Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 374

Last change on this file since 374 was 374, checked in by raasch, 15 years ago
mrun adjustments for ice2 and for ssh calls of restart jobs
Property svn:keywords set to `Id Rev`
File size: 44.0 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 374 2009-08-28 10:50:21Z 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
108	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
109	delete_dayfile=false
110	email_notification=none
111	group_number=none
112	locat=normal
113	no_default_queue=none
114	no_submit=false
115	job_catalog="~/job_queue"
116	job_name=none
117	local_user=$LOGNAME
118	node_usage=shared
119	numprocs=1
120	punkte="..........................................................."
121	submcom=qsub
122	queue=default
123	remote_host=none
124	remote_user=""
125	verify=true
126
127	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
128	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
129	typeset -L20 spalte1
130	typeset -R40 spalte2
131	typeset -L60 spalte3
132
133
134
135	# FEHLERBEHANDLUNG
136	# BEI EXIT:
137	trap 'if [[ $locat != normal ]]
138	then
139	case $locat in
140	(option) printf "\n --> available optios can be displayed"
141	printf " by typing:"
142	printf "\n \"subjob ?\" \n";;
143	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
144	(*) printf "\n +++ unknown error"
145	printf "\n please inform S. Raasch!\n"
146	esac
147	[[ -f $job_to_send ]] && rm $job_to_send
148	printf "\n\n+++ SUBJOB killed \n\n"
149	fi' exit
150
151
152	# BEI TERMINAL-BREAK:
153	trap '[[ -f $job_to_send ]] && rm $job_to_send
154	printf "\n\n+++ SUBJOB killed \n\n"
155	exit
156	' 2
157
158
159
160
161	# LOKALEN HOSTNAMEN FESTSTELLEN
162	local_host=$(hostname)
163
164
165
166	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
167	# UEBERHAUPT ZULAESSIG IST
168	case $local_host in
169	(atmos) local_addres=172.20.25.35; local_host=lcide;;
170	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
171	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
172	(bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
173	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
174	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
175	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
176	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
177	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
178	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
179	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
180	(cs*) local_addres=136.172.44.131; local_host=nech;;
181	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
182	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
183	(node*) local_addres=165.132.26.61 local_host=lck;;
184	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
185	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
186	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
187	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
188	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
189	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
190	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
191	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
192	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
193	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
194	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
195	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
196	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
197	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
198	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
199	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
200	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
201	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
202	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
203	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
204	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
205	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
206	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
207	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
208	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
209	(*) printf "\n +++ \"$local_host\" unknown";
210	printf "\n please inform S. Raasch!";
211	locat=parameter; exit;;
212	esac
213
214
215
216	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
217	remote_host=$local_host
218
219
220
221
222	# PROZEDUROPTIONEN EINLESEN
223	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
224	do
225	case $option in
226	(c) job_catalog=$OPTARG;;
227	(d) delete_dayfile=true;;
228	(D) no_submit=true;;
229	(e) email_notification=$OPTARG;;
230	(g) group_number=$OPTARG;;
231	(h) remote_host=$OPTARG;;
232	(m) memory=$OPTARG;;
233	(n) job_name=$OPTARG;;
234	(N) node_usage=$OPTARG;;
235	(O) threads_per_task=$OPTARG;;
236	(q) no_default_queue=$OPTARG;;
237	(t) cputime=$OPTARG;;
238	(T) tasks_per_node=$OPTARG;;
239	(u) remote_user=$OPTARG;;
240	(v) verify=false;;
241	(X) numprocs=$OPTARG;;
242	(\?) printf "\n +++ Option $OPTARG unknown \n";
243	locat=option; exit;;
244	esac
245	done
246
247
248	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
249	shift OPTIND-1; file_to_send=$1
250
251
252	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
253	if [ "$1" = "?" ]
254	then
255	(printf "\n *** subjob can be called as follows:\n"
256	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
257	printf "\n Description of available options:\n"
258	printf "\n Option Description Default-Value"
259	printf "\n -c job-input- and output-catalog ~/job_queue"
260	printf "\n -d no job-protocol will be created ---"
261	printf "\n -D only the job-file will be created ---"
262	printf "\n -h execution host, available hosts: $remote_host"
263	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
264	printf "\n lctit, nech, necriam, unics"
265	printf "\n -m memory demand per process in MByte ---"
266	printf "\n -n jobname <jobdatei>"
267	printf "\n -O threads per task (for OpenMP usage) 1"
268	printf "\n -q job-queue to be used default"
269	printf "\n -t allowed cpu-time in seconds ---"
270	printf "\n -T tasks per node (on parallel hosts) ---"
271	printf "\n -u username on execution host from .netrc"
272	printf "\n -v no prompt for confirmation ---"
273	printf "\n -X # of processors (on parallel hosts) 1"
274	printf "\n "
275	printf "\n The only possible positional parameter is <jobfile>:"
276	printf "\n The complete NQS-job must be provided here."
