Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 47

Last change on this file since 47 was 35, checked in by raasch, 18 years ago
small bugfix in mbuild, subjob adjusted for RIAM machines
File size: 37.7 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	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
82
83
84	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
85	delete_dayfile=false
86	filetransfer_protocol=scp
87	locat=normal
88	no_default_queue=none
89	no_submit=false
90	job_catalog="~/job_queue"
91	job_name=none
92	local_user=$LOGNAME
93	node_usage=shared
94	numprocs=1
95	punkte="..........................................................."
96	submcom=qsub
97	queue=default
98	remote_host=none
99	remote_user=""
100	verify=true
101
102	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
103	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
104	typeset -L20 spalte1
105	typeset -R40 spalte2
106	typeset -L60 spalte3
107
108
109
110	# FEHLERBEHANDLUNG
111	# BEI EXIT:
112	trap 'if [[ $locat != normal ]]
113	then
114	case $locat in
115	(option) printf "\n --> available optios can be displayed"
116	printf " by typing:"
117	printf "\n \"subjob ?\" \n";;
118	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
119	(*) printf "\n +++ unknown error"
120	printf "\n please inform S. Raasch!\n"
121	esac
122	[[ -f $job_to_send ]] && rm $job_to_send
123	printf "\n\n+++ SUBJOB killed \n\n"
124	fi' exit
125
126
127	# BEI TERMINAL-BREAK:
128	trap '[[ -f $job_to_send ]] && rm $job_to_send
129	printf "\n\n+++ SUBJOB killed \n\n"
130	exit
131	' 2
132
133
134
135
136	# LOKALEN HOSTNAMEN FESTSTELLEN
137	local_host=$(hostname)
138
139
140
141	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
142	# UEBERHAUPT ZULAESSIG IST
143	case $local_host in
144	(atmos) local_addres=172.20.25.35; local_host=lcide;;
145	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
146	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
147	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
148	(cs*) local_addres=136.172.44.131; local_host=nech;;
149	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
150	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
151	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
152	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
153	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
154	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
155	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
156	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
157	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
158	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
159	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
160	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
161	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
162	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
163	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
164	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
165	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
166	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
167	(*) printf "\n +++ \"$local_host\" unknown";
168	printf "\n please inform S. Raasch!";
169	locat=parameter; exit;;
170	esac
171
172
173
174	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
175	remote_host=$local_host
176
177
178
179
180	# PROZEDUROPTIONEN EINLESEN
181	while getopts :c:dDf:h:m:n:N:O:q:t:T:u:vX: option
182	do
183	case $option in
184	(c) job_catalog=$OPTARG;;
185	(d) delete_dayfile=true;;
186	(D) no_submit=true;;
187	(f) filetransfer_protocol=$OPTARG;;
188	(h) remote_host=$OPTARG;;
189	(m) memory=$OPTARG;;
190	(n) job_name=$OPTARG;;
191	(N) node_usage=$OPTARG;;
192	(O) threads_per_task=$OPTARG;;
193	(q) no_default_queue=$OPTARG;;
194	(t) cputime=$OPTARG;;
195	(T) tasks_per_node=$OPTARG;;
196	(u) remote_user=$OPTARG;;
197	(v) verify=false;;
198	(X) numprocs=$OPTARG;;
199	(\?) printf "\n +++ Option $OPTARG unknown \n";
200	locat=option; exit;;
201	esac
202	done
203
204
205	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
206	shift OPTIND-1; file_to_send=$1
207
208
209	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
210	if [ "$1" = "?" ]
211	then
212	(printf "\n *** subjob can be called as follows:\n"
213	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
214	printf "\n Description of available options:\n"
215	printf "\n Option Description Default-Value"
216	printf "\n -c job-input- and output-catalog ~/job_queue"
217	printf "\n -d no job-protocol will be created ---"
218	printf "\n -D only the job-file will be created ---"
219	printf "\n -f filetransfer protocol scp"
220	printf "\n -h execution host, available hosts: $remote_host"
221	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
222	printf "\n lctit, nech, neck, unics"
223	printf "\n -m memory demand per process in MByte ---"
224	printf "\n -n jobname <jobdatei>"
225	printf "\n -O threads per task (for OpenMP usage) 1"
226	printf "\n -q job-queue to be used default"
227	printf "\n -t allowed cpu-time in seconds ---"
228	printf "\n -T tasks per node (on parallel hosts) ---"
229	printf "\n -u username on execution host from .netrc"
230	printf "\n -v no prompt for confirmation ---"
231	printf "\n -X # of processors (on parallel hosts) 1"
232	printf "\n "
233	printf "\n The only possible positional parameter is <jobfile>:"
234	printf "\n The complete NQS-job must be provided here."
