Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 79

Last change on this file since 79 was 69, checked in by raasch, 18 years ago
scripts are running under pdksh
Property svn:keywords set to `Rev`
File size: 37.9 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0 13/02/07
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
84
85	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
86	delete_dayfile=false
87	filetransfer_protocol=scp
88	locat=normal
89	no_default_queue=none
90	no_submit=false
91	job_catalog="~/job_queue"
92	job_name=none
93	local_user=$LOGNAME
94	node_usage=shared
95	numprocs=1
96	punkte="..........................................................."
97	submcom=qsub
98	queue=default
99	remote_host=none
100	remote_user=""
101	verify=true
102
103	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
104	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
105	typeset -L20 spalte1
106	typeset -R40 spalte2
107	typeset -L60 spalte3
108
109
110
111	# FEHLERBEHANDLUNG
112	# BEI EXIT:
113	trap 'if [[ $locat != normal ]]
114	then
115	case $locat in
116	(option) printf "\n --> available optios can be displayed"
117	printf " by typing:"
118	printf "\n \"subjob ?\" \n";;
119	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
120	(*) printf "\n +++ unknown error"
121	printf "\n please inform S. Raasch!\n"
122	esac
123	[[ -f $job_to_send ]] && rm $job_to_send
124	printf "\n\n+++ SUBJOB killed \n\n"
125	fi' exit
126
127
128	# BEI TERMINAL-BREAK:
129	trap '[[ -f $job_to_send ]] && rm $job_to_send
130	printf "\n\n+++ SUBJOB killed \n\n"
131	exit
132	' 2
133
134
135
136
137	# LOKALEN HOSTNAMEN FESTSTELLEN
138	local_host=$(hostname)
139
140
141
142	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
143	# UEBERHAUPT ZULAESSIG IST
144	case $local_host in
145	(atmos) local_addres=172.20.25.35; local_host=lcide;;
146	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
147	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
148	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
149	(cs*) local_addres=136.172.44.131; local_host=nech;;
150	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
151	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
152	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
153	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
154	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
155	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
156	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
157	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
158	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
159	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
160	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
161	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
162	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
163	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
164	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
165	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
166	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
167	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
168	(*) printf "\n +++ \"$local_host\" unknown";
169	printf "\n please inform S. Raasch!";
170	locat=parameter; exit;;
171	esac
172
173
174
175	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
176	remote_host=$local_host
177
178
179
180
181	# PROZEDUROPTIONEN EINLESEN
182	while getopts :c:dDf:h:m:n:N:O:q:t:T:u:vX: option
183	do
184	case $option in
185	(c) job_catalog=$OPTARG;;
186	(d) delete_dayfile=true;;
187	(D) no_submit=true;;
188	(f) filetransfer_protocol=$OPTARG;;
189	(h) remote_host=$OPTARG;;
190	(m) memory=$OPTARG;;
191	(n) job_name=$OPTARG;;
192	(N) node_usage=$OPTARG;;
193	(O) threads_per_task=$OPTARG;;
194	(q) no_default_queue=$OPTARG;;
195	(t) cputime=$OPTARG;;
196	(T) tasks_per_node=$OPTARG;;
197	(u) remote_user=$OPTARG;;
198	(v) verify=false;;
199	(X) numprocs=$OPTARG;;
200	(\?) printf "\n +++ Option $OPTARG unknown \n";
201	locat=option; exit;;
202	esac
203	done
204
205
206	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
207	shift OPTIND-1; file_to_send=$1
208
209
210	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
211	if [ "$1" = "?" ]
212	then
213	(printf "\n *** subjob can be called as follows:\n"
214	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
215	printf "\n Description of available options:\n"
216	printf "\n Option Description Default-Value"
217	printf "\n -c job-input- and output-catalog ~/job_queue"
218	printf "\n -d no job-protocol will be created ---"
219	printf "\n -D only the job-file will be created ---"
220	printf "\n -f filetransfer protocol scp"
221	printf "\n -h execution host, available hosts: $remote_host"
222	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
223	printf "\n lctit, nech, neck, unics"
224	printf "\n -m memory demand per process in MByte ---"
225	printf "\n -n jobname <jobdatei>"
226	printf "\n -O threads per task (for OpenMP usage) 1"
227	printf "\n -q job-queue to be used default"
228	printf "\n -t allowed cpu-time in seconds ---"
229	printf "\n -T tasks per node (on parallel hosts) ---"
230	printf "\n -u username on execution host from .netrc"
231	printf "\n -v no prompt for confirmation ---"
232	printf "\n -X # of processors (on parallel hosts) 1"
233	printf "\n "
234	printf "\n The only possible positional parameter is <jobfile>:"
235	printf "\n The complete NQS-job must be provided here."
