Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 179

Last change on this file since 179 was 179, checked in by raasch, 16 years ago
preliminary adjustments for HLRN-2 system
Property svn:keywords set to `Id`
File size: 38.8 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 179 2008-07-16 11:47:48Z 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	# 14/07/08 - Siggi - adjustments for lcsgih
95
96
97	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
98	delete_dayfile=false
99	group_number=none
100	locat=normal
101	no_default_queue=none
102	no_submit=false
103	job_catalog="~/job_queue"
104	job_name=none
105	local_user=$LOGNAME
106	node_usage=shared
107	numprocs=1
108	punkte="..........................................................."
109	submcom=qsub
110	queue=default
111	remote_host=none
112	remote_user=""
113	verify=true
114
115	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
116	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
117	typeset -L20 spalte1
118	typeset -R40 spalte2
119	typeset -L60 spalte3
120
121
122
123	# FEHLERBEHANDLUNG
124	# BEI EXIT:
125	trap 'if [[ $locat != normal ]]
126	then
127	case $locat in
128	(option) printf "\n --> available optios can be displayed"
129	printf " by typing:"
130	printf "\n \"subjob ?\" \n";;
131	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
132	(*) printf "\n +++ unknown error"
133	printf "\n please inform S. Raasch!\n"
134	esac
135	[[ -f $job_to_send ]] && rm $job_to_send
136	printf "\n\n+++ SUBJOB killed \n\n"
137	fi' exit
138
139
140	# BEI TERMINAL-BREAK:
141	trap '[[ -f $job_to_send ]] && rm $job_to_send
142	printf "\n\n+++ SUBJOB killed \n\n"
143	exit
144	' 2
145
146
147
148
149	# LOKALEN HOSTNAMEN FESTSTELLEN
150	local_host=$(hostname)
151
152
153
154	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
155	# UEBERHAUPT ZULAESSIG IST
156	case $local_host in
157	(atmos) local_addres=172.20.25.35; local_host=lcide;;
158	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
159	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
160	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
161	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
162	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
163	(cs*) local_addres=136.172.44.131; local_host=nech;;
164	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
165	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
166	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
167	(gate\|n-sx) local_addres=133.5.178.11; local_host=neck;;
168	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
169	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
170	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
171	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
172	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
173	(hreg-en0\|hanni-en0) local_addres=130.75.4.10; local_host=ibmh;;
174	(h01*) local_addres=10.152.0.7; local_host=lcsgih;;
175	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
176	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
177	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
178	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
179	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
180	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
181	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
182	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
183	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
184	(service100) local_addres=10.152.0.7; local_host=lcsgih;;
185	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
186	(r1*) local_addres=10.152.0.7; local_host=lcsgih;;
187	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
188	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
189	(*) printf "\n +++ \"$local_host\" unknown";
190	printf "\n please inform S. Raasch!";
191	locat=parameter; exit;;
192	esac
193
194
195
196	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
197	remote_host=$local_host
198
199
200
201
202	# PROZEDUROPTIONEN EINLESEN
203	while getopts :c:dDg:h:m:n:N:O:q:t:T:u:vX: option
204	do
205	case $option in
206	(c) job_catalog=$OPTARG;;
207	(d) delete_dayfile=true;;
208	(D) no_submit=true;;
209	(g) group_number=$OPTARG;;
210	(h) remote_host=$OPTARG;;
211	(m) memory=$OPTARG;;
212	(n) job_name=$OPTARG;;
213	(N) node_usage=$OPTARG;;
214	(O) threads_per_task=$OPTARG;;
215	(q) no_default_queue=$OPTARG;;
216	(t) cputime=$OPTARG;;
217	(T) tasks_per_node=$OPTARG;;
218	(u) remote_user=$OPTARG;;
219	(v) verify=false;;
220	(X) numprocs=$OPTARG;;
221	(\?) printf "\n +++ Option $OPTARG unknown \n";
222	locat=option; exit;;
223	esac
224	done
225
226
227	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
228	shift OPTIND-1; file_to_send=$1
229
230
231	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
232	if [ "$1" = "?" ]
233	then
234	(printf "\n *** subjob can be called as follows:\n"
235	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
236	printf "\n Description of available options:\n"
237	printf "\n Option Description Default-Value"
238	printf "\n -c job-input- and output-catalog ~/job_queue"
239	printf "\n -d no job-protocol will be created ---"
240	printf "\n -D only the job-file will be created ---"
241	printf "\n -h execution host, available hosts: $remote_host"
242	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
243	printf "\n lctit, nech, neck, unics"
244	printf "\n -m memory demand per process in MByte ---"
245	printf "\n -n jobname <jobdatei>"
246	printf "\n -O threads per task (for OpenMP usage) 1"
247	printf "\n -q job-queue to be used default"
248	printf "\n -t allowed cpu-time in seconds ---"
249	printf "\n -T tasks per node (on parallel hosts) ---"
250	printf "\n -u username on execution host from .netrc"
251	printf "\n -v no prompt for confirmation ---"
252	printf "\n -X # of processors (on parallel hosts) 1"
253	printf "\n "
254	printf "\n The only possible positional parameter is <jobfile>:"
255	printf "\n The complete NQS-job must be provided here."