235	printf "\n <jobfile>=? creates this outline\n\n") \| more
236	exit
237	fi
238
239
240
241	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
242	if [[ "$file_to_send" = "" ]]
243	then
244	printf "\n +++ job-file missing"
245	locat=parameter; exit
246	else
247	if [[ -f $file_to_send ]]
248	then
249	true
250	else
251	printf "\n +++ job-file: "
252	printf "\n $file_to_send"
253	printf "\n does not exist"
254	locat=parameter; exit
255	fi
256	fi
257
258
259
260	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
261	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
262	if [[ $job_name = none ]]
263	then
264	job_name=$file_to_send
265	fi
266	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
267	then
268	printf "\n +++ job-file name: "
269	printf "\n $job_name"
270	printf "\n must not contain \"/\"-characters"
271	locat=parameter; exit
272	fi
273
274
275
276
277	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
278	# ODER WENN HOST NICHT ANGEGEBEN WURDE
279	if [[ $remote_host = none ]]
280	then
281	printf "\n +++ host missing"
282	locat=option; exit
283	else
284	case $remote_host in
285	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
286	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
287	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
288	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
289	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
290	(lctit) queue=test; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
291	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
292	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
293	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
294	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
295	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
296	locat=parameter; exit;;
297	esac
298	fi
299
300
301	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
302	if [[ $no_default_queue != none ]]
303	then
304	error=false
305	ndq=$no_default_queue
306	case $remote_host in
307	(ibm) case $ndq in
308	(p690_express\|p690_standard\|p690_long) error=false;;
309	(*) error=true;;
310	esac;;
311	(ibmb) case $ndq in
312	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
313	(*) error=true;;
314	esac;;
315	(ibmh) case $ndq in
316	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
317	(*) error=true;;
318	esac;;
319	(ibms) case $ndq in
320	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
321	(*) error=true;;
322	esac;;
323	(ibmy) case $ndq in
324	(parallel) error=false;;
325	(*) error=true;;
326	esac;;
327	(lctit) case $ndq in
328	(cs\|default\|high\|test) error=false;;
329	(*) error=true;;
330	esac;;
331	(t3eb) case $ndq in
332	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
333	(*) error=true;;
334	esac;;
335	(t3eh) case $ndq in
336	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
337	(*) error=true;;
338	esac;;
339	(t3ej2\|t3ej5) case $ndq in
340	(low\|normal\|high) error=false;;
341	(*) error=true;;
342	esac;;
343	(t3es) case $ndq in
344	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
345	(*) error=true;;
346	esac;;
347	(unics) case $ndq in
348	(unics\|ht) error=false;;
349	(*) error=true;;
350	esac;;
351	esac
352	if [[ $error = true ]]
353	then
354	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
355	locat=parameter; exit
356	else
357	queue=$no_default_queue
358	fi
359	fi
360
361
362
363	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
364	if [[ $node_usage != shared && $queue = cdev ]]
365	then
366	node_usage=shared
367	fi
368
369
370
371	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
372	# AUFTEILEN
373	done=false
374	while [[ $done = false ]]
375	do
376	if (( $cputime <= 0 ))
377	then
378	printf "\n +++ wrong cpu-time or cpu-time missing"
379	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
380	printf "\n >>> "
381	read cputime 1>/dev/null 2>&1
382	else
383	done=true
384	fi
385	done
386	if [[ $remote_host = nech ]]
387	then
388	if (( tasks_per_node != 0 ))
389	then
390	(( cputime = cputime * tasks_per_node ))
391	elif [[ $numprocs != 0 ]]
392	then
393	(( cputime = cputime * numprocs ))
394	fi
395	fi
396	(( stunden = cputime / 3600 ))
397	(( resttime = cputime - stunden * 3600 ))
398	(( minuten = resttime / 60 ))
399	(( sekunden = resttime - minuten * 60 ))
400	timestring=${stunden}:${minuten}:${sekunden}
401
402
403
404	# PRUEFEN DER KERNSPEICHERANFORDERUNG
405	done=false
406	while [[ $done = false ]]
407	do
408	if (( memory <= 0 ))
409	then
410	printf "\n +++ wrong memory demand or memory demand missing"
411	printf "\n >>> Please type memory in MByte per process as INTEGER:"
412	printf "\n >>> "
413	read memory 1>/dev/null 2>&1
414	else
415	done=true
416	fi
417	done
418
419	if [[ $remote_host = nech \|\| $remote_host = neck ]]
420	then
421	if (( tasks_per_node != 0 ))
422	then
423	(( Memory = memory * tasks_per_node / 1000 ))
