Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 249

Last change on this file since 249 was 221, checked in by raasch, 16 years ago
new environment variable mpilib in configuration file, used by mrun, mbuild, and subjob; small bugfix concerning t<0 in init_cloud_physics
Property svn:keywords set to `Id Rev`
File size: 41.4 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 221 2009-01-12 15:32:23Z 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
98
99
100	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
101	delete_dayfile=false
102	group_number=none
103	locat=normal
104	no_default_queue=none
105	no_submit=false
106	job_catalog="~/job_queue"
107	job_name=none
108	local_user=$LOGNAME
109	node_usage=shared
110	numprocs=1
111	punkte="..........................................................."
112	submcom=qsub
113	queue=default
114	remote_host=none
115	remote_user=""
116	verify=true
117
118	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
119	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
120	typeset -L20 spalte1
121	typeset -R40 spalte2
122	typeset -L60 spalte3
123
124
125
126	# FEHLERBEHANDLUNG
127	# BEI EXIT:
128	trap 'if [[ $locat != normal ]]
129	then
130	case $locat in
131	(option) printf "\n --> available optios can be displayed"
132	printf " by typing:"
133	printf "\n \"subjob ?\" \n";;
134	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
135	(*) printf "\n +++ unknown error"
136	printf "\n please inform S. Raasch!\n"
137	esac
138	[[ -f $job_to_send ]] && rm $job_to_send
139	printf "\n\n+++ SUBJOB killed \n\n"
140	fi' exit
141
142
143	# BEI TERMINAL-BREAK:
144	trap '[[ -f $job_to_send ]] && rm $job_to_send
145	printf "\n\n+++ SUBJOB killed \n\n"
146	exit
147	' 2
148
149
150
151
152	# LOKALEN HOSTNAMEN FESTSTELLEN
153	local_host=$(hostname)
154
155
156
157	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
158	# UEBERHAUPT ZULAESSIG IST
159	case $local_host in
160	(atmos) local_addres=172.20.25.35; local_host=lcide;;
161	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
162	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
163	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
164	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
165	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
166	(b01*) local_addres=130.73.232.102; local_host=lcsgib;;
167	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
168	(cs*) local_addres=136.172.44.131; local_host=nech;;
169	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
170	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
171	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
172	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
173	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
174	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
175	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
176	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
177	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
178	(hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
179	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
180	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
181	(h01*) local_addres=130.75.4.102; local_host=lcsgih;;
182	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
183	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
184	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
185	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
186	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
187	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
188	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
189	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
190	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
191	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
192	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
193	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
194	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
195	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
196	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
197	(*) printf "\n +++ \"$local_host\" unknown";
198	printf "\n please inform S. Raasch!";
199	locat=parameter; exit;;
200	esac
201
202
203
204	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
205	remote_host=$local_host
206
207
208
209
210	# PROZEDUROPTIONEN EINLESEN
211	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
212	do
213	case $option in
214	(c) job_catalog=$OPTARG;;
215	(d) delete_dayfile=true;;
216	(D) no_submit=true;;
217	(g) group_number=$OPTARG;;
218	(h) remote_host=$OPTARG;;
219	(m) memory=$OPTARG;;
220	(n) job_name=$OPTARG;;
221	(N) node_usage=$OPTARG;;
222	(O) threads_per_task=$OPTARG;;
223	(q) no_default_queue=$OPTARG;;
224	(t) cputime=$OPTARG;;
225	(T) tasks_per_node=$OPTARG;;
226	(u) remote_user=$OPTARG;;
227	(v) verify=false;;
228	(X) numprocs=$OPTARG;;
229	(\?) printf "\n +++ Option $OPTARG unknown \n";
230	locat=option; exit;;
231	esac
232	done
233
234
235	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
236	shift OPTIND-1; file_to_send=$1
237
238
239	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
240	if [ "$1" = "?" ]
241	then
242	(printf "\n *** subjob can be called as follows:\n"
243	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
244	printf "\n Description of available options:\n"
245	printf "\n Option Description Default-Value"
246	printf "\n -c job-input- and output-catalog ~/job_queue"
247	printf "\n -d no job-protocol will be created ---"
248	printf "\n -D only the job-file will be created ---"
249	printf "\n -h execution host, available hosts: $remote_host"
250	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
251	printf "\n lctit, nech, neck, unics"
252	printf "\n -m memory demand per process in MByte ---"
253	printf "\n -n jobname <jobdatei>"
254	printf "\n -O threads per task (for OpenMP usage) 1"
255	printf "\n -q job-queue to be used default"
256	printf "\n -t allowed cpu-time in seconds ---"
257	printf "\n -T tasks per node (on parallel hosts) ---"
258	printf "\n -u username on execution host from .netrc"
259	printf "\n -v no prompt for confirmation ---"
260	printf "\n -X # of processors (on parallel hosts) 1"
261	printf "\n "
262	printf "\n The only possible positional parameter is <jobfile>:"
263	printf "\n The complete NQS-job must be provided here."
