Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 449

Last change on this file since 449 was 440, checked in by raasch, 15 years ago
lckyoto admitted for mrun and subjob
Property svn:keywords set to `Id Rev`
File size: 47.3 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 440 2010-02-01 11:55:04Z 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	# 23/09/08 - Gerald- paesano admitted
96	# 02/10/08 - Siggi - PBS adjustments for lcxt4
97	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
98	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
99	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
100	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
101	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
102	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
103	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
104	# now run on the data nodes (feature=data)
105	# 25/08/09 - BjornM- adapted for lck
106	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
107	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
108	# set in the config file by the user
109	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
110	# adding special1q
111	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
112	# 16/10/09 - Carolin-adding permq
113	# 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600)
114
115
116
117	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
118	delete_dayfile=false
119	email_notification=none
120	group_number=none
121	locat=normal
122	no_default_queue=none
123	no_submit=false
124	job_catalog="~/job_queue"
125	job_name=none
126	local_user=$LOGNAME
127	node_usage=shared
128	numprocs=1
129	punkte="..........................................................."
130	submcom=qsub
131	queue=default
132	remote_host=none
133	remote_user=""
134	verify=true
135
136	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
137	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
138	typeset -L20 spalte1
139	typeset -R40 spalte2
140	typeset -L60 spalte3
141
142
143
144	# FEHLERBEHANDLUNG
145	# BEI EXIT:
146	trap 'if [[ $locat != normal ]]
147	then
148	case $locat in
149	(option) printf "\n --> available optios can be displayed"
150	printf " by typing:"
151	printf "\n \"subjob ?\" \n";;
152	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
153	(*) printf "\n +++ unknown error"
154	printf "\n please inform S. Raasch!\n"
155	esac
156	[[ -f $job_to_send ]] && rm $job_to_send
157	printf "\n\n+++ SUBJOB killed \n\n"
158	fi' exit
159
160
161	# BEI TERMINAL-BREAK:
162	trap '[[ -f $job_to_send ]] && rm $job_to_send
163	printf "\n\n+++ SUBJOB killed \n\n"
164	exit
165	' 2
166
167
168
169
170	# LOKALEN HOSTNAMEN FESTSTELLEN
171	local_host=$(hostname)
172
173
174
175	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
176	# UEBERHAUPT ZULAESSIG IST
177	case $local_host in
178	(atmos) local_addres=172.20.25.35; local_host=lcide;;
179	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
180	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
181	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
182	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
183	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
184	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
185	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
186	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
187	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
188	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
189	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
190	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
191	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
192	(cs*) local_addres=136.172.44.131; local_host=nech;;
193	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
194	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
195	(node*) local_addres=165.132.26.61 local_host=lck;;
196	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
197	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
198	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
199	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
200	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
201	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
202	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
203	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
204	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
205	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
206	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
207	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
208	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
209	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
210	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
211	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
212	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
213	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
214	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
215	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
