Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1

Last change on this file since 1 was 1, checked in by raasch, 17 years ago
Initial repository layout and content
File size: 36.4 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0 28/11/06
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
81
82	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
83	delete_dayfile=false
84	filetransfer_protocol=scp
85	locat=normal
86	no_default_queue=none
87	no_submit=false
88	job_catalog="~/job_queue"
89	job_name=none
90	local_user=$LOGNAME
91	node_usage=shared
92	numprocs=1
93	punkte="..........................................................."
94	submcom=qsub
95	queue=default
96	remote_host=none
97	remote_user=""
98	verify=true
99
100	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
101	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
102	typeset -L20 spalte1
103	typeset -R40 spalte2
104	typeset -L60 spalte3
105
106
107
108	# FEHLERBEHANDLUNG
109	# BEI EXIT:
110	trap 'if [[ $locat != normal ]]
111	then
112	case $locat in
113	(option) printf "\n --> available optios can be displayed"
114	printf " by typing:"
115	printf "\n \"subjob ?\" \n";;
116	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
117	(*) printf "\n +++ unknown error"
118	printf "\n please inform S. Raasch!\n"
119	esac
120	[[ -f $job_to_send ]] && rm $job_to_send
121	printf "\n\n+++ SUBJOB killed \n\n"
122	fi' exit
123
124
125	# BEI TERMINAL-BREAK:
126	trap '[[ -f $job_to_send ]] && rm $job_to_send
127	printf "\n\n+++ SUBJOB killed \n\n"
128	exit
129	' 2
130
131
132
133
134	# LOKALEN HOSTNAMEN FESTSTELLEN
135	local_host=$(hostname)
136
137
138
139	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
140	# UEBERHAUPT ZULAESSIG IST
141	case $local_host in
142	(atmos) local_addres=172.20.25.35; local_host=lcide;;
143	(bora) local_addres=130.75.105.103; local_host=hpmuk;;
144	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
145	(breva) local_addres=130.75.105.98; local_host=hpmuk;;
146	(caurus) local_addres=130.75.105.19; local_host=hpmuk;;
147	(chinook) local_addres=130.75.105.100; local_host=hpmuk;;
148	(cs*) local_addres=136.172.44.131; local_host=nech;;
149	(elephanta) local_addres=130.75.105.6; local_host=hpmuk;;
150	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
151	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
152	(gregale) local_addres=130.75.105.109; local_host=hpmuk;;
153	(hababai) local_addres=130.75.105.108; local_host=hpmuk;;
154	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
155	(irifi) local_addres=130.75.105.104; local_host=hpmuk;;
156	(levanto) local_addres=130.75.105.45; local_host=hpmuk;;
157	(maestro) local_addres=130.75.105.2; local_host=hpmuk;;
158	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
159	(orkan) local_addres=130.75.105.3; local_host=hpmuk;;
160	(ostria) local_addres=130.75.105.106; local_host=hpmuk;;
161	(quanero) local_addres=130.75.105.107; local_host=hpmuk;;
162	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
163	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
164	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
165	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
166	(*) printf "\n +++ \"$local_host\" unknown";
167	printf "\n please inform S. Raasch!";
168	locat=parameter; exit;;
169	esac
170
171
172
173	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
174	remote_host=$local_host
175
176
177
178
179	# PROZEDUROPTIONEN EINLESEN
180	while getopts :c:dDf:h:m:n:N:O:q:t:T:u:vX: option
181	do
182	case $option in
183	(c) job_catalog=$OPTARG;;
184	(d) delete_dayfile=true;;
185	(D) no_submit=true;;
186	(f) filetransfer_protocol=$OPTARG;;
187	(h) remote_host=$OPTARG;;
188	(m) memory=$OPTARG;;
189	(n) job_name=$OPTARG;;
190	(N) node_usage=$OPTARG;;
191	(O) threads_per_task=$OPTARG;;
192	(q) no_default_queue=$OPTARG;;
193	(t) cputime=$OPTARG;;
194	(T) tasks_per_node=$OPTARG;;
195	(u) remote_user=$OPTARG;;
196	(v) verify=false;;
197	(X) numprocs=$OPTARG;;
