Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 9

Last change on this file since 9 was 5, checked in by raasch, 18 years ago
html-documentation added
File size: 35.9 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0 13/02/07
3
4	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
5	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
6	# Host zurueckgeschickt werden sollen
7
8
9	# letzte Aenderung:
10	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
11	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
12	# 06/02/98 - Siggi - berte validiert
13	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
14	# moeglich
15	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
16	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
17	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
18	# allgemeine Anpassungen fuer ibm-Rechner
19	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
20	# Ruecktransfer des Jobprotokolls fuer diese
21	# Maschine abgeschaltet
22	# 31/10/02 - Siggi - berni validiert
23	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
24	# 08/11/02 - Siggi - quanero validiert
25	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
26	# 23/01/03 - Siggi - hostname nobel changed to nobela
27	# 06/02/03 - Siggi - gregale validated
28	# 12/02/03 - Siggi - orkan and maestro validated
29	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
30	# 12/03/03 - Siggi - nec at DKRZ validated
31	# 13/03/03 - Siggi - new nqs resource variable Memory
32	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
33	# node
34	# 11/04/03 - Siggi - network on ibms has attribute "shared"
35	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
36	# cxxl added to ibmh
37	# 29/08/03 - Siggi - changes in job queues and communication system on
38	# ibms
39	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
40	# 30/10/03 - Siggi - nech is not supported any more
41	# 10/11/03 - Siggi - nech2 renamed to nech
42	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
43	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
44	# 31/03/04 - Siggi - new option -N for node usage
45	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
46	# due to error in ssh installation (otherwise a prompt
47	# for the password appears)
48	# 23/07/04 - Siggi - changes due to the new berni configuration
49	# (federation switch)
50	# 01/09/04 - Gerald new job-classes on hanni
51	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
52	# 23/11/04 - Siggi - new job class cdata on hanni and berni
53	# 03/12/04 - Siggi - notification on ibm switched of in case of
54	# delete_dayfile = true, node usage in cdev set to
55	# shared
56	# 16/02/05 - Gerald hababai validated
57	# 29/03/05 - Micha - new job class channi on hanni
58	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
59	# on ibms
60	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
61	# p_normal
62	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
63	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
64	# "sn_all", new job class cexp
65	# 08/09/05 - Siggi - IP-address of gfdl3 changed
66	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
67	# hurricane (with -X0)
68	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
69	# added
70	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
71	# 09/02/06 - Siggi - ibmy admitted
72	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
73	# 13/04/06 - Siggi - ostria admitted
74	# 18/04/06 - Siggi - new option -O for OpenMP usage
75	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
76	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
77	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
78	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
79	# 28/11/06 - Siggi - levanto admitted
80	# 13/02/07 - Siggi - hpmuk releated code removed
81
82
83	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
84	delete_dayfile=false
85	filetransfer_protocol=scp
86	locat=normal
87	no_default_queue=none
88	no_submit=false
89	job_catalog="~/job_queue"
90	job_name=none
91	local_user=$LOGNAME
92	node_usage=shared
93	numprocs=1
94	punkte="..........................................................."