264	printf "\n <jobfile>=? creates this outline\n\n") \| more
265	exit
266	fi
267
268
269
270	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
271	if [[ "$file_to_send" = "" ]]
272	then
273	printf "\n +++ job-file missing"
274	locat=parameter; exit
275	else
276	if [[ -f $file_to_send ]]
277	then
278	true
279	else
280	printf "\n +++ job-file: "
281	printf "\n $file_to_send"
282	printf "\n does not exist"
283	locat=parameter; exit
284	fi
285	fi
286
287
288
289	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
290	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
291	if [[ $job_name = none ]]
292	then
293	job_name=$file_to_send
294	fi
295	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
296	then
297	printf "\n +++ job-file name: "
298	printf "\n $job_name"
299	printf "\n must not contain \"/\"-characters"
300	locat=parameter; exit
301	fi
302
303
304
305
306	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
307	# ODER WENN HOST NICHT ANGEGEBEN WURDE
308	if [[ $remote_host = none ]]
309	then
310	printf "\n +++ host missing"
311	locat=option; exit
312	else
313	case $remote_host in
314	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
315	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
316	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
317	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
318	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
319	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
320	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
321	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
322	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
323	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
324	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
325	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
326	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
327	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
328	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
329	locat=parameter; exit;;
330	esac
331	fi
332
333
334	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
335	if [[ $no_default_queue != none ]]
336	then
337	error=false
338	ndq=$no_default_queue
339	case $remote_host in
340	(ibm) case $ndq in
341	(p690_express\|p690_standard\|p690_long) error=false;;
342	(*) error=true;;
343	esac;;
344	(ibmb) case $ndq in
345	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
346	(*) error=true;;
347	esac;;
348	(ibmh) case $ndq in
349	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
350	(*) error=true;;
351	esac;;
352	(ibms) case $ndq in
353	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
354	(*) error=true;;
355	esac;;
356	(ibmy) case $ndq in
357	(parallel) error=false;;
358	(*) error=true;;
359	esac;;
360	(lcsgib\|lcsgih) case $ndq in
361	(testq\|smallq\|bigq) error=false;;
362	(*) error=true;;
363	esac;;
364	(lctit) case $ndq in
365	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
366	(*) error=true;;
367	esac;;
368	(t3eb) case $ndq in
369	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
370	(*) error=true;;
371	esac;;
372	(t3eh) case $ndq in
373	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
374	(*) error=true;;
375	esac;;
376	(t3ej2\|t3ej5) case $ndq in
377	(low\|normal\|high) error=false;;
378	(*) error=true;;
379	esac;;
380	(t3es) case $ndq in