256	printf "\n <jobfile>=? creates this outline\n\n") \| more
257	exit
258	fi
259
260
261
262	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
263	if [[ "$file_to_send" = "" ]]
264	then
265	printf "\n +++ job-file missing"
266	locat=parameter; exit
267	else
268	if [[ -f $file_to_send ]]
269	then
270	true
271	else
272	printf "\n +++ job-file: "
273	printf "\n $file_to_send"
274	printf "\n does not exist"
275	locat=parameter; exit
276	fi
277	fi
278
279
280
281	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
282	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
283	if [[ $job_name = none ]]
284	then
285	job_name=$file_to_send
286	fi
287	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
288	then
289	printf "\n +++ job-file name: "
290	printf "\n $job_name"
291	printf "\n must not contain \"/\"-characters"
292	locat=parameter; exit
293	fi
294
295
296
297
298	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
299	# ODER WENN HOST NICHT ANGEGEBEN WURDE
300	if [[ $remote_host = none ]]
301	then
302	printf "\n +++ host missing"
303	locat=option; exit
304	else
305	case $remote_host in
306	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
307	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
308	(ibmh) queue=cpar; remote_addres=130.75.4.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
309	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
310	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
311	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
312	(lcsgih) remote_addres=10.152.0.7; submcom=/opt/moab/bin/msub;;
313	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
314	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
315	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
316	(neck) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=133.5.178.11; submcom="/usr/bin/nqsII/qsub";;
317	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
318	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
319	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
320	locat=parameter; exit;;
321	esac
322	fi
323
324
325	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
326	if [[ $no_default_queue != none ]]
327	then
328	error=false
329	ndq=$no_default_queue
330	case $remote_host in
331	(ibm) case $ndq in
332	(p690_express\|p690_standard\|p690_long) error=false;;
333	(*) error=true;;
334	esac;;
335	(ibmb) case $ndq in
336	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
337	(*) error=true;;
338	esac;;
339	(ibmh) case $ndq in
340	(cdata\|cdev\|cexp\|channi\|cxxl\|c1\|cshare\|csolo\|csoloh3h\|csoloh6h\|csoloh12h\|csoloh1d\|csoloh2d\|cspec) error=false;;
341	(*) error=true;;
342	esac;;
343	(ibms) case $ndq in
344	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
345	(*) error=true;;
346	esac;;
347	(ibmy) case $ndq in
348	(parallel) error=false;;
349	(*) error=true;;
350	esac;;
351	(lctit) case $ndq in
352	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
353	(*) error=true;;
354	esac;;
355	(t3eb) case $ndq in
356	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
357	(*) error=true;;
358	esac;;
359	(t3eh) case $ndq in
360	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
361	(*) error=true;;
362	esac;;
363	(t3ej2\|t3ej5) case $ndq in
364	(low\|normal\|high) error=false;;
365	(*) error=true;;
366	esac;;
367	(t3es) case $ndq in