95	submcom=qsub
96	queue=default
97	remote_host=none
98	remote_user=""
99	verify=true
100
101	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
102	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
103	typeset -L20 spalte1
104	typeset -R40 spalte2
105	typeset -L60 spalte3
106
107
108
109	# FEHLERBEHANDLUNG
110	# BEI EXIT:
111	trap 'if [[ $locat != normal ]]
112	then
113	case $locat in
114	(option) printf "\n --> available optios can be displayed"
115	printf " by typing:"
116	printf "\n \"subjob ?\" \n";;
117	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
118	(*) printf "\n +++ unknown error"
119	printf "\n please inform S. Raasch!\n"
120	esac
121	[[ -f $job_to_send ]] && rm $job_to_send
122	printf "\n\n+++ SUBJOB killed \n\n"
123	fi' exit
124
125
126	# BEI TERMINAL-BREAK:
127	trap '[[ -f $job_to_send ]] && rm $job_to_send
128	printf "\n\n+++ SUBJOB killed \n\n"
129	exit
130	' 2
131
132
133
134
135	# LOKALEN HOSTNAMEN FESTSTELLEN
136	local_host=$(hostname)
137
138
139
140	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
141	# UEBERHAUPT ZULAESSIG IST
142	case $local_host in
143	(atmos) local_addres=172.20.25.35; local_host=lcide;;
144	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
145	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
146	(cs*) local_addres=136.172.44.131; local_host=nech;;
147	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
148	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
149	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
150	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
151	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
152	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
153	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
154	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
155	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
156	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
157	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
158	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
159	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
160	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
161	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
162	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
163	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
164	(*) printf "\n +++ \"$local_host\" unknown";
165	printf "\n please inform S. Raasch!";
166	locat=parameter; exit;;
167	esac
168
169
170
171	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
172	remote_host=$local_host
173
174
175
176
177	# PROZEDUROPTIONEN EINLESEN
178	while getopts :c:dDf:h:m:n:N:O:q:t:T:u:vX: option
179	do
180	case $option in
181	(c) job_catalog=$OPTARG;;
182	(d) delete_dayfile=true;;
183	(D) no_submit=true;;
184	(f) filetransfer_protocol=$OPTARG;;
185	(h) remote_host=$OPTARG;;
186	(m) memory=$OPTARG;;
187	(n) job_name=$OPTARG;;
188	(N) node_usage=$OPTARG;;
189	(O) threads_per_task=$OPTARG;;
190	(q) no_default_queue=$OPTARG;;
191	(t) cputime=$OPTARG;;
192	(T) tasks_per_node=$OPTARG;;
193	(u) remote_user=$OPTARG;;
194	(v) verify=false;;
195	(X) numprocs=$OPTARG;;
196	(\?) printf "\n +++ Option $OPTARG unknown \n";
197	locat=option; exit;;
198	esac
199	done
200
201
202	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
203	shift OPTIND-1; file_to_send=$1
204
205
206	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
207	if [ "$1" = "?" ]
208	then
209	(printf "\n *** subjob can be called as follows:\n"
210	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
211	printf "\n Description of available options:\n"
212	printf "\n Option Description Default-Value"
213	printf "\n -c job-input- and output-catalog ~/job_queue"
214	printf "\n -d no job-protocol will be created ---"
215	printf "\n -D only the job-file will be created ---"
216	printf "\n -f filetransfer protocol scp"
217	printf "\n -h execution host, available hosts: $remote_host"
218	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
219	printf "\n unics"
220	printf "\n -m memory demand per process in MByte ---"
221	printf "\n -n jobname <jobdatei>"
222	printf "\n -O threads per task (for OpenMP usage) 1"
223	printf "\n -q job-queue to be used default"
224	printf "\n -t allowed cpu-time in seconds ---"
225	printf "\n -T tasks per node (on parallel hosts) ---"
226	printf "\n -u username on execution host from .netrc"
227	printf "\n -v no prompt for confirmation ---"
228	printf "\n -X # of processors (on parallel hosts) 1"
229	printf "\n "
230	printf "\n The only possible positional parameter is <jobfile>:"
231	printf "\n The complete NQS-job must be provided here."
