Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 164

Last change on this file since 164 was 164, checked in by raasch, 17 years ago
optimization of transpositions for 2D decompositions, workaround for using -env option with mpiexec, adjustments for lcxt4
Property svn:keywords set to `Rev`
File size: 37.6 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	# 30/10/07 - Marcus- further adjustments for queues on lctit
92	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
93	# Science)
94
95
96	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
97	delete_dayfile=false
98	group_number=none
99	locat=normal
100	no_default_queue=none
101	no_submit=false
102	job_catalog="~/job_queue"
103	job_name=none
104	local_user=$LOGNAME
105	node_usage=shared
106	numprocs=1
107	punkte="..........................................................."
108	submcom=qsub
109	queue=default
110	remote_host=none
111	remote_user=""
112	verify=true
113
114	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
115	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
116	typeset -L20 spalte1
117	typeset -R40 spalte2
118	typeset -L60 spalte3
119
120
121
122	# FEHLERBEHANDLUNG
123	# BEI EXIT:
124	trap 'if [[ $locat != normal ]]
125	then
126	case $locat in
127	(option) printf "\n --> available optios can be displayed"
128	printf " by typing:"
129	printf "\n \"subjob ?\" \n";;
130	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
131	(*) printf "\n +++ unknown error"
132	printf "\n please inform S. Raasch!\n"
133	esac
134	[[ -f $job_to_send ]] && rm $job_to_send
135	printf "\n\n+++ SUBJOB killed \n\n"
136	fi' exit
137
138
139	# BEI TERMINAL-BREAK:
140	trap '[[ -f $job_to_send ]] && rm $job_to_send
141	printf "\n\n+++ SUBJOB killed \n\n"
142	exit
143	' 2
144
145
146
147
148	# LOKALEN HOSTNAMEN FESTSTELLEN
149	local_host=$(hostname)
150
151
152
153	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
154	# UEBERHAUPT ZULAESSIG IST
155	case $local_host in
156	(atmos) local_addres=172.20.25.35; local_host=lcide;;
157	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
158	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
159	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
160	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
161	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
162	(cs*) local_addres=136.172.44.131; local_host=nech;;
163	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
164	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
165	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
166	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
167	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
168	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
169	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
170	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
171	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
172	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
173	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
174	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
175	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
176	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
177	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
178	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
179	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
180	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
181	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
182	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
183	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
184	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
185	(*) printf "\n +++ \"$local_host\" unknown";
186	printf "\n please inform S. Raasch!";
187	locat=parameter; exit;;
188	esac
189
190
191
192	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
193	remote_host=$local_host
194
195
196
197
198	# PROZEDUROPTIONEN EINLESEN
199	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
200	do
201	case $option in
202	(c) job_catalog=$OPTARG;;
203	(d) delete_dayfile=true;;
204	(D) no_submit=true;;
205	(g) group_number=$OPTARG;;
206	(h) remote_host=$OPTARG;;
207	(m) memory=$OPTARG;;
208	(n) job_name=$OPTARG;;
209	(N) node_usage=$OPTARG;;
210	(O) threads_per_task=$OPTARG;;
211	(q) no_default_queue=$OPTARG;;
212	(t) cputime=$OPTARG;;
213	(T) tasks_per_node=$OPTARG;;
214	(u) remote_user=$OPTARG;;
215	(v) verify=false;;
216	(X) numprocs=$OPTARG;;
217	(\?) printf "\n +++ Option $OPTARG unknown \n";
218	locat=option; exit;;
219	esac
220	done
221
222
223	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
224	shift OPTIND-1; file_to_send=$1
225
226
227	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
228	if [ "$1" = "?" ]
229	then
230	(printf "\n *** subjob can be called as follows:\n"
231	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
232	printf "\n Description of available options:\n"
233	printf "\n Option Description Default-Value"
234	printf "\n -c job-input- and output-catalog ~/job_queue"
235	printf "\n -d no job-protocol will be created ---"
236	printf "\n -D only the job-file will be created ---"
237	printf "\n -h execution host, available hosts: $remote_host"
238	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
239	printf "\n lctit, nech, neck, unics"
240	printf "\n -m memory demand per process in MByte ---"
241	printf "\n -n jobname <jobdatei>"
242	printf "\n -O threads per task (for OpenMP usage) 1"
243	printf "\n -q job-queue to be used default"
244	printf "\n -t allowed cpu-time in seconds ---"
245	printf "\n -T tasks per node (on parallel hosts) ---"
246	printf "\n -u username on execution host from .netrc"
247	printf "\n -v no prompt for confirmation ---"
248	printf "\n -X # of processors (on parallel hosts) 1"
249	printf "\n "
250	printf "\n The only possible positional parameter is <jobfile>:"
251	printf "\n The complete NQS-job must be provided here."
