Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 377

Last change on this file since 377 was 377, checked in by raasch, 15 years ago
adjustments for openmp usage on sgi-ice
Property svn:keywords set to `Id Rev`
File size: 44.4 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 377 2009-09-04 11:09:00Z raasch $
4
5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
7	# Host zurueckgeschickt werden sollen
8
9
10	# letzte Aenderung:
11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
13	# 06/02/98 - Siggi - berte validiert
14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
15	# moeglich
16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
19	# allgemeine Anpassungen fuer ibm-Rechner
20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
21	# Ruecktransfer des Jobprotokolls fuer diese
22	# Maschine abgeschaltet
23	# 31/10/02 - Siggi - berni validiert
24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
25	# 08/11/02 - Siggi - quanero validiert
26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
27	# 23/01/03 - Siggi - hostname nobel changed to nobela
28	# 06/02/03 - Siggi - gregale validated
29	# 12/02/03 - Siggi - orkan and maestro validated
30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
31	# 12/03/03 - Siggi - nec at DKRZ validated
32	# 13/03/03 - Siggi - new nqs resource variable Memory
33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
34	# node
35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
37	# cxxl added to ibmh
38	# 29/08/03 - Siggi - changes in job queues and communication system on
39	# ibms
40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
41	# 30/10/03 - Siggi - nech is not supported any more
42	# 10/11/03 - Siggi - nech2 renamed to nech
43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
45	# 31/03/04 - Siggi - new option -N for node usage
46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
47	# due to error in ssh installation (otherwise a prompt
48	# for the password appears)
49	# 23/07/04 - Siggi - changes due to the new berni configuration
50	# (federation switch)
51	# 01/09/04 - Gerald new job-classes on hanni
52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
54	# 03/12/04 - Siggi - notification on ibm switched of in case of
55	# delete_dayfile = true, node usage in cdev set to
56	# shared
57	# 16/02/05 - Gerald hababai validated
58	# 29/03/05 - Micha - new job class channi on hanni
59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
60	# on ibms
61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
62	# p_normal
63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
65	# "sn_all", new job class cexp
66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
68	# hurricane (with -X0)
69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
70	# added
71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
72	# 09/02/06 - Siggi - ibmy admitted
73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
74	# 13/04/06 - Siggi - ostria admitted
75	# 18/04/06 - Siggi - new option -O for OpenMP usage
76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
80	# 28/11/06 - Siggi - levanto admitted
81	# 13/02/07 - Siggi - hpmuk releated code removed
82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
84	# 27/07/07 - Siggi - autan admitted
85	# 03/08/07 - Marcus- lcfimm admitted
86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
88	# suggestions
89	# 19/10/07 - Marcus- add new optional argument -g group_number
90	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
91	# 30/10/07 - Marcus- further adjustments for queues on lctit
92	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
93	# Science)
94	# 14/07/08 - Siggi - adjustments for lcsgih
95	# 23/09/08 - Gerald- paesano admitted
96	# 02/10/08 - Siggi - PBS adjustments for lcxt4
97	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
104	# now run on the data nodes (feature=data)
105	# 25/08/09 - BjornM- adapted for lck
106	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
107	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
108	# set in the config file by the user
109
110	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
111	delete_dayfile=false
112	email_notification=none
113	group_number=none
114	locat=normal
115	no_default_queue=none
116	no_submit=false
117	job_catalog="~/job_queue"
118	job_name=none
119	local_user=$LOGNAME
120	node_usage=shared
121	numprocs=1
122	punkte="..........................................................."