232	printf "\n <jobfile>=? creates this outline\n\n") \| more
233	exit
234	fi
235
236
237
238	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
239	if [[ "$file_to_send" = "" ]]
240	then
241	printf "\n +++ job-file missing"
242	locat=parameter; exit
243	else
244	if [[ -f $file_to_send ]]
245	then
246	true
247	else
248	printf "\n +++ job-file: "
249	printf "\n $file_to_send"
250	printf "\n does not exist"
251	locat=parameter; exit
252	fi
253	fi
254
255
256
257	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
258	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
259	if [[ $job_name = none ]]
260	then
261	job_name=$file_to_send
262	fi
263	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
264	then
265	printf "\n +++ job-file name: "
266	printf "\n $job_name"
267	printf "\n must not contain \"/\"-characters"
268	locat=parameter; exit
269	fi
270
271
272
273
274	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
275	# ODER WENN HOST NICHT ANGEGEBEN WURDE
276	if [[ $remote_host = none ]]
277	then
278	printf "\n +++ host missing"
279	locat=option; exit
280	else
281	case $remote_host in
282	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
283	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
284	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
285	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
286	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
287	(lctit) queue=test; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
288	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
289	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
290	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
291	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
292	locat=parameter; exit;;
293	esac
294	fi
295
296
297	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
298	if [[ $no_default_queue != none ]]
299	then
300	error=false
301	ndq=$no_default_queue
302	case $remote_host in
303	(ibm) case $ndq in
304	(p690_express\|p690_standard\|p690_long) error=false;;
305	(*) error=true;;
306	esac;;
307	(ibmb) case $ndq in
308	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
309	(*) error=true;;
310	esac;;
311	(ibmh) case $ndq in
312	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
313	(*) error=true;;
314	esac;;
315	(ibms) case $ndq in
316	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
317	(*) error=true;;
318	esac;;
319	(ibmy) case $ndq in
320	(parallel) error=false;;
321	(*) error=true;;
322	esac;;
323	(lctit) case $ndq in
324	(cs\|default\|high\|test) error=false;;
325	(*) error=true;;
326	esac;;
327	(t3eb) case $ndq in
328	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
329	(*) error=true;;
330	esac;;
331	(t3eh) case $ndq in
332	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
333	(*) error=true;;
334	esac;;
335	(t3ej2\|t3ej5) case $ndq in
336	(low\|normal\|high) error=false;;
337	(*) error=true;;
338	esac;;
339	(t3es) case $ndq in
340	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
341	(*) error=true;;
342	esac;;
343	(unics) case $ndq in
344	(unics\|ht) error=false;;
345	(*) error=true;;
346	esac;;
347	esac
348	if [[ $error = true ]]
349	then
350	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
351	locat=parameter; exit
352	else
353	queue=$no_default_queue
354	fi
355	fi
356
357
358
359	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
360	if [[ $node_usage != shared && $queue = cdev ]]
361	then
362	node_usage=shared
363	fi
364
365
366
367	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
368	# AUFTEILEN
369	done=false
370	while [[ $done = false ]]
371	do
372	if (( $cputime <= 0 ))
373	then
374	printf "\n +++ wrong cpu-time or cpu-time missing"
375	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
376	printf "\n >>> "
377	read cputime 1>/dev/null 2>&1
378	else
379	done=true
380	fi
381	done
382	if [[ $remote_host = nech ]]
383	then
384	if (( tasks_per_node != 0 ))
385	then
386	(( cputime = cputime * tasks_per_node ))
387	elif [[ $numprocs != 0 ]]
388	then
389	(( cputime = cputime * numprocs ))
390	fi
391	fi
392	(( stunden = cputime / 3600 ))
393	(( resttime = cputime - stunden * 3600 ))
394	(( minuten = resttime / 60 ))
395	(( sekunden = resttime - minuten * 60 ))
396	timestring=${stunden}:${minuten}:${sekunden}
397
398
399
400	# PRUEFEN DER KERNSPEICHERANFORDERUNG
401	done=false
402	while [[ $done = false ]]
403	do
404	if (( memory <= 0 ))
405	then
406	printf "\n +++ wrong memory demand or memory demand missing"
407	printf "\n >>> Please type memory in MByte per process as INTEGER:"
