Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 125

Last change on this file since 125 was 125, checked in by raasch, 17 years ago
small bugfix in subjob
Property svn:keywords set to `Rev`
File size: 36.9 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 54 2007-03-08 00:00:02Z raasch $
4
5	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
6	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
7	# Host zurueckgeschickt werden sollen
8
9
10	# letzte Aenderung:
11	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
12	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
13	# 06/02/98 - Siggi - berte validiert
14	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
15	# moeglich
16	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
17	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
18	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
19	# allgemeine Anpassungen fuer ibm-Rechner
20	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
21	# Ruecktransfer des Jobprotokolls fuer diese
22	# Maschine abgeschaltet
23	# 31/10/02 - Siggi - berni validiert
24	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
25	# 08/11/02 - Siggi - quanero validiert
26	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
27	# 23/01/03 - Siggi - hostname nobel changed to nobela
28	# 06/02/03 - Siggi - gregale validated
29	# 12/02/03 - Siggi - orkan and maestro validated
30	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
31	# 12/03/03 - Siggi - nec at DKRZ validated
32	# 13/03/03 - Siggi - new nqs resource variable Memory
33	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
34	# node
35	# 11/04/03 - Siggi - network on ibms has attribute "shared"
36	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
37	# cxxl added to ibmh
38	# 29/08/03 - Siggi - changes in job queues and communication system on
39	# ibms
40	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
41	# 30/10/03 - Siggi - nech is not supported any more
42	# 10/11/03 - Siggi - nech2 renamed to nech
43	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
44	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
45	# 31/03/04 - Siggi - new option -N for node usage
46	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
47	# due to error in ssh installation (otherwise a prompt
48	# for the password appears)
49	# 23/07/04 - Siggi - changes due to the new berni configuration
50	# (federation switch)
51	# 01/09/04 - Gerald new job-classes on hanni
52	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
53	# 23/11/04 - Siggi - new job class cdata on hanni and berni
54	# 03/12/04 - Siggi - notification on ibm switched of in case of
55	# delete_dayfile = true, node usage in cdev set to
56	# shared
57	# 16/02/05 - Gerald hababai validated
58	# 29/03/05 - Micha - new job class channi on hanni
59	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
60	# on ibms
61	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
62	# p_normal
63	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
64	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
65	# "sn_all", new job class cexp
66	# 08/09/05 - Siggi - IP-address of gfdl3 changed
67	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
68	# hurricane (with -X0)
69	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
70	# added
71	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
72	# 09/02/06 - Siggi - ibmy admitted
73	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
74	# 13/04/06 - Siggi - ostria admitted
75	# 18/04/06 - Siggi - new option -O for OpenMP usage
76	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
77	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
78	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
79	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
80	# 28/11/06 - Siggi - levanto admitted
81	# 13/02/07 - Siggi - hpmuk releated code removed
82	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
83	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
84	# 27/07/07 - Siggi - autan admitted
85	# 03/08/07 - Marcus- lcfimm admitted
86	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
87	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
88	# suggestions
89	# 19/10/07 - Marcus- add new optional argument -g group_number
90	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
91
92
93	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
94	delete_dayfile=false
95	group_number=none
96	locat=normal
97	no_default_queue=none
98	no_submit=false
99	job_catalog="~/job_queue"
100	job_name=none
101	local_user=$LOGNAME
102	node_usage=shared
103	numprocs=1
104	punkte="..........................................................."
