Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 107

Last change on this file since 107 was 102, checked in by raasch, 17 years ago
preliminary version for coupled runs
Property svn:keywords set to `Rev`
File size: 35.4 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 54 2007-03-08 00:00:02Z 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
86
87	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
88	delete_dayfile=false
89	locat=normal
90	no_default_queue=none
91	no_submit=false
92	job_catalog="~/job_queue"
93	job_name=none
94	local_user=$LOGNAME
95	node_usage=shared
96	numprocs=1
97	punkte="..........................................................."
98	submcom=qsub
99	queue=default
100	remote_host=none
101	remote_user=""
102	verify=true
103
104	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
105	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
106	typeset -L20 spalte1
107	typeset -R40 spalte2
108	typeset -L60 spalte3
109
110
111
112	# FEHLERBEHANDLUNG
113	# BEI EXIT:
114	trap 'if [[ $locat != normal ]]
115	then
116	case $locat in
117	(option) printf "\n --> available optios can be displayed"
118	printf " by typing:"
119	printf "\n \"subjob ?\" \n";;
120	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
121	(*) printf "\n +++ unknown error"
122	printf "\n please inform S. Raasch!\n"
123	esac
124	[[ -f $job_to_send ]] && rm $job_to_send
125	printf "\n\n+++ SUBJOB killed \n\n"
126	fi' exit
127
128
129	# BEI TERMINAL-BREAK:
130	trap '[[ -f $job_to_send ]] && rm $job_to_send
131	printf "\n\n+++ SUBJOB killed \n\n"
132	exit
133	' 2
134
135
136
137
138	# LOKALEN HOSTNAMEN FESTSTELLEN
139	local_host=$(hostname)
140
141
142
143	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
144	# UEBERHAUPT ZULAESSIG IST
145	case $local_host in
146	(atmos) local_addres=172.20.25.35; local_host=lcide;;
147	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
148	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
149	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
150	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
151	(cs*) local_addres=136.172.44.131; local_host=nech;;
152	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
153	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
154	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
155	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
156	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
157	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
158	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
159	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
160	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
161	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
162	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
163	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
164	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
165	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
166	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
167	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
168	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
169	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
170	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
171	(*) printf "\n +++ \"$local_host\" unknown";
172	printf "\n please inform S. Raasch!";
173	locat=parameter; exit;;
174	esac
175
176
177
178	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
179	remote_host=$local_host
180
181
182
183
184	# PROZEDUROPTIONEN EINLESEN
185	while getopts :c:dDh:m:n:N:O:q:t:T:u:vX: option
186	do
187	case $option in
188	(c) job_catalog=$OPTARG;;
189	(d) delete_dayfile=true;;
190	(D) no_submit=true;;
191	(h) remote_host=$OPTARG;;
192	(m) memory=$OPTARG;;
193	(n) job_name=$OPTARG;;
194	(N) node_usage=$OPTARG;;
195	(O) threads_per_task=$OPTARG;;
196	(q) no_default_queue=$OPTARG;;
197	(t) cputime=$OPTARG;;
198	(T) tasks_per_node=$OPTARG;;
199	(u) remote_user=$OPTARG;;
200	(v) verify=false;;
201	(X) numprocs=$OPTARG;;
202	(\?) printf "\n +++ Option $OPTARG unknown \n";
203	locat=option; exit;;
204	esac
205	done
206
207
208	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
209	shift OPTIND-1; file_to_send=$1
210
211
212	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
213	if [ "$1" = "?" ]
214	then
215	(printf "\n *** subjob can be called as follows:\n"
216	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
217	printf "\n Description of available options:\n"
218	printf "\n Option Description Default-Value"
219	printf "\n -c job-input- and output-catalog ~/job_queue"
220	printf "\n -d no job-protocol will be created ---"
221	printf "\n -D only the job-file will be created ---"
222	printf "\n -h execution host, available hosts: $remote_host"
223	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
224	printf "\n lctit, nech, neck, unics"
225	printf "\n -m memory demand per process in MByte ---"
226	printf "\n -n jobname <jobdatei>"
227	printf "\n -O threads per task (for OpenMP usage) 1"
228	printf "\n -q job-queue to be used default"
229	printf "\n -t allowed cpu-time in seconds ---"
230	printf "\n -T tasks per node (on parallel hosts) ---"
231	printf "\n -u username on execution host from .netrc"
232	printf "\n -v no prompt for confirmation ---"
233	printf "\n -X # of processors (on parallel hosts) 1"
234	printf "\n "
235	printf "\n The only possible positional parameter is <jobfile>:"
236	printf "\n The complete NQS-job must be provided here."