216	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
217	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
218	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
219	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
220	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
221	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
222	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
223	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
224	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
225	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
226	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
227	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
228	(*) printf "\n +++ \"$local_host\" unknown";
229	printf "\n please inform S. Raasch!";
230	locat=parameter; exit;;
231	esac
232
233
234
235	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
236	remote_host=$local_host
237
238
239
240
241	# PROZEDUROPTIONEN EINLESEN
242	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
243	do
244	case $option in
245	(c) job_catalog=$OPTARG;;
246	(d) delete_dayfile=true;;
247	(D) no_submit=true;;
248	(e) email_notification=$OPTARG;;
249	(g) group_number=$OPTARG;;
250	(h) remote_host=$OPTARG;;
251	(m) memory=$OPTARG;;
252	(n) job_name=$OPTARG;;
253	(N) node_usage=$OPTARG;;
254	(O) threads_per_task=$OPTARG;;
255	(q) no_default_queue=$OPTARG;;
256	(t) cputime=$OPTARG;;
257	(T) tasks_per_node=$OPTARG;;
258	(u) remote_user=$OPTARG;;
259	(v) verify=false;;
260	(X) numprocs=$OPTARG;;
261	(\?) printf "\n +++ Option $OPTARG unknown \n";
262	locat=option; exit;;
263	esac
264	done
265
266
267	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
268	shift OPTIND-1; file_to_send=$1
269
270
271	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
272	if [ "$1" = "?" ]
273	then
274	(printf "\n *** subjob can be called as follows:\n"
275	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
276	printf "\n Description of available options:\n"
277	printf "\n Option Description Default-Value"
278	printf "\n -c job-input- and output-catalog ~/job_queue"
279	printf "\n -d no job-protocol will be created ---"
280	printf "\n -D only the job-file will be created ---"
281	printf "\n -h execution host, available hosts: $remote_host"
282	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
283	printf "\n lctit, nech, necriam, unics"
284	printf "\n -m memory demand per process in MByte ---"
285	printf "\n -n jobname <jobdatei>"
286	printf "\n -O threads per task (for OpenMP usage) 1"
287	printf "\n -q job-queue to be used default"
288	printf "\n -t allowed cpu-time in seconds ---"
289	printf "\n -T tasks per node (on parallel hosts) ---"
290	printf "\n -u username on execution host from .netrc"
291	printf "\n -v no prompt for confirmation ---"
292	printf "\n -X # of processors (on parallel hosts) 1"
293	printf "\n "
294	printf "\n The only possible positional parameter is <jobfile>:"
295	printf "\n The complete NQS-job must be provided here."
296	printf "\n <jobfile>=? creates this outline\n\n") \| more
297	exit
298	fi
299
300
301
302	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
303	if [[ "$file_to_send" = "" ]]
304	then
305	printf "\n +++ job-file missing"
306	locat=parameter; exit
307	else
308	if [[ -f $file_to_send ]]
309	then
310	true
311	else
312	printf "\n +++ job-file: "
313	printf "\n $file_to_send"
314	printf "\n does not exist"
315	locat=parameter; exit
316	fi
317	fi
318
319
320
321	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
322	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
323	if [[ $job_name = none ]]
324	then
325	job_name=$file_to_send
326	fi
327	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
328	then
329	printf "\n +++ job-file name: "
330	printf "\n $job_name"
331	printf "\n must not contain \"/\"-characters"
332	locat=parameter; exit
333	fi
334
335
336
337
338	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
339	# ODER WENN HOST NICHT ANGEGEBEN WURDE
340	if [[ $remote_host = none ]]
341	then
342	printf "\n +++ host missing"
343	locat=option; exit
344	else
345	case $remote_host in
346	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
347	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
348	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
349	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
350	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
351	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
352	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
353	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
354	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
355	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
356	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
357	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
358	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
359	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
360	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
361	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
362	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