105	submcom=qsub
106	queue=default
107	remote_host=none
108	remote_user=""
109	verify=true
110
111	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
112	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
113	typeset -L20 spalte1
114	typeset -R40 spalte2
115	typeset -L60 spalte3
116
117
118
119	# FEHLERBEHANDLUNG
120	# BEI EXIT:
121	trap 'if [[ $locat != normal ]]
122	then
123	case $locat in
124	(option) printf "\n --> available optios can be displayed"
125	printf " by typing:"
126	printf "\n \"subjob ?\" \n";;
127	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
128	(*) printf "\n +++ unknown error"
129	printf "\n please inform S. Raasch!\n"
130	esac
131	[[ -f $job_to_send ]] && rm $job_to_send
132	printf "\n\n+++ SUBJOB killed \n\n"
133	fi' exit
134
135
136	# BEI TERMINAL-BREAK:
137	trap '[[ -f $job_to_send ]] && rm $job_to_send
138	printf "\n\n+++ SUBJOB killed \n\n"
139	exit
140	' 2
141
142
143
144
145	# LOKALEN HOSTNAMEN FESTSTELLEN
146	local_host=$(hostname)
147
148
149
150	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
151	# UEBERHAUPT ZULAESSIG IST
152	case $local_host in
153	(atmos) local_addres=172.20.25.35; local_host=lcide;;
154	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
155	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
156	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
157	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
158	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
159	(cs*) local_addres=136.172.44.131; local_host=nech;;
160	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
161	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
162	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
163	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
164	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
165	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
166	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
167	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
168	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
169	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
170	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
171	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
172	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
173	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
174	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
175	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
176	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
177	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
178	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
179	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
180	(*) printf "\n +++ \"$local_host\" unknown";
181	printf "\n please inform S. Raasch!";
182	locat=parameter; exit;;
183	esac
184
185
186
187	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
188	remote_host=$local_host
189
190
191
192
193	# PROZEDUROPTIONEN EINLESEN
194	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
195	do
196	case $option in
197	(c) job_catalog=$OPTARG;;
198	(d) delete_dayfile=true;;
199	(D) no_submit=true;;
200	(g) group_number=$OPTARG;;
201	(h) remote_host=$OPTARG;;
202	(m) memory=$OPTARG;;
203	(n) job_name=$OPTARG;;
204	(N) node_usage=$OPTARG;;
205	(O) threads_per_task=$OPTARG;;
206	(q) no_default_queue=$OPTARG;;
207	(t) cputime=$OPTARG;;
208	(T) tasks_per_node=$OPTARG;;
209	(u) remote_user=$OPTARG;;
210	(v) verify=false;;
211	(X) numprocs=$OPTARG;;
212	(\?) printf "\n +++ Option $OPTARG unknown \n";
213	locat=option; exit;;
214	esac
215	done
216
217
218	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
219	shift OPTIND-1; file_to_send=$1
220
221
222	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
223	if [ "$1" = "?" ]
224	then
225	(printf "\n *** subjob can be called as follows:\n"
226	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
227	printf "\n Description of available options:\n"
228	printf "\n Option Description Default-Value"
229	printf "\n -c job-input- and output-catalog ~/job_queue"
230	printf "\n -d no job-protocol will be created ---"
231	printf "\n -D only the job-file will be created ---"
232	printf "\n -h execution host, available hosts: $remote_host"
233	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
234	printf "\n lctit, nech, neck, unics"
235	printf "\n -m memory demand per process in MByte ---"
236	printf "\n -n jobname <jobdatei>"
237	printf "\n -O threads per task (for OpenMP usage) 1"
238	printf "\n -q job-queue to be used default"
239	printf "\n -t allowed cpu-time in seconds ---"
240	printf "\n -T tasks per node (on parallel hosts) ---"
241	printf "\n -u username on execution host from .netrc"
242	printf "\n -v no prompt for confirmation ---"
243	printf "\n -X # of processors (on parallel hosts) 1"
244	printf "\n "
245	printf "\n The only possible positional parameter is <jobfile>:"
246	printf "\n The complete NQS-job must be provided here."
