mpi = csss,shared,us"
582	elif [[ $remote_host = ibmy ]]
583	then
584	network_to_use=""
585	else
586	network_to_use="# @ network.mpi = sn_all,shared,us"
587	data_limit="# @ data_limit = 1.76gb"
588	fi
589	fi
590	fi
591
592	cat > $job_to_send << %%END%%
593	#!/bin/ksh
594	# @ shell = /bin/ksh
595
596	# @ job_type = parallel
597	# @ job_name = palm
598	# @ wall_clock_limit = ${timestring},$timestring
599	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
600	# @ output = $remote_dayfile
601	# @ error = $remote_dayfile
602	# @ image_size = 50
603	$class
604	$environment
605	$network_to_use
606	$data_limit
607	$notification
608
609	%%END%%
610
611	if (( nodes > 0 ))
612	then
613
614	cat >> $job_to_send << %%END%%
615	# @ node = $nodes
616	# @ tasks_per_node = $tasks_per_node
617	# @ node_usage = $node_usage
618	# @ queue
619
620	%%END%%
621
622	else
623
624	if [[ $remote_host != ibmy ]]
625	then
626
627	cat >> $job_to_send << %%END%%
628	# @ blocking = unlimited
629	# @ total_tasks = $numprocs
630	# @ node_usage = $node_usage
631	# @ queue
632
633	%%END%%
634
635	else
636
637	cat >> $job_to_send << %%END%%
638	# @ node = 1
639	# @ total_tasks = $numprocs
640	# @ queue
641
642	%%END%%
643
644	fi
645
646	fi
647
648	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
649	then
650
651	cat > $job_to_send << %%END%%
652	#!/bin/ksh
653
654	# @ job_type = serial
655	# @ node_usage = $node_usage
656	# @ job_name = palm
657	# @ wall_clock_limit = ${timestring},$timestring
658	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
659	# @ output = $remote_dayfile
660	# @ error = $remote_dayfile
661	$class
662	$notification
663
664	# @ queue
665
666	%%END%%
667
668	elif [[ $remote_host = lcfimm ]]
669	then
670
671	if [[ $numprocs != 0 ]]
672	then
673	cat > $job_to_send << %%END%%
674	#!/bin/ksh
675	#PBS -N $job_name
676	#PBS -A geofysisk
677	#PBS -l walltime=$timestring
678	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
679	#PBS -l pmem=${memory}mb
680	#PBS -m abe
681	#PBS -M maronga@muk.uni-hannover.de
682	#PBS -o $remote_dayfile
683	#PBS -j oe
684	mpd &
685
686	%%END%%
687
688	else
689	cat > $job_to_send << %%END%%
690	#!/bin/ksh
691	#PBS -N $job_name
692	#PBS -A geofysisk
693	#PBS -l walltime=$timestring
694	#PBS -l ncpus=1
695	#PBS -l pmem=${memory}mb
696	#PBS -m abe
697	#PBS -M maronga@muk.uni-hannover.de
698	#PBS -o $remote_dayfile
699	#PBS -j oe
700
701	%%END%%
702
703	fi
704
705	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
706	then
707
708	if [[ $queue = testq ]]
709	then
710	queue_directive="#PBS -q $queue"
711	else
712	queue_directive=""
713	fi
714
715	if [[ $email_notification = none ]]
716	then
717	email_directive=""
718	else
719	email_directive="#PBS -M $email_notification"
720	fi
721
722	if [[ $numprocs != 0 ]]
723	then
724	cat > $job_to_send << %%END%%
725	#!/bin/bash
726	#PBS -N $job_name
727	#PBS -l walltime=$timestring
728	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
729	#PBS -l feature=$sgi_feature
730	#PBS -l naccesspolicy=$node_usage
731	#PBS -o $remote_dayfile
732	#PBS -j oe
733	$queue_directive
734	$email_directive
735
736	. /usr/share/modules/init/bash
737	module load $mpilib
738	# module load mvapich2/1.2rc1-intel
739	# module load mvapich2
740	module load netcdf
741	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
742	echo ld_library_path=\$LD_LIBRARY_PATH
743
744	%%END%%
745
746	else
747	cat > $job_to_send << %%END%%
748	#PBS -S /bin/bash
749	#PBS -N $job_name
750	#PBS -l walltime=$timestring
751	#PBS -l ncpus=1
752	#PBS -o $remote_dayfile
753	#PBS -e $remote_dayfile
754
755	%%END%%
756
757	fi
758
759	elif [[ $remote_host = lcxt4 ]]
760	then
761
762	if [[ $numprocs != 0 ]]
763	then
764	cat > $job_to_send << %%END%%
765	#!/bin/ksh
766	#PBS -S /bin/ksh
767	#PBS -N $job_name
768	#PBS -A geofysisk
769	#PBS -j oe
770	#PBS -l walltime=$timestring
771	#PBS -l mppwidth=${numprocs}
772	#PBS -l mppnppn=${tasks_per_node}
773	#PBS -m abe
774	#PBS -M maronga@muk.uni-hannover.de
775	#PBS -o $remote_dayfile
776
777	. /opt/modules/default/init/ksh
778	module load modules/3.1.6
779	module load PrgEnv-pgi/2.1.56HD
780	module load Base-opts/2.1.56HD
781	module load moab/5.2.5-s13441
782	module load torque/2.3.4-snap.200809221601
783
784	%%END%%
785
786	else