368	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
369	(*) error=true;;
370	esac;;
371	(unics) case $ndq in
372	(unics\|ht) error=false;;
373	(*) error=true;;
374	esac;;
375	esac
376	if [[ $error = true ]]
377	then
378	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
379	locat=parameter; exit
380	else
381	queue=$no_default_queue
382	fi
383	fi
384
385
386
387	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
388	if [[ $node_usage != shared && $queue = cdev ]]
389	then
390	node_usage=shared
391	fi
392
393
394
395	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
396	# AUFTEILEN
397	done=false
398	while [[ $done = false ]]
399	do
400	if (( $cputime <= 0 ))
401	then
402	printf "\n +++ wrong cpu-time or cpu-time missing"
403	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
404	printf "\n >>> "
405	read cputime 1>/dev/null 2>&1
406	else
407	done=true
408	fi
409	done
410	if [[ $remote_host = nech ]]
411	then
412	if (( tasks_per_node != 0 ))
413	then
414	(( cputime = cputime * tasks_per_node ))
415	elif [[ $numprocs != 0 ]]
416	then
417	(( cputime = cputime * numprocs ))
418	fi
419	fi
420	(( stunden = cputime / 3600 ))
421	(( resttime = cputime - stunden * 3600 ))
422	(( minuten = resttime / 60 ))
423	(( sekunden = resttime - minuten * 60 ))
424	timestring=${stunden}:${minuten}:${sekunden}
425
426
427
428	# PRUEFEN DER KERNSPEICHERANFORDERUNG
429	done=false
430	while [[ $done = false ]]
431	do
432	if (( memory <= 0 ))
433	then
434	printf "\n +++ wrong memory demand or memory demand missing"
435	printf "\n >>> Please type memory in MByte per process as INTEGER:"
436	printf "\n >>> "
437	read memory 1>/dev/null 2>&1
438	else
439	done=true
440	fi
441	done
442
443	if [[ $remote_host = nech \|\| $remote_host = neck ]]
444	then
445	if (( tasks_per_node != 0 ))
446	then
447	(( Memory = memory * tasks_per_node / 1000 ))
448	elif [[ $numprocs != 0 ]]
449	then
450	(( Memory = memory * numprocs / 1000 ))
451	else
452	(( Memory = memory / 1000 ))
453	fi
454	fi
455
456
457	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
458	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
459	then
460	(( memory = memory * threads_per_task ))
461	fi
462
463
464	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
465	if (( tasks_per_node != 0 ))
466	then
467	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
468	fi
469
470
471
472	# HEADER-AUSGABE
473	if [[ $verify = true ]]
474	then
475	printf "\n\n"
476	printf "#--------------------------------------------------------------# \n"
477	spalte1=SUBJOB;spalte2=$(date)
478	printf "\| $spalte1$spalte2 \| \n"
479	printf "\| \| \n"
480	printf "\| values of parameters/options: \| \n"
481	spalte1=local_host$punkte; spalte2=$punkte$local_host
482	printf "\| $spalte1$spalte2 \| \n"
483	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
484	printf "\| $spalte1$spalte2 \| \n"
485	spalte1=queue$punkte; spalte2=$punkte$queue
486	printf "\| $spalte1$spalte2 \| \n"
487	spalte1=memory$punkte; spalte2="$punkte$memory mb"
488	printf "\| $spalte1$spalte2 \| \n"
489	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
490	printf "\| $spalte1$spalte2 \| \n"
491	spalte1=job_name$punkte; spalte2="$punkte$job_name"
492	printf "\| $spalte1$spalte2 \| \n"
493	printf "#--------------------------------------------------------------# \n\n"
494
495
496	# KONTROLLABFRAGE, OB ALLES O.K.