237	printf "\n <jobfile>=? creates this outline\n\n") \| more
238	exit
239	fi
240
241
242
243	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
244	if [[ "$file_to_send" = "" ]]
245	then
246	printf "\n +++ job-file missing"
247	locat=parameter; exit
248	else
249	if [[ -f $file_to_send ]]
250	then
251	true
252	else
253	printf "\n +++ job-file: "
254	printf "\n $file_to_send"
255	printf "\n does not exist"
256	locat=parameter; exit
257	fi
258	fi
259
260
261
262	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
263	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
264	if [[ $job_name = none ]]
265	then
266	job_name=$file_to_send
267	fi
268	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
269	then
270	printf "\n +++ job-file name: "
271	printf "\n $job_name"
272	printf "\n must not contain \"/\"-characters"
273	locat=parameter; exit
274	fi
275
276
277
278
279	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
280	# ODER WENN HOST NICHT ANGEGEBEN WURDE
281	if [[ $remote_host = none ]]
282	then
283	printf "\n +++ host missing"
284	locat=option; exit
285	else
286	case $remote_host in
287	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
288	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
289	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
290	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
291	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
292	(lctit) queue=default; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
293	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
294	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
295	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
296	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
297	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
298	locat=parameter; exit;;
299	esac
300	fi
301
302
303	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
304	if [[ $no_default_queue != none ]]
305	then
306	error=false
307	ndq=$no_default_queue
308	case $remote_host in
309	(ibm) case $ndq in
310	(p690_express\|p690_standard\|p690_long) error=false;;
311	(*) error=true;;
312	esac;;
313	(ibmb) case $ndq in
314	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
315	(*) error=true;;
316	esac;;
317	(ibmh) case $ndq in
318	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
319	(*) error=true;;
320	esac;;
321	(ibms) case $ndq in
322	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
323	(*) error=true;;
324	esac;;
325	(ibmy) case $ndq in
326	(parallel) error=false;;
327	(*) error=true;;
328	esac;;
329	(lctit) case $ndq in
330	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
331	(*) error=true;;
332	esac;;
333	(t3eb) case $ndq in
334	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
335	(*) error=true;;
336	esac;;
337	(t3eh) case $ndq in
338	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
339	(*) error=true;;
340	esac;;
341	(t3ej2\|t3ej5) case $ndq in
342	(low\|normal\|high) error=false;;
343	(*) error=true;;
344	esac;;
345	(t3es) case $ndq in
346	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
347	(*) error=true;;
348	esac;;
349	(unics) case $ndq in
350	(unics\|ht) error=false;;
351	(*) error=true;;
352	esac;;