252	printf "\n <jobfile>=? creates this outline\n\n") \| more
253	exit
254	fi
255
256
257
258	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
259	if [[ "$file_to_send" = "" ]]
260	then
261	printf "\n +++ job-file missing"
262	locat=parameter; exit
263	else
264	if [[ -f $file_to_send ]]
265	then
266	true
267	else
268	printf "\n +++ job-file: "
269	printf "\n $file_to_send"
270	printf "\n does not exist"
271	locat=parameter; exit
272	fi
273	fi
274
275
276
277	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
278	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
279	if [[ $job_name = none ]]
280	then
281	job_name=$file_to_send
282	fi
283	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
284	then
285	printf "\n +++ job-file name: "
286	printf "\n $job_name"
287	printf "\n must not contain \"/\"-characters"
288	locat=parameter; exit
289	fi
290
291
292
293
294	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
295	# ODER WENN HOST NICHT ANGEGEBEN WURDE
296	if [[ $remote_host = none ]]
297	then
298	printf "\n +++ host missing"
299	locat=option; exit
300	else
301	case $remote_host in
302	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
303	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
304	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
305	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
306	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
307	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
308	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
309	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
310	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
311	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
312	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
313	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
314	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
315	locat=parameter; exit;;
316	esac
317	fi
318
319
320	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
321	if [[ $no_default_queue != none ]]
322	then
323	error=false
324	ndq=$no_default_queue
325	case $remote_host in
326	(ibm) case $ndq in
327	(p690_express\|p690_standard\|p690_long) error=false;;
328	(*) error=true;;
329	esac;;
330	(ibmb) case $ndq in
331	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
332	(*) error=true;;
333	esac;;
334	(ibmh) case $ndq in
335	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
336	(*) error=true;;
337	esac;;
338	(ibms) case $ndq in
339	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
340	(*) error=true;;
341	esac;;
342	(ibmy) case $ndq in
343	(parallel) error=false;;
344	(*) error=true;;
345	esac;;
346	(lctit) case $ndq in
347	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
348	(*) error=true;;
349	esac;;
350	(t3eb) case $ndq in
351	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
352	(*) error=true;;
353	esac;;
354	(t3eh) case $ndq in
355	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
356	(*) error=true;;
357	esac;;
358	(t3ej2\|t3ej5) case $ndq in
359	(low\|normal\|high) error=false;;
360	(*) error=true;;
361	esac;;
362	(t3es) case $ndq in
363	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
364	(*) error=true;;
365	esac;;
366	(unics) case $ndq in
367	(unics\|ht) error=false;;
368	(*) error=true;;
369	esac;;
370	esac
371	if [[ $error = true ]]
372	then
373	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
374	locat=parameter; exit
375	else
376	queue=$no_default_queue
377	fi
378	fi
379
380
381
382	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
383	if [[ $node_usage != shared && $queue = cdev ]]
384	then
385	node_usage=shared
386	fi
387
388
389
390	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
391	# AUFTEILEN
392	done=false
393	while [[ $done = false ]]
394	do
395	if (( $cputime <= 0 ))
396	then
397	printf "\n +++ wrong cpu-time or cpu-time missing"
398	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
399	printf "\n >>> "
400	read cputime 1>/dev/null 2>&1
401	else
402	done=true
403	fi
404	done
405	if [[ $remote_host = nech ]]
406	then
407	if (( tasks_per_node != 0 ))
408	then
409	(( cputime = cputime * tasks_per_node ))
410	elif [[ $numprocs != 0 ]]
411	then
412	(( cputime = cputime * numprocs ))
413	fi
414	fi
415	(( stunden = cputime / 3600 ))
416	(( resttime = cputime - stunden * 3600 ))
417	(( minuten = resttime / 60 ))