497	antwort="dummy"
498	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
499	do
500	read antwort?" >>> continue (y/n) ? "
501	done
502	if [[ $antwort = n \|\| $antwort = N ]]
503	then
504	locat=verify; exit
505	fi
506	printf "\n"
507	fi
508
509
510
511	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
512	kennung=$RANDOM
513	job_on_remhost=${job_name}_${kennung}_$local_host
514	job_to_send=job_to_send_$kennung
515	if [[ $delete_dayfile = false ]]
516	then
517	remote_dayfile=${local_host}_${job_name}_result_$kennung
518	local_dayfile=${remote_host}_${job_name}
519	else
520	remote_dayfile=/dev/null
521	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
522	then
523	notification='# @ notification = never'
524	fi
525	fi
526
527
528
529	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
530	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
531	then
532
533	if [[ $remote_host = ibmy ]]
534	then
535	consumable_memory=""
536	else
537	consumable_memory="ConsumableMemory($memory mb)"
538	fi
539	if [[ $queue = cdev ]]
540	then
541	data_limit="# @ data_limit = 1.76gb"
542	network_to_use="# @ network.mpi = sn_all,shared,ip"
543	else
544	if [[ $remote_host = ibms ]]
545	then
546	network_to_use="# @ network.mpi = csss,shared,us"
547	elif [[ $remote_host = ibmy ]]
548	then
549	network_to_use=""
550	else
551	network_to_use="# @ network.mpi = sn_all,shared,us"
552	data_limit="# @ data_limit = 1.76gb"
553	fi
554	fi
555
556	cat > $job_to_send << %%END%%
557	#!/bin/ksh
558
559	# @ job_type = parallel
560	# @ job_name = palm
561	# @ wall_clock_limit = ${timestring},$timestring
562	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
563	# @ output = $remote_dayfile
564	# @ error = $remote_dayfile
565	# @ image_size = 50
566	# @ class = $queue
567	# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes
568	$network_to_use
569	$data_limit
570	$notification
571
572	%%END%%
573
574	if (( nodes > 0 ))
575	then
576
577	cat >> $job_to_send << %%END%%
578	# @ node = $nodes
579	# @ tasks_per_node = $tasks_per_node
580	# @ node_usage = $node_usage
581	# @ queue
582
583	%%END%%
584
585	else
586
587	if [[ $remote_host != ibmy ]]
588	then
589
590	cat >> $job_to_send << %%END%%
591	# @ blocking = unlimited
592	# @ total_tasks = $numprocs
593	# @ node_usage = $node_usage
594	# @ queue
595
596	%%END%%
597
598	else
599
600	cat >> $job_to_send << %%END%%
601	# @ node = 1
602	# @ total_tasks = $numprocs
603	# @ queue
604
605	%%END%%
606
607	fi
608
609	fi
610
611	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
612	then
613
614	cat > $job_to_send << %%END%%
615	#!/bin/ksh
616
617	# @ job_type = serial
618	# @ node_usage = $node_usage
619	# @ job_name = palm
620	# @ wall_clock_limit = ${timestring},$timestring
621	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
622	# @ output = $remote_dayfile
623	# @ error = $remote_dayfile
624	# @ class = $queue
625	$notification
626
627	# @ queue
628
629	%%END%%
630
631	elif [[ $remote_host = lcfimm ]]
632	then
633
634	if [[ $numprocs != 0 ]]
635	then
636	cat > $job_to_send << %%END%%
637	#!/bin/ksh
638	#PBS -N $job_name
639	#PBS -A nersc
640	#PBS -l walltime=$timestring
641	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
642	#PBS -l pmem=${memory}mb
643	#PBS -m abe
644	#PBS -M igore@nersc.no
645	#PBS -o $remote_dayfile
646	#PBS -j oe
647	mpd &
648
649	%%END%%
650
651	else
652	cat > $job_to_send << %%END%%
653	#!/bin/ksh
654	#PBS -N $job_name
655	#PBS -A nersc
656	#PBS -l walltime=$timestring
657	#PBS -l ncpus=1
658	#PBS -l pmem=${memory}mb
659	#PBS -m abe
660	#PBS -M igore@nersc.no
661	#PBS -o $remote_dayfile
662	#PBS -j oe
663
664	%%END%%
665
666	fi
667
668	elif [[ $remote_host = lcsgih ]]
669	then
670
671	if [[ $numprocs != 0 ]]
672	then
673	cat > $job_to_send << %%END%%
674	#PBS -S /bin/bash
675	#PBS -N $job_name
676	#PBS -l walltime=$timestring
677	#PBS -l nodes=$nodes:ppn=${tasks_per_node}:xe
678	#PBS -l naccesspolicy=singlejob
679	#PBS -o $remote_dayfile
680	#PBS -j oe
681
682	. /usr/share/modules/init/bash
683	module load mvapich/mvapich-0.9.9-intel
684	module load netcdf