236	printf "\n <jobfile>=? creates this outline\n\n") \| more
237	exit
238	fi
239
240
241
242	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
243	if [[ "$file_to_send" = "" ]]
244	then
245	printf "\n +++ job-file missing"
246	locat=parameter; exit
247	else
248	if [[ -f $file_to_send ]]
249	then
250	true
251	else
252	printf "\n +++ job-file: "
253	printf "\n $file_to_send"
254	printf "\n does not exist"
255	locat=parameter; exit
256	fi
257	fi
258
259
260
261	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
262	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
263	if [[ $job_name = none ]]
264	then
265	job_name=$file_to_send
266	fi
267	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
268	then
269	printf "\n +++ job-file name: "
270	printf "\n $job_name"
271	printf "\n must not contain \"/\"-characters"
272	locat=parameter; exit
273	fi
274
275
276
277
278	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
279	# ODER WENN HOST NICHT ANGEGEBEN WURDE
280	if [[ $remote_host = none ]]
281	then
282	printf "\n +++ host missing"
283	locat=option; exit
284	else
285	case $remote_host in
286	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
287	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
288	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
289	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
290	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
291	(lctit) queue=default; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
292	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
293	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
294	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
295	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
296	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
297	locat=parameter; exit;;
298	esac
299	fi
300
301
302	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
303	if [[ $no_default_queue != none ]]
304	then
305	error=false
306	ndq=$no_default_queue
307	case $remote_host in
308	(ibm) case $ndq in
309	(p690_express\|p690_standard\|p690_long) error=false;;
310	(*) error=true;;
311	esac;;
312	(ibmb) case $ndq in
313	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
314	(*) error=true;;
315	esac;;
316	(ibmh) case $ndq in
317	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
318	(*) error=true;;
319	esac;;
320	(ibms) case $ndq in
321	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
322	(*) error=true;;
323	esac;;
324	(ibmy) case $ndq in
325	(parallel) error=false;;
326	(*) error=true;;
327	esac;;
328	(lctit) case $ndq in
329	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
330	(*) error=true;;
331	esac;;
332	(t3eb) case $ndq in
333	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
334	(*) error=true;;
335	esac;;
336	(t3eh) case $ndq in
337	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
338	(*) error=true;;
339	esac;;
340	(t3ej2\|t3ej5) case $ndq in
341	(low\|normal\|high) error=false;;
342	(*) error=true;;
343	esac;;
344	(t3es) case $ndq in
345	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
346	(*) error=true;;
347	esac;;
348	(unics) case $ndq in
349	(unics\|ht) error=false;;
350	(*) error=true;;
351	esac;;
352	esac
353	if [[ $error = true ]]
354	then
355	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
356	locat=parameter; exit
357	else
358	queue=$no_default_queue
359	fi
360	fi
361
362
363
364	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
365	if [[ $node_usage != shared && $queue = cdev ]]
366	then
367	node_usage=shared
368	fi
369
370
371
372	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
373	# AUFTEILEN
374	done=false
375	while [[ $done = false ]]
376	do
377	if (( $cputime <= 0 ))
378	then
379	printf "\n +++ wrong cpu-time or cpu-time missing"
380	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
381	printf "\n >>> "
382	read cputime 1>/dev/null 2>&1
383	else
384	done=true
385	fi
386	done
387	if [[ $remote_host = nech ]]
388	then
389	if (( tasks_per_node != 0 ))
390	then
391	(( cputime = cputime * tasks_per_node ))
392	elif [[ $numprocs != 0 ]]
393	then
394	(( cputime = cputime * numprocs ))
395	fi
396	fi
397	(( stunden = cputime / 3600 ))
398	(( resttime = cputime - stunden * 3600 ))
399	(( minuten = resttime / 60 ))
400	(( sekunden = resttime - minuten * 60 ))
401	timestring=${stunden}:${minuten}:${sekunden}
402
403
404
405	# PRUEFEN DER KERNSPEICHERANFORDERUNG
406	done=false
407	while [[ $done = false ]]
408	do
409	if (( memory <= 0 ))
410	then
411	printf "\n +++ wrong memory demand or memory demand missing"
412	printf "\n >>> Please type memory in MByte per process as INTEGER:"
413	printf "\n >>> "
414	read memory 1>/dev/null 2>&1
415	else
416	done=true
417	fi
418	done
419
420	if [[ $remote_host = nech \|\| $remote_host = neck ]]
421	then
422	if (( tasks_per_node != 0 ))
423	then
424	(( Memory = memory * tasks_per_node / 1000 ))