418	(( sekunden = resttime - minuten * 60 ))
419	timestring=${stunden}:${minuten}:${sekunden}
420
421
422
423	# PRUEFEN DER KERNSPEICHERANFORDERUNG
424	done=false
425	while [[ $done = false ]]
426	do
427	if (( memory <= 0 ))
428	then
429	printf "\n +++ wrong memory demand or memory demand missing"
430	printf "\n >>> Please type memory in MByte per process as INTEGER:"
431	printf "\n >>> "
432	read memory 1>/dev/null 2>&1
433	else
434	done=true
435	fi
436	done
437
438	if [[ $remote_host = nech \|\| $remote_host = neck ]]
439	then
440	if (( tasks_per_node != 0 ))
441	then
442	(( Memory = memory * tasks_per_node / 1000 ))
443	elif [[ $numprocs != 0 ]]
444	then
445	(( Memory = memory * numprocs / 1000 ))
446	else
447	(( Memory = memory / 1000 ))
448	fi
449	fi
450
451
452	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
453	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
454	then
455	(( memory = memory * threads_per_task ))
456	fi
457
458
459	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
460	if (( tasks_per_node != 0 ))
461	then
462	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
463	fi
464
465
466
467	# HEADER-AUSGABE
468	if [[ $verify = true ]]
469	then
470	printf "\n\n"
471	printf "#--------------------------------------------------------------# \n"
472	spalte1=SUBJOB;spalte2=$(date)
473	printf "\| $spalte1$spalte2 \| \n"
474	printf "\| \| \n"
475	printf "\| values of parameters/options: \| \n"
476	spalte1=local_host$punkte; spalte2=$punkte$local_host
477	printf "\| $spalte1$spalte2 \| \n"
478	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
479	printf "\| $spalte1$spalte2 \| \n"
480	spalte1=queue$punkte; spalte2=$punkte$queue
481	printf "\| $spalte1$spalte2 \| \n"
482	spalte1=memory$punkte; spalte2="$punkte$memory mb"
483	printf "\| $spalte1$spalte2 \| \n"
484	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
485	printf "\| $spalte1$spalte2 \| \n"
486	spalte1=job_name$punkte; spalte2="$punkte$job_name"
487	printf "\| $spalte1$spalte2 \| \n"
488	printf "#--------------------------------------------------------------# \n\n"
489
490
491	# KONTROLLABFRAGE, OB ALLES O.K.
492	antwort="dummy"
493	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
494	do
495	read antwort?" >>> continue (y/n) ? "
496	done
497	if [[ $antwort = n \|\| $antwort = N ]]
498	then
499	locat=verify; exit
500	fi
501	printf "\n"
502	fi
503
504
505
506	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
507	kennung=$RANDOM
508	job_on_remhost=${job_name}_${kennung}_$local_host
509	job_to_send=job_to_send_$kennung
510	if [[ $delete_dayfile = false ]]
511	then
512	remote_dayfile=${local_host}_${job_name}_result_$kennung
513	local_dayfile=${remote_host}_${job_name}
514	else
515	remote_dayfile=/dev/null
516	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
517	then
518	notification='# @ notification = never'
519	fi
520	fi
521
522
523
524	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
525	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
526	then
527
528	if [[ $remote_host = ibmy ]]
529	then
530	consumable_memory=""
531	else
532	consumable_memory="ConsumableMemory($memory mb)"
533	fi
534	if [[ $queue = cdev ]]
535	then
536	data_limit="# @ data_limit = 1.76gb"
537	network_to_use="# @ network.mpi = sn_all,shared,ip"
538	else
539	if [[ $remote_host = ibms ]]
540	then
541	network_to_use="# @ network.mpi = csss,shared,us"
542	elif [[ $remote_host = ibmy ]]
543	then
544	network_to_use=""
545	else
546	network_to_use="# @ network.mpi = sn_all,shared,us"
547	data_limit="# @ data_limit = 1.76gb"
548	fi
549	fi
550
551	cat > $job_to_send << %%END%%
552	#!/bin/ksh
553
554	# @ job_type = parallel
555	# @ job_name = palm
556	# @ wall_clock_limit = ${timestring},$timestring
557	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
558	# @ output = $remote_dayfile
559	# @ error = $remote_dayfile
560	# @ image_size = 50
561	# @ class = $queue
562	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
563	$network_to_use
564	$data_limit
565	$notification
566
567	%%END%%
568
569	if (( nodes > 0 ))
570	then
571
572	cat >> $job_to_send << %%END%%
573	# @ node = $nodes
574	# @ tasks_per_node = $tasks_per_node
575	# @ node_usage = $node_usage
576	# @ queue
577
578	%%END%%
579
580	else
581
582	if [[ $remote_host != ibmy ]]
583	then
584
585	cat >> $job_to_send << %%END%%