408	printf "\n >>> "
409	read memory 1>/dev/null 2>&1
410	else
411	done=true
412	fi
413	done
414
415	if [[ $remote_host = nech ]]
416	then
417	if (( tasks_per_node != 0 ))
418	then
419	(( Memory = memory * tasks_per_node / 1000 ))
420	elif [[ $numprocs != 0 ]]
421	then
422	(( Memory = memory * numprocs / 1000 ))
423	else
424	(( Memory = memory / 1000 ))
425	fi
426	fi
427
428
429	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
430	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
431	then
432	(( memory = memory * threads_per_task ))
433	fi
434
435
436	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
437	if (( tasks_per_node != 0 ))
438	then
439	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
440	fi
441
442
443
444	# HEADER-AUSGABE
445	if [[ $verify = true ]]
446	then
447	printf "\n\n"
448	printf "#--------------------------------------------------------------# \n"
449	spalte1=SUBJOB;spalte2=$(date)
450	printf "\| $spalte1$spalte2 \| \n"
451	printf "\| \| \n"
452	printf "\| values of parameters/options: \| \n"
453	spalte1=local_host$punkte; spalte2=$punkte$local_host
454	printf "\| $spalte1$spalte2 \| \n"
455	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
456	printf "\| $spalte1$spalte2 \| \n"
457	spalte1=queue$punkte; spalte2=$punkte$queue
458	printf "\| $spalte1$spalte2 \| \n"
459	spalte1=memory$punkte; spalte2="$punkte$memory mb"
460	printf "\| $spalte1$spalte2 \| \n"
461	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
462	printf "\| $spalte1$spalte2 \| \n"
463	spalte1=job_name$punkte; spalte2="$punkte$job_name"
464	printf "\| $spalte1$spalte2 \| \n"
465	printf "#--------------------------------------------------------------# \n\n"
466
467
468	# KONTROLLABFRAGE, OB ALLES O.K.
469	antwort="dummy"
470	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
471	do
472	read antwort?" >>> continue (y/n) ? "
473	done
474	if [[ $antwort = n \|\| $antwort = N ]]
475	then
476	locat=verify; exit
477	fi
478	printf "\n"
479	fi
480
481
482
483	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
484	kennung=$RANDOM
485	job_on_remhost=${job_name}_${kennung}_$local_host
486	job_to_send=job_to_send_$kennung
487	if [[ $delete_dayfile = false ]]
488	then
489	remote_dayfile=${local_host}_${job_name}_result_$kennung
490	local_dayfile=${remote_host}_${job_name}
491	else
492	remote_dayfile=/dev/null
493	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
494	then
495	notification='# @ notification = never'
496	fi
497	fi
498
499
500
501	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
502	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
503	then
504
505	if [[ $remote_host = ibmy ]]
506	then
507	consumable_memory=""
508	else
509	consumable_memory="ConsumableMemory($memory mb)"
510	fi
511	if [[ $queue = cdev ]]
512	then
513	data_limit="# @ data_limit = 1.76gb"
514	network_to_use="# @ network.mpi = sn_all,shared,ip"
515	else
516	if [[ $remote_host = ibms ]]
517	then
518	network_to_use="# @ network.mpi = csss,shared,us"
519	elif [[ $remote_host = ibmy ]]
520	then
521	network_to_use=""
522	else
523	network_to_use="# @ network.mpi = sn_all,shared,us"
524	data_limit="# @ data_limit = 1.76gb"
525	fi
526	fi
527
528	cat > $job_to_send << %%END%%
529	#!/bin/ksh
530
531	# @ job_type = parallel
532	# @ job_name = palm
533	# @ wall_clock_limit = ${timestring},$timestring
534	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
535	# @ output = $remote_dayfile
536	# @ error = $remote_dayfile
537	# @ image_size = 50
538	# @ class = $queue
539	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
540	$network_to_use
541	$data_limit
542	$notification
543
544	%%END%%
545
546	if (( nodes > 0 ))
547	then
548
549	cat >> $job_to_send << %%END%%
550	# @ node = $nodes
551	# @ tasks_per_node = $tasks_per_node
552	# @ node_usage = $node_usage
553	# @ queue
554
555	%%END%%
556
557	else
558
559	if [[ $remote_host != ibmy ]]
560	then
561
562	cat >> $job_to_send << %%END%%
563	# @ blocking = unlimited
564	# @ total_tasks = $numprocs
565	# @ node_usage = $node_usage
566	# @ queue
567
568	%%END%%
569
570	else
571
572	cat >> $job_to_send << %%END%%
573	# @ node = 1
574	# @ total_tasks = $numprocs
575	# @ queue
576
577	%%END%%
578
579	fi
580
581	fi
582
583	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
584	then
585
586	cat > $job_to_send << %%END%%
587	#!/bin/ksh
588
589	# @ job_type = serial
590	# @ node_usage = $node_usage
591	# @ job_name = palm
592	# @ wall_clock_limit = ${timestring},$timestring
593	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