425	elif [[ $numprocs != 0 ]]
426	then
427	(( Memory = memory * numprocs / 1000 ))
428	else
429	(( Memory = memory / 1000 ))
430	fi
431	fi
432
433
434	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
435	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
436	then
437	(( memory = memory * threads_per_task ))
438	fi
439
440
441	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
442	if (( tasks_per_node != 0 ))
443	then
444	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
445	fi
446
447
448
449	# HEADER-AUSGABE
450	if [[ $verify = true ]]
451	then
452	printf "\n\n"
453	printf "#--------------------------------------------------------------# \n"
454	spalte1=SUBJOB;spalte2=$(date)
455	printf "\| $spalte1$spalte2 \| \n"
456	printf "\| \| \n"
457	printf "\| values of parameters/options: \| \n"
458	spalte1=local_host$punkte; spalte2=$punkte$local_host
459	printf "\| $spalte1$spalte2 \| \n"
460	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
461	printf "\| $spalte1$spalte2 \| \n"
462	spalte1=queue$punkte; spalte2=$punkte$queue
463	printf "\| $spalte1$spalte2 \| \n"
464	spalte1=memory$punkte; spalte2="$punkte$memory mb"
465	printf "\| $spalte1$spalte2 \| \n"
466	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
467	printf "\| $spalte1$spalte2 \| \n"
468	spalte1=job_name$punkte; spalte2="$punkte$job_name"
469	printf "\| $spalte1$spalte2 \| \n"
470	printf "#--------------------------------------------------------------# \n\n"
471
472
473	# KONTROLLABFRAGE, OB ALLES O.K.
474	antwort="dummy"
475	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
476	do
477	read antwort?" >>> continue (y/n) ? "
478	done
479	if [[ $antwort = n \|\| $antwort = N ]]
480	then
481	locat=verify; exit
482	fi
483	printf "\n"
484	fi
485
486
487
488	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
489	kennung=$RANDOM
490	job_on_remhost=${job_name}_${kennung}_$local_host
491	job_to_send=job_to_send_$kennung
492	if [[ $delete_dayfile = false ]]
493	then
494	remote_dayfile=${local_host}_${job_name}_result_$kennung
495	local_dayfile=${remote_host}_${job_name}
496	else
497	remote_dayfile=/dev/null
498	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
499	then
500	notification='# @ notification = never'
501	fi
502	fi
503
504
505
506	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
507	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
508	then
509
510	if [[ $remote_host = ibmy ]]
511	then
512	consumable_memory=""
513	else
514	consumable_memory="ConsumableMemory($memory mb)"
515	fi
516	if [[ $queue = cdev ]]
517	then
518	data_limit="# @ data_limit = 1.76gb"
519	network_to_use="# @ network.mpi = sn_all,shared,ip"
520	else
521	if [[ $remote_host = ibms ]]
522	then
523	network_to_use="# @ network.mpi = csss,shared,us"
524	elif [[ $remote_host = ibmy ]]
525	then
526	network_to_use=""
527	else
528	network_to_use="# @ network.mpi = sn_all,shared,us"
529	data_limit="# @ data_limit = 1.76gb"
530	fi
531	fi
532
533	cat > $job_to_send << %%END%%
534	#!/bin/ksh
535
536	# @ job_type = parallel
537	# @ job_name = palm
538	# @ wall_clock_limit = ${timestring},$timestring
539	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
540	# @ output = $remote_dayfile
541	# @ error = $remote_dayfile
542	# @ image_size = 50
543	# @ class = $queue
544	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
545	$network_to_use
546	$data_limit
547	$notification
548
549	%%END%%
550
551	if (( nodes > 0 ))
552	then
553
554	cat >> $job_to_send << %%END%%
555	# @ node = $nodes
556	# @ tasks_per_node = $tasks_per_node
557	# @ node_usage = $node_usage
558	# @ queue
559
560	%%END%%
561
562	else
563
564	if [[ $remote_host != ibmy ]]
565	then
566
567	cat >> $job_to_send << %%END%%
568	# @ blocking = unlimited
569	# @ total_tasks = $numprocs
570	# @ node_usage = $node_usage
571	# @ queue
572
573	%%END%%
574
575	else
576
577	cat >> $job_to_send << %%END%%
578	# @ node = 1
579	# @ total_tasks = $numprocs
580	# @ queue
581
582	%%END%%
583
584	fi
585
586	fi
587
588	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
589	then
590
591	cat > $job_to_send << %%END%%
592	#!/bin/ksh
593
594	# @ job_type = serial
595	# @ node_usage = $node_usage
596	# @ job_name = palm
597	# @ wall_clock_limit = ${timestring},$timestring
598	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
599	# @ output = $remote_dayfile
600	# @ error = $remote_dayfile
601	# @ class = $queue
602	$notification
603
604	# @ queue
605
606	%%END%%
607
608	elif [[ $remote_host = nech ]]
609	then
610
611	if (( nodes > 1 ))
612	then
613	cat > $job_to_send << %%END%%
614	#!/bin/ksh
615	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
616	#PBS -l ${qsubmem}=${Memory}gb
617	#PBS -b $nodes
618	#PBS -o $remote_dayfile
619	#PBS -N palm
620	#PBS -j o
621	#PBS -T mpisx
622
623	%%END%%
624
625	elif [[ $numprocs != 0 ]]
626	then
627	cat > $job_to_send << %%END%%
628	#!/bin/ksh