424	elif [[ $numprocs != 0 ]]
425	then
426	(( Memory = memory * numprocs / 1000 ))
427	else
428	(( Memory = memory / 1000 ))
429	fi
430	fi
431
432
433	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
434	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
435	then
436	(( memory = memory * threads_per_task ))
437	fi
438
439
440	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
441	if (( tasks_per_node != 0 ))
442	then
443	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
444	fi
445
446
447
448	# HEADER-AUSGABE
449	if [[ $verify = true ]]
450	then
451	printf "\n\n"
452	printf "#--------------------------------------------------------------# \n"
453	spalte1=SUBJOB;spalte2=$(date)
454	printf "\| $spalte1$spalte2 \| \n"
455	printf "\| \| \n"
456	printf "\| values of parameters/options: \| \n"
457	spalte1=local_host$punkte; spalte2=$punkte$local_host
458	printf "\| $spalte1$spalte2 \| \n"
459	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
460	printf "\| $spalte1$spalte2 \| \n"
461	spalte1=queue$punkte; spalte2=$punkte$queue
462	printf "\| $spalte1$spalte2 \| \n"
463	spalte1=memory$punkte; spalte2="$punkte$memory mb"
464	printf "\| $spalte1$spalte2 \| \n"
465	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
466	printf "\| $spalte1$spalte2 \| \n"
467	spalte1=job_name$punkte; spalte2="$punkte$job_name"
468	printf "\| $spalte1$spalte2 \| \n"
469	printf "#--------------------------------------------------------------# \n\n"
470
471
472	# KONTROLLABFRAGE, OB ALLES O.K.
473	antwort="dummy"
474	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
475	do
476	read antwort?" >>> continue (y/n) ? "
477	done
478	if [[ $antwort = n \|\| $antwort = N ]]
479	then
480	locat=verify; exit
481	fi
482	printf "\n"
483	fi
484
485
486
487	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
488	kennung=$RANDOM
489	job_on_remhost=${job_name}_${kennung}_$local_host
490	job_to_send=job_to_send_$kennung
491	if [[ $delete_dayfile = false ]]
492	then
493	remote_dayfile=${local_host}_${job_name}_result_$kennung
494	local_dayfile=${remote_host}_${job_name}
495	else
496	remote_dayfile=/dev/null
497	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
498	then
499	notification='# @ notification = never'
500	fi
501	fi
502
503
504
505	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
506	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
507	then
508
509	if [[ $remote_host = ibmy ]]
510	then
511	consumable_memory=""
512	else
513	consumable_memory="ConsumableMemory($memory mb)"
514	fi
515	if [[ $queue = cdev ]]
516	then
517	data_limit="# @ data_limit = 1.76gb"
518	network_to_use="# @ network.mpi = sn_all,shared,ip"
519	else
520	if [[ $remote_host = ibms ]]
521	then
522	network_to_use="# @ network.mpi = csss,shared,us"
523	elif [[ $remote_host = ibmy ]]
524	then
525	network_to_use=""
526	else
527	network_to_use="# @ network.mpi = sn_all,shared,us"
528	data_limit="# @ data_limit = 1.76gb"
529	fi
530	fi
531
532	cat > $job_to_send << %%END%%
533	#!/bin/ksh
534
535	# @ job_type = parallel
536	# @ job_name = palm
537	# @ wall_clock_limit = ${timestring},$timestring
538	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
539	# @ output = $remote_dayfile
540	# @ error = $remote_dayfile
541	# @ image_size = 50
542	# @ class = $queue
543	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
544	$network_to_use
545	$data_limit
546	$notification
547
548	%%END%%
549
550	if (( nodes > 0 ))
551	then
552
553	cat >> $job_to_send << %%END%%
554	# @ node = $nodes
555	# @ tasks_per_node = $tasks_per_node
556	# @ node_usage = $node_usage
557	# @ queue
558
559	%%END%%
560
561	else
562
563	if [[ $remote_host != ibmy ]]
564	then
565
566	cat >> $job_to_send << %%END%%
567	# @ blocking = unlimited
568	# @ total_tasks = $numprocs
569	# @ node_usage = $node_usage
570	# @ queue
571
572	%%END%%
573
574	else
575
576	cat >> $job_to_send << %%END%%
577	# @ node = 1
578	# @ total_tasks = $numprocs
579	# @ queue
580
581	%%END%%
582
583	fi
584
585	fi
586
587	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
588	then
589
590	cat > $job_to_send << %%END%%
591	#!/bin/ksh
592
593	# @ job_type = serial
594	# @ node_usage = $node_usage
595	# @ job_name = palm
596	# @ wall_clock_limit = ${timestring},$timestring
597	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
598	# @ output = $remote_dayfile
599	# @ error = $remote_dayfile
600	# @ class = $queue
601	$notification
602
603	# @ queue
604
605	%%END%%
606
607	elif [[ $remote_host = nech ]]
608	then
609
610	if (( nodes > 1 ))
611	then
612	cat > $job_to_send << %%END%%
613	#!/bin/ksh
614	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
615	#PBS -l ${qsubmem}=${Memory}gb
616	#PBS -b $nodes
617	#PBS -o $remote_dayfile
618	#PBS -N palm
619	#PBS -j o
620	#PBS -T mpisx
621
622	%%END%%
623
624	elif [[ $numprocs != 0 ]]