123	submcom=qsub
124	queue=default
125	remote_host=none
126	remote_user=""
127	verify=true
128
129	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
130	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
131	typeset -L20 spalte1
132	typeset -R40 spalte2
133	typeset -L60 spalte3
134
135
136
137	# FEHLERBEHANDLUNG
138	# BEI EXIT:
139	trap 'if [[ $locat != normal ]]
140	then
141	case $locat in
142	(option) printf "\n --> available optios can be displayed"
143	printf " by typing:"
144	printf "\n \"subjob ?\" \n";;
145	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
146	(*) printf "\n +++ unknown error"
147	printf "\n please inform S. Raasch!\n"
148	esac
149	[[ -f $job_to_send ]] && rm $job_to_send
150	printf "\n\n+++ SUBJOB killed \n\n"
151	fi' exit
152
153
154	# BEI TERMINAL-BREAK:
155	trap '[[ -f $job_to_send ]] && rm $job_to_send
156	printf "\n\n+++ SUBJOB killed \n\n"
157	exit
158	' 2
159
160
161
162
163	# LOKALEN HOSTNAMEN FESTSTELLEN
164	local_host=$(hostname)
165
166
167
168	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
169	# UEBERHAUPT ZULAESSIG IST
170	case $local_host in
171	(atmos) local_addres=172.20.25.35; local_host=lcide;;
172	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
173	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
174	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
175	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
176	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
177	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
178	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
179	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
180	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
181	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
182	(cs*) local_addres=136.172.44.131; local_host=nech;;
183	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
184	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
185	(node*) local_addres=165.132.26.61 local_host=lck;;
186	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
187	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
188	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
189	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
190	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
191	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
192	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
193	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
194	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
195	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
196	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
197	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
198	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
199	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
200	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
201	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
202	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
203	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
204	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
205	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
206	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
207	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
208	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
209	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
210	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
211	(*) printf "\n +++ \"$local_host\" unknown";
212	printf "\n please inform S. Raasch!";
213	locat=parameter; exit;;
214	esac
215
216
217
218	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
219	remote_host=$local_host
220
221
222
223
224	# PROZEDUROPTIONEN EINLESEN
225	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
226	do
227	case $option in
228	(c) job_catalog=$OPTARG;;
229	(d) delete_dayfile=true;;
230	(D) no_submit=true;;
231	(e) email_notification=$OPTARG;;
232	(g) group_number=$OPTARG;;
233	(h) remote_host=$OPTARG;;
234	(m) memory=$OPTARG;;
235	(n) job_name=$OPTARG;;
236	(N) node_usage=$OPTARG;;
237	(O) threads_per_task=$OPTARG;;
238	(q) no_default_queue=$OPTARG;;
239	(t) cputime=$OPTARG;;
240	(T) tasks_per_node=$OPTARG;;
241	(u) remote_user=$OPTARG;;
242	(v) verify=false;;
243	(X) numprocs=$OPTARG;;
244	(\?) printf "\n +++ Option $OPTARG unknown \n";
245	locat=option; exit;;
246	esac
247	done
248
249
250	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
251	shift OPTIND-1; file_to_send=$1
252
253
254	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
255	if [ "$1" = "?" ]
256	then
257	(printf "\n *** subjob can be called as follows:\n"
258	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
259	printf "\n Description of available options:\n"
260	printf "\n Option Description Default-Value"
261	printf "\n -c job-input- and output-catalog ~/job_queue"
262	printf "\n -d no job-protocol will be created ---"
263	printf "\n -D only the job-file will be created ---"
264	printf "\n -h execution host, available hosts: $remote_host"
265	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
266	printf "\n lctit, nech, necriam, unics"
267	printf "\n -m memory demand per process in MByte ---"
268	printf "\n -n jobname <jobdatei>"
269	printf "\n -O threads per task (for OpenMP usage) 1"
270	printf "\n -q job-queue to be used default"
271	printf "\n -t allowed cpu-time in seconds ---"
272	printf "\n -T tasks per node (on parallel hosts) ---"
273	printf "\n -u username on execution host from .netrc"
274	printf "\n -v no prompt for confirmation ---"
275	printf "\n -X # of processors (on parallel hosts) 1"
276	printf "\n "
277	printf "\n The only possible positional parameter is <jobfile>:"
278	printf "\n The complete NQS-job must be provided here."