629	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
630	#PBS -l ${qsubmem}=${Memory}gb
631	#PBS -o $remote_dayfile
632	#PBS -N palm
633	#PBS -j o
634
635	%%END%%
636
637	else
638	cat > $job_to_send << %%END%%
639	#!/bin/ksh
640	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
641	#PBS -o $remote_dayfile
642	#PBS -j o
643
644	%%END%%
645
646	fi
647
648	elif [[ $remote_host = neck ]]
649	then
650
651	if (( nodes > 1 ))
652	then
653	cat > $job_to_send << %%END%%
654	#!/bin/ksh
655	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
656	#PBS -l ${qsubmem}=${Memory}gb
657	#PBS -b $nodes
658	#PBS -o $remote_dayfile
659	#PBS -N palm
660	#PBS -j o
661	#PBS -T mpisx
662
663	%%END%%
664
665	elif [[ $numprocs != 0 ]]
666	then
667	cat > $job_to_send << %%END%%
668	#!/bin/ksh
669	#PBS -q ${queue}
670	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
671	#PBS -l ${qsubmem}=${Memory}gb
672	#PBS -o $remote_dayfile
673	#PBS -N palm
674	#PBS -j o
675	#PBS -T mpisx
676
677	%%END%%
678
679	else
680	cat > $job_to_send << %%END%%
681	#!/bin/ksh
682	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
683	#PBS -o $remote_dayfile
684	#PBS -j o
685
686	%%END%%
687
688	fi
689
690	elif [[ $remote_host = lctit ]]
691	then
692	cat > $job_to_send << %%END%%
693	#!/bin/bash
694	#$ -S /bin/bash
695	cd $job_catalog
696	export PATH=$PALM_BIN:\$PATH
697	echo \$PATH
698
699	%%END%%
700
701	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
702	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
703
704	else
705
706	cat > $job_to_send << %%END%%
707	# @\$-q ${queue}
708	# @\$-l${qsubtime} $timestring
709	# @\$-l${qsubmem} ${memory}mb
710	# @\$-o $remote_dayfile
711	# @\$-eo
712
713	%%END%%
714
715	fi
716
717
718	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
719	# BEI EXIT VERANLASSEN
720	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
721	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
722	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
723	then
724	echo "set +vx" >> $job_to_send
725	echo "trap '" >> $job_to_send
726	echo "set +vx" >> $job_to_send
727	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
728	then
729	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
730	then
731	return_queue=c1
732	elif [[ $remote_host = ibms ]]
733	then
734	return_queue=p_normal
735	elif [[ $remote_host = ibmy ]]
736	then
737	return_queue=serial
738	elif [[ $remote_host = lctit ]]
739	then
740	return_queue=default
741	elif [[ $remote_host = neck ]]
742	then
743	return_queue=S
744	else
745	return_queue=unknown
746	fi
747
748	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
749	then
750
751	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
752	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
753	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
754	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
755	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
756	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
757	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
758	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
759	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
760	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
761
762	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
763	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
764
765	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
766	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
767	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
768
769	elif [[ $remote_host = nech ]]
770	then
771	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
772	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
773	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
774	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
775	echo "#PBS -j o" >> $job_to_send
776	echo " " >> $job_to_send
777	echo "set -x" >> $job_to_send
778	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
779	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
780	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
781	echo "%%END%%" >> $job_to_send
782
783	elif [[ $remote_host = neck ]]
784	then
785	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
786	echo "#PBS -q $return_queue" >> $job_to_send
787	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
788	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
789	echo "#PBS -j o" >> $job_to_send
790	echo " " >> $job_to_send
791	echo "set -x" >> $job_to_send
792	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
793	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
794	echo "%%END%%" >> $job_to_send
795
796	elif [[ $remote_host = lctit ]]
797	then
798	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
799	echo "set -x" >> $job_to_send
800	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
801	echo "cd $job_catalog" >> $job_to_send