787	cat > $job_to_send << %%END%%
788	#!/bin/ksh
789	#PBS -S /bin/ksh
790	#PBS -N $job_name
791	#PBS -A geofysisk
792	#PBS -j oe
793	#PBS -l walltime=$timestring
794	#PBS -l ncpus=1
795	#PBS -l pmem=${memory}mb
796	#PBS -m abe
797	#PBS -M maronga@muk.uni-hannover.de
798	#PBS -o $remote_dayfile
799
800	. /opt/modules/default/init/ksh
801	module load modules/3.1.6
802	module load PrgEnv-pgi/2.1.56HD
803	module load Base-opts/2.1.56HD
804	module load moab/5.2.5-s13441
805	module load torque/2.3.4-snap.200809221601
806
807	%%END%%
808
809	fi
810
811	elif [[ $remote_host = nech ]]
812	then
813
814	if (( nodes > 1 ))
815	then
816	cat > $job_to_send << %%END%%
817	#!/bin/ksh
818	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
819	#PBS -l ${qsubmem}=${Memory}gb
820	#PBS -b $nodes
821	#PBS -o $remote_dayfile
822	#PBS -N palm
823	#PBS -j o
824	#PBS -T mpisx
825
826	%%END%%
827
828	elif [[ $numprocs != 0 ]]
829	then
830	cat > $job_to_send << %%END%%
831	#!/bin/ksh
832	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
833	#PBS -l ${qsubmem}=${Memory}gb
834	#PBS -o $remote_dayfile
835	#PBS -N palm
836	#PBS -j o
837
838	%%END%%
839
840	else
841	cat > $job_to_send << %%END%%
842	#!/bin/ksh
843	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
844	#PBS -o $remote_dayfile
845	#PBS -j o
846
847	%%END%%
848
849	fi
850
851	elif [[ $remote_host = necriam ]]
852	then
853
854	if (( nodes > 1 ))
855	then
856	cat > $job_to_send << %%END%%
857	#!/bin/ksh
858	#PBS -b $nodes
859	#PBS -o $remote_dayfile
860	#PBS -N $job_name
861	#PBS -j o
862	#PBS -v MPIPROGINV=YES
863
864	%%END%%
865
866	elif [[ $numprocs != 0 ]]
867	then
868	cat > $job_to_send << %%END%%
869	#!/bin/ksh
870	#PBS -q ${queue}
871	#PBS -o $remote_dayfile
872	#PBS -N $job_name
873	#PBS -j o
874	#PBS -v MPIPROGINV=YES
875
876	%%END%%
877
878	else
879	cat > $job_to_send << %%END%%
880	#!/bin/ksh
881	#PBS -o $remote_dayfile
882	#PBS -j o
883
884	%%END%%
885
886	fi
887
888	elif [[ $remote_host = lctit ]]
889	then
890	cat > $job_to_send << %%END%%
891	#!/bin/bash
892	#$ -S /bin/bash
893	cd $job_catalog
894	export PATH=$PALM_BIN:\$PATH
895	echo \$PATH
896
897	%%END%%
898
899	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
900	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
901	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
902
903	else
904
905	cat > $job_to_send << %%END%%
906	# @\$-q ${queue}
907	# @\$-l${qsubtime} $timestring
908	# @\$-l${qsubmem} ${memory}mb
909	# @\$-o $remote_dayfile
910	# @\$-eo
911
912	%%END%%
913
914	fi
915
916
917	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
918	# BEI EXIT VERANLASSEN
919	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
920	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
921	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
922	then
923	echo "set +vx" >> $job_to_send
924	echo "trap '" >> $job_to_send
925	echo "set +vx" >> $job_to_send
926	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 ]]
927	then
928	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
929	then
930	return_queue=c1
931	elif [[ $remote_host = ibms ]]
932	then
933	return_queue=p_normal
934	elif [[ $remote_host = ibmy ]]
935	then
936	return_queue=serial
937	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
938	then
939	return_queue=serialq
940	elif [[ $remote_host = lctit ]]
941	then
942	return_queue=sla3
943	elif [[ $remote_host = necriam ]]
944	then
945	return_queue=SP
946	else
947	return_queue=unknown
948	fi
949
950	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
951	then
952
953	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
954	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
955	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
956	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
957	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
958	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
959	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
960	if [[ $host != "ibmh" ]]
961	then
962	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
963	fi
964	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
965	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
966
967	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