381	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
382	(*) error=true;;
383	esac;;
384	(unics) case $ndq in
385	(unics\|ht) error=false;;
386	(*) error=true;;
387	esac;;
388	esac
389	if [[ $error = true ]]
390	then
391	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
392	locat=parameter; exit
393	else
394	queue=$no_default_queue
395	fi
396	fi
397
398
399
400	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
401	if [[ $node_usage != shared && $queue = cdev ]]
402	then
403	node_usage=shared
404	fi
405
406
407
408	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
409	# AUFTEILEN
410	done=false
411	while [[ $done = false ]]
412	do
413	if (( $cputime <= 0 ))
414	then
415	printf "\n +++ wrong cpu-time or cpu-time missing"
416	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
417	printf "\n >>> "
418	read cputime 1>/dev/null 2>&1
419	else
420	done=true
421	fi
422	done
423	if [[ $remote_host = nech ]]
424	then
425	if (( tasks_per_node != 0 ))
426	then
427	(( cputime = cputime * tasks_per_node ))
428	elif [[ $numprocs != 0 ]]
429	then
430	(( cputime = cputime * numprocs ))
431	fi
432	fi
433	(( stunden = cputime / 3600 ))
434	(( resttime = cputime - stunden * 3600 ))
435	(( minuten = resttime / 60 ))
436	(( sekunden = resttime - minuten * 60 ))
437	timestring=${stunden}:${minuten}:${sekunden}
438
439
440
441	# PRUEFEN DER KERNSPEICHERANFORDERUNG
442	done=false
443	while [[ $done = false ]]
444	do
445	if (( memory <= 0 ))
446	then
447	printf "\n +++ wrong memory demand or memory demand missing"
448	printf "\n >>> Please type memory in MByte per process as INTEGER:"
449	printf "\n >>> "
450	read memory 1>/dev/null 2>&1
451	else
452	done=true
453	fi
454	done
455
456	if [[ $remote_host = nech \|\| $remote_host = neck ]]
457	then
458	if (( tasks_per_node != 0 ))
459	then
460	(( Memory = memory * tasks_per_node / 1000 ))
461	elif [[ $numprocs != 0 ]]
462	then
463	(( Memory = memory * numprocs / 1000 ))
464	else
465	(( Memory = memory / 1000 ))
466	fi
467	fi
468
469
470	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
471	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
472	then
473	(( memory = memory * threads_per_task ))
474	fi
475
476
477	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
478	if (( tasks_per_node != 0 ))
479	then
480	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
481	fi
482
483
484
485	# HEADER-AUSGABE
486	if [[ $verify = true ]]
487	then
488	printf "\n\n"
489	printf "#--------------------------------------------------------------# \n"
490	spalte1=SUBJOB;spalte2=$(date)
491	printf "\| $spalte1$spalte2 \| \n"
492	printf "\| \| \n"
493	printf "\| values of parameters/options: \| \n"
494	spalte1=local_host$punkte; spalte2=$punkte$local_host
495	printf "\| $spalte1$spalte2 \| \n"
496	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
497	printf "\| $spalte1$spalte2 \| \n"
498	spalte1=queue$punkte; spalte2=$punkte$queue
499	printf "\| $spalte1$spalte2 \| \n"
500	spalte1=memory$punkte; spalte2="$punkte$memory mb"
501	printf "\| $spalte1$spalte2 \| \n"
502	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
503	printf "\| $spalte1$spalte2 \| \n"
504	spalte1=job_name$punkte; spalte2="$punkte$job_name"
505	printf "\| $spalte1$spalte2 \| \n"
506	printf "#--------------------------------------------------------------# \n\n"
507
508
509	# KONTROLLABFRAGE, OB ALLES O.K.