247	printf "\n <jobfile>=? creates this outline\n\n") \| more
248	exit
249	fi
250
251
252
253	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
254	if [[ "$file_to_send" = "" ]]
255	then
256	printf "\n +++ job-file missing"
257	locat=parameter; exit
258	else
259	if [[ -f $file_to_send ]]
260	then
261	true
262	else
263	printf "\n +++ job-file: "
264	printf "\n $file_to_send"
265	printf "\n does not exist"
266	locat=parameter; exit
267	fi
268	fi
269
270
271
272	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
273	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
274	if [[ $job_name = none ]]
275	then
276	job_name=$file_to_send
277	fi
278	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
279	then
280	printf "\n +++ job-file name: "
281	printf "\n $job_name"
282	printf "\n must not contain \"/\"-characters"
283	locat=parameter; exit
284	fi
285
286
287
288
289	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
290	# ODER WENN HOST NICHT ANGEGEBEN WURDE
291	if [[ $remote_host = none ]]
292	then
293	printf "\n +++ host missing"
294	locat=option; exit
295	else
296	case $remote_host in
297	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
298	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
299	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
300	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
301	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
302	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
303	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
304	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
305	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
306	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
307	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
308	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
309	locat=parameter; exit;;
310	esac
311	fi
312
313
314	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
315	if [[ $no_default_queue != none ]]
316	then
317	error=false
318	ndq=$no_default_queue
319	case $remote_host in
320	(ibm) case $ndq in
321	(p690_express\|p690_standard\|p690_long) error=false;;
322	(*) error=true;;
323	esac;;
324	(ibmb) case $ndq in
325	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
326	(*) error=true;;
327	esac;;
328	(ibmh) case $ndq in
329	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
330	(*) error=true;;
331	esac;;
332	(ibms) case $ndq in
333	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
334	(*) error=true;;
335	esac;;
336	(ibmy) case $ndq in
337	(parallel) error=false;;
338	(*) error=true;;
339	esac;;
340	(lctit) case $ndq in
341	(default\|high\|bes1\|bes2\|sla1\|sla2) error=false;;
342	(*) error=true;;
343	esac;;
344	(t3eb) case $ndq in
345	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
346	(*) error=true;;
347	esac;;
348	(t3eh) case $ndq in
349	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
350	(*) error=true;;
351	esac;;
352	(t3ej2\|t3ej5) case $ndq in
353	(low\|normal\|high) error=false;;
354	(*) error=true;;
355	esac;;
356	(t3es) case $ndq in
357	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
358	(*) error=true;;
359	esac;;
360	(unics) case $ndq in
361	(unics\|ht) error=false;;
362	(*) error=true;;
363	esac;;
364	esac
365	if [[ $error = true ]]
366	then
367	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
368	locat=parameter; exit
369	else
370	queue=$no_default_queue
371	fi
372	fi
373
374
375
376	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
377	if [[ $node_usage != shared && $queue = cdev ]]
378	then
379	node_usage=shared
380	fi
381
382
383
384	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
385	# AUFTEILEN
386	done=false
387	while [[ $done = false ]]
388	do
389	if (( $cputime <= 0 ))
390	then
391	printf "\n +++ wrong cpu-time or cpu-time missing"
392	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
393	printf "\n >>> "
394	read cputime 1>/dev/null 2>&1
395	else
396	done=true
397	fi
398	done
399	if [[ $remote_host = nech ]]
400	then
401	if (( tasks_per_node != 0 ))
402	then
403	(( cputime = cputime * tasks_per_node ))
404	elif [[ $numprocs != 0 ]]
405	then
406	(( cputime = cputime * numprocs ))
407	fi
408	fi
409	(( stunden = cputime / 3600 ))
410	(( resttime = cputime - stunden * 3600 ))
411	(( minuten = resttime / 60 ))
