Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 426

Last change on this file since 426 was 416, checked in by weinreis, 15 years ago
mrun and subjob modified for archiving
Property svn:keywords set to `Id Rev`
File size: 45.5 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 416 2009-12-18 08:42:25Z 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	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
110	# adding special1q
111	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
112	# 16/10/09 - Carolin-adding permq
113	#
114
115	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
116	delete_dayfile=false
117	email_notification=none
118	group_number=none
119	locat=normal
120	no_default_queue=none
121	no_submit=false
122	job_catalog="~/job_queue"
123	job_name=none
124	local_user=$LOGNAME
125	node_usage=shared
126	numprocs=1
127	punkte="..........................................................."
128	submcom=qsub
129	queue=default
130	remote_host=none
131	remote_user=""
132	verify=true
133
134	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
135	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
136	typeset -L20 spalte1
137	typeset -R40 spalte2
138	typeset -L60 spalte3
139
140
141
142	# FEHLERBEHANDLUNG
143	# BEI EXIT:
144	trap 'if [[ $locat != normal ]]
145	then
146	case $locat in
147	(option) printf "\n --> available optios can be displayed"
148	printf " by typing:"
149	printf "\n \"subjob ?\" \n";;
150	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
151	(*) printf "\n +++ unknown error"
152	printf "\n please inform S. Raasch!\n"
153	esac
154	[[ -f $job_to_send ]] && rm $job_to_send
155	printf "\n\n+++ SUBJOB killed \n\n"
156	fi' exit
157
158
159	# BEI TERMINAL-BREAK:
160	trap '[[ -f $job_to_send ]] && rm $job_to_send
161	printf "\n\n+++ SUBJOB killed \n\n"
162	exit
163	' 2
164
165
166
167
168	# LOKALEN HOSTNAMEN FESTSTELLEN
169	local_host=$(hostname)
170
171
172
173	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
174	# UEBERHAUPT ZULAESSIG IST
175	case $local_host in
176	(atmos) local_addres=172.20.25.35; local_host=lcide;;
177	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
178	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
179	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
180	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
181	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
182	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
183	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
184	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
185	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
186	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
187	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
188	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
189	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
190	(cs*) local_addres=136.172.44.131; local_host=nech;;
191	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
192	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
193	(node*) local_addres=165.132.26.61 local_host=lck;;
194	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
195	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
196	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
197	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
198	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
199	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
200	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
201	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
202	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
203	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
204	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
205	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
206	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
207	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
208	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
209	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
210	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
211	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
212	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
213	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
214	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
215	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
216	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
217	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
218	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
219	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
220	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
221	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