279	printf "\n <jobfile>=? creates this outline\n\n") \| more
280	exit
281	fi
282
283
284
285	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
286	if [[ "$file_to_send" = "" ]]
287	then
288	printf "\n +++ job-file missing"
289	locat=parameter; exit
290	else
291	if [[ -f $file_to_send ]]
292	then
293	true
294	else
295	printf "\n +++ job-file: "
296	printf "\n $file_to_send"
297	printf "\n does not exist"
298	locat=parameter; exit
299	fi
300	fi
301
302
303
304	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
305	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
306	if [[ $job_name = none ]]
307	then
308	job_name=$file_to_send
309	fi
310	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
311	then
312	printf "\n +++ job-file name: "
313	printf "\n $job_name"
314	printf "\n must not contain \"/\"-characters"
315	locat=parameter; exit
316	fi
317
318
319
320
321	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
322	# ODER WENN HOST NICHT ANGEGEBEN WURDE
323	if [[ $remote_host = none ]]
324	then
325	printf "\n +++ host missing"
326	locat=option; exit
327	else
328	case $remote_host in
329	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
331	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
332	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
333	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
334	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
335	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
336	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
337	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
338	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
339	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
340	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
341	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
342	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
343	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
344	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
345	locat=parameter; exit;;
346	esac
347	fi
348
349
350	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
351	if [[ $no_default_queue != none ]]
352	then
353	error=false
354	ndq=$no_default_queue
355	case $remote_host in
356	(ibm) case $ndq in
357	(p690_express\|p690_standard\|p690_long) error=false;;
358	(*) error=true;;
359	esac;;
360	(ibmb) case $ndq in
361	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
362	(*) error=true;;
363	esac;;
364	(ibmh) case $ndq in
365	(no_class) error=false;;
366	(*) error=true;;
367	esac;;
368	(ibms) case $ndq in
369	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
370	(*) error=true;;
371	esac;;
372	(ibmy) case $ndq in
373	(parallel) error=false;;
374	(*) error=true;;
375	esac;;
376	(lcsgib\|lcsgih) case $ndq in
377	(testq\|serialq\|smallq\|bigq\|workq) error=false;;
378	(*) error=true;;
379	esac;;
380	(lctit) case $ndq in
381	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
382	(*) error=true;;
383	esac;;
384	(t3eb) case $ndq in
385	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
386	(*) error=true;;
387	esac;;
388	(necriam) case $ndq in
389	(SP\|SS\|P6) error=false;;
390	(*) error=true;;
391	esac;;
392	(t3eh) case $ndq in
393	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
394	(*) error=true;;
395	esac;;
396	(t3ej2\|t3ej5) case $ndq in
397	(low\|normal\|high) error=false;;
398	(*) error=true;;
399	esac;;
400	(t3es) case $ndq in
401	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
402	(*) error=true;;
403	esac;;
404	(unics) case $ndq in
405	(unics\|ht) error=false;;
406	(*) error=true;;
407	esac;;
408	esac
409	if [[ $error = true ]]
410	then
411	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
412	locat=parameter; exit
413	else
414	queue=$no_default_queue
415	fi
416	fi
417
418
419
420	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
421	if [[ $node_usage != shared && $queue = cdev ]]
422	then
423	node_usage=shared
424	fi
425
426
427
428	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
429	# AUFTEILEN
430	done=false
431	while [[ $done = false ]]
432	do
433	if (( $cputime <= 0 ))
434	then
435	printf "\n +++ wrong cpu-time or cpu-time missing"
436	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
437	printf "\n >>> "
438	read cputime 1>/dev/null 2>&1
439	else
440	done=true
441	fi
442	done
443	if [[ $remote_host = nech ]]
444	then