363	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
364	locat=parameter; exit;;
365	esac
366	fi
367
368
369	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
370	if [[ $no_default_queue != none ]]
371	then
372	error=false
373	ndq=$no_default_queue
374	case $remote_host in
375	(ibm) case $ndq in
376	(p690_express\|p690_standard\|p690_long) error=false;;
377	(*) error=true;;
378	esac;;
379	(ibmb) case $ndq in
380	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
381	(*) error=true;;
382	esac;;
383	(ibmh) case $ndq in
384	(no_class) error=false;;
385	(*) error=true;;
386	esac;;
387	(ibms) case $ndq in
388	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
389	(*) error=true;;
390	esac;;
391	(ibmy) case $ndq in
392	(parallel) error=false;;
393	(*) error=true;;
394	esac;;
395	(lckyoto) case $ndq in
396	(eh\|ph) error=false;;
397	(*) error=true;;
398	esac;;
399	(lcsgib\|lcsgih) case $ndq in
400	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
401	(*) error=true;;
402	esac;;
403	(lctit) case $ndq in
404	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
405	(*) error=true;;
406	esac;;
407	(t3eb) case $ndq in
408	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
409	(*) error=true;;
410	esac;;
411	(necriam) case $ndq in
412	(SP\|SS\|P6) error=false;;
413	(*) error=true;;
414	esac;;
415	(t3eh) case $ndq in
416	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
417	(*) error=true;;
418	esac;;
419	(t3ej2\|t3ej5) case $ndq in
420	(low\|normal\|high) error=false;;
421	(*) error=true;;
422	esac;;
423	(t3es) case $ndq in
424	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
425	(*) error=true;;
426	esac;;
427	(unics) case $ndq in
428	(unics\|ht) error=false;;
429	(*) error=true;;
430	esac;;
431	esac
432	if [[ $error = true ]]
433	then
434	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
435	locat=parameter; exit
436	else
437	queue=$no_default_queue
438	fi
439	fi
440
441
442
443	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
444	if [[ $node_usage != shared && $queue = cdev ]]
445	then
446	node_usage=shared
447	fi
448
449
450
451	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
452	# AUFTEILEN
453	done=false
454	while [[ $done = false ]]
455	do
456	if (( $cputime <= 0 ))
457	then
458	printf "\n +++ wrong cpu-time or cpu-time missing"
459	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
460	printf "\n >>> "
461	read cputime 1>/dev/null 2>&1
462	else
463	done=true
464	fi
465	done
466	if [[ $remote_host = nech ]]
467	then
468	if (( tasks_per_node != 0 ))
469	then
470	(( cputime = cputime * tasks_per_node ))
471	elif [[ $numprocs != 0 ]]
472	then
473	(( cputime = cputime * numprocs ))
474	fi
475	fi
476	(( stunden = cputime / 3600 ))
477	(( resttime = cputime - stunden * 3600 ))
478	(( minuten = resttime / 60 ))
479	(( sekunden = resttime - minuten * 60 ))
480	timestring=${stunden}:${minuten}:${sekunden}
481
482
483
484	# PRUEFEN DER KERNSPEICHERANFORDERUNG
485	done=false
486	while [[ $done = false ]]
487	do
488	if (( memory <= 0 ))
489	then
490	printf "\n +++ wrong memory demand or memory demand missing"
491	printf "\n >>> Please type memory in MByte per process as INTEGER:"
492	printf "\n >>> "
493	read memory 1>/dev/null 2>&1
494	else
495	done=true
496	fi
497	done
498
499	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
500	then
501	if (( tasks_per_node != 0 ))
502	then
503	(( Memory = memory * tasks_per_node / 1000 ))
504	elif [[ $numprocs != 0 ]]
505	then
506	(( Memory = memory * numprocs / 1000 ))
507	else
508	(( Memory = memory / 1000 ))
509	fi
510	fi
511
512
513	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
514	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
515	then
516	(( memory = memory * threads_per_task ))
517	fi
518
519
520	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
521	if (( tasks_per_node != 0 ))
522	then
523	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
524	then
525	(( nodes = numprocs / tasks_per_node ))
526	else
527	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
528	fi
529	fi
530
531
532
533	# HEADER-AUSGABE
534	if [[ $verify = true ]]
535	then
536	printf "\n\n"
537	printf "#--------------------------------------------------------------# \n"
538	spalte1=SUBJOB;spalte2=$(date)
539	printf "\| $spalte1$spalte2 \| \n"
540	printf "\| \| \n"
541	printf "\| values of parameters/options: \| \n"
542	spalte1=local_host$punkte; spalte2=$punkte$local_host
543	printf "\| $spalte1$spalte2 \| \n"
544	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
545	printf "\| $spalte1$spalte2 \| \n"
546	spalte1=queue$punkte; spalte2=$punkte$queue
547	printf "\| $spalte1$spalte2 \| \n"
548	spalte1=memory$punkte; spalte2="$punkte$memory mb"
549	printf "\| $spalte1$spalte2 \| \n"
550	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
551	printf "\| $spalte1$spalte2 \| \n"
552	spalte1=job_name$punkte; spalte2="$punkte$job_name"
553	printf "\| $spalte1$spalte2 \| \n"
554	printf "#--------------------------------------------------------------# \n\n"
555
556
557	# KONTROLLABFRAGE, OB ALLES O.K.