222	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
223	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
224	(*) printf "\n +++ \"$local_host\" unknown";
225	printf "\n please inform S. Raasch!";
226	locat=parameter; exit;;
227	esac
228
229
230
231	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
232	remote_host=$local_host
233
234
235
236
237	# PROZEDUROPTIONEN EINLESEN
238	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
239	do
240	case $option in
241	(c) job_catalog=$OPTARG;;
242	(d) delete_dayfile=true;;
243	(D) no_submit=true;;
244	(e) email_notification=$OPTARG;;
245	(g) group_number=$OPTARG;;
246	(h) remote_host=$OPTARG;;
247	(m) memory=$OPTARG;;
248	(n) job_name=$OPTARG;;
249	(N) node_usage=$OPTARG;;
250	(O) threads_per_task=$OPTARG;;
251	(q) no_default_queue=$OPTARG;;
252	(t) cputime=$OPTARG;;
253	(T) tasks_per_node=$OPTARG;;
254	(u) remote_user=$OPTARG;;
255	(v) verify=false;;
256	(X) numprocs=$OPTARG;;
257	(\?) printf "\n +++ Option $OPTARG unknown \n";
258	locat=option; exit;;
259	esac
260	done
261
262
263	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
264	shift OPTIND-1; file_to_send=$1
265
266
267	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
268	if [ "$1" = "?" ]
269	then
270	(printf "\n *** subjob can be called as follows:\n"
271	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
272	printf "\n Description of available options:\n"
273	printf "\n Option Description Default-Value"
274	printf "\n -c job-input- and output-catalog ~/job_queue"
275	printf "\n -d no job-protocol will be created ---"
276	printf "\n -D only the job-file will be created ---"
277	printf "\n -h execution host, available hosts: $remote_host"
278	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
279	printf "\n lctit, nech, necriam, unics"
280	printf "\n -m memory demand per process in MByte ---"
281	printf "\n -n jobname <jobdatei>"
282	printf "\n -O threads per task (for OpenMP usage) 1"
283	printf "\n -q job-queue to be used default"
284	printf "\n -t allowed cpu-time in seconds ---"
285	printf "\n -T tasks per node (on parallel hosts) ---"
286	printf "\n -u username on execution host from .netrc"
287	printf "\n -v no prompt for confirmation ---"
288	printf "\n -X # of processors (on parallel hosts) 1"
289	printf "\n "
290	printf "\n The only possible positional parameter is <jobfile>:"
291	printf "\n The complete NQS-job must be provided here."
292	printf "\n <jobfile>=? creates this outline\n\n") \| more
293	exit
294	fi
295
296
297
298	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
299	if [[ "$file_to_send" = "" ]]
300	then
301	printf "\n +++ job-file missing"
302	locat=parameter; exit
303	else
304	if [[ -f $file_to_send ]]
305	then
306	true
307	else
308	printf "\n +++ job-file: "
309	printf "\n $file_to_send"
310	printf "\n does not exist"
311	locat=parameter; exit
312	fi
313	fi
314
315
316
317	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
318	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
319	if [[ $job_name = none ]]
320	then
321	job_name=$file_to_send
322	fi
323	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
324	then
325	printf "\n +++ job-file name: "
326	printf "\n $job_name"
327	printf "\n must not contain \"/\"-characters"
328	locat=parameter; exit
329	fi
330
331
332
333
334	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
335	# ODER WENN HOST NICHT ANGEGEBEN WURDE
336	if [[ $remote_host = none ]]
337	then
338	printf "\n +++ host missing"
339	locat=option; exit
340	else
341	case $remote_host in
342	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
343	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
344	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
345	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
346	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
347	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
348	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
349	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
350	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
351	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
352	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
353	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
354	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
355	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
356	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
357	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
358	locat=parameter; exit;;
359	esac
360	fi
361
362
363	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
364	if [[ $no_default_queue != none ]]
365	then
366	error=false
367	ndq=$no_default_queue
368	case $remote_host in