353	esac
354	if [[ $error = true ]]
355	then
356	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
357	locat=parameter; exit
358	else
359	queue=$no_default_queue
360	fi
361	fi
362
363
364
365	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
366	if [[ $node_usage != shared && $queue = cdev ]]
367	then
368	node_usage=shared
369	fi
370
371
372
373	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
374	# AUFTEILEN
375	done=false
376	while [[ $done = false ]]
377	do
378	if (( $cputime <= 0 ))
379	then
380	printf "\n +++ wrong cpu-time or cpu-time missing"
381	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
382	printf "\n >>> "
383	read cputime 1>/dev/null 2>&1
384	else
385	done=true
386	fi
387	done
388	if [[ $remote_host = nech ]]
389	then
390	if (( tasks_per_node != 0 ))
391	then
392	(( cputime = cputime * tasks_per_node ))
393	elif [[ $numprocs != 0 ]]
394	then
395	(( cputime = cputime * numprocs ))
396	fi
397	fi
398	(( stunden = cputime / 3600 ))
399	(( resttime = cputime - stunden * 3600 ))
400	(( minuten = resttime / 60 ))
401	(( sekunden = resttime - minuten * 60 ))
402	timestring=${stunden}:${minuten}:${sekunden}
403
404
405
406	# PRUEFEN DER KERNSPEICHERANFORDERUNG
407	done=false
408	while [[ $done = false ]]
409	do
410	if (( memory <= 0 ))
411	then
412	printf "\n +++ wrong memory demand or memory demand missing"
413	printf "\n >>> Please type memory in MByte per process as INTEGER:"
414	printf "\n >>> "
415	read memory 1>/dev/null 2>&1
416	else
417	done=true
418	fi
419	done
420
421	if [[ $remote_host = nech \|\| $remote_host = neck ]]
422	then
423	if (( tasks_per_node != 0 ))
424	then
425	(( Memory = memory * tasks_per_node / 1000 ))
426	elif [[ $numprocs != 0 ]]
427	then
428	(( Memory = memory * numprocs / 1000 ))
429	else
430	(( Memory = memory / 1000 ))
431	fi
432	fi
433
434
435	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
436	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
437	then
438	(( memory = memory * threads_per_task ))
439	fi
440
441
442	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
443	if (( tasks_per_node != 0 ))
444	then
445	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
446	fi
447
448
449
450	# HEADER-AUSGABE
451	if [[ $verify = true ]]
452	then
453	printf "\n\n"
454	printf "#--------------------------------------------------------------# \n"
455	spalte1=SUBJOB;spalte2=$(date)
456	printf "\| $spalte1$spalte2 \| \n"
457	printf "\| \| \n"
458	printf "\| values of parameters/options: \| \n"
459	spalte1=local_host$punkte; spalte2=$punkte$local_host
460	printf "\| $spalte1$spalte2 \| \n"
461	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
462	printf "\| $spalte1$spalte2 \| \n"
463	spalte1=queue$punkte; spalte2=$punkte$queue
464	printf "\| $spalte1$spalte2 \| \n"
465	spalte1=memory$punkte; spalte2="$punkte$memory mb"
466	printf "\| $spalte1$spalte2 \| \n"
467	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
468	printf "\| $spalte1$spalte2 \| \n"
469	spalte1=job_name$punkte; spalte2="$punkte$job_name"
470	printf "\| $spalte1$spalte2 \| \n"
471	printf "#--------------------------------------------------------------# \n\n"
472
473
474	# KONTROLLABFRAGE, OB ALLES O.K.