198	(\?) printf "\n +++ Option $OPTARG unknown \n";
199	locat=option; exit;;
200	esac
201	done
202
203
204	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
205	shift OPTIND-1; file_to_send=$1
206
207
208	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
209	if [ "$1" = "?" ]
210	then
211	(printf "\n *** subjob can be called as follows:\n"
212	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
213	printf "\n Description of available options:\n"
214	printf "\n Option Description Default-Value"
215	printf "\n -c job-input- and output-catalog ~/job_queue"
216	printf "\n -d no job-protocol will be created ---"
217	printf "\n -D only the job-file will be created ---"
218	printf "\n -f filetransfer protocol scp"
219	printf "\n -h execution host, available hosts: $remote_host"
220	printf "\n hpmuk, ibm, ibmb, ibmh, ibms, ibmy,"
221	printf "\n unics"
222	printf "\n -m memory demand per process in MByte ---"
223	printf "\n -n jobname <jobdatei>"
224	printf "\n -O threads per task (for OpenMP usage) 1"
225	printf "\n -q job-queue to be used default"
226	printf "\n -t allowed cpu-time in seconds ---"
227	printf "\n -T tasks per node (on parallel hosts) ---"
228	printf "\n -u username on execution host from .netrc"
229	printf "\n -v no prompt for confirmation ---"
230	printf "\n -X # of processors (on parallel hosts) 1"
231	printf "\n "
232	printf "\n The only possible positional parameter is <jobfile>:"
233	printf "\n The complete NQS-job must be provided here."
234	printf "\n <jobfile>=? creates this outline\n\n") \| more
235	exit
236	fi
237
238
239
240	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
241	if [[ "$file_to_send" = "" ]]
242	then
243	printf "\n +++ job-file missing"
244	locat=parameter; exit
245	else
246	if [[ -f $file_to_send ]]
247	then
248	true
249	else
250	printf "\n +++ job-file: "
251	printf "\n $file_to_send"
252	printf "\n does not exist"
253	locat=parameter; exit
254	fi
255	fi
256
257
258
259	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
260	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
261	if [[ $job_name = none ]]
262	then
263	job_name=$file_to_send
264	fi
265	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
266	then
267	printf "\n +++ job-file name: "
268	printf "\n $job_name"
269	printf "\n must not contain \"/\"-characters"
270	locat=parameter; exit
271	fi
272
273
274
275
276	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
277	# ODER WENN HOST NICHT ANGEGEBEN WURDE
278	if [[ $remote_host = none ]]
279	then
280	printf "\n +++ host missing"
281	locat=option; exit
282	else
283	case $remote_host in
284	(hpmuk) qsubmem=d; qsubtime=t; queue=hpmuk; remote_addres=130.75.105.19; submcom=/opt/nqs/bin/qsub;;
285	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
286	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
287	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
288	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
289	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
290	(lctit) queue=test; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
291	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
292	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
293	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
294	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
295	locat=parameter; exit;;
296	esac
297	fi
298
299
300	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
301	if [[ $no_default_queue != none ]]
302	then
303	error=false
304	ndq=$no_default_queue
305	case $remote_host in
306	(hpmuk) case $ndq in
307	(hpmuk) error=false;;
308	(*) error=true;;
309	esac;;
310	(ibm) case $ndq in
311	(p690_express\|p690_standard\|p690_long) error=false;;
312	(*) error=true;;
313	esac;;
314	(ibmb) case $ndq in
315	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
316	(*) error=true;;
317	esac;;
318	(ibmh) case $ndq in
319	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|cspec) error=false;;
320	(*) error=true;;
321	esac;;