369	(ibm) case $ndq in
370	(p690_express\|p690_standard\|p690_long) error=false;;
371	(*) error=true;;
372	esac;;
373	(ibmb) case $ndq in
374	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
375	(*) error=true;;
376	esac;;
377	(ibmh) case $ndq in
378	(no_class) error=false;;
379	(*) error=true;;
380	esac;;
381	(ibms) case $ndq in
382	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
383	(*) error=true;;
384	esac;;
385	(ibmy) case $ndq in
386	(parallel) error=false;;
387	(*) error=true;;
388	esac;;
389	(lcsgib\|lcsgih) case $ndq in
390	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
391	(*) error=true;;
392	esac;;
393	(lctit) case $ndq in
394	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
395	(*) error=true;;
396	esac;;
397	(t3eb) case $ndq in
398	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
399	(*) error=true;;
400	esac;;
401	(necriam) case $ndq in
402	(SP\|SS\|P6) error=false;;
403	(*) error=true;;
404	esac;;
405	(t3eh) case $ndq in
406	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
407	(*) error=true;;
408	esac;;
409	(t3ej2\|t3ej5) case $ndq in
410	(low\|normal\|high) error=false;;
411	(*) error=true;;
412	esac;;
413	(t3es) case $ndq in
414	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
415	(*) error=true;;
416	esac;;
417	(unics) case $ndq in
418	(unics\|ht) error=false;;
419	(*) error=true;;
420	esac;;
421	esac
422	if [[ $error = true ]]
423	then
424	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
425	locat=parameter; exit
426	else
427	queue=$no_default_queue
428	fi
429	fi
430
431
432
433	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
434	if [[ $node_usage != shared && $queue = cdev ]]
435	then
436	node_usage=shared
437	fi
438
439
440
441	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
442	# AUFTEILEN
443	done=false
444	while [[ $done = false ]]
445	do
446	if (( $cputime <= 0 ))
447	then
448	printf "\n +++ wrong cpu-time or cpu-time missing"
449	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
450	printf "\n >>> "
451	read cputime 1>/dev/null 2>&1
452	else
453	done=true
454	fi
455	done
456	if [[ $remote_host = nech ]]
457	then
458	if (( tasks_per_node != 0 ))
459	then
460	(( cputime = cputime * tasks_per_node ))
461	elif [[ $numprocs != 0 ]]
462	then
463	(( cputime = cputime * numprocs ))
464	fi
465	fi
466	(( stunden = cputime / 3600 ))
467	(( resttime = cputime - stunden * 3600 ))
468	(( minuten = resttime / 60 ))
469	(( sekunden = resttime - minuten * 60 ))
470	timestring=${stunden}:${minuten}:${sekunden}
471
472
473
474	# PRUEFEN DER KERNSPEICHERANFORDERUNG
475	done=false
476	while [[ $done = false ]]
477	do
478	if (( memory <= 0 ))
479	then
480	printf "\n +++ wrong memory demand or memory demand missing"
481	printf "\n >>> Please type memory in MByte per process as INTEGER:"
482	printf "\n >>> "
483	read memory 1>/dev/null 2>&1
484	else
485	done=true
486	fi
487	done
488
489	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
490	then
491	if (( tasks_per_node != 0 ))
492	then
493	(( Memory = memory * tasks_per_node / 1000 ))
494	elif [[ $numprocs != 0 ]]
495	then
496	(( Memory = memory * numprocs / 1000 ))
497	else
498	(( Memory = memory / 1000 ))
499	fi
500	fi
501
502
503	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
504	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
505	then
506	(( memory = memory * threads_per_task ))
507	fi
508
509
510	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
511	if (( tasks_per_node != 0 ))
512	then
513	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
514	then
515	(( nodes = numprocs / tasks_per_node ))
516	else
517	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
518	fi
519	fi
520
521
522
523	# HEADER-AUSGABE
524	if [[ $verify = true ]]
525	then
526	printf "\n\n"
527	printf "#--------------------------------------------------------------# \n"
528	spalte1=SUBJOB;spalte2=$(date)
529	printf "\| $spalte1$spalte2 \| \n"
530	printf "\| \| \n"
531	printf "\| values of parameters/options: \| \n"
532	spalte1=local_host$punkte; spalte2=$punkte$local_host
533	printf "\| $spalte1$spalte2 \| \n"
534	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
535	printf "\| $spalte1$spalte2 \| \n"
536	spalte1=queue$punkte; spalte2=$punkte$queue
537	printf "\| $spalte1$spalte2 \| \n"
538	spalte1=memory$punkte; spalte2="$punkte$memory mb"
539	printf "\| $spalte1$spalte2 \| \n"
540	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
541	printf "\| $spalte1$spalte2 \| \n"
542	spalte1=job_name$punkte; spalte2="$punkte$job_name"
543	printf "\| $spalte1$spalte2 \| \n"
544	printf "#--------------------------------------------------------------# \n\n"
545
546
547	# KONTROLLABFRAGE, OB ALLES O.K.