510	antwort="dummy"
511	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
512	do
513	read antwort?" >>> continue (y/n) ? "
514	done
515	if [[ $antwort = n \|\| $antwort = N ]]
516	then
517	locat=verify; exit
518	fi
519	printf "\n"
520	fi
521
522
523
524	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
525	kennung=$RANDOM
526	job_on_remhost=${job_name}_${kennung}_$local_host
527	job_to_send=job_to_send_$kennung
528	if [[ $delete_dayfile = false ]]
529	then
530	remote_dayfile=${local_host}_${job_name}_result_$kennung
531	local_dayfile=${remote_host}_${job_name}
532	else
533	remote_dayfile=/dev/null
534	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
535	then
536	notification='# @ notification = never'
537	fi
538	fi
539
540
541
542	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
543	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
544	then
545
546	if [[ $remote_host = ibmy ]]
547	then
548	consumable_memory=""
549	else
550	consumable_memory="ConsumableMemory($memory mb)"
551	fi
552	if [[ $queue = cdev ]]
553	then
554	data_limit="# @ data_limit = 1.76gb"
555	network_to_use="# @ network.mpi = sn_all,shared,ip"
556	else
557	if [[ $remote_host = ibms ]]
558	then
559	network_to_use="# @ network.mpi = csss,shared,us"
560	elif [[ $remote_host = ibmy ]]
561	then
562	network_to_use=""
563	else
564	network_to_use="# @ network.mpi = sn_all,shared,us"
565	data_limit="# @ data_limit = 1.76gb"
566	fi
567	fi
568
569	cat > $job_to_send << %%END%%
570	#!/bin/ksh
571
572	# @ job_type = parallel
573	# @ job_name = palm
574	# @ wall_clock_limit = ${timestring},$timestring
575	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
576	# @ output = $remote_dayfile
577	# @ error = $remote_dayfile
578	# @ image_size = 50
579	# @ class = $queue
580	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
581	$network_to_use
582	$data_limit
583	$notification
584
585	%%END%%
586
587	if (( nodes > 0 ))
588	then
589
590	cat >> $job_to_send << %%END%%
591	# @ node = $nodes
592	# @ tasks_per_node = $tasks_per_node
593	# @ node_usage = $node_usage
594	# @ queue
595
596	%%END%%
597
598	else
599
600	if [[ $remote_host != ibmy ]]
601	then
602
603	cat >> $job_to_send << %%END%%
604	# @ blocking = unlimited
605	# @ total_tasks = $numprocs
606	# @ node_usage = $node_usage
607	# @ queue
608
609	%%END%%
610
611	else
612
613	cat >> $job_to_send << %%END%%
614	# @ node = 1
615	# @ total_tasks = $numprocs
616	# @ queue
617
618	%%END%%
619
620	fi
621
622	fi
623
624	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
625	then
626
627	cat > $job_to_send << %%END%%
628	#!/bin/ksh
629
630	# @ job_type = serial
631	# @ node_usage = $node_usage
632	# @ job_name = palm
633	# @ wall_clock_limit = ${timestring},$timestring
634	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
635	# @ output = $remote_dayfile
636	# @ error = $remote_dayfile
637	# @ class = $queue
638	$notification
639
640	# @ queue
641
642	%%END%%
643
644	elif [[ $remote_host = lcfimm ]]
645	then
646
647	if [[ $numprocs != 0 ]]
648	then
649	cat > $job_to_send << %%END%%
650	#!/bin/ksh
651	#PBS -N $job_name
652	#PBS -A geofysisk
653	#PBS -l walltime=$timestring
654	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
655	#PBS -l pmem=${memory}mb
656	#PBS -m abe
657	#PBS -M bjorn.maronga@student.uib.no
658	#PBS -o $remote_dayfile
659	#PBS -j oe
660	mpd &
661
662	%%END%%
663
664	else
665	cat > $job_to_send << %%END%%
666	#!/bin/ksh
667	#PBS -N $job_name
668	#PBS -A geofysisk
669	#PBS -l walltime=$timestring
670	#PBS -l ncpus=1
671	#PBS -l pmem=${memory}mb
672	#PBS -m abe
673	#PBS -M bjorn.maronga@student.uib.no
674	#PBS -o $remote_dayfile
675	#PBS -j oe
676
677	%%END%%
678
679	fi
680
681	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
682	then
683
684	if [[ $queue = testq ]]
685	then
686	queue_directive="#PBS -q $queue"
687	else
688	queue_directive=""
689	fi
690
691	if [[ $numprocs != 0 ]]
692	then
693	cat > $job_to_send << %%END%%
694	#!/bin/bash
695	#PBS -N $job_name
696	#PBS -l walltime=$timestring
697	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
698	#PBS -l feature=$sgi_feature
699	#PBS -l naccesspolicy=$node_usage
700	#PBS -o $remote_dayfile
701	#PBS -j oe
702	$queue_directive
703
704	. /usr/share/modules/init/bash
705	module load $mpilib
706	# module load mvapich2/1.2rc1-intel
707	# module load mvapich2
708	module load netcdf
709	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