558	antwort="dummy"
559	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
560	do
561	read antwort?" >>> continue (y/n) ? "
562	done
563	if [[ $antwort = n \|\| $antwort = N ]]
564	then
565	locat=verify; exit
566	fi
567	printf "\n"
568	fi
569
570
571
572	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
573	kennung=$RANDOM
574	job_on_remhost=${job_name}_${kennung}_$local_host
575	job_to_send=job_to_send_$kennung
576	if [[ $delete_dayfile = false ]]
577	then
578	remote_dayfile=${local_host}_${job_name}_result_$kennung
579	local_dayfile=${remote_host}_${job_name}
580	else
581	remote_dayfile=/dev/null
582	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
583	then
584	notification='# @ notification = never'
585	fi
586	fi
587
588
589
590	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
591	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
592	then
593
594	if [[ $remote_host = ibmy ]]
595	then
596	consumable_memory=""
597	else
598	consumable_memory="ConsumableMemory($memory mb)"
599	fi
600
601	if [[ $remote_host = ibmh ]]
602	then
603	data_limit=""
604	network_to_use=""
605	class=""
606	environment=""
607	else
608	class="# @ class = $queue"
609	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
610	if [[ $queue = cdev ]]
611	then
612	data_limit="# @ data_limit = 1.76gb"
613	network_to_use="# @ network.mpi = sn_all,shared,ip"
614	else
615	if [[ $remote_host = ibms ]]
616	then
617	network_to_use="# @ network.mpi = csss,shared,us"
618	elif [[ $remote_host = ibmy ]]
619	then
620	network_to_use=""
621	else
622	network_to_use="# @ network.mpi = sn_all,shared,us"
623	data_limit="# @ data_limit = 1.76gb"
624	fi
625	fi
626	fi
627
628	cat > $job_to_send << %%END%%
629	#!/bin/ksh
630	# @ shell = /bin/ksh
631
632	# @ job_type = parallel
633	# @ job_name = palm
634	# @ wall_clock_limit = ${timestring},$timestring
635	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
636	# @ output = $remote_dayfile
637	# @ error = $remote_dayfile
638	# @ image_size = 50
639	$class
640	$environment
641	$network_to_use
642	$data_limit
643	$notification
644
645	%%END%%
646
647	if (( nodes > 0 ))
648	then
649
650	cat >> $job_to_send << %%END%%
651	# @ node = $nodes
652	# @ tasks_per_node = $tasks_per_node
653	# @ node_usage = $node_usage
654	# @ queue
655
656	%%END%%
657
658	else
659
660	if [[ $remote_host != ibmy ]]
661	then
662
663	cat >> $job_to_send << %%END%%
664	# @ blocking = unlimited
665	# @ total_tasks = $numprocs
666	# @ node_usage = $node_usage
667	# @ queue
668
669	%%END%%
670
671	else
672
673	cat >> $job_to_send << %%END%%
674	# @ node = 1
675	# @ total_tasks = $numprocs
676	# @ queue
677
678	%%END%%
679
680	fi
681
682	fi
683
684	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
685	then
686
687	cat > $job_to_send << %%END%%
688	#!/bin/ksh
689
690	# @ job_type = serial
691	# @ node_usage = $node_usage
692	# @ job_name = palm
693	# @ wall_clock_limit = ${timestring},$timestring
694	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
695	# @ output = $remote_dayfile
696	# @ error = $remote_dayfile
697	$class
698	$notification
699
700	# @ queue
701
702	%%END%%
703
704	elif [[ $remote_host = lcfimm ]]
705	then
706
707	if [[ $numprocs != 0 ]]
708	then
709	cat > $job_to_send << %%END%%
710	#!/bin/ksh