277	printf "\n <jobfile>=? creates this outline\n\n") \| more
278	exit
279	fi
280
281
282
283	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
284	if [[ "$file_to_send" = "" ]]
285	then
286	printf "\n +++ job-file missing"
287	locat=parameter; exit
288	else
289	if [[ -f $file_to_send ]]
290	then
291	true
292	else
293	printf "\n +++ job-file: "
294	printf "\n $file_to_send"
295	printf "\n does not exist"
296	locat=parameter; exit
297	fi
298	fi
299
300
301
302	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
303	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
304	if [[ $job_name = none ]]
305	then
306	job_name=$file_to_send
307	fi
308	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
309	then
310	printf "\n +++ job-file name: "
311	printf "\n $job_name"
312	printf "\n must not contain \"/\"-characters"
313	locat=parameter; exit
314	fi
315
316
317
318
319	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
320	# ODER WENN HOST NICHT ANGEGEBEN WURDE
321	if [[ $remote_host = none ]]
322	then
323	printf "\n +++ host missing"
324	locat=option; exit
325	else
326	case $remote_host in
327	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
328	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
329	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
330	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
331	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
332	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
333	(lcsgib) queue=smallq; remote_addres=130.73.232.103; submcom=/opt/moab/bin/msub;;
334	(lcsgih) queue=smallq; remote_addres=130.75.4.102; submcom=/opt/moab/bin/msub;;
335	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
336	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
337	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
338	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
339	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
340	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
341	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
342	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
343	locat=parameter; exit;;
344	esac
345	fi
346
347
348	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
349	if [[ $no_default_queue != none ]]
350	then
351	error=false
352	ndq=$no_default_queue
353	case $remote_host in
354	(ibm) case $ndq in
355	(p690_express\|p690_standard\|p690_long) error=false;;
356	(*) error=true;;
357	esac;;
358	(ibmb) case $ndq in
359	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
360	(*) error=true;;
361	esac;;
362	(ibmh) case $ndq in
363	(no_class) error=false;;
364	(*) error=true;;
365	esac;;
366	(ibms) case $ndq in
367	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
368	(*) error=true;;
369	esac;;
370	(ibmy) case $ndq in
371	(parallel) error=false;;
372	(*) error=true;;
373	esac;;
374	(lcsgib\|lcsgih) case $ndq in
375	(testq\|serialq\|smallq\|bigq\|workq) error=false;;
376	(*) error=true;;
377	esac;;
378	(lctit) case $ndq in
379	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
380	(*) error=true;;
381	esac;;
382	(t3eb) case $ndq in
383	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
384	(*) error=true;;
385	esac;;
386	(necriam) case $ndq in
387	(SP\|SS\|P6) error=false;;
388	(*) error=true;;
389	esac;;
390	(t3eh) case $ndq in
391	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
392	(*) error=true;;
393	esac;;
394	(t3ej2\|t3ej5) case $ndq in
395	(low\|normal\|high) error=false;;
396	(*) error=true;;
397	esac;;
398	(t3es) case $ndq in
399	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
400	(*) error=true;;
401	esac;;
402	(unics) case $ndq in
403	(unics\|ht) error=false;;
404	(*) error=true;;
405	esac;;
406	esac
407	if [[ $error = true ]]
408	then
409	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
410	locat=parameter; exit
411	else
412	queue=$no_default_queue
413	fi
414	fi
415
416
417
418	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
419	if [[ $node_usage != shared && $queue = cdev ]]
420	then
421	node_usage=shared
422	fi
423
424
425
426	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
427	# AUFTEILEN
428	done=false
429	while [[ $done = false ]]
430	do
431	if (( $cputime <= 0 ))
432	then
433	printf "\n +++ wrong cpu-time or cpu-time missing"
434	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
435	printf "\n >>> "