548	antwort="dummy"
549	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
550	do
551	read antwort?" >>> continue (y/n) ? "
552	done
553	if [[ $antwort = n \|\| $antwort = N ]]
554	then
555	locat=verify; exit
556	fi
557	printf "\n"
558	fi
559
560
561
562	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
563	kennung=$RANDOM
564	job_on_remhost=${job_name}_${kennung}_$local_host
565	job_to_send=job_to_send_$kennung
566	if [[ $delete_dayfile = false ]]
567	then
568	remote_dayfile=${local_host}_${job_name}_result_$kennung
569	local_dayfile=${remote_host}_${job_name}
570	else
571	remote_dayfile=/dev/null
572	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
573	then
574	notification='# @ notification = never'
575	fi
576	fi
577
578
579
580	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
581	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
582	then
583
584	if [[ $remote_host = ibmy ]]
585	then
586	consumable_memory=""
587	else
588	consumable_memory="ConsumableMemory($memory mb)"
589	fi
590
591	if [[ $remote_host = ibmh ]]
592	then
593	data_limit=""
594	network_to_use=""
595	class=""
596	environment=""
597	else
598	class="# @ class = $queue"
599	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
600	if [[ $queue = cdev ]]
601	then
602	data_limit="# @ data_limit = 1.76gb"
603	network_to_use="# @ network.mpi = sn_all,shared,ip"
604	else
605	if [[ $remote_host = ibms ]]
606	then
607	network_to_use="# @ network.mpi = csss,shared,us"
608	elif [[ $remote_host = ibmy ]]
609	then
610	network_to_use=""
611	else
612	network_to_use="# @ network.mpi = sn_all,shared,us"
613	data_limit="# @ data_limit = 1.76gb"
614	fi
615	fi
616	fi
617
618	cat > $job_to_send << %%END%%
619	#!/bin/ksh
620	# @ shell = /bin/ksh
621
622	# @ job_type = parallel
623	# @ job_name = palm
624	# @ wall_clock_limit = ${timestring},$timestring
625	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
626	# @ output = $remote_dayfile
627	# @ error = $remote_dayfile
628	# @ image_size = 50
629	$class
630	$environment
631	$network_to_use
632	$data_limit
633	$notification
634
635	%%END%%
636
637	if (( nodes > 0 ))
638	then
639
640	cat >> $job_to_send << %%END%%
641	# @ node = $nodes
642	# @ tasks_per_node = $tasks_per_node
643	# @ node_usage = $node_usage
644	# @ queue
645
646	%%END%%
647
648	else
649
650	if [[ $remote_host != ibmy ]]
651	then
652
653	cat >> $job_to_send << %%END%%
654	# @ blocking = unlimited
655	# @ total_tasks = $numprocs
656	# @ node_usage = $node_usage
657	# @ queue
658
659	%%END%%
660
661	else
662
663	cat >> $job_to_send << %%END%%
664	# @ node = 1
665	# @ total_tasks = $numprocs
666	# @ queue
667
668	%%END%%
669
670	fi
671
672	fi
673
674	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
675	then
676
677	cat > $job_to_send << %%END%%
678	#!/bin/ksh
679
680	# @ job_type = serial
681	# @ node_usage = $node_usage
682	# @ job_name = palm
683	# @ wall_clock_limit = ${timestring},$timestring
684	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
685	# @ output = $remote_dayfile
686	# @ error = $remote_dayfile
687	$class
688	$notification
689
690	# @ queue
691
692	%%END%%
693
694	elif [[ $remote_host = lcfimm ]]
695	then
696
697	if [[ $numprocs != 0 ]]
698	then
699	cat > $job_to_send << %%END%%
700	#!/bin/ksh
701	#PBS -N $job_name
702	#PBS -A geofysisk
703	#PBS -l walltime=$timestring
704	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
705	#PBS -l pmem=${memory}mb
706	#PBS -m abe
707	#PBS -M maronga@muk.uni-hannover.de
708	#PBS -o $remote_dayfile
709	#PBS -j oe
710	mpd &
711
712	%%END%%
713
714	else
715	cat > $job_to_send << %%END%%
716	#!/bin/ksh
717	#PBS -N $job_name
718	#PBS -A geofysisk
719	#PBS -l walltime=$timestring
720	#PBS -l ncpus=1
721	#PBS -l pmem=${memory}mb
722	#PBS -m abe
723	#PBS -M maronga@muk.uni-hannover.de
724	#PBS -o $remote_dayfile
725	#PBS -j oe
726
727	%%END%%
728
729	fi
730
731	elif [[ $remote_host = lck ]]