711	#PBS -N $job_name
712	#PBS -A geofysisk
713	#PBS -l walltime=$timestring
714	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
715	#PBS -l pmem=${memory}mb
716	#PBS -m abe
717	#PBS -M maronga@muk.uni-hannover.de
718	#PBS -o $remote_dayfile
719	#PBS -j oe
720	mpd &
721
722	%%END%%
723
724	else
725	cat > $job_to_send << %%END%%
726	#!/bin/ksh
727	#PBS -N $job_name
728	#PBS -A geofysisk
729	#PBS -l walltime=$timestring
730	#PBS -l ncpus=1
731	#PBS -l pmem=${memory}mb
732	#PBS -m abe
733	#PBS -M maronga@muk.uni-hannover.de
734	#PBS -o $remote_dayfile
735	#PBS -j oe
736
737	%%END%%
738
739	fi
740
741	elif [[ $remote_host = lck ]]
742	then
743
744	if [[ $numprocs != 0 ]]
745	then
746	cat > $job_to_send << %%END%%
747	#!/bin/ksh
748	#PBS -N $job_name
749	#PBS -l walltime=$timestring
750	#PBS -l ncpus=$numprocs
751	#PBS -l pmem=${memory}mb
752	#PBS -o $remote_dayfile
753	#PBS -j oe
754
755	mpd &
756
757	%%END%%
758
759	else
760	cat > $job_to_send << %%END%%
761	#!/bin/ksh
762	#PBS -N $job_name
763	#PBS -l walltime=$timestring
764	#PBS -l ncpus=1
765	#PBS -l pmem=${memory}mb
766	#PBS -o $remote_dayfile
767	#PBS -j oe
768
769	%%END%%
770
771	fi
772
773	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
774	then
775
776	if [[ "$sgi_feature" != "" ]]
777	then
778	feature_directive="#PBS -l feature=$sgi_feature"
779	else
780	feature_directive=""
781	fi
782
783	if [[ $queue = dataq \|\| $queue = permq ]]
784	then
785	feature_directive="#PBS -l feature=data"
786	fi
787
788	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
789	then
790	queue_directive="#PBS -q $queue"
791	else
792	queue_directive=""
793	fi
794
795	if [[ $email_notification = none ]]
796	then
797	email_directive=""
798	else
799	email_directive="#PBS -M $email_notification"
800	fi
801
802	if [[ $numprocs != 0 ]]
803	then
804	cat > $job_to_send << %%END%%
805	#!/bin/bash
806	#PBS -N $job_name
807	#PBS -l walltime=$timestring
808	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
809	#PBS -l naccesspolicy=$node_usage
810	#PBS -o $remote_dayfile
811	#PBS -j oe
812	$feature_directive
813	$queue_directive
814	$email_directive
815
816	. /usr/share/modules/init/bash
817	module load ifort/11.0.069
818	module load $mpilib
819	module load netcdf
820
821	echo ld_library_path=\$LD_LIBRARY_PATH
822
823	%%END%%
824
825	else
826	cat > $job_to_send << %%END%%
827	#PBS -S /bin/bash
828	#PBS -N $job_name
829	#PBS -l walltime=$timestring
830	#PBS -l ncpus=1
831	#PBS -o $remote_dayfile
832	#PBS -e $remote_dayfile
833	$queue_directive
834	$email_directive
835
836	%%END%%
837
838	fi
839
840	elif [[ $remote_host = lcxt4 ]]
841	then
842
843	if [[ $numprocs != 0 ]]
844	then
845	cat > $job_to_send << %%END%%
846	#!/bin/ksh
847	#PBS -S /bin/ksh
848	#PBS -N $job_name
849	#PBS -A geofysisk
850	#PBS -j oe
851	#PBS -l walltime=$timestring
852	#PBS -l mppwidth=${numprocs}
853	#PBS -l mppnppn=${tasks_per_node}
854	#PBS -m abe
855	#PBS -M maronga@muk.uni-hannover.de
856	#PBS -o $remote_dayfile
857
858	. /opt/modules/default/init/ksh
859	module load modules/3.1.6
860	module load PrgEnv-pgi/2.1.56HDA
861	module load Base-opts/2.1.56HDA
862	module load moab/5.2.5-s13570
863	module load torque/2.3.4-snap.200809221601
864
865
866	%%END%%
867
868	else
869	cat > $job_to_send << %%END%%