685	export LD_LIBRARY_PATH=/sw/intel/fce/10.1.015/lib:/sw/dataformats/netcdf/3.6.2/lib:$LD_LIBRARY_PATH
686
687	echo ld_library_path=$LD_LIBRARY_PATH
688
689	%%END%%
690
691	else
692	cat > $job_to_send << %%END%%
693	#PBS -S /bin/bash
694	#PBS -N $job_name
695	#PBS -l walltime=$timestring
696	#PBS -l ncpus=1
697	#PBS -o $remote_dayfile
698	#PBS -e $remote_dayfile
699
700	%%END%%
701
702	fi
703
704	elif [[ $remote_host = lcxt4 ]]
705	then
706
707	if [[ $numprocs != 0 ]]
708	then
709	cat > $job_to_send << %%END%%
710	#!/bin/ksh
711	#PBS -S /bin/ksh
712	#PBS -N $job_name
713	#PBS -A nersc
714	#PBS -l walltime=$timestring
715	#PBS -l mppwidth=${numprocs}
716	#PBS -l mppnppn=${tasks_per_node}
717	#PBS -m abe
718	#PBS -M igore@nersc.no
719	#PBS -o $remote_dayfile
720	#PBS -e $remote_dayfile
721
722	%%END%%
723
724	else
725	cat > $job_to_send << %%END%%
726	#!/bin/ksh
727	#PBS -S /bin/ksh
728	#PBS -N $job_name
729	#PBS -A nersc
730	#PBS -l walltime=$timestring
731	#PBS -l ncpus=1
732	#PBS -l pmem=${memory}mb
733	#PBS -m abe
734	#PBS -M igore@nersc.no
735	#PBS -o $remote_dayfile
736	#PBS -e $remote_dayfile
737
738	%%END%%
739
740	fi
741
742	elif [[ $remote_host = nech ]]
743	then
744
745	if (( nodes > 1 ))
746	then
747	cat > $job_to_send << %%END%%
748	#!/bin/ksh
749	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
750	#PBS -l ${qsubmem}=${Memory}gb
751	#PBS -b $nodes
752	#PBS -o $remote_dayfile
753	#PBS -N palm
754	#PBS -j o
755	#PBS -T mpisx
756
757	%%END%%
758
759	elif [[ $numprocs != 0 ]]
760	then
761	cat > $job_to_send << %%END%%
762	#!/bin/ksh
763	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
764	#PBS -l ${qsubmem}=${Memory}gb
765	#PBS -o $remote_dayfile
766	#PBS -N palm
767	#PBS -j o
768
769	%%END%%
770
771	else
772	cat > $job_to_send << %%END%%
773	#!/bin/ksh
774	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
775	#PBS -o $remote_dayfile
776	#PBS -j o
777
778	%%END%%
779
780	fi
781
782	elif [[ $remote_host = neck ]]
783	then
784
785	if (( nodes > 1 ))
786	then
787	cat > $job_to_send << %%END%%
788	#!/bin/ksh
789	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
790	#PBS -l ${qsubmem}=${Memory}gb
791	#PBS -b $nodes
792	#PBS -o $remote_dayfile
793	#PBS -N palm
794	#PBS -j o
795	#PBS -T mpisx
796
797	%%END%%
798
799	elif [[ $numprocs != 0 ]]
800	then
801	cat > $job_to_send << %%END%%
802	#!/bin/ksh
803	#PBS -q ${queue}
804	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
805	#PBS -l ${qsubmem}=${Memory}gb
806	#PBS -o $remote_dayfile
807	#PBS -N palm
808	#PBS -j o
809	#PBS -T mpisx
810
811	%%END%%
812
813	else
814	cat > $job_to_send << %%END%%
815	#!/bin/ksh
816	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
817	#PBS -o $remote_dayfile
818	#PBS -j o
819
820	%%END%%
821
822	fi
823
824	elif [[ $remote_host = lctit ]]
825	then
826	cat > $job_to_send << %%END%%
827	#!/bin/bash
828	#$ -S /bin/bash
829	cd $job_catalog
830	export PATH=$PALM_BIN:\$PATH
831	echo \$PATH
832
833	%%END%%
834
835	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
836	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
837	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
838
839	else
840
841	cat > $job_to_send << %%END%%
842	# @\$-q ${queue}
843	# @\$-l${qsubtime} $timestring
844	# @\$-l${qsubmem} ${memory}mb
845	# @\$-o $remote_dayfile
846	# @\$-eo
847
848	%%END%%
849
850	fi
851
852
853	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
854	# BEI EXIT VERANLASSEN
855	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
856	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
857	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
858	then
859	echo "set +vx" >> $job_to_send
860	echo "trap '" >> $job_to_send
861	echo "set +vx" >> $job_to_send
862	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
863	then
864	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
865	then
866	return_queue=c1
867	elif [[ $remote_host = ibms ]]
868	then
869	return_queue=p_normal
870	elif [[ $remote_host = ibmy ]]
871	then
872	return_queue=serial
873	elif [[ $remote_host = lctit ]]
874	then
875	return_queue=sla3
876	elif [[ $remote_host = neck ]]
877	then
878	return_queue=S
879	else