445	if (( tasks_per_node != 0 ))
446	then
447	(( cputime = cputime * tasks_per_node ))
448	elif [[ $numprocs != 0 ]]
449	then
450	(( cputime = cputime * numprocs ))
451	fi
452	fi
453	(( stunden = cputime / 3600 ))
454	(( resttime = cputime - stunden * 3600 ))
455	(( minuten = resttime / 60 ))
456	(( sekunden = resttime - minuten * 60 ))
457	timestring=${stunden}:${minuten}:${sekunden}
458
459
460
461	# PRUEFEN DER KERNSPEICHERANFORDERUNG
462	done=false
463	while [[ $done = false ]]
464	do
465	if (( memory <= 0 ))
466	then
467	printf "\n +++ wrong memory demand or memory demand missing"
468	printf "\n >>> Please type memory in MByte per process as INTEGER:"
469	printf "\n >>> "
470	read memory 1>/dev/null 2>&1
471	else
472	done=true
473	fi
474	done
475
476	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
477	then
478	if (( tasks_per_node != 0 ))
479	then
480	(( Memory = memory * tasks_per_node / 1000 ))
481	elif [[ $numprocs != 0 ]]
482	then
483	(( Memory = memory * numprocs / 1000 ))
484	else
485	(( Memory = memory / 1000 ))
486	fi
487	fi
488
489
490	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
491	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
492	then
493	(( memory = memory * threads_per_task ))
494	fi
495
496
497	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
498	if (( tasks_per_node != 0 ))
499	then
500	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
501	then
502	(( nodes = numprocs / tasks_per_node ))
503	else
504	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
505	fi
506	fi
507
508
509
510	# HEADER-AUSGABE
511	if [[ $verify = true ]]
512	then
513	printf "\n\n"
514	printf "#--------------------------------------------------------------# \n"
515	spalte1=SUBJOB;spalte2=$(date)
516	printf "\| $spalte1$spalte2 \| \n"
517	printf "\| \| \n"
518	printf "\| values of parameters/options: \| \n"
519	spalte1=local_host$punkte; spalte2=$punkte$local_host
520	printf "\| $spalte1$spalte2 \| \n"
521	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
522	printf "\| $spalte1$spalte2 \| \n"
523	spalte1=queue$punkte; spalte2=$punkte$queue
524	printf "\| $spalte1$spalte2 \| \n"
525	spalte1=memory$punkte; spalte2="$punkte$memory mb"
526	printf "\| $spalte1$spalte2 \| \n"
527	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
528	printf "\| $spalte1$spalte2 \| \n"
529	spalte1=job_name$punkte; spalte2="$punkte$job_name"
530	printf "\| $spalte1$spalte2 \| \n"
531	printf "#--------------------------------------------------------------# \n\n"
532
533
534	# KONTROLLABFRAGE, OB ALLES O.K.
535	antwort="dummy"
536	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
537	do
538	read antwort?" >>> continue (y/n) ? "
539	done
540	if [[ $antwort = n \|\| $antwort = N ]]
541	then
542	locat=verify; exit
543	fi
544	printf "\n"
545	fi
546
547
548
549	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
550	kennung=$RANDOM
551	job_on_remhost=${job_name}_${kennung}_$local_host
552	job_to_send=job_to_send_$kennung
553	if [[ $delete_dayfile = false ]]
554	then
555	remote_dayfile=${local_host}_${job_name}_result_$kennung
556	local_dayfile=${remote_host}_${job_name}
557	else
558	remote_dayfile=/dev/null
559	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
560	then
561	notification='# @ notification = never'
562	fi
563	fi
564
565
566
567	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
568	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
569	then
570
571	if [[ $remote_host = ibmy ]]
572	then
573	consumable_memory=""
574	else
575	consumable_memory="ConsumableMemory($memory mb)"
576	fi
577
578	if [[ $remote_host = ibmh ]]
579	then
580	data_limit=""
581	network_to_use=""
582	class=""
583	environment=""
584	else
585	class="# @ class = $queue"
586	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
587	if [[ $queue = cdev ]]
588	then
589	data_limit="# @ data_limit = 1.76gb"
590	network_to_use="# @ network.mpi = sn_all,shared,ip"
591	else
592	if [[ $remote_host = ibms ]]
593	then
594	network_to_use="# @ network.mpi = csss,shared,us"
595	elif [[ $remote_host = ibmy ]]
596	then
597	network_to_use=""
598	else
599	network_to_use="# @ network.mpi = sn_all,shared,us"
600	data_limit="# @ data_limit = 1.76gb"
601	fi
602	fi
603	fi
604
605	cat > $job_to_send << %%END%%
606	#!/bin/ksh
607	# @ shell = /bin/ksh
608
609	# @ job_type = parallel
610	# @ job_name = palm