968	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
969
970	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
971	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
972	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
973
974	elif [[ $remote_host = nech ]]
975	then
976	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
977	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
978	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
979	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
980	echo "#PBS -j o" >> $job_to_send
981	echo " " >> $job_to_send
982	echo "set -x" >> $job_to_send
983	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
984	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
985	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
986	echo "%%END%%" >> $job_to_send
987
988	elif [[ $remote_host = necriam ]]
989	then
990	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
991	echo "#PBS -q $return_queue" >> $job_to_send
992	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
993	echo "#PBS -j o" >> $job_to_send
994	echo " " >> $job_to_send
995	echo "set -x" >> $job_to_send
996	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
997	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
998	echo "%%END%%" >> $job_to_send
999
1000	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1001	then
1002	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1003	echo "#!/bin/bash" >> $job_to_send
1004	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1005	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1006	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1007	echo "#PBS -l feature=xe" >> $job_to_send
1008	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1009	echo "#PBS -j oe" >> $job_to_send
1010	echo " " >> $job_to_send
1011	echo ". /usr/share/modules/init/bash" >> $job_to_send
1012	echo "set -x" >> $job_to_send
1013	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1014	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1015	echo "%%END%%" >> $job_to_send
1016
1017	elif [[ $remote_host = lctit ]]
1018	then
1019	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1020	echo "set -x" >> $job_to_send
1021	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1022	echo "cd $job_catalog" >> $job_to_send
1023	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1024	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1025	echo "%%END%%" >> $job_to_send
1026
1027	else
1028
1029	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1030	echo "# @\\\$-q $return_queue" >> $job_to_send
1031	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1032	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1033	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1034	then
1035	echo "# @\$-l mpp_p=0" >> $job_to_send
1036	fi
1037	echo '# @\$-lF 10mb' >> $job_to_send
1038	# echo '# @\$-o /dev/null' >> $job_to_send
1039	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1040	echo '# @\\\$-eo' >> $job_to_send
1041	echo " " >> $job_to_send
1042	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1043	then
1044	echo "set +vx" >> $job_to_send
1045	echo ". .profile" >> $job_to_send
1046	fi
1047	echo "set -x" >> $job_to_send
1048	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1049	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1050	echo "%%END%%" >> $job_to_send
1051	fi
1052
1053	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1054	then
1055	echo "llsubmit scpjob.$kennung" >> $job_to_send
1056	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1057	then
1058	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1059	echo "msub scpjob.$kennung" >> $job_to_send
1060	elif [[ $remote_host = lctit ]]
1061	then
1062	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1063	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1064	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1065	echo "sleep 10" >> $job_to_send
1066	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1067	then
1068	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1069	elif [[ $remote_host = t3es ]]
1070	then
1071	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1072	else
1073	echo "qsub scpjob.$kennung" >> $job_to_send
1074	fi
1075	echo "rm scpjob.$kennung" >> $job_to_send