475	antwort="dummy"
476	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
477	do
478	read antwort?" >>> continue (y/n) ? "
479	done
480	if [[ $antwort = n \|\| $antwort = N ]]
481	then
482	locat=verify; exit
483	fi
484	printf "\n"
485	fi
486
487
488
489	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
490	kennung=$RANDOM
491	job_on_remhost=${job_name}_${kennung}_$local_host
492	job_to_send=job_to_send_$kennung
493	if [[ $delete_dayfile = false ]]
494	then
495	remote_dayfile=${local_host}_${job_name}_result_$kennung
496	local_dayfile=${remote_host}_${job_name}
497	else
498	remote_dayfile=/dev/null
499	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
500	then
501	notification='# @ notification = never'
502	fi
503	fi
504
505
506
507	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
508	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
509	then
510
511	if [[ $remote_host = ibmy ]]
512	then
513	consumable_memory=""
514	else
515	consumable_memory="ConsumableMemory($memory mb)"
516	fi
517	if [[ $queue = cdev ]]
518	then
519	data_limit="# @ data_limit = 1.76gb"
520	network_to_use="# @ network.mpi = sn_all,shared,ip"
521	else
522	if [[ $remote_host = ibms ]]
523	then
524	network_to_use="# @ network.mpi = csss,shared,us"
525	elif [[ $remote_host = ibmy ]]
526	then
527	network_to_use=""
528	else
529	network_to_use="# @ network.mpi = sn_all,shared,us"
530	data_limit="# @ data_limit = 1.76gb"
531	fi
532	fi
533
534	cat > $job_to_send << %%END%%
535	#!/bin/ksh
536
537	# @ job_type = parallel
538	# @ job_name = palm
539	# @ wall_clock_limit = ${timestring},$timestring
540	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
541	# @ output = $remote_dayfile
542	# @ error = $remote_dayfile
543	# @ image_size = 50
544	# @ class = $queue
545	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
546	$network_to_use
547	$data_limit
548	$notification
549
550	%%END%%
551
552	if (( nodes > 0 ))
553	then
554
555	cat >> $job_to_send << %%END%%
556	# @ node = $nodes
557	# @ tasks_per_node = $tasks_per_node
558	# @ node_usage = $node_usage
559	# @ queue
560
561	%%END%%
562
563	else
564
565	if [[ $remote_host != ibmy ]]
566	then
567
568	cat >> $job_to_send << %%END%%
569	# @ blocking = unlimited
570	# @ total_tasks = $numprocs
571	# @ node_usage = $node_usage
572	# @ queue
573
574	%%END%%
575
576	else
577
578	cat >> $job_to_send << %%END%%
579	# @ node = 1
580	# @ total_tasks = $numprocs
581	# @ queue
582
583	%%END%%
584
585	fi
586
587	fi
588
589	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
590	then
591
592	cat > $job_to_send << %%END%%
593	#!/bin/ksh
594
595	# @ job_type = serial
596	# @ node_usage = $node_usage
597	# @ job_name = palm
598	# @ wall_clock_limit = ${timestring},$timestring
599	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
600	# @ output = $remote_dayfile
601	# @ error = $remote_dayfile
602	# @ class = $queue
603	$notification
604
605	# @ queue
606
607	%%END%%
608
609	elif [[ $remote_host = nech ]]
610	then
611
612	if (( nodes > 1 ))
613	then
614	cat > $job_to_send << %%END%%
615	#!/bin/ksh
616	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
617	#PBS -l ${qsubmem}=${Memory}gb
618	#PBS -b $nodes
619	#PBS -o $remote_dayfile
620	#PBS -N palm
621	#PBS -j o
622	#PBS -T mpisx
623
624	%%END%%
625
626	elif [[ $numprocs != 0 ]]
627	then
628	cat > $job_to_send << %%END%%
629	#!/bin/ksh
630	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
631	#PBS -l ${qsubmem}=${Memory}gb
632	#PBS -o $remote_dayfile
633	#PBS -N palm
634	#PBS -j o
635
636	%%END%%
637
638	else
639	cat > $job_to_send << %%END%%
640	#!/bin/ksh
641	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
642	#PBS -o $remote_dayfile
643	#PBS -j o
644
645	%%END%%
646
647	fi
648
649	elif [[ $remote_host = neck ]]
650	then
651
652	if (( nodes > 1 ))
653	then
654	cat > $job_to_send << %%END%%
655	#!/bin/ksh
656	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
657	#PBS -l ${qsubmem}=${Memory}gb
658	#PBS -b $nodes
659	#PBS -o $remote_dayfile
660	#PBS -N palm
661	#PBS -j o
662	#PBS -T mpisx
663
664	%%END%%
665
666	elif [[ $numprocs != 0 ]]
667	then
668	cat > $job_to_send << %%END%%
669	#!/bin/ksh
670	#PBS -q ${queue}
671	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
672	#PBS -l ${qsubmem}=${Memory}gb
673	#PBS -o $remote_dayfile
674	#PBS -N palm
675	#PBS -j o
676	#PBS -T mpisx
677
678	%%END%%
679
680	else
681	cat > $job_to_send << %%END%%
682	#!/bin/ksh
683	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
684	#PBS -o $remote_dayfile
685	#PBS -j o
686
687	%%END%%
688
689	fi
690
691	elif [[ $remote_host = lctit ]]
692	then
693	cat > $job_to_send << %%END%%
694	#!/bin/bash
695	#$ -S /bin/bash