436	read cputime 1>/dev/null 2>&1
437	else
438	done=true
439	fi
440	done
441	if [[ $remote_host = nech ]]
442	then
443	if (( tasks_per_node != 0 ))
444	then
445	(( cputime = cputime * tasks_per_node ))
446	elif [[ $numprocs != 0 ]]
447	then
448	(( cputime = cputime * numprocs ))
449	fi
450	fi
451	(( stunden = cputime / 3600 ))
452	(( resttime = cputime - stunden * 3600 ))
453	(( minuten = resttime / 60 ))
454	(( sekunden = resttime - minuten * 60 ))
455	timestring=${stunden}:${minuten}:${sekunden}
456
457
458
459	# PRUEFEN DER KERNSPEICHERANFORDERUNG
460	done=false
461	while [[ $done = false ]]
462	do
463	if (( memory <= 0 ))
464	then
465	printf "\n +++ wrong memory demand or memory demand missing"
466	printf "\n >>> Please type memory in MByte per process as INTEGER:"
467	printf "\n >>> "
468	read memory 1>/dev/null 2>&1
469	else
470	done=true
471	fi
472	done
473
474	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
475	then
476	if (( tasks_per_node != 0 ))
477	then
478	(( Memory = memory * tasks_per_node / 1000 ))
479	elif [[ $numprocs != 0 ]]
480	then
481	(( Memory = memory * numprocs / 1000 ))
482	else
483	(( Memory = memory / 1000 ))
484	fi
485	fi
486
487
488	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
489	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
490	then
491	(( memory = memory * threads_per_task ))
492	fi
493
494
495	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
496	if (( tasks_per_node != 0 ))
497	then
498	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
499	fi
500
501
502
503	# HEADER-AUSGABE
504	if [[ $verify = true ]]
505	then
506	printf "\n\n"
507	printf "#--------------------------------------------------------------# \n"
508	spalte1=SUBJOB;spalte2=$(date)
509	printf "\| $spalte1$spalte2 \| \n"
510	printf "\| \| \n"
511	printf "\| values of parameters/options: \| \n"
512	spalte1=local_host$punkte; spalte2=$punkte$local_host
513	printf "\| $spalte1$spalte2 \| \n"
514	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
515	printf "\| $spalte1$spalte2 \| \n"
516	spalte1=queue$punkte; spalte2=$punkte$queue
517	printf "\| $spalte1$spalte2 \| \n"
518	spalte1=memory$punkte; spalte2="$punkte$memory mb"
519	printf "\| $spalte1$spalte2 \| \n"
520	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
521	printf "\| $spalte1$spalte2 \| \n"
522	spalte1=job_name$punkte; spalte2="$punkte$job_name"
523	printf "\| $spalte1$spalte2 \| \n"
524	printf "#--------------------------------------------------------------# \n\n"
525
526
527	# KONTROLLABFRAGE, OB ALLES O.K.
528	antwort="dummy"
529	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
530	do
531	read antwort?" >>> continue (y/n) ? "
532	done
533	if [[ $antwort = n \|\| $antwort = N ]]
534	then
535	locat=verify; exit
536	fi
537	printf "\n"
538	fi
539
540
541
542	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
543	kennung=$RANDOM
544	job_on_remhost=${job_name}_${kennung}_$local_host
545	job_to_send=job_to_send_$kennung
546	if [[ $delete_dayfile = false ]]
547	then
548	remote_dayfile=${local_host}_${job_name}_result_$kennung
549	local_dayfile=${remote_host}_${job_name}
550	else
551	remote_dayfile=/dev/null
552	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
553	then
554	notification='# @ notification = never'
555	fi
556	fi
557
558
559
560	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
561	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
562	then
563
564	if [[ $remote_host = ibmy ]]
565	then
566	consumable_memory=""
567	else
568	consumable_memory="ConsumableMemory($memory mb)"
569	fi
570
571	if [[ $remote_host = ibmh ]]
572	then
573	data_limit=""
574	network_to_use=""
575	class=""
576	environment=""
577	else
578	class="# @ class = $queue"
579	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
580	if [[ $queue = cdev ]]
581	then
582	data_limit="# @ data_limit = 1.76gb"
583	network_to_use="# @ network.mpi = sn_all,shared,ip"
584	else
585	if [[ $remote_host = ibms ]]
586	then
587	network_to_use="# @ network.mpi = csss,shared,us"
588	elif [[ $remote_host = ibmy ]]
589	then
590	network_to_use=""
591	else
592	network_to_use="# @ network.mpi = sn_all,shared,us"
593	data_limit="# @ data_limit = 1.76gb"
594	fi
595	fi
596	fi
597
598	cat > $job_to_send << %%END%%
599	#!/bin/ksh
600	# @ shell = /bin/ksh
601
602	# @ job_type = parallel