412	(( sekunden = resttime - minuten * 60 ))
413	timestring=${stunden}:${minuten}:${sekunden}
414
415
416
417	# PRUEFEN DER KERNSPEICHERANFORDERUNG
418	done=false
419	while [[ $done = false ]]
420	do
421	if (( memory <= 0 ))
422	then
423	printf "\n +++ wrong memory demand or memory demand missing"
424	printf "\n >>> Please type memory in MByte per process as INTEGER:"
425	printf "\n >>> "
426	read memory 1>/dev/null 2>&1
427	else
428	done=true
429	fi
430	done
431
432	if [[ $remote_host = nech \|\| $remote_host = neck ]]
433	then
434	if (( tasks_per_node != 0 ))
435	then
436	(( Memory = memory * tasks_per_node / 1000 ))
437	elif [[ $numprocs != 0 ]]
438	then
439	(( Memory = memory * numprocs / 1000 ))
440	else
441	(( Memory = memory / 1000 ))
442	fi
443	fi
444
445
446	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
447	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
448	then
449	(( memory = memory * threads_per_task ))
450	fi
451
452
453	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
454	if (( tasks_per_node != 0 ))
455	then
456	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
457	fi
458
459
460
461	# HEADER-AUSGABE
462	if [[ $verify = true ]]
463	then
464	printf "\n\n"
465	printf "#--------------------------------------------------------------# \n"
466	spalte1=SUBJOB;spalte2=$(date)
467	printf "\| $spalte1$spalte2 \| \n"
468	printf "\| \| \n"
469	printf "\| values of parameters/options: \| \n"
470	spalte1=local_host$punkte; spalte2=$punkte$local_host
471	printf "\| $spalte1$spalte2 \| \n"
472	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
473	printf "\| $spalte1$spalte2 \| \n"
474	spalte1=queue$punkte; spalte2=$punkte$queue
475	printf "\| $spalte1$spalte2 \| \n"
476	spalte1=memory$punkte; spalte2="$punkte$memory mb"
477	printf "\| $spalte1$spalte2 \| \n"
478	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
479	printf "\| $spalte1$spalte2 \| \n"
480	spalte1=job_name$punkte; spalte2="$punkte$job_name"
481	printf "\| $spalte1$spalte2 \| \n"
482	printf "#--------------------------------------------------------------# \n\n"
483
484
485	# KONTROLLABFRAGE, OB ALLES O.K.
486	antwort="dummy"
487	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
488	do
489	read antwort?" >>> continue (y/n) ? "
490	done
491	if [[ $antwort = n \|\| $antwort = N ]]
492	then
493	locat=verify; exit
494	fi
495	printf "\n"
496	fi
497
498
499
500	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
501	kennung=$RANDOM
502	job_on_remhost=${job_name}_${kennung}_$local_host
503	job_to_send=job_to_send_$kennung
504	if [[ $delete_dayfile = false ]]
505	then
506	remote_dayfile=${local_host}_${job_name}_result_$kennung
507	local_dayfile=${remote_host}_${job_name}
508	else
509	remote_dayfile=/dev/null
510	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
511	then
512	notification='# @ notification = never'
513	fi
514	fi
515
516
517
518	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
519	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
520	then
521
522	if [[ $remote_host = ibmy ]]
523	then
524	consumable_memory=""
525	else
526	consumable_memory="ConsumableMemory($memory mb)"
527	fi
528	if [[ $queue = cdev ]]
529	then
530	data_limit="# @ data_limit = 1.76gb"
531	network_to_use="# @ network.mpi = sn_all,shared,ip"
532	else
533	if [[ $remote_host = ibms ]]
534	then
535	network_to_use="# @ network.mpi = csss,shared,us"
536	elif [[ $remote_host = ibmy ]]
537	then
538	network_to_use=""
539	else
540	network_to_use="# @ network.mpi = sn_all,shared,us"
541	data_limit="# @ data_limit = 1.76gb"
542	fi
543	fi
544
545	cat > $job_to_send << %%END%%
546	#!/bin/ksh
547
548	# @ job_type = parallel
549	# @ job_name = palm
550	# @ wall_clock_limit = ${timestring},$timestring
551	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
552	# @ output = $remote_dayfile
553	# @ error = $remote_dayfile
554	# @ image_size = 50
555	# @ class = $queue
556	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
557	$network_to_use
558	$data_limit
559	$notification
560
561	%%END%%
562
563	if (( nodes > 0 ))
564	then
565
566	cat >> $job_to_send << %%END%%
567	# @ node = $nodes
568	# @ tasks_per_node = $tasks_per_node
569	# @ node_usage = $node_usage
570	# @ queue
571
572	%%END%%
573
574	else
575
576	if [[ $remote_host != ibmy ]]