322	(ibms) case $ndq in
323	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
324	(*) error=true;;
325	esac;;
326	(ibmy) case $ndq in
327	(parallel) error=false;;
328	(*) error=true;;
329	esac;;
330	(lctit) case $ndq in
331	(cs\|default\|high\|test) error=false;;
332	(*) error=true;;
333	esac;;
334	(t3eb) case $ndq in
335	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
336	(*) error=true;;
337	esac;;
338	(t3eh) case $ndq in
339	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
340	(*) error=true;;
341	esac;;
342	(t3ej2\|t3ej5) case $ndq in
343	(low\|normal\|high) error=false;;
344	(*) error=true;;
345	esac;;
346	(t3es) case $ndq in
347	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
348	(*) error=true;;
349	esac;;
350	(unics) case $ndq in
351	(unics\|ht) error=false;;
352	(*) error=true;;
353	esac;;
354	esac
355	if [[ $error = true ]]
356	then
357	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
358	locat=parameter; exit
359	else
360	queue=$no_default_queue
361	fi
362	fi
363
364
365
366	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
367	if [[ $node_usage != shared && $queue = cdev ]]
368	then
369	node_usage=shared
370	fi
371
372
373
374	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
375	# AUFTEILEN
376	done=false
377	while [[ $done = false ]]
378	do
379	if (( $cputime <= 0 ))
380	then
381	printf "\n +++ wrong cpu-time or cpu-time missing"
382	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
383	printf "\n >>> "
384	read cputime 1>/dev/null 2>&1
385	else
386	done=true
387	fi
388	done
389	if [[ $remote_host = nech ]]
390	then
391	if (( tasks_per_node != 0 ))
392	then
393	(( cputime = cputime * tasks_per_node ))
394	elif [[ $numprocs != 0 ]]
395	then
396	(( cputime = cputime * numprocs ))
397	fi
398	fi
399	(( stunden = cputime / 3600 ))
400	(( resttime = cputime - stunden * 3600 ))
401	(( minuten = resttime / 60 ))
402	(( sekunden = resttime - minuten * 60 ))
403	timestring=${stunden}:${minuten}:${sekunden}
404
405
406
407	# PRUEFEN DER KERNSPEICHERANFORDERUNG
408	done=false
409	while [[ $done = false ]]
410	do
411	if (( memory <= 0 ))
412	then
413	printf "\n +++ wrong memory demand or memory demand missing"
414	printf "\n >>> Please type memory in MByte per process as INTEGER:"
415	printf "\n >>> "
416	read memory 1>/dev/null 2>&1
417	else
418	done=true
419	fi
420	done
421
422	if [[ $remote_host = nech ]]
423	then
424	if (( tasks_per_node != 0 ))
425	then
426	(( Memory = memory * tasks_per_node / 1000 ))
427	elif [[ $numprocs != 0 ]]
428	then
429	(( Memory = memory * numprocs / 1000 ))
430	else
431	(( Memory = memory / 1000 ))
432	fi
433	fi
434
435
436	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
437	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
438	then
439	(( memory = memory * threads_per_task ))
440	fi
441
442
443	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
444	if (( tasks_per_node != 0 ))
445	then
446	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
447	fi
448
449
450
451	# HEADER-AUSGABE
452	if [[ $verify = true ]]
453	then
454	printf "\n\n"
455	printf "#--------------------------------------------------------------# \n"
456	spalte1=SUBJOB;spalte2=$(date)
457	printf "\| $spalte1$spalte2 \| \n"
458	printf "\| \| \n"
459	printf "\| values of parameters/options: \| \n"
460	spalte1=local_host$punkte; spalte2=$punkte$local_host
461	printf "\| $spalte1$spalte2 \| \n"
462	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
463	printf "\| $spalte1$spalte2 \| \n"
464	spalte1=queue$punkte; spalte2=$punkte$queue
465	printf "\| $spalte1$spalte2 \| \n"
466	spalte1=memory$punkte; spalte2="$punkte$memory mb"
467	printf "\| $spalte1$spalte2 \| \n"
468	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
469	printf "\| $spalte1$spalte2 \| \n"
470	spalte1=job_name$punkte; spalte2="$punkte$job_name"
471	printf "\| $spalte1$spalte2 \| \n"
472	printf "#--------------------------------------------------------------# \n\n"
473
474
475	# KONTROLLABFRAGE, OB ALLES O.K.