603	# @ job_name = palm
604	# @ wall_clock_limit = ${timestring},$timestring
605	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
606	# @ output = $remote_dayfile
607	# @ error = $remote_dayfile
608	# @ image_size = 50
609	$class
610	$environment
611	$network_to_use
612	$data_limit
613	$notification
614
615	%%END%%
616
617	if (( nodes > 0 ))
618	then
619
620	cat >> $job_to_send << %%END%%
621	# @ node = $nodes
622	# @ tasks_per_node = $tasks_per_node
623	# @ node_usage = $node_usage
624	# @ queue
625
626	%%END%%
627
628	else
629
630	if [[ $remote_host != ibmy ]]
631	then
632
633	cat >> $job_to_send << %%END%%
634	# @ blocking = unlimited
635	# @ total_tasks = $numprocs
636	# @ node_usage = $node_usage
637	# @ queue
638
639	%%END%%
640
641	else
642
643	cat >> $job_to_send << %%END%%
644	# @ node = 1
645	# @ total_tasks = $numprocs
646	# @ queue
647
648	%%END%%
649
650	fi
651
652	fi
653
654	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
655	then
656
657	cat > $job_to_send << %%END%%
658	#!/bin/ksh
659
660	# @ job_type = serial
661	# @ node_usage = $node_usage
662	# @ job_name = palm
663	# @ wall_clock_limit = ${timestring},$timestring
664	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
665	# @ output = $remote_dayfile
666	# @ error = $remote_dayfile
667	$class
668	$notification
669
670	# @ queue
671
672	%%END%%
673
674	elif [[ $remote_host = lcfimm ]]
675	then
676
677	if [[ $numprocs != 0 ]]
678	then
679	cat > $job_to_send << %%END%%
680	#!/bin/ksh
681	#PBS -N $job_name
682	#PBS -A geofysisk
683	#PBS -l walltime=$timestring
684	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
685	#PBS -l pmem=${memory}mb
686	#PBS -m abe
687	#PBS -M maronga@muk.uni-hannover.de
688	#PBS -o $remote_dayfile
689	#PBS -j oe
690	mpd &
691
692	%%END%%
693
694	else
695	cat > $job_to_send << %%END%%
696	#!/bin/ksh
697	#PBS -N $job_name
698	#PBS -A geofysisk
699	#PBS -l walltime=$timestring
700	#PBS -l ncpus=1
701	#PBS -l pmem=${memory}mb
702	#PBS -m abe
703	#PBS -M maronga@muk.uni-hannover.de
704	#PBS -o $remote_dayfile
705	#PBS -j oe
706
707	%%END%%
708
709	fi
710
711	elif [[ $remote_host = lck ]]
712	then
713
714	if [[ $numprocs != 0 ]]
715	then
716	cat > $job_to_send << %%END%%
717	#!/bin/ksh
718	#PBS -N $job_name
719	#PBS -l walltime=$timestring
720	#PBS -l ncpus=$numprocs
721	#PBS -l pmem=${memory}mb
722	#PBS -o $remote_dayfile
723	#PBS -j oe
724
725	mpd &
726
727	%%END%%
728
729	else
730	cat > $job_to_send << %%END%%
731	#!/bin/ksh
732	#PBS -N $job_name
733	#PBS -l walltime=$timestring
734	#PBS -l ncpus=1
735	#PBS -l pmem=${memory}mb
736	#PBS -o $remote_dayfile
737	#PBS -j oe
738
739	%%END%%
740
741	fi
742
743	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
744	then
745
746	if [[ $queue = testq \|\| $queue = workq ]]
747	then
748	queue_directive="#PBS -q $queue"
749	else
750	queue_directive=""
751	fi
752
753	if [[ $email_notification = none ]]
754	then
755	email_directive=""
756	else
757	email_directive="#PBS -M $email_notification"
758	fi
759
760	if [[ $numprocs != 0 ]]
761	then
762	cat > $job_to_send << %%END%%
763	#!/bin/bash
764	#PBS -N $job_name
765	#PBS -l walltime=$timestring
766	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
767	#PBS -l feature=$sgi_feature
768	#PBS -l naccesspolicy=$node_usage
769	#PBS -o $remote_dayfile
770	#PBS -j oe
771	$queue_directive
772	$email_directive
773
774	. /usr/share/modules/init/bash
775	module load ifort/11.0.069
776	module load $mpilib
777	module load netcdf
778
779	echo ld_library_path=\$LD_LIBRARY_PATH
780
781	%%END%%
782
783	else
784	cat > $job_to_send << %%END%%
785	#PBS -S /bin/bash
786	#PBS -N $job_name
787	#PBS -l walltime=$timestring
788	#PBS -l ncpus=1
789	#PBS -o $remote_dayfile
790	#PBS -e $remote_dayfile
791
792	%%END%%
793
794	fi
795
796	elif [[ $remote_host = lcxt4 ]]
797	then
798
799	if [[ $numprocs != 0 ]]
800	then
801	cat > $job_to_send << %%END%%
802	#!/bin/ksh
803	#PBS -S /bin/ksh
804	#PBS -N $job_name
805	#PBS -A geofysisk
806	#PBS -j oe
807	#PBS -l walltime=$timestring
808	#PBS -l mppwidth=${numprocs}
809	#PBS -l mppnppn=${tasks_per_node}
810	#PBS -m abe
811	#PBS -M maronga@muk.uni-hannover.de
812	#PBS -o $remote_dayfile
813
814	. /opt/modules/default/init/ksh
815	module load modules/3.1.6
816	module load PrgEnv-pgi/2.1.56HD
817	module load Base-opts/2.1.56HD
818	module load moab/5.2.5-s13441