710	echo ld_library_path=\$LD_LIBRARY_PATH
711
712	%%END%%
713
714	else
715	cat > $job_to_send << %%END%%
716	#PBS -S /bin/bash
717	#PBS -N $job_name
718	#PBS -l walltime=$timestring
719	#PBS -l ncpus=1
720	#PBS -o $remote_dayfile
721	#PBS -e $remote_dayfile
722
723	%%END%%
724
725	fi
726
727	elif [[ $remote_host = lcxt4 ]]
728	then
729
730	if [[ $numprocs != 0 ]]
731	then
732	cat > $job_to_send << %%END%%
733	#!/bin/ksh
734	#PBS -S /bin/ksh
735	#PBS -N $job_name
736	#PBS -A geofysisk
737	#PBS -j oe
738	#PBS -l walltime=$timestring
739	#PBS -l mppwidth=${numprocs}
740	#PBS -l mppnppn=${tasks_per_node}
741	#PBS -m abe
742	#PBS -M bjorn.maronga@student.uib.no
743	#PBS -o $remote_dayfile
744
745	%%END%%
746
747	else
748	cat > $job_to_send << %%END%%
749	#!/bin/ksh
750	#PBS -S /bin/ksh
751	#PBS -N $job_name
752	#PBS -A geofysisk
753	#PBS -j oe
754	#PBS -l walltime=$timestring
755	#PBS -l ncpus=1
756	#PBS -l pmem=${memory}mb
757	#PBS -m abe
758	#PBS -M bjorn.maronga@student.uib.no
759	#PBS -o $remote_dayfile
760
761	%%END%%
762
763	fi
764
765	elif [[ $remote_host = nech ]]
766	then
767
768	if (( nodes > 1 ))
769	then
770	cat > $job_to_send << %%END%%
771	#!/bin/ksh
772	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
773	#PBS -l ${qsubmem}=${Memory}gb
774	#PBS -b $nodes
775	#PBS -o $remote_dayfile
776	#PBS -N palm
777	#PBS -j o
778	#PBS -T mpisx
779
780	%%END%%
781
782	elif [[ $numprocs != 0 ]]
783	then
784	cat > $job_to_send << %%END%%
785	#!/bin/ksh
786	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
787	#PBS -l ${qsubmem}=${Memory}gb
788	#PBS -o $remote_dayfile
789	#PBS -N palm
790	#PBS -j o
791
792	%%END%%
793
794	else
795	cat > $job_to_send << %%END%%
796	#!/bin/ksh
797	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
798	#PBS -o $remote_dayfile
799	#PBS -j o
800
801	%%END%%
802
803	fi
804
805	elif [[ $remote_host = neck ]]
806	then
807
808	if (( nodes > 1 ))
809	then
810	cat > $job_to_send << %%END%%
811	#!/bin/ksh
812	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
813	#PBS -l ${qsubmem}=${Memory}gb
814	#PBS -b $nodes
815	#PBS -o $remote_dayfile
816	#PBS -N palm
817	#PBS -j o
818	#PBS -T mpisx
819
820	%%END%%
821
822	elif [[ $numprocs != 0 ]]
823	then
824	cat > $job_to_send << %%END%%
825	#!/bin/ksh
826	#PBS -q ${queue}
827	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
828	#PBS -l ${qsubmem}=${Memory}gb
829	#PBS -o $remote_dayfile
830	#PBS -N palm
831	#PBS -j o
832	#PBS -T mpisx
833
834	%%END%%
835
836	else
837	cat > $job_to_send << %%END%%
838	#!/bin/ksh
839	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
840	#PBS -o $remote_dayfile
841	#PBS -j o
842
843	%%END%%
844
845	fi
846
847	elif [[ $remote_host = lctit ]]
848	then
849	cat > $job_to_send << %%END%%
850	#!/bin/bash
851	#$ -S /bin/bash
852	cd $job_catalog
853	export PATH=$PALM_BIN:\$PATH
854	echo \$PATH
855
856	%%END%%
857
858	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
859	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
860	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
861
862	else
863
864	cat > $job_to_send << %%END%%
865	# @\$-q ${queue}
866	# @\$-l${qsubtime} $timestring
867	# @\$-l${qsubmem} ${memory}mb
868	# @\$-o $remote_dayfile
869	# @\$-eo
870
871	%%END%%
872
873	fi
874
875
876	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
877	# BEI EXIT VERANLASSEN
878	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
879	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
880	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
881	then
882	echo "set +vx" >> $job_to_send
883	echo "trap '" >> $job_to_send
884	echo "set +vx" >> $job_to_send
885	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 ]]
886	then
887	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
888	then
889	return_queue=c1
890	elif [[ $remote_host = ibms ]]
891	then
892	return_queue=p_normal
893	elif [[ $remote_host = ibmy ]]
894	then
895	return_queue=serial
896	elif [[ $remote_host = lctit ]]
897	then
898	return_queue=sla3
899	elif [[ $remote_host = neck ]]
900	then
901	return_queue=S
902	else
903	return_queue=unknown
904	fi
905
906	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
907	then
908
909	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
910	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
911	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
912	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