586	# @ blocking = unlimited
587	# @ total_tasks = $numprocs
588	# @ node_usage = $node_usage
589	# @ queue
590
591	%%END%%
592
593	else
594
595	cat >> $job_to_send << %%END%%
596	# @ node = 1
597	# @ total_tasks = $numprocs
598	# @ queue
599
600	%%END%%
601
602	fi
603
604	fi
605
606	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
607	then
608
609	cat > $job_to_send << %%END%%
610	#!/bin/ksh
611
612	# @ job_type = serial
613	# @ node_usage = $node_usage
614	# @ job_name = palm
615	# @ wall_clock_limit = ${timestring},$timestring
616	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
617	# @ output = $remote_dayfile
618	# @ error = $remote_dayfile
619	# @ class = $queue
620	$notification
621
622	# @ queue
623
624	%%END%%
625
626	elif [[ $remote_host = lcfimm ]]
627	then
628
629	if [[ $numprocs != 0 ]]
630	then
631	cat > $job_to_send << %%END%%
632	#!/bin/ksh
633	#PBS -N $job_name
634	#PBS -A nersc
635	#PBS -l walltime=$timestring
636	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
637	#PBS -l pmem=${memory}mb
638	#PBS -m abe
639	#PBS -M igore@nersc.no
640	#PBS -o $remote_dayfile
641	#PBS -j oe
642	mpd &
643
644	%%END%%
645
646	elif [[ $remote_host = lcxt4 ]]
647	then
648
649	if [[ $numprocs != 0 ]]
650	then
651	cat > $job_to_send << %%END%%
652	#!/bin/ksh
653	#PBS -S /bin/ksh
654	#PBS -N $job_name
655	#PBS -A nersc
656	#PBS -l walltime=$timestring
657	#PBS -l mppwidth=${numprocs}
658	#PBS -l mppnppn=${tasks_per_node}
659	#PBS -m abe
660	#PBS -M igore@nersc.no
661	#PBS -o $remote_dayfile
662	#PBS -e $remote_dayfile
663
664	%%END%%
665
666	else
667	cat > $job_to_send << %%END%%
668	#!/bin/ksh
669	#PBS -N $job_name
670	#PBS -A nersc
671	#PBS -l walltime=$timestring
672	#PBS -l ncpus=1
673	#PBS -l pmem=${memory}mb
674	#PBS -m abe
675	#PBS -M igore@nersc.no
676	#PBS -o $remote_dayfile
677	#PBS -j oe
678
679	%%END%%
680
681	fi
682
683	elif [[ $remote_host = nech ]]
684	then
685
686	if (( nodes > 1 ))
687	then
688	cat > $job_to_send << %%END%%
689	#!/bin/ksh
690	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
691	#PBS -l ${qsubmem}=${Memory}gb
692	#PBS -b $nodes
693	#PBS -o $remote_dayfile
694	#PBS -N palm
695	#PBS -j o
696	#PBS -T mpisx
697
698	%%END%%
699
700	elif [[ $numprocs != 0 ]]
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 -o $remote_dayfile
707	#PBS -N palm
708	#PBS -j o
709
710	%%END%%
711
712	else
713	cat > $job_to_send << %%END%%
714	#!/bin/ksh
715	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
716	#PBS -o $remote_dayfile
717	#PBS -j o
718
719	%%END%%
720
721	fi
722
723	elif [[ $remote_host = neck ]]
724	then
725
726	if (( nodes > 1 ))
727	then
728	cat > $job_to_send << %%END%%
729	#!/bin/ksh
730	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
731	#PBS -l ${qsubmem}=${Memory}gb
732	#PBS -b $nodes
733	#PBS -o $remote_dayfile
734	#PBS -N palm
735	#PBS -j o
736	#PBS -T mpisx
737
738	%%END%%
739
740	elif [[ $numprocs != 0 ]]
741	then
742	cat > $job_to_send << %%END%%
743	#!/bin/ksh
744	#PBS -q ${queue}
745	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
746	#PBS -l ${qsubmem}=${Memory}gb
747	#PBS -o $remote_dayfile
748	#PBS -N palm
749	#PBS -j o
750	#PBS -T mpisx
751
752	%%END%%
753
754	else
755	cat > $job_to_send << %%END%%
756	#!/bin/ksh
757	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
758	#PBS -o $remote_dayfile
759	#PBS -j o
760
761	%%END%%
762
763	fi
764
765	elif [[ $remote_host = lctit ]]
766	then
767	cat > $job_to_send << %%END%%
768	#!/bin/bash
769	#$ -S /bin/bash
770	cd $job_catalog
771	export PATH=$PALM_BIN:\$PATH
772	echo \$PATH
773
774	%%END%%
775
776	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
777	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
778	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
779
780	else
781
782	cat > $job_to_send << %%END%%
783	# @\$-q ${queue}
784	# @\$-l${qsubtime} $timestring
785	# @\$-l${qsubmem} ${memory}mb
786	# @\$-o $remote_dayfile
787	# @\$-eo
788
789	%%END%%
790
791	fi
792
793
794	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
795	# BEI EXIT VERANLASSEN
796	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
797	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
798	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
799	then
800	echo "set +vx" >> $job_to_send
801	echo "trap '" >> $job_to_send