819	module load torque/2.3.4-snap.200809221601
820
821	%%END%%
822
823	else
824	cat > $job_to_send << %%END%%
825	#!/bin/ksh
826	#PBS -S /bin/ksh
827	#PBS -N $job_name
828	#PBS -A geofysisk
829	#PBS -j oe
830	#PBS -l walltime=$timestring
831	#PBS -l ncpus=1
832	#PBS -l pmem=${memory}mb
833	#PBS -m abe
834	#PBS -M maronga@muk.uni-hannover.de
835	#PBS -o $remote_dayfile
836
837	. /opt/modules/default/init/ksh
838	module load modules/3.1.6
839	module load PrgEnv-pgi/2.1.56HD
840	module load Base-opts/2.1.56HD
841	module load moab/5.2.5-s13441
842	module load torque/2.3.4-snap.200809221601
843
844	%%END%%
845
846	fi
847
848	elif [[ $remote_host = nech ]]
849	then
850
851	if (( nodes > 1 ))
852	then
853	cat > $job_to_send << %%END%%
854	#!/bin/ksh
855	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
856	#PBS -l ${qsubmem}=${Memory}gb
857	#PBS -b $nodes
858	#PBS -o $remote_dayfile
859	#PBS -N palm
860	#PBS -j o
861	#PBS -T mpisx
862
863	%%END%%
864
865	elif [[ $numprocs != 0 ]]
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 -o $remote_dayfile
872	#PBS -N palm
873	#PBS -j o
874
875	%%END%%
876
877	else
878	cat > $job_to_send << %%END%%
879	#!/bin/ksh
880	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
881	#PBS -o $remote_dayfile
882	#PBS -j o
883
884	%%END%%
885
886	fi
887
888	elif [[ $remote_host = necriam ]]
889	then
890
891	if (( nodes > 1 ))
892	then
893	cat > $job_to_send << %%END%%
894	#!/bin/ksh
895	#PBS -b $nodes
896	#PBS -o $remote_dayfile
897	#PBS -N $job_name
898	#PBS -j o
899	#PBS -v MPIPROGINV=YES
900
901	%%END%%
902
903	elif [[ $numprocs != 0 ]]
904	then
905	cat > $job_to_send << %%END%%
906	#!/bin/ksh
907	#PBS -q ${queue}
908	#PBS -o $remote_dayfile
909	#PBS -N $job_name
910	#PBS -j o
911	#PBS -v MPIPROGINV=YES
912
913	%%END%%
914
915	else
916	cat > $job_to_send << %%END%%
917	#!/bin/ksh
918	#PBS -o $remote_dayfile
919	#PBS -j o
920
921	%%END%%
922
923	fi
924
925	elif [[ $remote_host = lctit ]]
926	then
927	cat > $job_to_send << %%END%%
928	#!/bin/bash
929	#$ -S /bin/bash
930	cd $job_catalog
931	export PATH=$PALM_BIN:\$PATH
932	echo \$PATH
933
934	%%END%%
935
936	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
937	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
938	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
939
940	else
941
942	cat > $job_to_send << %%END%%
943	# @\$-q ${queue}
944	# @\$-l${qsubtime} $timestring
945	# @\$-l${qsubmem} ${memory}mb
946	# @\$-o $remote_dayfile
947	# @\$-eo
948
949	%%END%%
950
951	fi
952
953
954	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
955	# BEI EXIT VERANLASSEN
956	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
957	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
958	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
959	then
960	echo "set +vx" >> $job_to_send
961	echo "trap '" >> $job_to_send
962	echo "set +vx" >> $job_to_send
963	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 ]]
964	then
965	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
966	then
967	return_queue=c1
968	elif [[ $remote_host = ibms ]]
969	then
970	return_queue=p_normal
971	elif [[ $remote_host = ibmy ]]
972	then
973	return_queue=serial
974	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
975	then
976	return_queue=serialq
977	elif [[ $remote_host = lctit ]]
978	then
979	return_queue=sla3
980	elif [[ $remote_host = necriam ]]
981	then
982	return_queue=SP
983	else
984	return_queue=unknown
985	fi
986
987	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
988	then
989
990	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
991	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
992	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
993	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
994	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
995	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
996	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
997	if [[ $host != "ibmh" ]]
998	then
999	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1000	fi
1001	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1002	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1003
1004	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1005	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1006