880	return_queue=unknown
881	fi
882
883	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
884	then
885
886	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
887	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
888	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
889	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
890	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
891	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
892	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
893	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
894	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
895	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
896
897	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
898	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
899
900	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
901	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
902	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
903
904	elif [[ $remote_host = nech ]]
905	then
906	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
907	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
908	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
909	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
910	echo "#PBS -j o" >> $job_to_send
911	echo " " >> $job_to_send
912	echo "set -x" >> $job_to_send
913	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
914	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
915	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
916	echo "%%END%%" >> $job_to_send
917
918	elif [[ $remote_host = neck ]]
919	then
920	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
921	echo "#PBS -q $return_queue" >> $job_to_send
922	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
923	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
924	echo "#PBS -j o" >> $job_to_send
925	echo " " >> $job_to_send
926	echo "set -x" >> $job_to_send
927	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
928	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
929	echo "%%END%%" >> $job_to_send
930
931	elif [[ $remote_host = lctit ]]
932	then
933	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
934	echo "set -x" >> $job_to_send
935	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
936	echo "cd $job_catalog" >> $job_to_send
937	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
938	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
939	echo "%%END%%" >> $job_to_send
940
941	else
942
943	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
944	echo "# @\\\$-q $return_queue" >> $job_to_send
945	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
946	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
947	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
948	then
949	echo "# @\$-l mpp_p=0" >> $job_to_send
950	fi
951	echo '# @\$-lF 10mb' >> $job_to_send
952	# echo '# @\$-o /dev/null' >> $job_to_send
953	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
954	echo '# @\\\$-eo' >> $job_to_send
955	echo " " >> $job_to_send
956	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
957	then
958	echo "set +vx" >> $job_to_send
959	echo ". .profile" >> $job_to_send
960	fi
961	echo "set -x" >> $job_to_send
962	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
963	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
964	echo "%%END%%" >> $job_to_send
965	fi
966
967	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
968	then
969	echo "llsubmit scpjob.$kennung" >> $job_to_send
970	elif [[ $remote_host = lctit ]]
971	then
972	echo "chmod u+x scpjob.$kennung" >> $job_to_send
973	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
974	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
975	echo "sleep 10" >> $job_to_send
976	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
977	then
978	echo "qsub -J n scpjob.$kennung" >> $job_to_send
979	elif [[ $remote_host = t3es ]]
980	then
981	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
982	else
983	echo "qsub scpjob.$kennung" >> $job_to_send
984	fi
985	echo "rm scpjob.$kennung" >> $job_to_send
986	if [[ $remote_host = nech ]]