476	antwort="dummy"
477	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
478	do
479	read antwort?" >>> continue (y/n) ? "
480	done
481	if [[ $antwort = n \|\| $antwort = N ]]
482	then
483	locat=verify; exit
484	fi
485	printf "\n"
486	fi
487
488
489
490	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
491	kennung=$RANDOM
492	job_on_remhost=${job_name}_${kennung}_$local_host
493	job_to_send=job_to_send_$kennung
494	if [[ $delete_dayfile = false ]]
495	then
496	remote_dayfile=${local_host}_${job_name}_result_$kennung
497	local_dayfile=${remote_host}_${job_name}
498	else
499	remote_dayfile=/dev/null
500	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
501	then
502	notification='# @ notification = never'
503	fi
504	fi
505
506
507
508	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
509	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
510	then
511
512	if [[ $remote_host = ibmy ]]
513	then
514	consumable_memory=""
515	else
516	consumable_memory="ConsumableMemory($memory mb)"
517	fi
518	if [[ $queue = cdev ]]
519	then
520	data_limit="# @ data_limit = 1.76gb"
521	network_to_use="# @ network.mpi = sn_all,shared,ip"
522	else
523	if [[ $remote_host = ibms ]]
524	then
525	network_to_use="# @ network.mpi = csss,shared,us"
526	elif [[ $remote_host = ibmy ]]
527	then
528	network_to_use=""
529	else
530	network_to_use="# @ network.mpi = sn_all,shared,us"
531	data_limit="# @ data_limit = 1.76gb"
532	fi
533	fi
534
535	cat > $job_to_send << %%END%%
536	#!/bin/ksh
537
538	# @ job_type = parallel
539	# @ job_name = palm
540	# @ wall_clock_limit = ${timestring},$timestring
541	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
542	# @ output = $remote_dayfile
543	# @ error = $remote_dayfile
544	# @ image_size = 50
545	# @ class = $queue
546	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
547	$network_to_use
548	$data_limit
549	$notification
550
551	%%END%%
552
553	if (( nodes > 0 ))
554	then
555
556	cat >> $job_to_send << %%END%%
557	# @ node = $nodes
558	# @ tasks_per_node = $tasks_per_node
559	# @ node_usage = $node_usage
560	# @ queue
561
562	%%END%%
563
564	else
565
566	if [[ $remote_host != ibmy ]]
567	then
568
569	cat >> $job_to_send << %%END%%
570	# @ blocking = unlimited
571	# @ total_tasks = $numprocs
572	# @ node_usage = $node_usage
573	# @ queue
574
575	%%END%%
576
577	else
578
579	cat >> $job_to_send << %%END%%
580	# @ node = 1
581	# @ total_tasks = $numprocs
582	# @ queue
583
584	%%END%%
585
586	fi
587
588	fi
589
590	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
591	then
592
593	cat > $job_to_send << %%END%%
594	#!/bin/ksh
595
596	# @ job_type = serial
597	# @ node_usage = $node_usage
598	# @ job_name = palm
599	# @ wall_clock_limit = ${timestring},$timestring
600	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
601	# @ output = $remote_dayfile
602	# @ error = $remote_dayfile
603	# @ class = $queue
604	$notification
605
606	# @ queue
607
608	%%END%%
609
610	elif [[ $remote_host = nech ]]
611	then
612
613	if (( nodes > 1 ))
614	then
615	cat > $job_to_send << %%END%%
616	#!/bin/ksh
617	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
618	#PBS -l ${qsubmem}=${Memory}gb
619	#PBS -b $nodes
620	#PBS -o $remote_dayfile
621	#PBS -N palm
622	#PBS -j o
623	#PBS -T mpisx
624
625	%%END%%
626
627	elif [[ $numprocs != 0 ]]
628	then
629	cat > $job_to_send << %%END%%
630	#!/bin/ksh
631	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
632	#PBS -l ${qsubmem}=${Memory}gb
633	#PBS -o $remote_dayfile
634	#PBS -N palm
635	#PBS -j o
636
637	%%END%%
638
639	else
640	cat > $job_to_send << %%END%%
641	#!/bin/ksh
642	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
643	#PBS -o $remote_dayfile
644	#PBS -j o
645
646	%%END%%
647
648	fi
649
650	elif [[ $remote_host = lctit ]]
651	then
652	cat > $job_to_send << %%END%%
653	#!/bin/bash
654	#$ -S /bin/bash
655	cd $job_catalog
656	PATH=$job_catalog:${job_catalog}/../pub:\$PATH
657	export PATH
658
659	%%END%%
660
661	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
662	submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
663
664	else
665
666	cat > $job_to_send << %%END%%
667	# @\$-q ${queue}
668	# @\$-l${qsubtime} $timestring
669	# @\$-l${qsubmem} ${memory}mb
670	# @\$-o $remote_dayfile
671	# @\$-eo
672
673	%%END%%
674
675	fi
676
677
678	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
679	# BEI EXIT VERANLASSEN
680	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
681	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
682	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
683	then
684	echo "set +vx" >> $job_to_send
685	echo "trap '" >> $job_to_send
686	echo "set +vx" >> $job_to_send
687	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = nech \|\| $remote_host = lctit ]]
688	then
689	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
690	then
691	return_queue=c1
692	elif [[ $remote_host = ibms ]]
693	then
694	return_queue=p_normal
695	elif [[ $remote_host = ibmy ]]
696	then
697	return_queue=serial
698	elif [[ $remote_host = lctit ]]
699	then
700	return_queue=default
701	elif [[ $remote_host = t3eb ]]
702	then
703	return_queue=berte
704	elif [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
705	then
706	return_queue=normal
707	elif [[ $remote_host = t3es ]]
708	then
709	return_queue=batch
710	else
711	return_queue=unknown
712	fi
713
714	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
715	then
716
717	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
718	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
719	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
720	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
721	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
722	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
723	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
724	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
725	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