611	# @ wall_clock_limit = ${timestring},$timestring
612	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
613	# @ output = $remote_dayfile
614	# @ error = $remote_dayfile
615	# @ image_size = 50
616	$class
617	$environment
618	$network_to_use
619	$data_limit
620	$notification
621
622	%%END%%
623
624	if (( nodes > 0 ))
625	then
626
627	cat >> $job_to_send << %%END%%
628	# @ node = $nodes
629	# @ tasks_per_node = $tasks_per_node
630	# @ node_usage = $node_usage
631	# @ queue
632
633	%%END%%
634
635	else
636
637	if [[ $remote_host != ibmy ]]
638	then
639
640	cat >> $job_to_send << %%END%%
641	# @ blocking = unlimited
642	# @ total_tasks = $numprocs
643	# @ node_usage = $node_usage
644	# @ queue
645
646	%%END%%
647
648	else
649
650	cat >> $job_to_send << %%END%%
651	# @ node = 1
652	# @ total_tasks = $numprocs
653	# @ queue
654
655	%%END%%
656
657	fi
658
659	fi
660
661	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
662	then
663
664	cat > $job_to_send << %%END%%
665	#!/bin/ksh
666
667	# @ job_type = serial
668	# @ node_usage = $node_usage
669	# @ job_name = palm
670	# @ wall_clock_limit = ${timestring},$timestring
671	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
672	# @ output = $remote_dayfile
673	# @ error = $remote_dayfile
674	$class
675	$notification
676
677	# @ queue
678
679	%%END%%
680
681	elif [[ $remote_host = lcfimm ]]
682	then
683
684	if [[ $numprocs != 0 ]]
685	then
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 nodes=${nodes}:ppn=$tasks_per_node
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	mpd &
698
699	%%END%%
700
701	else
702	cat > $job_to_send << %%END%%
703	#!/bin/ksh
704	#PBS -N $job_name
705	#PBS -A geofysisk
706	#PBS -l walltime=$timestring
707	#PBS -l ncpus=1
708	#PBS -l pmem=${memory}mb
709	#PBS -m abe
710	#PBS -M maronga@muk.uni-hannover.de
711	#PBS -o $remote_dayfile
712	#PBS -j oe
713
714	%%END%%
715
716	fi
717
718	elif [[ $remote_host = lck ]]
719	then
720
721	if [[ $numprocs != 0 ]]
722	then
723	cat > $job_to_send << %%END%%
724	#!/bin/ksh
725	#PBS -N $job_name
726	#PBS -l walltime=$timestring
727	#PBS -l ncpus=$numprocs
728	#PBS -l pmem=${memory}mb
729	#PBS -o $remote_dayfile
730	#PBS -j oe
731
732	mpd &
733
734	%%END%%
735
736	else
737	cat > $job_to_send << %%END%%
738	#!/bin/ksh
739	#PBS -N $job_name
740	#PBS -l walltime=$timestring
741	#PBS -l ncpus=1
742	#PBS -l pmem=${memory}mb
743	#PBS -o $remote_dayfile
744	#PBS -j oe
745
746	%%END%%
747
748	fi
749
750	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
751	then
752
753	if [[ "$sgi_feature" != "" ]]
754	then
755	feature_directive="#PBS -l feature=$sgi_feature"
756	else
757	feature_directive=""
758	fi
759
760	if [[ $queue = testq \|\| $queue = workq ]]
761	then
762	queue_directive="#PBS -q $queue"
763	else
764	queue_directive=""
765	fi
766
767	if [[ $email_notification = none ]]
768	then
769	email_directive=""
770	else
771	email_directive="#PBS -M $email_notification"
772	fi
773
774	if [[ $numprocs != 0 ]]
775	then
776	cat > $job_to_send << %%END%%
777	#!/bin/bash
778	#PBS -N $job_name
779	#PBS -l walltime=$timestring
780	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
781	#PBS -l naccesspolicy=$node_usage
782	#PBS -o $remote_dayfile
783	#PBS -j oe
784	$feature_directive
785	$queue_directive
786	$email_directive
787
788	. /usr/share/modules/init/bash
789	module load ifort/11.0.069
790	module load $mpilib
791	module load netcdf
792
793	echo ld_library_path=\$LD_LIBRARY_PATH
794
795	%%END%%
796
797	else
798	cat > $job_to_send << %%END%%
799	#PBS -S /bin/bash
800	#PBS -N $job_name
801	#PBS -l walltime=$timestring
802	#PBS -l ncpus=1
803	#PBS -o $remote_dayfile
804	#PBS -e $remote_dayfile
805
806	%%END%%
807
808	fi
809
810	elif [[ $remote_host = lcxt4 ]]
811	then
812
813	if [[ $numprocs != 0 ]]
814	then
815	cat > $job_to_send << %%END%%
816	#!/bin/ksh
817	#PBS -S /bin/ksh
818	#PBS -N $job_name
819	#PBS -A geofysisk
820	#PBS -j oe
821	#PBS -l walltime=$timestring
822	#PBS -l mppwidth=${numprocs}
823	#PBS -l mppnppn=${tasks_per_node}
824	#PBS -m abe
825	#PBS -M maronga@muk.uni-hannover.de
826	#PBS -o $remote_dayfile
827
828	. /opt/modules/default/init/ksh
829	module load modules/3.1.6
830	module load PrgEnv-pgi/2.1.56HD
831	module load Base-opts/2.1.56HD
832	module load moab/5.2.5-s13441
833	module load torque/2.3.4-snap.200809221601
834