802	echo "set +vx" >> $job_to_send
803	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
804	then
805	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
806	then
807	return_queue=c1
808	elif [[ $remote_host = ibms ]]
809	then
810	return_queue=p_normal
811	elif [[ $remote_host = ibmy ]]
812	then
813	return_queue=serial
814	elif [[ $remote_host = lctit ]]
815	then
816	return_queue=sla3
817	elif [[ $remote_host = neck ]]
818	then
819	return_queue=S
820	else
821	return_queue=unknown
822	fi
823
824	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
825	then
826
827	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
828	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
829	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
830	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
831	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
832	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
833	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
834	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
835	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
836	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
837
838	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
839	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
840
841	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
842	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
843	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
844
845	elif [[ $remote_host = nech ]]
846	then
847	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
848	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
849	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
850	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
851	echo "#PBS -j o" >> $job_to_send
852	echo " " >> $job_to_send
853	echo "set -x" >> $job_to_send
854	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
855	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
856	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
857	echo "%%END%%" >> $job_to_send
858
859	elif [[ $remote_host = neck ]]
860	then
861	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
862	echo "#PBS -q $return_queue" >> $job_to_send
863	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
864	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
865	echo "#PBS -j o" >> $job_to_send
866	echo " " >> $job_to_send
867	echo "set -x" >> $job_to_send
868	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
869	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
870	echo "%%END%%" >> $job_to_send
871
872	elif [[ $remote_host = lctit ]]
873	then
874	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
875	echo "set -x" >> $job_to_send
876	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
877	echo "cd $job_catalog" >> $job_to_send
878	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
879	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
880	echo "%%END%%" >> $job_to_send
881
882	else
883
884	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
885	echo "# @\\\$-q $return_queue" >> $job_to_send
886	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
887	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
888	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
889	then
890	echo "# @\$-l mpp_p=0" >> $job_to_send
891	fi
892	echo '# @\$-lF 10mb' >> $job_to_send
893	# echo '# @\$-o /dev/null' >> $job_to_send
894	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
895	echo '# @\\\$-eo' >> $job_to_send
896	echo " " >> $job_to_send
897	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
898	then
899	echo "set +vx" >> $job_to_send
900	echo ". .profile" >> $job_to_send
901	fi
902	echo "set -x" >> $job_to_send
903	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
904	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
905	echo "%%END%%" >> $job_to_send
906	fi
907
908	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
909	then
910	echo "llsubmit scpjob.$kennung" >> $job_to_send
911	elif [[ $remote_host = lctit ]]
912	then
913	echo "chmod u+x scpjob.$kennung" >> $job_to_send
914	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
915	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
916	echo "sleep 10" >> $job_to_send
917	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
918	then
919	echo "qsub -J n scpjob.$kennung" >> $job_to_send