726	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
727
728	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
729	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
730
731	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
732	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
733	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
734
735	elif [[ $remote_host = nech ]]
736	then
737	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
738	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
739	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
740	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
741	echo "#PBS -j o" >> $job_to_send
742	echo " " >> $job_to_send
743	echo "set -x" >> $job_to_send
744	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
745	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
746	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
747	echo "%%END%%" >> $job_to_send
748
749	elif [[ $remote_host = lctit ]]
750	then
751	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
752	echo "set -x" >> $job_to_send
753	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
754	echo "cd $job_catalog" >> $job_to_send
755	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
756	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
757	echo "%%END%%" >> $job_to_send
758
759	else
760
761	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
762	echo "# @\\\$-q $return_queue" >> $job_to_send
763	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
764	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
765	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
766	then
767	echo "# @\$-l mpp_p=0" >> $job_to_send
768	fi
769	echo '# @\$-lF 10mb' >> $job_to_send
770	# echo '# @\$-o /dev/null' >> $job_to_send
771	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
772	echo '# @\\\$-eo' >> $job_to_send
773	echo " " >> $job_to_send
774	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
775	then
776	echo "set +vx" >> $job_to_send
777	echo ". .profile" >> $job_to_send
778	fi
779	echo "set -x" >> $job_to_send
780	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
781	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
782	echo "%%END%%" >> $job_to_send
783	fi
784
785	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
786	then
787	echo "llsubmit scpjob.$kennung" >> $job_to_send
788	elif [[ $remote_host = lctit ]]
789	then
790	echo "chmod u+x scpjob.$kennung" >> $job_to_send
791	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
792	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
793	echo "sleep 10" >> $job_to_send
794	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
795	then
796	echo "qsub -J n scpjob.$kennung" >> $job_to_send
797	elif [[ $remote_host = t3es ]]
798	then
799	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
800	else
801	echo "qsub scpjob.$kennung" >> $job_to_send
802	fi
803	echo "rm scpjob.$kennung" >> $job_to_send
804	if [[ $remote_host = nech ]]
805	then
806	echo "cd -" >> $job_to_send
807	fi
808	else
809	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
810	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
811	fi
812	echo "set -x" >> $job_to_send
813	echo " ' exit" >> $job_to_send
814	echo "set -x" >> $job_to_send
815	fi
816
817
818
819	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
820	cat $file_to_send >> $job_to_send
821	if [[ $remote_host = ibm ]]
822	then
823	echo " " >> $job_to_send
824	echo "exit" >> $job_to_send
825	fi
826
827
828
829	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
830	if [[ -z $remote_user ]]
831	then
832	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
833	then
834	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
835	fi
836	fi
837
838
839
840	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
841	if [[ $no_submit = false ]]
842	then
843	if [[ $remote_host != $local_host ]]
844	then
845	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
846	if [[ $filetransfer_protocol = scp ]]
847	then
848	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
849	then
850	job_catalog_save=$job_catalog
851	job_catalog=job_queue
852	elif [[ $remote_host = nech ]]
853	then
854	job_catalog_save=$job_catalog
855	job_catalog=/hpf/b/${remote_user}/job_queue
856	fi
857	if [[ $local_host = decalpha ]]
858	then
859	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
860	# PASSWORT AB
861	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
862	elif [[ $remote_host = nech ]]
863	then
864	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
865	# TRANSFERIERT WERDEN
866	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
867	else
868	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
869	fi
870	if [[ $? = 1 ]]
871	then
872	locat=scp; exit
873	fi
874	if [[ $remote_host = ibms ]]
875	then
876	job_catalog=$job_catalog_save
877	fi
878	else
879	ftpcopy -s -o $remote_addres $job_to_send "$job_catalog" $job_on_remhost
880	if [[ $? = 1 ]]
881	then
882	locat=ftpcopy; exit
883	fi
884	fi
885	[[ $verify = true ]] && printf "\n >>> finished\n"
886	else
887	eval job_catalog=$job_catalog
888	cp $job_to_send ${job_catalog}/$job_on_remhost
889	fi
890
891
892
893	# NQS- BZW. LOADLEVELER-JOB STARTEN
894	if [[ $remote_host != $local_host ]]
895	then
896	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
897	if [[ $filetransfer_protocol = scp ]]
898	then
899	if [[ $remote_host != lctit ]]
900	then
901	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
902	else
903	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
904	# MIT SSH, DESHALB AUFRUF PER PIPE
905	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
906	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
907	echo "first try" > send_protocol
908	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
909	do
910	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
911	then
912	printf "\n trying to submit job to TIT cluster..."