913	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
914	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
915	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
916	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
917	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
918	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
919
920	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
921	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
922
923	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
924	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
925	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
926
927	elif [[ $remote_host = nech ]]
928	then
929	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
930	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
931	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
932	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
933	echo "#PBS -j o" >> $job_to_send
934	echo " " >> $job_to_send
935	echo "set -x" >> $job_to_send
936	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
937	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
938	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
939	echo "%%END%%" >> $job_to_send
940
941	elif [[ $remote_host = neck ]]
942	then
943	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
944	echo "#PBS -q $return_queue" >> $job_to_send
945	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
946	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
947	echo "#PBS -j o" >> $job_to_send
948	echo " " >> $job_to_send
949	echo "set -x" >> $job_to_send
950	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
951	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
952	echo "%%END%%" >> $job_to_send
953
954	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
955	then
956	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
957	echo "#!/bin/bash" >> $job_to_send
958	echo "#PBS -N job_protocol_transfer" >> $job_to_send
959	echo "#PBS -l walltime=00:02:00" >> $job_to_send
960	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
961	echo "#PBS -l feature=xe" >> $job_to_send
962	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
963	echo "#PBS -j oe" >> $job_to_send
964	echo " " >> $job_to_send
965	echo ". /usr/share/modules/init/bash" >> $job_to_send
966	echo "set -x" >> $job_to_send
967	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
968	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
969	echo "%%END%%" >> $job_to_send
970
971	elif [[ $remote_host = lctit ]]
972	then
973	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
974	echo "set -x" >> $job_to_send
975	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
976	echo "cd $job_catalog" >> $job_to_send
977	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
978	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
979	echo "%%END%%" >> $job_to_send
980
981	else
982
983	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
984	echo "# @\\\$-q $return_queue" >> $job_to_send
985	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
986	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
987	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
988	then
989	echo "# @\$-l mpp_p=0" >> $job_to_send
990	fi
991	echo '# @\$-lF 10mb' >> $job_to_send
992	# echo '# @\$-o /dev/null' >> $job_to_send
993	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
994	echo '# @\\\$-eo' >> $job_to_send
995	echo " " >> $job_to_send
996	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
997	then
998	echo "set +vx" >> $job_to_send
999	echo ". .profile" >> $job_to_send
1000	fi
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 > /dev/null" >> $job_to_send
1003	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1004	echo "%%END%%" >> $job_to_send
1005	fi
1006
1007	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1008	then
1009	echo "llsubmit scpjob.$kennung" >> $job_to_send
1010	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1011	then
1012	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1013	echo "msub scpjob.$kennung" >> $job_to_send
1014	elif [[ $remote_host = lctit ]]
1015	then
1016	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1017	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1018	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1019	echo "sleep 10" >> $job_to_send
1020	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1021	then