594	# @ output = $remote_dayfile
595	# @ error = $remote_dayfile
596	# @ class = $queue
597	$notification
598
599	# @ queue
600
601	%%END%%
602
603	elif [[ $remote_host = nech ]]
604	then
605
606	if (( nodes > 1 ))
607	then
608	cat > $job_to_send << %%END%%
609	#!/bin/ksh
610	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
611	#PBS -l ${qsubmem}=${Memory}gb
612	#PBS -b $nodes
613	#PBS -o $remote_dayfile
614	#PBS -N palm
615	#PBS -j o
616	#PBS -T mpisx
617
618	%%END%%
619
620	elif [[ $numprocs != 0 ]]
621	then
622	cat > $job_to_send << %%END%%
623	#!/bin/ksh
624	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
625	#PBS -l ${qsubmem}=${Memory}gb
626	#PBS -o $remote_dayfile
627	#PBS -N palm
628	#PBS -j o
629
630	%%END%%
631
632	else
633	cat > $job_to_send << %%END%%
634	#!/bin/ksh
635	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
636	#PBS -o $remote_dayfile
637	#PBS -j o
638
639	%%END%%
640
641	fi
642
643	elif [[ $remote_host = lctit ]]
644	then
645	cat > $job_to_send << %%END%%
646	#!/bin/bash
647	#$ -S /bin/bash
648	cd $job_catalog
649	PATH=$job_catalog:${job_catalog}/../pub:\$PATH
650	export PATH
651
652	%%END%%
653
654	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
655	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
656
657	else
658
659	cat > $job_to_send << %%END%%
660	# @\$-q ${queue}
661	# @\$-l${qsubtime} $timestring
662	# @\$-l${qsubmem} ${memory}mb
663	# @\$-o $remote_dayfile
664	# @\$-eo
665
666	%%END%%
667
668	fi
669
670
671	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
672	# BEI EXIT VERANLASSEN
673	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
674	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
675	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
676	then
677	echo "set +vx" >> $job_to_send
678	echo "trap '" >> $job_to_send
679	echo "set +vx" >> $job_to_send
680	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = nech \|\| $remote_host = lctit ]]
681	then
682	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
683	then
684	return_queue=c1
685	elif [[ $remote_host = ibms ]]
686	then
687	return_queue=p_normal
688	elif [[ $remote_host = ibmy ]]
689	then
690	return_queue=serial
691	elif [[ $remote_host = lctit ]]
692	then
693	return_queue=default
694	elif [[ $remote_host = t3eb ]]
695	then
696	return_queue=berte
697	elif [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
698	then
699	return_queue=normal
700	elif [[ $remote_host = t3es ]]
701	then
702	return_queue=batch
703	else
704	return_queue=unknown
705	fi
706
707	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
708	then
709
710	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
711	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
712	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
713	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
714	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
715	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
716	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
717	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
718	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
719	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
720
721	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
722	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
723
724	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
725	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
726	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
727
728	elif [[ $remote_host = nech ]]
729	then
730	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
731	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
732	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
733	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
734	echo "#PBS -j o" >> $job_to_send
735	echo " " >> $job_to_send
736	echo "set -x" >> $job_to_send
737	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
738	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
739	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
740	echo "%%END%%" >> $job_to_send
741
742	elif [[ $remote_host = lctit ]]
743	then
744	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
745	echo "set -x" >> $job_to_send
746	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
747	echo "cd $job_catalog" >> $job_to_send
748	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
749	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
750	echo "%%END%%" >> $job_to_send
751
752	else
753
754	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