625	then
626	cat > $job_to_send << %%END%%
627	#!/bin/ksh
628	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
629	#PBS -l ${qsubmem}=${Memory}gb
630	#PBS -o $remote_dayfile
631	#PBS -N palm
632	#PBS -j o
633
634	%%END%%
635
636	else
637	cat > $job_to_send << %%END%%
638	#!/bin/ksh
639	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
640	#PBS -o $remote_dayfile
641	#PBS -j o
642
643	%%END%%
644
645	fi
646
647	elif [[ $remote_host = neck ]]
648	then
649
650	if (( nodes > 1 ))
651	then
652	cat > $job_to_send << %%END%%
653	#!/bin/ksh
654	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
655	#PBS -l ${qsubmem}=${Memory}gb
656	#PBS -b $nodes
657	#PBS -o $remote_dayfile
658	#PBS -N palm
659	#PBS -j o
660	#PBS -T mpisx
661
662	%%END%%
663
664	elif [[ $numprocs != 0 ]]
665	then
666	cat > $job_to_send << %%END%%
667	#!/bin/ksh
668	#PBS -q ${queue}
669	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
670	#PBS -l ${qsubmem}=${Memory}gb
671	#PBS -o $remote_dayfile
672	#PBS -N palm
673	#PBS -j o
674	#PBS -T mpisx
675
676	%%END%%
677
678	else
679	cat > $job_to_send << %%END%%
680	#!/bin/ksh
681	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
682	#PBS -o $remote_dayfile
683	#PBS -j o
684
685	%%END%%
686
687	fi
688
689	elif [[ $remote_host = lctit ]]
690	then
691	cat > $job_to_send << %%END%%
692	#!/bin/bash
693	#$ -S /bin/bash
694	cd $job_catalog
695	PATH=$job_catalog:${job_catalog}/../pub:\$PATH
696	export PATH
697
698	%%END%%
699
700	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
701	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
702
703	else
704
705	cat > $job_to_send << %%END%%
706	# @\$-q ${queue}
707	# @\$-l${qsubtime} $timestring
708	# @\$-l${qsubmem} ${memory}mb
709	# @\$-o $remote_dayfile
710	# @\$-eo
711
712	%%END%%
713
714	fi
715
716
717	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
718	# BEI EXIT VERANLASSEN
719	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
720	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
721	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
722	then
723	echo "set +vx" >> $job_to_send
724	echo "trap '" >> $job_to_send
725	echo "set +vx" >> $job_to_send
726	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
727	then
728	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
729	then
730	return_queue=c1
731	elif [[ $remote_host = ibms ]]
732	then
733	return_queue=p_normal
734	elif [[ $remote_host = ibmy ]]
735	then
736	return_queue=serial
737	elif [[ $remote_host = lctit ]]
738	then
739	return_queue=default
740	elif [[ $remote_host = neck ]]
741	then
742	return_queue=S
743	else
744	return_queue=unknown
745	fi
746
747	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
748	then
749
750	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
751	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
752	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
753	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
754	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
755	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
756	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
757	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
758	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
759	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
760
761	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
762	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
763
764	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
765	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
766	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
767
768	elif [[ $remote_host = nech ]]
769	then
770	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
771	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
772	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
773	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
774	echo "#PBS -j o" >> $job_to_send
775	echo " " >> $job_to_send
776	echo "set -x" >> $job_to_send
777	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
778	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
779	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
780	echo "%%END%%" >> $job_to_send
781
782	elif [[ $remote_host = neck ]]
783	then
784	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
785	echo "#PBS -q $return_queue" >> $job_to_send
786	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
787	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
788	echo "#PBS -j o" >> $job_to_send
789	echo " " >> $job_to_send
790	echo "set -x" >> $job_to_send
791	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
792	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
793	echo "%%END%%" >> $job_to_send
794
795	elif [[ $remote_host = lctit ]]
796	then
797	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
798	echo "set -x" >> $job_to_send