913	else
914	printf "\n +++ failed ... trying to submit again ..."
915	fi
916	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
917	cat send_protocol
918	done
919	sleep 10
920	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
921	echo "$submcom $job_on_remhost"
922	rm send_protocol
923	fi
924	else
925	if [[ $remote_host = ibms ]]
926	then
927	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
928	echo "echo \"$remote_user\" " > telnet_script
929	echo "sleep 5" >> telnet_script
930	echo "echo \"$password\" " >> telnet_script
931	echo "sleep 10" >> telnet_script
932	echo "echo \"\" " >> telnet_script
933	echo "sleep 2" >> telnet_script
934	echo "echo \"cd $job_catalog; $submcom $job_on_remhost; rm -rf $job_on_remhost\" " >> telnet_script
935	echo "sleep 10" >> telnet_script
936	echo "echo \"exit\" " >> telnet_script
937
938	chmod u+x telnet_script
939	telnet_script \| telnet $remote_addres 2>&1 \| grep "llsubmit:"
940	rm telnet_script
941	elif [[ $remote_host = t3es ]]
942	then
943	password=`cat ~/.netrc \| grep $remote_addres \| awk '{print $6}'`
944	echo "printf \"$remote_user\\\n\" " > telnet_script
945	echo "sleep 10" >> telnet_script
946	echo "printf \"$password\\\n\" " >> telnet_script
947	echo "sleep 10" >> telnet_script
948	echo "printf \"\\\n\" " >> telnet_script
949	echo "sleep 5" >> telnet_script
950	echo "printf \"\\\n\" " >> telnet_script
951	echo "sleep 5" >> telnet_script
952	echo "printf \"cd $job_catalog; $submcom $job_on_remhost; rm -f $job_on_remhost\\\n\" " >> telnet_script
953	echo "sleep 10" >> telnet_script
954	echo "printf \"exit\\\n\" " >> telnet_script
955
956	chmod u+x telnet_script
957	cat telnet_script
958	telnet_script \| telnet $remote_addres \| grep $remote_user
959	rm telnet_script
960	else
961	remcom.x $remote_addres "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
962	fi
963	fi
964	[[ $verify = true ]] && printf " >>> o.k.\n"
965	else
966	cd $job_catalog
967	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
968	then
969	eval $submcom $job_on_remhost
970	elif [[ $local_host = lctit ]]
971	then
972	eval $submcom $job_on_remhost
973	echo "$submcom $job_on_remhost"
974	chmod u+x $job_on_remhost
975	sleep 5 # WEIL N1GE IM HINTERGRUND ARBEITET UND DATEI SONST EVTL.
976	# SCHON WIEDER GELOESCT IST
977	elif [[ $local_host = nech ]]
978	then
979	if [[ $queue = default ]]
980	then
981	eval $submcom $job_on_remhost
982	else
983	eval $submcom -q $queue $job_on_remhost
984	fi
985	else
986	qsub $job_on_remhost
987	fi
988	rm $job_on_remhost
989	cd - > /dev/null
990	fi
991	fi
992
993
994
995	# ABSCHLUSSARBEITEN
996
997	if [[ $no_submit = false ]]
998	then
999	rm $job_to_send
1000	fi
1001	[[ $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 |