696	cd $job_catalog
697	export PATH=$PALM_BIN:\$PATH
698	echo \$PATH
699
700	%%END%%
701
702	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
703	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
704
705	else
706
707	cat > $job_to_send << %%END%%
708	# @\$-q ${queue}
709	# @\$-l${qsubtime} $timestring
710	# @\$-l${qsubmem} ${memory}mb
711	# @\$-o $remote_dayfile
712	# @\$-eo
713
714	%%END%%
715
716	fi
717
718
719	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
720	# BEI EXIT VERANLASSEN
721	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
722	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
723	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
724	then
725	echo "set +vx" >> $job_to_send
726	echo "trap '" >> $job_to_send
727	echo "set +vx" >> $job_to_send
728	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
729	then
730	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
731	then
732	return_queue=c1
733	elif [[ $remote_host = ibms ]]
734	then
735	return_queue=p_normal
736	elif [[ $remote_host = ibmy ]]
737	then
738	return_queue=serial
739	elif [[ $remote_host = lctit ]]
740	then
741	return_queue=default
742	elif [[ $remote_host = neck ]]
743	then
744	return_queue=S
745	else
746	return_queue=unknown
747	fi
748
749	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
750	then
751
752	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
753	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
754	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
755	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
756	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
757	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
758	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
759	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
760	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
761	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
762
763	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
764	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
765
766	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
767	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
768	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
769
770	elif [[ $remote_host = nech ]]
771	then
772	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
773	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
774	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
775	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
776	echo "#PBS -j o" >> $job_to_send
777	echo " " >> $job_to_send
778	echo "set -x" >> $job_to_send
779	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
780	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
781	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
782	echo "%%END%%" >> $job_to_send
783
784	elif [[ $remote_host = neck ]]
785	then
786	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
787	echo "#PBS -q $return_queue" >> $job_to_send
788	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
789	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
790	echo "#PBS -j o" >> $job_to_send
791	echo " " >> $job_to_send
792	echo "set -x" >> $job_to_send
793	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
794	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
795	echo "%%END%%" >> $job_to_send
796
797	elif [[ $remote_host = lctit ]]
798	then
799	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
800	echo "set -x" >> $job_to_send
801	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
802	echo "cd $job_catalog" >> $job_to_send
803	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
804	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
805	echo "%%END%%" >> $job_to_send
806
807	else
808
809	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
810	echo "# @\\\$-q $return_queue" >> $job_to_send
811	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
812	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
813	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
814	then
815	echo "# @\$-l mpp_p=0" >> $job_to_send
816	fi
817	echo '# @\$-lF 10mb' >> $job_to_send
818	# echo '# @\$-o /dev/null' >> $job_to_send
819	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
820	echo '# @\\\$-eo' >> $job_to_send
821	echo " " >> $job_to_send
822	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
823	then
824	echo "set +vx" >> $job_to_send
825	echo ". .profile" >> $job_to_send
826	fi
827	echo "set -x" >> $job_to_send
828	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
829	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
830	echo "%%END%%" >> $job_to_send
831	fi
832
833	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