577	then
578
579	cat >> $job_to_send << %%END%%
580	# @ blocking = unlimited
581	# @ total_tasks = $numprocs
582	# @ node_usage = $node_usage
583	# @ queue
584
585	%%END%%
586
587	else
588
589	cat >> $job_to_send << %%END%%
590	# @ node = 1
591	# @ total_tasks = $numprocs
592	# @ queue
593
594	%%END%%
595
596	fi
597
598	fi
599
600	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
601	then
602
603	cat > $job_to_send << %%END%%
604	#!/bin/ksh
605
606	# @ job_type = serial
607	# @ node_usage = $node_usage
608	# @ job_name = palm
609	# @ wall_clock_limit = ${timestring},$timestring
610	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
611	# @ output = $remote_dayfile
612	# @ error = $remote_dayfile
613	# @ class = $queue
614	$notification
615
616	# @ queue
617
618	%%END%%
619
620	elif [[ $remote_host = lcfimm ]]
621	then
622
623	if [[ $numprocs != 0 ]]
624	then
625	cat > $job_to_send << %%END%%
626	#!/bin/ksh
627	#PBS -N $job_name
628	#PBS -A nersc
629	#PBS -l walltime=$timestring
630	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
631	#PBS -l pmem=${memory}mb
632	#PBS -m abe
633	#PBS -M igore@nersc.no
634	#PBS -o $remote_dayfile
635	#PBS -j oe
636	mpd &
637
638	%%END%%
639
640	else
641	cat > $job_to_send << %%END%%
642	#!/bin/ksh
643	#PBS -N $job_name
644	#PBS -A nersc
645	#PBS -l walltime=$timestring
646	#PBS -l ncpus=1
647	#PBS -l pmem=${memory}mb
648	#PBS -m abe
649	#PBS -M igore@nersc.no
650	#PBS -o $remote_dayfile
651	#PBS -j oe
652
653	%%END%%
654
655	fi
656
657	elif [[ $remote_host = nech ]]
658	then
659
660	if (( nodes > 1 ))
661	then
662	cat > $job_to_send << %%END%%
663	#!/bin/ksh
664	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
665	#PBS -l ${qsubmem}=${Memory}gb
666	#PBS -b $nodes
667	#PBS -o $remote_dayfile
668	#PBS -N palm
669	#PBS -j o
670	#PBS -T mpisx
671
672	%%END%%
673
674	elif [[ $numprocs != 0 ]]
675	then
676	cat > $job_to_send << %%END%%
677	#!/bin/ksh
678	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
679	#PBS -l ${qsubmem}=${Memory}gb
680	#PBS -o $remote_dayfile
681	#PBS -N palm
682	#PBS -j o
683
684	%%END%%
685
686	else
687	cat > $job_to_send << %%END%%
688	#!/bin/ksh
689	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
690	#PBS -o $remote_dayfile
691	#PBS -j o
692
693	%%END%%
694
695	fi
696
697	elif [[ $remote_host = neck ]]
698	then
699
700	if (( nodes > 1 ))
701	then
702	cat > $job_to_send << %%END%%
703	#!/bin/ksh
704	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
705	#PBS -l ${qsubmem}=${Memory}gb
706	#PBS -b $nodes
707	#PBS -o $remote_dayfile
708	#PBS -N palm
709	#PBS -j o
710	#PBS -T mpisx
711
712	%%END%%
713
714	elif [[ $numprocs != 0 ]]
715	then
716	cat > $job_to_send << %%END%%
717	#!/bin/ksh
718	#PBS -q ${queue}
719	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
720	#PBS -l ${qsubmem}=${Memory}gb
721	#PBS -o $remote_dayfile
722	#PBS -N palm
723	#PBS -j o
724	#PBS -T mpisx
725
726	%%END%%
727
728	else
729	cat > $job_to_send << %%END%%
730	#!/bin/ksh
731	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
732	#PBS -o $remote_dayfile
733	#PBS -j o
734
735	%%END%%
736
737	fi
738
739	elif [[ $remote_host = lctit ]]
740	then
741	cat > $job_to_send << %%END%%
742	#!/bin/bash
743	#$ -S /bin/bash
744	cd $job_catalog
745	export PATH=$PALM_BIN:\$PATH
746	echo \$PATH
747
748	%%END%%
749
750	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
751	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
752	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
753
754	else
755
756	cat > $job_to_send << %%END%%
757	# @\$-q ${queue}
758	# @\$-l${qsubtime} $timestring
759	# @\$-l${qsubmem} ${memory}mb
760	# @\$-o $remote_dayfile
761	# @\$-eo
762
763	%%END%%
764
765	fi
766
767
768	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
769	# BEI EXIT VERANLASSEN
770	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
771	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
772	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
773	then
774	echo "set +vx" >> $job_to_send
775	echo "trap '" >> $job_to_send
776	echo "set +vx" >> $job_to_send
777	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
778	then
779	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
780	then
781	return_queue=c1
782	elif [[ $remote_host = ibms ]]