1022	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1023	elif [[ $remote_host = t3es ]]
1024	then
1025	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1026	else
1027	echo "qsub scpjob.$kennung" >> $job_to_send
1028	fi
1029	echo "rm scpjob.$kennung" >> $job_to_send
1030	if [[ $remote_host = nech ]]
1031	then
1032	echo "cd -" >> $job_to_send
1033	fi
1034	else
1035	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1036	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1037	fi
1038	echo "set -x" >> $job_to_send
1039	echo " ' exit" >> $job_to_send
1040	echo "set -x" >> $job_to_send
1041	fi
1042
1043
1044
1045	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1046	cat $file_to_send >> $job_to_send
1047
1048	if [[ $remote_host = ibm ]]
1049	then
1050	echo " " >> $job_to_send
1051	echo "exit" >> $job_to_send
1052	fi
1053	if [[ $remote_host = lctit ]]
1054	then
1055	echo " " >> $job_to_send
1056	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1057	fi
1058
1059
1060
1061	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1062	if [[ -z $remote_user ]]
1063	then
1064	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1065	then
1066	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1067	fi
1068	fi
1069
1070
1071
1072	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1073	if [[ $no_submit = false ]]
1074	then
1075	if [[ $remote_host != $local_host ]]
1076	then
1077	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1078	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1079	then
1080	job_catalog_save=$job_catalog
1081	job_catalog=job_queue
1082	elif [[ $remote_host = nech ]]
1083	then
1084	job_catalog_save=$job_catalog
1085	job_catalog=/hpf/b/${remote_user}/job_queue
1086	fi
1087	if [[ $local_host = decalpha ]]
1088	then
1089	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1090	# PASSWORT AB
1091	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1092	elif [[ $remote_host = nech ]]
1093	then
1094	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1095	# TRANSFERIERT WERDEN
1096	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1097	else
1098	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1099	fi
1100	if [[ $? = 1 ]]
1101	then
1102	locat=scp; exit
1103	fi
1104	if [[ $remote_host = ibms ]]
1105	then
1106	job_catalog=$job_catalog_save
1107	fi
1108	[[ $verify = true ]] && printf "\n >>> finished\n"
1109	else
1110	eval job_catalog=$job_catalog
1111	cp $job_to_send ${job_catalog}/$job_on_remhost
1112	fi
1113
1114
1115
1116	# NQS- BZW. LOADLEVELER-JOB STARTEN
1117	if [[ $remote_host != $local_host ]]
1118	then
1119	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1120	if [[ $remote_host != lctit ]]
1121	then
1122	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1123	else
1124	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1125	# MIT SSH, DESHALB AUFRUF PER PIPE
1126	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1127	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1128	echo "first try" > send_protocol
1129	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1130	do
1131	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1132	then
1133	printf "\n trying to submit job to TIT cluster..."
1134	else
1135	printf "\n +++ failed ... trying to submit again ..."
1136	fi
1137	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1138	cat send_protocol
1139	done
1140	sleep 10
1141	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1142	echo "$submcom $job_on_remhost"
1143	rm send_protocol
1144	fi
1145
1146	[[ $verify = true ]] && printf " >>> o.k.\n"
1147	else
1148	cd $job_catalog
1149	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1150	then
1151	eval $submcom $job_on_remhost
1152	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1153	then
1154	eval $submcom $job_on_remhost
1155	echo "$submcom $job_on_remhost"
1156	chmod u+x $job_on_remhost
1157	elif [[ $local_host = nech ]]
1158	then
1159	if [[ $queue = default ]]
1160	then
1161	eval $submcom $job_on_remhost
1162	else
1163	eval $submcom -q $queue $job_on_remhost
1164	fi
1165	else
1166	qsub $job_on_remhost
1167	fi
1168	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1169	[[ $local_host != lctit ]] && rm $job_on_remhost
1170	cd - > /dev/null
1171	fi
1172	fi
1173
1174
1175
1176	# ABSCHLUSSARBEITEN
1177
1178	if [[ $no_submit = false ]]
1179	then
1180	rm $job_to_send
1181	fi
1182	[[ $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 |