732	then
733
734	if [[ $numprocs != 0 ]]
735	then
736	cat > $job_to_send << %%END%%
737	#!/bin/ksh
738	#PBS -N $job_name
739	#PBS -l walltime=$timestring
740	#PBS -l ncpus=$numprocs
741	#PBS -l pmem=${memory}mb
742	#PBS -o $remote_dayfile
743	#PBS -j oe
744
745	mpd &
746
747	%%END%%
748
749	else
750	cat > $job_to_send << %%END%%
751	#!/bin/ksh
752	#PBS -N $job_name
753	#PBS -l walltime=$timestring
754	#PBS -l ncpus=1
755	#PBS -l pmem=${memory}mb
756	#PBS -o $remote_dayfile
757	#PBS -j oe
758
759	%%END%%
760
761	fi
762
763	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
764	then
765
766	if [[ "$sgi_feature" != "" ]]
767	then
768	feature_directive="#PBS -l feature=$sgi_feature"
769	else
770	feature_directive=""
771	fi
772
773	if [[ $queue = dataq \|\| $queue = permq ]]
774	then
775	feature_directive="#PBS -l feature=data"
776	fi
777
778	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
779	then
780	queue_directive="#PBS -q $queue"
781	else
782	queue_directive=""
783	fi
784
785	if [[ $email_notification = none ]]
786	then
787	email_directive=""
788	else
789	email_directive="#PBS -M $email_notification"
790	fi
791
792	if [[ $numprocs != 0 ]]
793	then
794	cat > $job_to_send << %%END%%
795	#!/bin/bash
796	#PBS -N $job_name
797	#PBS -l walltime=$timestring
798	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
799	#PBS -l naccesspolicy=$node_usage
800	#PBS -o $remote_dayfile
801	#PBS -j oe
802	$feature_directive
803	$queue_directive
804	$email_directive
805
806	. /usr/share/modules/init/bash
807	module load ifort/11.0.069
808	module load $mpilib
809	module load netcdf
810
811	echo ld_library_path=\$LD_LIBRARY_PATH
812
813	%%END%%
814
815	else
816	cat > $job_to_send << %%END%%
817	#PBS -S /bin/bash
818	#PBS -N $job_name
819	#PBS -l walltime=$timestring
820	#PBS -l ncpus=1
821	#PBS -o $remote_dayfile
822	#PBS -e $remote_dayfile
823	$queue_directive
824	$email_directive
825
826	%%END%%
827
828	fi
829
830	elif [[ $remote_host = lcxt4 ]]
831	then
832
833	if [[ $numprocs != 0 ]]
834	then
835	cat > $job_to_send << %%END%%
836	#!/bin/ksh
837	#PBS -S /bin/ksh
838	#PBS -N $job_name
839	#PBS -A geofysisk
840	#PBS -j oe
841	#PBS -l walltime=$timestring
842	#PBS -l mppwidth=${numprocs}
843	#PBS -l mppnppn=${tasks_per_node}
844	#PBS -m abe
845	#PBS -M maronga@muk.uni-hannover.de
846	#PBS -o $remote_dayfile
847
848	. /opt/modules/default/init/ksh
849	module load modules/3.1.6
850	module load PrgEnv-pgi/2.1.56HDA
851	module load Base-opts/2.1.56HDA
852	module load moab/5.2.5-s13570
853	module load torque/2.3.4-snap.200809221601
854
855
856	%%END%%
857
858	else
859	cat > $job_to_send << %%END%%
860	#!/bin/ksh
861	#PBS -S /bin/ksh
862	#PBS -N $job_name
863	#PBS -A geofysisk
864	#PBS -j oe
865	#PBS -l walltime=$timestring
866	#PBS -l ncpus=1
867	#PBS -l pmem=${memory}mb
868	#PBS -m abe
869	#PBS -M maronga@muk.uni-hannover.de
870	#PBS -o $remote_dayfile
871
872	. /opt/modules/default/init/ksh
873	module load modules/3.1.6
874	module load PrgEnv-pgi/2.1.56HD
875	module load Base-opts/2.1.56HD
876	module load moab/5.2.5-s13441
877	module load torque/2.3.4-snap.200809221601
878
879	%%END%%
880
881	fi
882
883	elif [[ $remote_host = nech ]]
884	then
885
886	if (( nodes > 1 ))
887	then
888	cat > $job_to_send << %%END%%
889	#!/bin/ksh
890	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
891	#PBS -l ${qsubmem}=${Memory}gb
892	#PBS -b $nodes
893	#PBS -o $remote_dayfile
894	#PBS -N palm
895	#PBS -j o
896	#PBS -T mpisx
897
898	%%END%%
899
900	elif [[ $numprocs != 0 ]]
901	then
902	cat > $job_to_send << %%END%%
903	#!/bin/ksh
904	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
905	#PBS -l ${qsubmem}=${Memory}gb
906	#PBS -o $remote_dayfile
907	#PBS -N palm
908	#PBS -j o
909
910	%%END%%
911
912	else
913	cat > $job_to_send << %%END%%
914	#!/bin/ksh
915	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
916	#PBS -o $remote_dayfile
917	#PBS -j o
918