835	%%END%%
836
837	else
838	cat > $job_to_send << %%END%%
839	#!/bin/ksh
840	#PBS -S /bin/ksh
841	#PBS -N $job_name
842	#PBS -A geofysisk
843	#PBS -j oe
844	#PBS -l walltime=$timestring
845	#PBS -l ncpus=1
846	#PBS -l pmem=${memory}mb
847	#PBS -m abe
848	#PBS -M maronga@muk.uni-hannover.de
849	#PBS -o $remote_dayfile
850
851	. /opt/modules/default/init/ksh
852	module load modules/3.1.6
853	module load PrgEnv-pgi/2.1.56HD
854	module load Base-opts/2.1.56HD
855	module load moab/5.2.5-s13441
856	module load torque/2.3.4-snap.200809221601
857
858	%%END%%
859
860	fi
861
862	elif [[ $remote_host = nech ]]
863	then
864
865	if (( nodes > 1 ))
866	then
867	cat > $job_to_send << %%END%%
868	#!/bin/ksh
869	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
870	#PBS -l ${qsubmem}=${Memory}gb
871	#PBS -b $nodes
872	#PBS -o $remote_dayfile
873	#PBS -N palm
874	#PBS -j o
875	#PBS -T mpisx
876
877	%%END%%
878
879	elif [[ $numprocs != 0 ]]
880	then
881	cat > $job_to_send << %%END%%
882	#!/bin/ksh
883	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
884	#PBS -l ${qsubmem}=${Memory}gb
885	#PBS -o $remote_dayfile
886	#PBS -N palm
887	#PBS -j o
888
889	%%END%%
890
891	else
892	cat > $job_to_send << %%END%%
893	#!/bin/ksh
894	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
895	#PBS -o $remote_dayfile
896	#PBS -j o
897
898	%%END%%
899
900	fi
901
902	elif [[ $remote_host = necriam ]]
903	then
904
905	if (( nodes > 1 ))
906	then
907	cat > $job_to_send << %%END%%
908	#!/bin/ksh
909	#PBS -b $nodes
910	#PBS -o $remote_dayfile
911	#PBS -N $job_name
912	#PBS -j o
913	#PBS -v MPIPROGINV=YES
914
915	%%END%%
916
917	elif [[ $numprocs != 0 ]]
918	then
919	cat > $job_to_send << %%END%%
920	#!/bin/ksh
921	#PBS -q ${queue}
922	#PBS -o $remote_dayfile
923	#PBS -N $job_name
924	#PBS -j o
925	#PBS -v MPIPROGINV=YES
926
927	%%END%%
928
929	else
930	cat > $job_to_send << %%END%%
931	#!/bin/ksh
932	#PBS -o $remote_dayfile
933	#PBS -j o
934
935	%%END%%
936
937	fi
938
939	elif [[ $remote_host = lctit ]]
940	then
941	cat > $job_to_send << %%END%%
942	#!/bin/bash
943	#$ -S /bin/bash
944	cd $job_catalog
945	export PATH=$PALM_BIN:\$PATH
946	echo \$PATH
947
948	%%END%%
949
950	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
951	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
952	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
953
954	else
955
956	cat > $job_to_send << %%END%%
957	# @\$-q ${queue}
958	# @\$-l${qsubtime} $timestring
959	# @\$-l${qsubmem} ${memory}mb
960	# @\$-o $remote_dayfile
961	# @\$-eo
962
963	%%END%%
964
965	fi
966
967
968	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
969	# BEI EXIT VERANLASSEN
970	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
971	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
972	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
973	then
974	echo "set +vx" >> $job_to_send
975	echo "trap '" >> $job_to_send
976	echo "set +vx" >> $job_to_send
977	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 ]]
978	then
979	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
980	then
981	return_queue=c1
982	elif [[ $remote_host = ibms ]]
983	then
984	return_queue=p_normal
985	elif [[ $remote_host = ibmy ]]
986	then
987	return_queue=serial
988	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
989	then
990	return_queue=serialq
991	elif [[ $remote_host = lctit ]]
992	then
993	return_queue=sla3
994	elif [[ $remote_host = necriam ]]
995	then
996	return_queue=SP
997	else
998	return_queue=unknown
999	fi
1000
1001	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1002	then
1003
1004	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1005	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1006	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1007	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1008	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1009	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1010	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1011	if [[ $host != "ibmh" ]]
1012	then
1013	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1014	fi
1015	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1016	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1017
1018	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1019	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1020