755	echo "# @\\\$-q $return_queue" >> $job_to_send
756	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
757	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
758	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
759	then
760	echo "# @\$-l mpp_p=0" >> $job_to_send
761	fi
762	echo '# @\$-lF 10mb' >> $job_to_send
763	# echo '# @\$-o /dev/null' >> $job_to_send
764	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
765	echo '# @\\\$-eo' >> $job_to_send
766	echo " " >> $job_to_send
767	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
768	then
769	echo "set +vx" >> $job_to_send
770	echo ". .profile" >> $job_to_send
771	fi
772	echo "set -x" >> $job_to_send
773	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
774	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
775	echo "%%END%%" >> $job_to_send
776	fi
777
778	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
779	then
780	echo "llsubmit scpjob.$kennung" >> $job_to_send
781	elif [[ $remote_host = lctit ]]
782	then
783	echo "chmod u+x scpjob.$kennung" >> $job_to_send
784	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
785	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
786	echo "sleep 10" >> $job_to_send
787	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
788	then
789	echo "qsub -J n scpjob.$kennung" >> $job_to_send
790	elif [[ $remote_host = t3es ]]
791	then
792	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
793	else
794	echo "qsub scpjob.$kennung" >> $job_to_send
795	fi
796	echo "rm scpjob.$kennung" >> $job_to_send
797	if [[ $remote_host = nech ]]
798	then
799	echo "cd -" >> $job_to_send
800	fi
801	else
802	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
803	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
804	fi
805	echo "set -x" >> $job_to_send
806	echo " ' exit" >> $job_to_send
807	echo "set -x" >> $job_to_send
808	fi
809
810
811
812	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
813	cat $file_to_send >> $job_to_send
814	if [[ $remote_host = ibm ]]
815	then
816	echo " " >> $job_to_send
817	echo "exit" >> $job_to_send
818	fi
819
820
821
822	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
823	if [[ -z $remote_user ]]
824	then
825	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
826	then
827	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
828	fi
829	fi
830
831
832
833	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
834	if [[ $no_submit = false ]]
835	then
836	if [[ $remote_host != $local_host ]]
837	then
838	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
839	if [[ $filetransfer_protocol = scp ]]
840	then
841	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
842	then
843	job_catalog_save=$job_catalog
844	job_catalog=job_queue
845	elif [[ $remote_host = nech ]]
846	then
847	job_catalog_save=$job_catalog
848	job_catalog=/hpf/b/${remote_user}/job_queue
849	fi
850	if [[ $local_host = decalpha ]]
851	then
852	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
853	# PASSWORT AB
854	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
855	elif [[ $remote_host = nech ]]
856	then
857	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
858	# TRANSFERIERT WERDEN
859	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
860	else
861	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
862	fi
863	if [[ $? = 1 ]]
864	then
865	locat=scp; exit
866	fi
867	if [[ $remote_host = ibms ]]
868	then
869	job_catalog=$job_catalog_save
870	fi
871	else
872	ftpcopy -s -o $remote_addres $job_to_send "$job_catalog" $job_on_remhost
873	if [[ $? = 1 ]]
874	then
875	locat=ftpcopy; exit
876	fi
877	fi
878	[[ $verify = true ]] && printf "\n >>> finished\n"
879	else
880	eval job_catalog=$job_catalog
881	cp $job_to_send ${job_catalog}/$job_on_remhost
882	fi
883
884
885
886	# NQS- BZW. LOADLEVELER-JOB STARTEN
887	if [[ $remote_host != $local_host ]]
888	then
889	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
890	if [[ $filetransfer_protocol = scp ]]
891	then
892	if [[ $remote_host != lctit ]]
893	then
894	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
895	else
896	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
897	# MIT SSH, DESHALB AUFRUF PER PIPE
898	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
899	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
900	echo "first try" > send_protocol
901	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
902	do
903	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
904	then
905	printf "\n trying to submit job to TIT cluster..."