919	%%END%%
920
921	fi
922
923	elif [[ $remote_host = necriam ]]
924	then
925
926	if (( nodes > 1 ))
927	then
928	cat > $job_to_send << %%END%%
929	#!/bin/ksh
930	#PBS -b $nodes
931	#PBS -o $remote_dayfile
932	#PBS -N $job_name
933	#PBS -j o
934	#PBS -v MPIPROGINV=YES
935
936	%%END%%
937
938	elif [[ $numprocs != 0 ]]
939	then
940	cat > $job_to_send << %%END%%
941	#!/bin/ksh
942	#PBS -q ${queue}
943	#PBS -o $remote_dayfile
944	#PBS -N $job_name
945	#PBS -j o
946	#PBS -v MPIPROGINV=YES
947
948	%%END%%
949
950	else
951	cat > $job_to_send << %%END%%
952	#!/bin/ksh
953	#PBS -o $remote_dayfile
954	#PBS -j o
955
956	%%END%%
957
958	fi
959
960	elif [[ $remote_host = lctit ]]
961	then
962	cat > $job_to_send << %%END%%
963	#!/bin/bash
964	#$ -S /bin/bash
965	cd $job_catalog
966	export PATH=$PALM_BIN:\$PATH
967	echo \$PATH
968
969	%%END%%
970
971	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
972	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
973	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
974
975	else
976
977	cat > $job_to_send << %%END%%
978	# @\$-q ${queue}
979	# @\$-l${qsubtime} $timestring
980	# @\$-l${qsubmem} ${memory}mb
981	# @\$-o $remote_dayfile
982	# @\$-eo
983
984	%%END%%
985
986	fi
987
988
989	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
990	# BEI EXIT VERANLASSEN
991	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
992	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
993	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
994	then
995	echo "set +vx" >> $job_to_send
996	echo "trap '" >> $job_to_send
997	echo "set +vx" >> $job_to_send
998	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 ]]
999	then
1000	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
1001	then
1002	return_queue=c1
1003	elif [[ $remote_host = ibms ]]
1004	then
1005	return_queue=p_normal
1006	elif [[ $remote_host = ibmy ]]
1007	then
1008	return_queue=serial
1009	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1010	then
1011	return_queue=serialq
1012	elif [[ $remote_host = lctit ]]
1013	then
1014	return_queue=sla3
1015	elif [[ $remote_host = necriam ]]
1016	then
1017	return_queue=SP
1018	else
1019	return_queue=unknown
1020	fi
1021
1022	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1023	then
1024
1025	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1026	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1027	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1028	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1029	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1030	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1031	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1032	if [[ $host != "ibmh" ]]
1033	then
1034	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1035	fi
1036	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1037	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1038
1039	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1040	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1041
1042	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1043	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1044	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1045
1046	elif [[ $remote_host = nech ]]
1047	then
1048	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1049	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1050	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1051	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1052	echo "#PBS -j o" >> $job_to_send
1053	echo " " >> $job_to_send
1054	echo "set -x" >> $job_to_send
1055	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1056	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1057	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1058	echo "%%END%%" >> $job_to_send
1059
1060	elif [[ $remote_host = necriam ]]
1061	then
1062	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1063	echo "#PBS -q $return_queue" >> $job_to_send