1007	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1008	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1009	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1010
1011	elif [[ $remote_host = nech ]]
1012	then
1013	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1014	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1015	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1016	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1017	echo "#PBS -j o" >> $job_to_send
1018	echo " " >> $job_to_send
1019	echo "set -x" >> $job_to_send
1020	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1021	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1022	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1023	echo "%%END%%" >> $job_to_send
1024
1025	elif [[ $remote_host = necriam ]]
1026	then
1027	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1028	echo "#PBS -q $return_queue" >> $job_to_send
1029	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1030	echo "#PBS -j o" >> $job_to_send
1031	echo " " >> $job_to_send
1032	echo "set -x" >> $job_to_send
1033	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1034	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1035	echo "%%END%%" >> $job_to_send
1036
1037	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1038	then
1039	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1040	echo "#!/bin/bash" >> $job_to_send
1041	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1042	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1043	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1044	echo "#PBS -l feature=data" >> $job_to_send
1045	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1046	echo "#PBS -j oe" >> $job_to_send
1047	echo " " >> $job_to_send
1048	echo ". /usr/share/modules/init/bash" >> $job_to_send
1049	echo "set -x" >> $job_to_send
1050	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1051	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1052	echo "%%END%%" >> $job_to_send
1053
1054	elif [[ $remote_host = lctit ]]
1055	then
1056	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1057	echo "set -x" >> $job_to_send
1058	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1059	echo "cd $job_catalog" >> $job_to_send
1060	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1061	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1062	echo "%%END%%" >> $job_to_send
1063
1064	else
1065
1066	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1067	echo "# @\\\$-q $return_queue" >> $job_to_send
1068	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1069	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1070	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1071	then
1072	echo "# @\$-l mpp_p=0" >> $job_to_send
1073	fi
1074	echo '# @\$-lF 10mb' >> $job_to_send
1075	# echo '# @\$-o /dev/null' >> $job_to_send
1076	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1077	echo '# @\\\$-eo' >> $job_to_send
1078	echo " " >> $job_to_send
1079	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1080	then
1081	echo "set +vx" >> $job_to_send
1082	echo ". .profile" >> $job_to_send
1083	fi
1084	echo "set -x" >> $job_to_send
1085	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1086	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1087	echo "%%END%%" >> $job_to_send
1088	fi
1089
1090	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1091	then
1092	echo "llsubmit scpjob.$kennung" >> $job_to_send
1093	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1094	then
1095	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1096	echo "msub scpjob.$kennung" >> $job_to_send
1097	elif [[ $remote_host = lctit ]]
1098	then
1099	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1100	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1101	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1102	echo "sleep 10" >> $job_to_send
1103	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1104	then
1105	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1106	elif [[ $remote_host = t3es ]]
1107	then
1108	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1109	else
1110	echo "qsub scpjob.$kennung" >> $job_to_send
1111	fi
1112	echo "rm scpjob.$kennung" >> $job_to_send
1113	if [[ $remote_host = nech ]]
1114	then
1115	echo "cd -" >> $job_to_send
1116	fi
1117	else