802	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
803	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
804	echo "%%END%%" >> $job_to_send
805
806	else
807
808	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
809	echo "# @\\\$-q $return_queue" >> $job_to_send
810	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
811	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
812	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
813	then
814	echo "# @\$-l mpp_p=0" >> $job_to_send
815	fi
816	echo '# @\$-lF 10mb' >> $job_to_send
817	# echo '# @\$-o /dev/null' >> $job_to_send
818	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
819	echo '# @\\\$-eo' >> $job_to_send
820	echo " " >> $job_to_send
821	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
822	then
823	echo "set +vx" >> $job_to_send
824	echo ". .profile" >> $job_to_send
825	fi
826	echo "set -x" >> $job_to_send
827	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
828	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
829	echo "%%END%%" >> $job_to_send
830	fi
831
832	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
833	then
834	echo "llsubmit scpjob.$kennung" >> $job_to_send
835	elif [[ $remote_host = lctit ]]
836	then
837	echo "chmod u+x scpjob.$kennung" >> $job_to_send
838	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
839	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
840	echo "sleep 10" >> $job_to_send
841	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
842	then
843	echo "qsub -J n scpjob.$kennung" >> $job_to_send
844	elif [[ $remote_host = t3es ]]
845	then
846	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
847	else
848	echo "qsub scpjob.$kennung" >> $job_to_send
849	fi
850	echo "rm scpjob.$kennung" >> $job_to_send
851	if [[ $remote_host = nech ]]
852	then
853	echo "cd -" >> $job_to_send
854	fi
855	else
856	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
857	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
858	fi
859	echo "set -x" >> $job_to_send
860	echo " ' exit" >> $job_to_send
861	echo "set -x" >> $job_to_send
862	fi
863
864
865
866	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
867	cat $file_to_send >> $job_to_send
868
869	if [[ $remote_host = ibm ]]
870	then
871	echo " " >> $job_to_send
872	echo "exit" >> $job_to_send
873	fi
874	if [[ $remote_host = lctit ]]
875	then
876	echo " " >> $job_to_send
877	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
878	fi
879
880
881
882	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
883	if [[ -z $remote_user ]]
884	then
885	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
886	then
887	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
888	fi
889	fi
890
891
892
893	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
894	if [[ $no_submit = false ]]
895	then
896	if [[ $remote_host != $local_host ]]
897	then
898	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
899	if [[ $filetransfer_protocol = scp ]]
900	then
901	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
902	then
903	job_catalog_save=$job_catalog
904	job_catalog=job_queue
905	elif [[ $remote_host = nech ]]
906	then
907	job_catalog_save=$job_catalog
908	job_catalog=/hpf/b/${remote_user}/job_queue
909	fi
910	if [[ $local_host = decalpha ]]
911	then
912	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
913	# PASSWORT AB
914	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
915	elif [[ $remote_host = nech ]]
916	then
917	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
918	# TRANSFERIERT WERDEN
919	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
920	else
921	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
922	fi
923	if [[ $? = 1 ]]
924	then
925	locat=scp; exit
926	fi
927	if [[ $remote_host = ibms ]]
928	then
929	job_catalog=$job_catalog_save
930	fi
931	else
932	ftpcopy -s -o $remote_addres $job_to_send "$job_catalog" $job_on_remhost
933	if [[ $? = 1 ]]
934	then
935	locat=ftpcopy; exit
936	fi
937	fi
938	[[ $verify = true ]] && printf "\n >>> finished\n"
939	else
940	eval job_catalog=$job_catalog
941	cp $job_to_send ${job_catalog}/$job_on_remhost
942	fi
943
944
945
946	# NQS- BZW. LOADLEVELER-JOB STARTEN
947	if [[ $remote_host != $local_host ]]
948	then
949	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
950	if [[ $filetransfer_protocol = scp ]]
951	then
952	if [[ $remote_host != lctit ]]
953	then
954	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
955	else
956	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
957	# MIT SSH, DESHALB AUFRUF PER PIPE
958	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
959	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
960	echo "first try" > send_protocol
961	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
962	do
963	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
964	then
965	printf "\n trying to submit job to TIT cluster..."