799	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
800	echo "cd $job_catalog" >> $job_to_send
801	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
802	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
803	echo "%%END%%" >> $job_to_send
804
805	else
806
807	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
808	echo "# @\\\$-q $return_queue" >> $job_to_send
809	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
810	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
811	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
812	then
813	echo "# @\$-l mpp_p=0" >> $job_to_send
814	fi
815	echo '# @\$-lF 10mb' >> $job_to_send
816	# echo '# @\$-o /dev/null' >> $job_to_send
817	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
818	echo '# @\\\$-eo' >> $job_to_send
819	echo " " >> $job_to_send
820	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
821	then
822	echo "set +vx" >> $job_to_send
823	echo ". .profile" >> $job_to_send
824	fi
825	echo "set -x" >> $job_to_send
826	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
827	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
828	echo "%%END%%" >> $job_to_send
829	fi
830
831	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
832	then
833	echo "llsubmit scpjob.$kennung" >> $job_to_send
834	elif [[ $remote_host = lctit ]]
835	then
836	echo "chmod u+x scpjob.$kennung" >> $job_to_send
837	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
838	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
839	echo "sleep 10" >> $job_to_send
840	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
841	then
842	echo "qsub -J n scpjob.$kennung" >> $job_to_send
843	elif [[ $remote_host = t3es ]]
844	then
845	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
846	else
847	echo "qsub scpjob.$kennung" >> $job_to_send
848	fi
849	echo "rm scpjob.$kennung" >> $job_to_send
850	if [[ $remote_host = nech ]]
851	then
852	echo "cd -" >> $job_to_send
853	fi
854	else
855	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
856	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
857	fi
858	echo "set -x" >> $job_to_send
859	echo " ' exit" >> $job_to_send
860	echo "set -x" >> $job_to_send
861	fi
862
863
864
865	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
866	cat $file_to_send >> $job_to_send
867	if [[ $remote_host = ibm ]]
868	then
869	echo " " >> $job_to_send
870	echo "exit" >> $job_to_send
871	fi
872
873
874
875	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
876	if [[ -z $remote_user ]]
877	then
878	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
879	then
880	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
881	fi
882	fi
883
884
885
886	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
887	if [[ $no_submit = false ]]
888	then
889	if [[ $remote_host != $local_host ]]
890	then
891	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
892	if [[ $filetransfer_protocol = scp ]]
893	then
894	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
895	then
896	job_catalog_save=$job_catalog
897	job_catalog=job_queue
898	elif [[ $remote_host = nech ]]
899	then
900	job_catalog_save=$job_catalog
901	job_catalog=/hpf/b/${remote_user}/job_queue
902	fi
903	if [[ $local_host = decalpha ]]
904	then
905	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
906	# PASSWORT AB
907	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
908	elif [[ $remote_host = nech ]]
909	then
910	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
911	# TRANSFERIERT WERDEN
912	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
913	else
914	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
915	fi
916	if [[ $? = 1 ]]
917	then
918	locat=scp; exit
919	fi
920	if [[ $remote_host = ibms ]]
921	then
922	job_catalog=$job_catalog_save
923	fi
924	else
925	ftpcopy -s -o $remote_addres $job_to_send "$job_catalog" $job_on_remhost
926	if [[ $? = 1 ]]
927	then
928	locat=ftpcopy; exit
929	fi
930	fi
931	[[ $verify = true ]] && printf "\n >>> finished\n"
932	else
933	eval job_catalog=$job_catalog
934	cp $job_to_send ${job_catalog}/$job_on_remhost
935	fi
936
937
938
939	# NQS- BZW. LOADLEVELER-JOB STARTEN
940	if [[ $remote_host != $local_host ]]
941	then
942	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
943	if [[ $filetransfer_protocol = scp ]]
944	then
945	if [[ $remote_host != lctit ]]
946	then
947	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
948	else
949	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
950	# MIT SSH, DESHALB AUFRUF PER PIPE
951	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
952	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
953	echo "first try" > send_protocol
954	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
955	do
956	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
957	then
958	printf "\n trying to submit job to TIT cluster..."