1064	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1065	echo "#PBS -j o" >> $job_to_send
1066	echo " " >> $job_to_send
1067	echo "set -x" >> $job_to_send
1068	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1069	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1070	echo "%%END%%" >> $job_to_send
1071
1072	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1073	then
1074	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1075	echo "#!/bin/bash" >> $job_to_send
1076	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1077	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1078	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1079	echo "#PBS -l feature=data" >> $job_to_send
1080	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1081	echo "#PBS -j oe" >> $job_to_send
1082	echo " " >> $job_to_send
1083	echo ". /usr/share/modules/init/bash" >> $job_to_send
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" >> $job_to_send
1086	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1087	echo "%%END%%" >> $job_to_send
1088
1089	elif [[ $remote_host = lctit ]]
1090	then
1091	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1092	echo "set -x" >> $job_to_send
1093	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1094	echo "cd $job_catalog" >> $job_to_send
1095	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1096	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1097	echo "%%END%%" >> $job_to_send
1098
1099	else
1100
1101	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1102	echo "# @\\\$-q $return_queue" >> $job_to_send
1103	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1104	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1105	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1106	then
1107	echo "# @\$-l mpp_p=0" >> $job_to_send
1108	fi
1109	echo '# @\$-lF 10mb' >> $job_to_send
1110	# echo '# @\$-o /dev/null' >> $job_to_send
1111	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1112	echo '# @\\\$-eo' >> $job_to_send
1113	echo " " >> $job_to_send
1114	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1115	then
1116	echo "set +vx" >> $job_to_send
1117	echo ". .profile" >> $job_to_send
1118	fi
1119	echo "set -x" >> $job_to_send
1120	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1121	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1122	echo "%%END%%" >> $job_to_send
1123	fi
1124
1125	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1126	then
1127	echo "llsubmit scpjob.$kennung" >> $job_to_send
1128	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1129	then
1130	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1131	echo "msub scpjob.$kennung" >> $job_to_send
1132	elif [[ $remote_host = lctit ]]
1133	then
1134	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1135	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1136	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1137	echo "sleep 10" >> $job_to_send
1138	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1139	then
1140	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1141	elif [[ $remote_host = t3es ]]
1142	then
1143	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1144	else
1145	echo "qsub scpjob.$kennung" >> $job_to_send
1146	fi
1147	echo "rm scpjob.$kennung" >> $job_to_send
1148	if [[ $remote_host = nech ]]
1149	then
1150	echo "cd -" >> $job_to_send
1151	fi
1152	else
1153	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1154	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1155	fi
1156	echo "set -x" >> $job_to_send
1157	echo " ' exit" >> $job_to_send
1158	echo "set -x" >> $job_to_send
1159	fi
1160
1161
1162
1163	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1164	cat $file_to_send >> $job_to_send
1165
1166	if [[ $remote_host = ibm ]]
1167	then
1168	echo " " >> $job_to_send
1169	echo "exit" >> $job_to_send
1170	fi