987	then
988	echo "cd -" >> $job_to_send
989	fi
990	else
991	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
992	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
993	fi
994	echo "set -x" >> $job_to_send
995	echo " ' exit" >> $job_to_send
996	echo "set -x" >> $job_to_send
997	fi
998
999
1000
1001	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1002	cat $file_to_send >> $job_to_send
1003
1004	if [[ $remote_host = ibm ]]
1005	then
1006	echo " " >> $job_to_send
1007	echo "exit" >> $job_to_send
1008	fi
1009	if [[ $remote_host = lctit ]]
1010	then
1011	echo " " >> $job_to_send
1012	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1013	fi
1014
1015
1016
1017	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1018	if [[ -z $remote_user ]]
1019	then
1020	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1021	then
1022	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1023	fi
1024	fi
1025
1026
1027
1028	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1029	if [[ $no_submit = false ]]
1030	then
1031	if [[ $remote_host != $local_host ]]
1032	then
1033	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1034	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1035	then
1036	job_catalog_save=$job_catalog
1037	job_catalog=job_queue
1038	elif [[ $remote_host = nech ]]
1039	then
1040	job_catalog_save=$job_catalog
1041	job_catalog=/hpf/b/${remote_user}/job_queue
1042	fi
1043	if [[ $local_host = decalpha ]]
1044	then
1045	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1046	# PASSWORT AB
1047	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1048	elif [[ $remote_host = nech ]]
1049	then
1050	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1051	# TRANSFERIERT WERDEN
1052	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1053	else
1054	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1055	fi
1056	if [[ $? = 1 ]]
1057	then
1058	locat=scp; exit
1059	fi
1060	if [[ $remote_host = ibms ]]
1061	then
1062	job_catalog=$job_catalog_save
1063	fi
1064	[[ $verify = true ]] && printf "\n >>> finished\n"
1065	else
1066	eval job_catalog=$job_catalog
1067	cp $job_to_send ${job_catalog}/$job_on_remhost
1068	fi
1069
1070
1071
1072	# NQS- BZW. LOADLEVELER-JOB STARTEN
1073	if [[ $remote_host != $local_host ]]
1074	then
1075	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1076	if [[ $remote_host != lctit ]]
1077	then
1078	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1079	else
1080	# TIT ERLAUBT NUR DIE AUSFÜHRUNG GANZ BESTIMMTER KOMMANDOS
1081	# MIT SSH, DESHALB AUFRUF PER PIPE
1082	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1083	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1084	echo "first try" > send_protocol
1085	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1086	do
1087	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1088	then
1089	printf "\n trying to submit job to TIT cluster..."
1090	else
1091	printf "\n +++ failed ... trying to submit again ..."
1092	fi
1093	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1094	cat send_protocol
1095	done
1096	sleep 10
1097	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1098	echo "$submcom $job_on_remhost"
1099	rm send_protocol
1100	fi
1101
1102	[[ $verify = true ]] && printf " >>> o.k.\n"
1103	else
1104	cd $job_catalog
1105	if [[ $(echo $local_host \| cut -c1-3) = ibm ]]
1106	then
1107	eval $submcom $job_on_remhost
1108	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 ]]
1109	then
1110	eval $submcom $job_on_remhost
1111	echo "$submcom $job_on_remhost"
1112	chmod u+x $job_on_remhost
1113	elif [[ $local_host = nech ]]
1114	then
1115	if [[ $queue = default ]]
1116	then
1117	eval $submcom $job_on_remhost
1118	else
1119	eval $submcom -q $queue $job_on_remhost
1120	fi
1121	else
1122	qsub $job_on_remhost
1123	fi
1124	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1125	[[ $local_host != lctit ]] && rm $job_on_remhost
1126	cd - > /dev/null
1127	fi
1128	fi
1129
1130
1131
1132	# ABSCHLUSSARBEITEN
1133
1134	if [[ $no_submit = false ]]
1135	then
1136	rm $job_to_send
1137	fi
1138	[[ $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 |