906	else
907	printf "\n +++ failed ... trying to submit again ..."
908	fi
909	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
910	cat send_protocol
911	done
912	sleep 10
913	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
914	echo "$submcom $job_on_remhost"
915	rm send_protocol
916	fi
917	else
918	if [[ $remote_host = ibms ]]
919	then
920	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
921	echo "echo \"$remote_user\" " > telnet_script
922	echo "sleep 5" >> telnet_script
923	echo "echo \"$password\" " >> telnet_script
924	echo "sleep 10" >> telnet_script
925	echo "echo \"\" " >> telnet_script
926	echo "sleep 2" >> telnet_script
927	echo "echo \"cd $job_catalog; $submcom $job_on_remhost; rm -rf $job_on_remhost\" " >> telnet_script
928	echo "sleep 10" >> telnet_script
929	echo "echo \"exit\" " >> telnet_script
930
931	chmod u+x telnet_script
932	telnet_script \| telnet $remote_addres 2>&1 \| grep "llsubmit:"
933	rm telnet_script
934	elif [[ $remote_host = t3es ]]
935	then
936	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
937	echo "printf \"$remote_user\\\n\" " > telnet_script
938	echo "sleep 10" >> telnet_script
939	echo "printf \"$password\\\n\" " >> telnet_script
940	echo "sleep 10" >> telnet_script
941	echo "printf \"\\\n\" " >> telnet_script
942	echo "sleep 5" >> telnet_script
943	echo "printf \"\\\n\" " >> telnet_script
944	echo "sleep 5" >> telnet_script
945	echo "printf \"cd $job_catalog; $submcom $job_on_remhost; rm -f $job_on_remhost\\\n\" " >> telnet_script
946	echo "sleep 10" >> telnet_script
947	echo "printf \"exit\\\n\" " >> telnet_script
948
949	chmod u+x telnet_script
950	cat telnet_script
951	telnet_script \| telnet $remote_addres \| grep $remote_user
952	rm telnet_script
953	else
954	remcom.x $remote_addres "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
955	fi
956	fi
957	[[ $verify = true ]] && printf " >>> o.k.\n"
958	else
959	cd $job_catalog
960	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
961	then
962	eval $submcom $job_on_remhost
963	elif [[ $local_host = lctit ]]
964	then
965	eval $submcom $job_on_remhost
966	echo "$submcom $job_on_remhost"
967	chmod u+x $job_on_remhost
968	sleep 5 # WEIL N1GE IM HINTERGRUND ARBEITET UND DATEI SONST EVTL.
969	# SCHON WIEDER GELOESCT IST
970	elif [[ $local_host = nech ]]
971	then
972	if [[ $queue = default ]]
973	then
974	eval $submcom $job_on_remhost
975	else
976	eval $submcom -q $queue $job_on_remhost
977	fi
978	else
979	qsub $job_on_remhost
980	fi
981	rm $job_on_remhost
982	cd - > /dev/null
983	fi
984	fi
985
986
987
988	# ABSCHLUSSARBEITEN
989
990	if [[ $no_submit = false ]]
991	then
992	rm $job_to_send
993	fi
994	[[ $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 |