966	else
967	printf "\n +++ failed ... trying to submit again ..."
968	fi
969	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
970	cat send_protocol
971	done
972	sleep 10
973	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
974	echo "$submcom $job_on_remhost"
975	rm send_protocol
976	fi
977	else
978	if [[ $remote_host = ibms ]]
979	then
980	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
981	echo "echo \"$remote_user\" " > telnet_script
982	echo "sleep 5" >> telnet_script
983	echo "echo \"$password\" " >> telnet_script
984	echo "sleep 10" >> telnet_script
985	echo "echo \"\" " >> telnet_script
986	echo "sleep 2" >> telnet_script
987	echo "echo \"cd $job_catalog; $submcom $job_on_remhost; rm -rf $job_on_remhost\" " >> telnet_script
988	echo "sleep 10" >> telnet_script
989	echo "echo \"exit\" " >> telnet_script
990
991	chmod u+x telnet_script
992	telnet_script \| telnet $remote_addres 2>&1 \| grep "llsubmit:"
993	rm telnet_script
994	elif [[ $remote_host = t3es ]]
995	then
996	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
997	echo "printf \"$remote_user\\\n\" " > telnet_script
998	echo "sleep 10" >> telnet_script
999	echo "printf \"$password\\\n\" " >> telnet_script
1000	echo "sleep 10" >> telnet_script
1001	echo "printf \"\\\n\" " >> telnet_script
1002	echo "sleep 5" >> telnet_script
1003	echo "printf \"\\\n\" " >> telnet_script
1004	echo "sleep 5" >> telnet_script
1005	echo "printf \"cd $job_catalog; $submcom $job_on_remhost; rm -f $job_on_remhost\\\n\" " >> telnet_script
1006	echo "sleep 10" >> telnet_script
1007	echo "printf \"exit\\\n\" " >> telnet_script
1008
1009	chmod u+x telnet_script
1010	cat telnet_script
1011	telnet_script \| telnet $remote_addres \| grep $remote_user
1012	rm telnet_script
1013	else
1014	remcom.x $remote_addres "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1015	fi
1016	fi
1017	[[ $verify = true ]] && printf " >>> o.k.\n"
1018	else
1019	cd $job_catalog
1020	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1021	then
1022	eval $submcom $job_on_remhost
1023	elif [[ $local_host = lctit ]]
1024	then
1025	eval $submcom $job_on_remhost
1026	echo "$submcom $job_on_remhost"
1027	chmod u+x $job_on_remhost
1028	elif [[ $local_host = nech ]]
1029	then
1030	if [[ $queue = default ]]
1031	then
1032	eval $submcom $job_on_remhost
1033	else
1034	eval $submcom -q $queue $job_on_remhost
1035	fi
1036	else
1037	qsub $job_on_remhost
1038	fi
1039	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1040	[[ $local_host != lctit ]] && rm $job_on_remhost
1041	cd - > /dev/null
1042	fi
1043	fi
1044
1045
1046
1047	# ABSCHLUSSARBEITEN
1048
1049	if [[ $no_submit = false ]]
1050	then
1051	rm $job_to_send
1052	fi
1053	[[ $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 |