1021	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1022	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1023	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1024
1025	elif [[ $remote_host = nech ]]
1026	then
1027	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1028	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1029	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1030	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1031	echo "#PBS -j o" >> $job_to_send
1032	echo " " >> $job_to_send
1033	echo "set -x" >> $job_to_send
1034	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1035	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1036	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1037	echo "%%END%%" >> $job_to_send
1038
1039	elif [[ $remote_host = necriam ]]
1040	then
1041	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1042	echo "#PBS -q $return_queue" >> $job_to_send
1043	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1044	echo "#PBS -j o" >> $job_to_send
1045	echo " " >> $job_to_send
1046	echo "set -x" >> $job_to_send
1047	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1048	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1049	echo "%%END%%" >> $job_to_send
1050
1051	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1052	then
1053	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1054	echo "#!/bin/bash" >> $job_to_send
1055	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1056	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1057	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1058	echo "#PBS -l feature=data" >> $job_to_send
1059	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1060	echo "#PBS -j oe" >> $job_to_send
1061	echo " " >> $job_to_send
1062	echo ". /usr/share/modules/init/bash" >> $job_to_send
1063	echo "set -x" >> $job_to_send
1064	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1065	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1066	echo "%%END%%" >> $job_to_send
1067
1068	elif [[ $remote_host = lctit ]]
1069	then
1070	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1071	echo "set -x" >> $job_to_send
1072	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1073	echo "cd $job_catalog" >> $job_to_send
1074	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1075	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1076	echo "%%END%%" >> $job_to_send
1077
1078	else
1079
1080	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1081	echo "# @\\\$-q $return_queue" >> $job_to_send
1082	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1083	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1084	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1085	then
1086	echo "# @\$-l mpp_p=0" >> $job_to_send
1087	fi
1088	echo '# @\$-lF 10mb' >> $job_to_send
1089	# echo '# @\$-o /dev/null' >> $job_to_send
1090	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1091	echo '# @\\\$-eo' >> $job_to_send
1092	echo " " >> $job_to_send
1093	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1094	then
1095	echo "set +vx" >> $job_to_send
1096	echo ". .profile" >> $job_to_send
1097	fi
1098	echo "set -x" >> $job_to_send
1099	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1100	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1101	echo "%%END%%" >> $job_to_send
1102	fi
1103
1104	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1105	then
1106	echo "llsubmit scpjob.$kennung" >> $job_to_send
1107	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1108	then
1109	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1110	echo "msub scpjob.$kennung" >> $job_to_send
1111	elif [[ $remote_host = lctit ]]
1112	then
1113	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1114	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1115	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1116	echo "sleep 10" >> $job_to_send
1117	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1118	then
1119	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1120	elif [[ $remote_host = t3es ]]
1121	then
1122	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1123	else
1124	echo "qsub scpjob.$kennung" >> $job_to_send
1125	fi
1126	echo "rm scpjob.$kennung" >> $job_to_send
1127	if [[ $remote_host = nech ]]
1128	then
1129	echo "cd -" >> $job_to_send
1130	fi
1131	else