1076	if [[ $remote_host = nech ]]
1077	then
1078	echo "cd -" >> $job_to_send
1079	fi
1080	else
1081	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1082	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1083	fi
1084	echo "set -x" >> $job_to_send
1085	echo " ' exit" >> $job_to_send
1086	echo "set -x" >> $job_to_send
1087	fi
1088
1089
1090
1091	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1092	cat $file_to_send >> $job_to_send
1093
1094	if [[ $remote_host = ibm ]]
1095	then
1096	echo " " >> $job_to_send
1097	echo "exit" >> $job_to_send
1098	fi
1099	if [[ $remote_host = lctit ]]
1100	then
1101	echo " " >> $job_to_send
1102	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1103	fi
1104
1105
1106
1107	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1108	if [[ -z $remote_user ]]
1109	then
1110	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1111	then
1112	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1113	fi
1114	fi
1115
1116
1117
1118	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1119	if [[ $no_submit = false ]]
1120	then
1121	if [[ $remote_host != $local_host ]]
1122	then
1123	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1124	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1125	then
1126	job_catalog_save=$job_catalog
1127	job_catalog=job_queue
1128	elif [[ $remote_host = nech ]]
1129	then
1130	job_catalog_save=$job_catalog
1131	job_catalog=/hpf/b/${remote_user}/job_queue
1132	fi
1133	if [[ $local_host = decalpha ]]
1134	then
1135	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1136	# PASSWORT AB
1137	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1138	elif [[ $remote_host = nech ]]
1139	then
1140	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1141	# TRANSFERIERT WERDEN
1142	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1143	else
1144	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1145	fi
1146	if [[ $? = 1 ]]
1147	then
1148	locat=scp; exit
1149	fi
1150	if [[ $remote_host = ibms ]]
1151	then
1152	job_catalog=$job_catalog_save
1153	fi
1154	[[ $verify = true ]] && printf "\n >>> finished\n"
1155	else
1156	eval job_catalog=$job_catalog
1157	cp $job_to_send ${job_catalog}/$job_on_remhost
1158	fi
1159
1160
1161
1162	# NQS- BZW. LOADLEVELER-JOB STARTEN
1163	if [[ $remote_host != $local_host ]]
1164	then
1165	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1166	if [[ $remote_host != lctit ]]
1167	then
1168	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1169	else
1170	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1171	# MIT SSH, DESHALB AUFRUF PER PIPE
1172	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1173	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1174	echo "first try" > send_protocol
1175	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1176	do
1177	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1178	then
1179	printf "\n trying to submit job to TIT cluster..."
1180	else
1181	printf "\n +++ failed ... trying to submit again ..."
1182	fi
1183	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1184	cat send_protocol
1185	done
1186	sleep 10
1187	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1188	echo "$submcom $job_on_remhost"
1189	rm send_protocol
1190	fi
1191
1192	[[ $verify = true ]] && printf " >>> o.k.\n"
1193	else
1194	cd $job_catalog
1195	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1196	then
1197	eval $submcom $job_on_remhost
1198	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1199	then
1200	eval $submcom $job_on_remhost
1201	echo "$submcom $job_on_remhost"
1202	chmod u+x $job_on_remhost
1203	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1204	then
1205	eval $submcom $job_on_remhost
1206	elif [[ $local_host = nech ]]
1207	then
1208	if [[ $queue = default ]]
1209	then
1210	eval $submcom $job_on_remhost
1211	else
1212	eval $submcom -q $queue $job_on_remhost
1213	fi
1214	else
1215	qsub $job_on_remhost
1216	fi
1217	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1218	[[ $local_host != lctit ]] && rm $job_on_remhost
1219	cd - > /dev/null
1220	fi
1221	fi
1222
1223
1224
1225	# ABSCHLUSSARBEITEN
1226
1227	if [[ $no_submit = false ]]
1228	then
1229	rm $job_to_send
1230	fi
1231	[[ $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 |