959	else
960	printf "\n +++ failed ... trying to submit again ..."
961	fi
962	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
963	cat send_protocol
964	done
965	sleep 10
966	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
967	echo "$submcom $job_on_remhost"
968	rm send_protocol
969	fi
970	else
971	if [[ $remote_host = ibms ]]
972	then
973	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
974	echo "echo \"$remote_user\" " > telnet_script
975	echo "sleep 5" >> telnet_script
976	echo "echo \"$password\" " >> telnet_script
977	echo "sleep 10" >> telnet_script
978	echo "echo \"\" " >> telnet_script
979	echo "sleep 2" >> telnet_script
980	echo "echo \"cd $job_catalog; $submcom $job_on_remhost; rm -rf $job_on_remhost\" " >> telnet_script
981	echo "sleep 10" >> telnet_script
982	echo "echo \"exit\" " >> telnet_script
983
984	chmod u+x telnet_script
985	telnet_script \| telnet $remote_addres 2>&1 \| grep "llsubmit:"
986	rm telnet_script
987	elif [[ $remote_host = t3es ]]
988	then
989	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
990	echo "printf \"$remote_user\\\n\" " > telnet_script
991	echo "sleep 10" >> telnet_script
992	echo "printf \"$password\\\n\" " >> telnet_script
993	echo "sleep 10" >> telnet_script
994	echo "printf \"\\\n\" " >> telnet_script
995	echo "sleep 5" >> telnet_script
996	echo "printf \"\\\n\" " >> telnet_script
997	echo "sleep 5" >> telnet_script
998	echo "printf \"cd $job_catalog; $submcom $job_on_remhost; rm -f $job_on_remhost\\\n\" " >> telnet_script
999	echo "sleep 10" >> telnet_script
1000	echo "printf \"exit\\\n\" " >> telnet_script
1001
1002	chmod u+x telnet_script
1003	cat telnet_script
1004	telnet_script \| telnet $remote_addres \| grep $remote_user
1005	rm telnet_script
1006	else
1007	remcom.x $remote_addres "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1008	fi
1009	fi
1010	[[ $verify = true ]] && printf " >>> o.k.\n"
1011	else
1012	cd $job_catalog
1013	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1014	then
1015	eval $submcom $job_on_remhost
1016	elif [[ $local_host = lctit ]]
1017	then
1018	eval $submcom $job_on_remhost
1019	echo "$submcom $job_on_remhost"
1020	chmod u+x $job_on_remhost
1021	sleep 5 # WEIL N1GE IM HINTERGRUND ARBEITET UND DATEI SONST EVTL.
1022	# SCHON WIEDER GELOESCT IST
1023	elif [[ $local_host = nech ]]
1024	then
1025	if [[ $queue = default ]]
1026	then
1027	eval $submcom $job_on_remhost
1028	else
1029	eval $submcom -q $queue $job_on_remhost
1030	fi
1031	else
1032	qsub $job_on_remhost
1033	fi
1034	rm $job_on_remhost
1035	cd - > /dev/null
1036	fi
1037	fi
1038
1039
1040
1041	# ABSCHLUSSARBEITEN
1042
1043	if [[ $no_submit = false ]]
1044	then
1045	rm $job_to_send
1046	fi
1047	[[ $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 |