1132	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1133	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1134	fi
1135	echo "set -x" >> $job_to_send
1136	echo " ' exit" >> $job_to_send
1137	echo "set -x" >> $job_to_send
1138	fi
1139
1140
1141
1142	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1143	cat $file_to_send >> $job_to_send
1144
1145	if [[ $remote_host = ibm ]]
1146	then
1147	echo " " >> $job_to_send
1148	echo "exit" >> $job_to_send
1149	fi
1150	if [[ $remote_host = lctit ]]
1151	then
1152	echo " " >> $job_to_send
1153	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1154	fi
1155
1156
1157
1158	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1159	if [[ -z $remote_user ]]
1160	then
1161	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1162	then
1163	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1164	fi
1165	fi
1166
1167
1168
1169	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1170	if [[ $no_submit = false ]]
1171	then
1172	if [[ $remote_host != $local_host ]]
1173	then
1174	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1175	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1176	then
1177	job_catalog_save=$job_catalog
1178	job_catalog=job_queue
1179	elif [[ $remote_host = nech ]]
1180	then
1181	job_catalog_save=$job_catalog
1182	job_catalog=/hpf/b/${remote_user}/job_queue
1183	fi
1184	if [[ $local_host = decalpha ]]
1185	then
1186	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1187	# PASSWORT AB
1188	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1189	elif [[ $remote_host = nech ]]
1190	then
1191	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1192	# TRANSFERIERT WERDEN
1193	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1194	else
1195	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1196	fi
1197	if [[ $? = 1 ]]
1198	then
1199	locat=scp; exit
1200	fi
1201	if [[ $remote_host = ibms ]]
1202	then
1203	job_catalog=$job_catalog_save
1204	fi
1205	[[ $verify = true ]] && printf "\n >>> finished\n"
1206	else
1207	eval job_catalog=$job_catalog
1208	cp $job_to_send ${job_catalog}/$job_on_remhost
1209	fi
1210
1211
1212
1213	# NQS- BZW. LOADLEVELER-JOB STARTEN
1214	if [[ $remote_host != $local_host ]]
1215	then
1216	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1217	if [[ $remote_host != lctit ]]
1218	then
1219	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1220	then
1221	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1222	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1223	else
1224	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1225	fi
1226	else
1227	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1228	# MIT SSH, DESHALB AUFRUF PER PIPE
1229	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1230	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1231	echo "first try" > send_protocol
1232	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1233	do
1234	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1235	then
1236	printf "\n trying to submit job to TIT cluster..."
1237	else
1238	printf "\n +++ failed ... trying to submit again ..."
1239	fi
1240	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1241	cat send_protocol
1242	done
1243	sleep 10
1244	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1245	echo "$submcom $job_on_remhost"
1246	rm send_protocol
1247	fi
1248
1249	[[ $verify = true ]] && printf " >>> o.k.\n"
1250	else
1251	cd $job_catalog
1252	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1253	then
1254	eval $submcom $job_on_remhost
1255	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1256	then
1257	eval $submcom $job_on_remhost
1258	echo "$submcom $job_on_remhost"
1259	chmod u+x $job_on_remhost
1260	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1261	then
1262	eval $submcom $job_on_remhost
1263	elif [[ $local_host = nech ]]
1264	then
1265	if [[ $queue = default ]]
1266	then
1267	eval $submcom $job_on_remhost
1268	else
1269	eval $submcom -q $queue $job_on_remhost
1270	fi
1271	else
1272	qsub $job_on_remhost
1273	fi
1274	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1275	[[ $local_host != lctit ]] && rm $job_on_remhost
1276	cd - > /dev/null
1277	fi
1278	fi
1279
1280
1281
1282	# ABSCHLUSSARBEITEN
1283
1284	if [[ $no_submit = false ]]
1285	then
1286	rm $job_to_send
1287	fi
1288	[[ $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 |