834	then
835	echo "llsubmit scpjob.$kennung" >> $job_to_send
836	elif [[ $remote_host = lctit ]]
837	then
838	echo "chmod u+x scpjob.$kennung" >> $job_to_send
839	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
840	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
841	echo "sleep 10" >> $job_to_send
842	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
843	then
844	echo "qsub -J n scpjob.$kennung" >> $job_to_send
845	elif [[ $remote_host = t3es ]]
846	then
847	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
848	else
849	echo "qsub scpjob.$kennung" >> $job_to_send
850	fi
851	echo "rm scpjob.$kennung" >> $job_to_send
852	if [[ $remote_host = nech ]]
853	then
854	echo "cd -" >> $job_to_send
855	fi
856	else
857	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
858	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
859	fi
860	echo "set -x" >> $job_to_send
861	echo " ' exit" >> $job_to_send
862	echo "set -x" >> $job_to_send
863	fi
864
865
866
867	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
868	cat $file_to_send >> $job_to_send
869
870	if [[ $remote_host = ibm ]]
871	then
872	echo " " >> $job_to_send
873	echo "exit" >> $job_to_send
874	fi
875	if [[ $remote_host = lctit ]]
876	then
877	echo " " >> $job_to_send
878	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
879	fi
880
881
882
883	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
884	if [[ -z $remote_user ]]
885	then
886	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
887	then
888	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
889	fi
890	fi
891
892
893
894	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
895	if [[ $no_submit = false ]]
896	then
897	if [[ $remote_host != $local_host ]]
898	then
899	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
900	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
901	then
902	job_catalog_save=$job_catalog
903	job_catalog=job_queue
904	elif [[ $remote_host = nech ]]
905	then
906	job_catalog_save=$job_catalog
907	job_catalog=/hpf/b/${remote_user}/job_queue
908	fi
909	if [[ $local_host = decalpha ]]
910	then
911	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
912	# PASSWORT AB
913	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
914	elif [[ $remote_host = nech ]]
915	then
916	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
917	# TRANSFERIERT WERDEN
918	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
919	else
920	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
921	fi
922	if [[ $? = 1 ]]
923	then
924	locat=scp; exit
925	fi
926	if [[ $remote_host = ibms ]]
927	then
928	job_catalog=$job_catalog_save
929	fi
930	[[ $verify = true ]] && printf "\n >>> finished\n"
931	else
932	eval job_catalog=$job_catalog
933	cp $job_to_send ${job_catalog}/$job_on_remhost
934	fi
935
936
937
938	# NQS- BZW. LOADLEVELER-JOB STARTEN
939	if [[ $remote_host != $local_host ]]
940	then
941	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
942	if [[ $remote_host != lctit ]]
943	then
944	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
945	else
946	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
947	# MIT SSH, DESHALB AUFRUF PER PIPE
948	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
949	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
950	echo "first try" > send_protocol
951	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
952	do
953	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
954	then
955	printf "\n trying to submit job to TIT cluster..."
956	else
957	printf "\n +++ failed ... trying to submit again ..."
958	fi
959	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
960	cat send_protocol
961	done
962	sleep 10
963	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
964	echo "$submcom $job_on_remhost"
965	rm send_protocol
966	fi
967
968	[[ $verify = true ]] && printf " >>> o.k.\n"
969	else
970	cd $job_catalog
971	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
972	then
973	eval $submcom $job_on_remhost
974	elif [[ $local_host = lctit ]]
975	then
976	eval $submcom $job_on_remhost
977	echo "$submcom $job_on_remhost"
978	chmod u+x $job_on_remhost
979	elif [[ $local_host = nech ]]
980	then
981	if [[ $queue = default ]]
982	then
983	eval $submcom $job_on_remhost
984	else
985	eval $submcom -q $queue $job_on_remhost
986	fi
987	else
988	qsub $job_on_remhost
989	fi
990	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
991	[[ $local_host != lctit ]] && rm $job_on_remhost
992	cd - > /dev/null
993	fi
994	fi
995
996
997
998	# ABSCHLUSSARBEITEN
999
1000	if [[ $no_submit = false ]]
1001	then
1002	rm $job_to_send
1003	fi
1004	[[ $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 |