1118	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1119	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1120	fi
1121	echo "set -x" >> $job_to_send
1122	echo " ' exit" >> $job_to_send
1123	echo "set -x" >> $job_to_send
1124	fi
1125
1126
1127
1128	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1129	cat $file_to_send >> $job_to_send
1130
1131	if [[ $remote_host = ibm ]]
1132	then
1133	echo " " >> $job_to_send
1134	echo "exit" >> $job_to_send
1135	fi
1136	if [[ $remote_host = lctit ]]
1137	then
1138	echo " " >> $job_to_send
1139	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1140	fi
1141
1142
1143
1144	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1145	if [[ -z $remote_user ]]
1146	then
1147	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1148	then
1149	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1150	fi
1151	fi
1152
1153
1154
1155	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1156	if [[ $no_submit = false ]]
1157	then
1158	if [[ $remote_host != $local_host ]]
1159	then
1160	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1161	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1162	then
1163	job_catalog_save=$job_catalog
1164	job_catalog=job_queue
1165	elif [[ $remote_host = nech ]]
1166	then
1167	job_catalog_save=$job_catalog
1168	job_catalog=/hpf/b/${remote_user}/job_queue
1169	fi
1170	if [[ $local_host = decalpha ]]
1171	then
1172	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1173	# PASSWORT AB
1174	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1175	elif [[ $remote_host = nech ]]
1176	then
1177	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1178	# TRANSFERIERT WERDEN
1179	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1180	else
1181	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1182	fi
1183	if [[ $? = 1 ]]
1184	then
1185	locat=scp; exit
1186	fi
1187	if [[ $remote_host = ibms ]]
1188	then
1189	job_catalog=$job_catalog_save
1190	fi
1191	[[ $verify = true ]] && printf "\n >>> finished\n"
1192	else
1193	eval job_catalog=$job_catalog
1194	cp $job_to_send ${job_catalog}/$job_on_remhost
1195	fi
1196
1197
1198
1199	# NQS- BZW. LOADLEVELER-JOB STARTEN
1200	if [[ $remote_host != $local_host ]]
1201	then
1202	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1203	if [[ $remote_host != lctit ]]
1204	then
1205	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1206	then
1207	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1208	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1209	else
1210	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1211	fi
1212	else
1213	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1214	# MIT SSH, DESHALB AUFRUF PER PIPE
1215	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1216	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1217	echo "first try" > send_protocol
1218	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1219	do
1220	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1221	then
1222	printf "\n trying to submit job to TIT cluster..."
1223	else
1224	printf "\n +++ failed ... trying to submit again ..."
1225	fi
1226	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1227	cat send_protocol
1228	done
1229	sleep 10
1230	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1231	echo "$submcom $job_on_remhost"
1232	rm send_protocol
1233	fi
1234
1235	[[ $verify = true ]] && printf " >>> o.k.\n"
1236	else
1237	cd $job_catalog
1238	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1239	then
1240	eval $submcom $job_on_remhost
1241	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1242	then
1243	eval $submcom $job_on_remhost
1244	echo "$submcom $job_on_remhost"
1245	chmod u+x $job_on_remhost
1246	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1247	then
1248	eval $submcom $job_on_remhost
1249	elif [[ $local_host = nech ]]
1250	then
1251	if [[ $queue = default ]]
1252	then
1253	eval $submcom $job_on_remhost
1254	else
1255	eval $submcom -q $queue $job_on_remhost
1256	fi
1257	else
1258	qsub $job_on_remhost
1259	fi
1260	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1261	[[ $local_host != lctit ]] && rm $job_on_remhost
1262	cd - > /dev/null
1263	fi
1264	fi
1265
1266
1267
1268	# ABSCHLUSSARBEITEN
1269
1270	if [[ $no_submit = false ]]
1271	then
1272	rm $job_to_send
1273	fi
1274	[[ $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 |