920	elif [[ $remote_host = t3es ]]
921	then
922	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
923	else
924	echo "qsub scpjob.$kennung" >> $job_to_send
925	fi
926	echo "rm scpjob.$kennung" >> $job_to_send
927	if [[ $remote_host = nech ]]
928	then
929	echo "cd -" >> $job_to_send
930	fi
931	else
932	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
933	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
934	fi
935	echo "set -x" >> $job_to_send
936	echo " ' exit" >> $job_to_send
937	echo "set -x" >> $job_to_send
938	fi
939
940
941
942	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
943	cat $file_to_send >> $job_to_send
944
945	if [[ $remote_host = ibm ]]
946	then
947	echo " " >> $job_to_send
948	echo "exit" >> $job_to_send
949	fi
950	if [[ $remote_host = lctit ]]
951	then
952	echo " " >> $job_to_send
953	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
954	fi
955
956
957
958	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
959	if [[ -z $remote_user ]]
960	then
961	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
962	then
963	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
964	fi
965	fi
966
967
968
969	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
970	if [[ $no_submit = false ]]
971	then
972	if [[ $remote_host != $local_host ]]
973	then
974	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
975	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
976	then
977	job_catalog_save=$job_catalog
978	job_catalog=job_queue
979	elif [[ $remote_host = nech ]]
980	then
981	job_catalog_save=$job_catalog
982	job_catalog=/hpf/b/${remote_user}/job_queue
983	fi
984	if [[ $local_host = decalpha ]]
985	then
986	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
987	# PASSWORT AB
988	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
989	elif [[ $remote_host = nech ]]
990	then
991	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
992	# TRANSFERIERT WERDEN
993	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
994	else
995	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
996	fi
997	if [[ $? = 1 ]]
998	then
999	locat=scp; exit
1000	fi
1001	if [[ $remote_host = ibms ]]
1002	then
1003	job_catalog=$job_catalog_save
1004	fi
1005	[[ $verify = true ]] && printf "\n >>> finished\n"
1006	else
1007	eval job_catalog=$job_catalog
1008	cp $job_to_send ${job_catalog}/$job_on_remhost
1009	fi
1010
1011
1012
1013	# NQS- BZW. LOADLEVELER-JOB STARTEN
1014	if [[ $remote_host != $local_host ]]
1015	then
1016	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1017	if [[ $remote_host != lctit ]]
1018	then
1019	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1020	else
1021	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
1022	# MIT SSH, DESHALB AUFRUF PER PIPE
1023	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1024	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1025	echo "first try" > send_protocol
1026	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1027	do
1028	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1029	then
1030	printf "\n trying to submit job to TIT cluster..."
1031	else
1032	printf "\n +++ failed ... trying to submit again ..."
1033	fi
1034	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1035	cat send_protocol
1036	done
1037	sleep 10
1038	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1039	echo "$submcom $job_on_remhost"
1040	rm send_protocol
1041	fi
1042
1043	[[ $verify = true ]] && printf " >>> o.k.\n"
1044	else
1045	cd $job_catalog
1046	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1047	then
1048	eval $submcom $job_on_remhost
1049	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1050	then
1051	eval $submcom $job_on_remhost
1052	echo "$submcom $job_on_remhost"
1053	chmod u+x $job_on_remhost
1054	elif [[ $local_host = nech ]]
1055	then
1056	if [[ $queue = default ]]
1057	then
1058	eval $submcom $job_on_remhost
1059	else
1060	eval $submcom -q $queue $job_on_remhost
1061	fi
1062	else
1063	qsub $job_on_remhost
1064	fi
1065	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1066	[[ $local_host != lctit ]] && rm $job_on_remhost
1067	cd - > /dev/null
1068	fi
1069	fi
1070
1071
1072
1073	# ABSCHLUSSARBEITEN
1074
1075	if [[ $no_submit = false ]]
1076	then
1077	rm $job_to_send
1078	fi
1079	[[ $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 |