783	then
784	return_queue=p_normal
785	elif [[ $remote_host = ibmy ]]
786	then
787	return_queue=serial
788	elif [[ $remote_host = lctit ]]
789	then
790	return_queue=sla3
791	elif [[ $remote_host = neck ]]
792	then
793	return_queue=S
794	else
795	return_queue=unknown
796	fi
797
798	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
799	then
800
801	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
802	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
803	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
804	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
805	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
806	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
807	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
808	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
809	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
810	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
811
812	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
813	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
814
815	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
816	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
817	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
818
819	elif [[ $remote_host = nech ]]
820	then
821	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
822	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
823	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
824	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
825	echo "#PBS -j o" >> $job_to_send
826	echo " " >> $job_to_send
827	echo "set -x" >> $job_to_send
828	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
829	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
830	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
831	echo "%%END%%" >> $job_to_send
832
833	elif [[ $remote_host = neck ]]
834	then
835	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
836	echo "#PBS -q $return_queue" >> $job_to_send
837	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
838	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
839	echo "#PBS -j o" >> $job_to_send
840	echo " " >> $job_to_send
841	echo "set -x" >> $job_to_send
842	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
843	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
844	echo "%%END%%" >> $job_to_send
845
846	elif [[ $remote_host = lctit ]]
847	then
848	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
849	echo "set -x" >> $job_to_send
850	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
851	echo "cd $job_catalog" >> $job_to_send
852	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
853	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
854	echo "%%END%%" >> $job_to_send
855
856	else
857
858	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
859	echo "# @\\\$-q $return_queue" >> $job_to_send
860	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
861	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
862	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
863	then
864	echo "# @\$-l mpp_p=0" >> $job_to_send
865	fi
866	echo '# @\$-lF 10mb' >> $job_to_send
867	# echo '# @\$-o /dev/null' >> $job_to_send
868	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
869	echo '# @\\\$-eo' >> $job_to_send
870	echo " " >> $job_to_send
871	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
872	then
873	echo "set +vx" >> $job_to_send
874	echo ". .profile" >> $job_to_send
875	fi
876	echo "set -x" >> $job_to_send
877	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
878	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
879	echo "%%END%%" >> $job_to_send
880	fi
881
882	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
883	then
884	echo "llsubmit scpjob.$kennung" >> $job_to_send
885	elif [[ $remote_host = lctit ]]
886	then
887	echo "chmod u+x scpjob.$kennung" >> $job_to_send
888	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
889	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
890	echo "sleep 10" >> $job_to_send
891	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
892	then
893	echo "qsub -J n scpjob.$kennung" >> $job_to_send
894	elif [[ $remote_host = t3es ]]
895	then
896	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
897	else
898	echo "qsub scpjob.$kennung" >> $job_to_send