1171	if [[ $remote_host = lctit ]]
1172	then
1173	echo " " >> $job_to_send
1174	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1175	fi
1176
1177
1178
1179	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1180	if [[ -z $remote_user ]]
1181	then
1182	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1183	then
1184	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1185	fi
1186	fi
1187
1188
1189
1190	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1191	if [[ $no_submit = false ]]
1192	then
1193	if [[ $remote_host != $local_host ]]
1194	then
1195	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1196	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1197	then
1198	job_catalog_save=$job_catalog
1199	job_catalog=job_queue
1200	elif [[ $remote_host = nech ]]
1201	then
1202	job_catalog_save=$job_catalog
1203	job_catalog=/hpf/b/${remote_user}/job_queue
1204	fi
1205	if [[ $local_host = decalpha ]]
1206	then
1207	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1208	# PASSWORT AB
1209	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1210	elif [[ $remote_host = nech ]]
1211	then
1212	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1213	# TRANSFERIERT WERDEN
1214	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1215	else
1216	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1217	fi
1218	if [[ $? = 1 ]]
1219	then
1220	locat=scp; exit
1221	fi
1222	if [[ $remote_host = ibms ]]
1223	then
1224	job_catalog=$job_catalog_save
1225	fi
1226	[[ $verify = true ]] && printf "\n >>> finished\n"
1227	else
1228	eval job_catalog=$job_catalog
1229	cp $job_to_send ${job_catalog}/$job_on_remhost
1230	fi
1231
1232
1233
1234	# NQS- BZW. LOADLEVELER-JOB STARTEN
1235	if [[ $remote_host != $local_host ]]
1236	then
1237	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1238	if [[ $remote_host != lctit ]]
1239	then
1240	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1241	then
1242	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1243	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1244	else
1245	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1246	fi
1247	else
1248	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1249	# MIT SSH, DESHALB AUFRUF PER PIPE
1250	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1251	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1252	echo "first try" > send_protocol
1253	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1254	do
1255	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1256	then
1257	printf "\n trying to submit job to TIT cluster..."
1258	else
1259	printf "\n +++ failed ... trying to submit again ..."
1260	fi
1261	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1262	cat send_protocol
1263	done
1264	sleep 10
1265	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1266	echo "$submcom $job_on_remhost"
1267	rm send_protocol
1268	fi
1269
1270	[[ $verify = true ]] && printf " >>> o.k.\n"
1271	else
1272	cd $job_catalog
1273	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1274	then
1275	eval $submcom $job_on_remhost
1276	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1277	then
1278	eval $submcom $job_on_remhost
1279	echo "$submcom $job_on_remhost"
1280	chmod u+x $job_on_remhost
1281
1282	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1283	then
1284	eval $submcom $job_on_remhost
1285	elif [[ $local_host = nech ]]
1286	then
1287	if [[ $queue = default ]]
1288	then
1289	eval $submcom $job_on_remhost
1290	else
1291	eval $submcom -q $queue $job_on_remhost
1292	fi
1293	else
1294	qsub $job_on_remhost
1295	fi
1296	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1297	[[ $local_host != lctit ]] && rm $job_on_remhost
1298	cd - > /dev/null
1299	fi
1300	fi
1301
1302
1303
1304	# ABSCHLUSSARBEITEN
1305
1306	if [[ $no_submit = false ]]
1307	then
1308	rm $job_to_send
1309	fi
1310	[[ $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 |