899	fi
900	echo "rm scpjob.$kennung" >> $job_to_send
901	if [[ $remote_host = nech ]]
902	then
903	echo "cd -" >> $job_to_send
904	fi
905	else
906	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
907	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
908	fi
909	echo "set -x" >> $job_to_send
910	echo " ' exit" >> $job_to_send
911	echo "set -x" >> $job_to_send
912	fi
913
914
915
916	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
917	cat $file_to_send >> $job_to_send
918
919	if [[ $remote_host = ibm ]]
920	then
921	echo " " >> $job_to_send
922	echo "exit" >> $job_to_send
923	fi
924	if [[ $remote_host = lctit ]]
925	then
926	echo " " >> $job_to_send
927	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
928	fi
929
930
931
932	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
933	if [[ -z $remote_user ]]
934	then
935	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
936	then
937	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
938	fi
939	fi
940
941
942
943	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
944	if [[ $no_submit = false ]]
945	then
946	if [[ $remote_host != $local_host ]]
947	then
948	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
949	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
950	then
951	job_catalog_save=$job_catalog
952	job_catalog=job_queue
953	elif [[ $remote_host = nech ]]
954	then
955	job_catalog_save=$job_catalog
956	job_catalog=/hpf/b/${remote_user}/job_queue
957	fi
958	if [[ $local_host = decalpha ]]
959	then
960	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
961	# PASSWORT AB
962	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
963	elif [[ $remote_host = nech ]]
964	then
965	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
966	# TRANSFERIERT WERDEN
967	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
968	else
969	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
970	fi
971	if [[ $? = 1 ]]
972	then
973	locat=scp; exit
974	fi
975	if [[ $remote_host = ibms ]]
976	then
977	job_catalog=$job_catalog_save
978	fi
979	[[ $verify = true ]] && printf "\n >>> finished\n"
980	else
981	eval job_catalog=$job_catalog
982	cp $job_to_send ${job_catalog}/$job_on_remhost
983	fi
984
985
986
987	# NQS- BZW. LOADLEVELER-JOB STARTEN
988	if [[ $remote_host != $local_host ]]
989	then
990	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
991	if [[ $remote_host != lctit ]]
992	then
993	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
994	else
995	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
996	# MIT SSH, DESHALB AUFRUF PER PIPE
997	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
998	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
999	echo "first try" > send_protocol
1000	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1001	do
1002	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1003	then
1004	printf "\n trying to submit job to TIT cluster..."
1005	else
1006	printf "\n +++ failed ... trying to submit again ..."
1007	fi
1008	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1009	cat send_protocol
1010	done
1011	sleep 10
1012	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1013	echo "$submcom $job_on_remhost"
1014	rm send_protocol
1015	fi
1016
1017	[[ $verify = true ]] && printf " >>> o.k.\n"
1018	else
1019	cd $job_catalog
1020	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1021	then
1022	eval $submcom $job_on_remhost
1023	elif [[ $local_host = lcfimm ]]
1024	then
1025	eval $submcom $job_on_remhost
1026	echo "$submcom $job_on_remhost"
1027	chmod u+x $job_on_remhost
1028	elif [[ $local_host = lctit ]]
1029	then
1030	eval $submcom $job_on_remhost
1031	echo "$submcom $job_on_remhost"
1032	chmod u+x $job_on_remhost
1033	elif [[ $local_host = nech ]]
1034	then
1035	if [[ $queue = default ]]
1036	then
1037	eval $submcom $job_on_remhost
1038	else
1039	eval $submcom -q $queue $job_on_remhost
1040	fi
1041	else
1042	qsub $job_on_remhost
1043	fi
1044	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1045	[[ $local_host != lctit ]] && rm $job_on_remhost
1046	cd - > /dev/null
1047	fi
1048	fi
1049
1050
1051
1052	# ABSCHLUSSARBEITEN
1053
1054	if [[ $no_submit = false ]]
1055	then
1056	rm $job_to_send
1057	fi
1058	[[ $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 |