870	#!/bin/ksh
871	#PBS -S /bin/ksh
872	#PBS -N $job_name
873	#PBS -A geofysisk
874	#PBS -j oe
875	#PBS -l walltime=$timestring
876	#PBS -l ncpus=1
877	#PBS -l pmem=${memory}mb
878	#PBS -m abe
879	#PBS -M maronga@muk.uni-hannover.de
880	#PBS -o $remote_dayfile
881
882	. /opt/modules/default/init/ksh
883	module load modules/3.1.6
884	module load PrgEnv-pgi/2.1.56HD
885	module load Base-opts/2.1.56HD
886	module load moab/5.2.5-s13441
887	module load torque/2.3.4-snap.200809221601
888
889	%%END%%
890
891	fi
892
893	elif [[ $remote_host = lckyoto ]]
894	then
895
896	cat > $job_to_send << %%END%%
897	#!/bin/ksh
898	# @\$-o $remote_dayfile
899	# @\$-eo -oi
900	# @\$-lP 16
901	# @\$-lp 1
902	# @\$-lm 28gb -llm unlimited -ls unlimited
903	# @\$-q $queue
904	# @\$-Pvn abs_pack
905	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
906	#. /thin/local/etc/setprofile/intel-11.0.sh
907	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
908	. ~/.myprofile
909	#. /home2/t/t51254/palm/current_version/myprofile
910	#. /thin/apps/pgi/mpi.sh
911	#
912	env
913	#
914	set -x
915
916	%%END%%
917
918	elif [[ $remote_host = lcxt5m ]]
919	then
920
921	if [[ $numprocs != 0 ]]
922	then
923	cat > $job_to_send << %%END%%
924	#!/bin/ksh
925	#PBS -S /bin/ksh
926	#PBS -N $job_name
927	#PBS -j oe
928	#PBS -l walltime=$timestring
929	#PBS -l mppwidth=${numprocs}
930	#PBS -l mppnppn=${tasks_per_node}
931	#PBS -m abe
932	#PBS -o $remote_dayfile
933
934	. /opt/modules/default/init/ksh
935	module load modules/3.1.6
936	module load netcdf/3.6.2
937	module load PrgEnv-pgi/2.2.41
938	module load Base-opts/2.2.41
939
940	%%END%%
941
942	else
943	cat > $job_to_send << %%END%%
944	#!/bin/ksh
945	#PBS -S /bin/ksh
946	#PBS -N $job_name
947	#PBS -j oe
948	#PBS -l walltime=$timestring
949	#PBS -l ncpus=1
950	#PBS -l pmem=${memory}mb
951	#PBS -m abe
952	#PBS -o $remote_dayfile
953
954	. /opt/modules/default/init/ksh
955	module load modules/3.1.6
956	module load netcdf/3.6.2
957	module load PrgEnv-pgi/2.2.41
958	module load Base-opts/2.2.41
959
960	%%END%%
961
962	fi
963
964	elif [[ $remote_host = nech ]]
965	then
966
967	if (( nodes > 1 ))
968	then
969	cat > $job_to_send << %%END%%
970	#!/bin/ksh
971	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
972	#PBS -l ${qsubmem}=${Memory}gb
973	#PBS -b $nodes
974	#PBS -o $remote_dayfile
975	#PBS -N palm
976	#PBS -j o
977	#PBS -T mpisx
978
979	%%END%%
980
981	elif [[ $numprocs != 0 ]]
982	then
983	cat > $job_to_send << %%END%%
984	#!/bin/ksh
985	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
986	#PBS -l ${qsubmem}=${Memory}gb
987	#PBS -o $remote_dayfile
988	#PBS -N palm
989	#PBS -j o
990
991	%%END%%
992
993	else
994	cat > $job_to_send << %%END%%
995	#!/bin/ksh
996	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
997	#PBS -o $remote_dayfile
998	#PBS -j o
999
1000	%%END%%
1001
1002	fi
1003
1004	elif [[ $remote_host = necriam ]]
1005	then
1006
1007	if (( nodes > 1 ))
1008	then
1009	cat > $job_to_send << %%END%%
1010	#!/bin/ksh
1011	#PBS -b $nodes
1012	#PBS -o $remote_dayfile
1013	#PBS -N $job_name
1014	#PBS -j o
1015	#PBS -v MPIPROGINV=YES
1016
1017	%%END%%
1018
1019	elif [[ $numprocs != 0 ]]
1020	then
1021	cat > $job_to_send << %%END%%
1022	#!/bin/ksh
1023	#PBS -q ${queue}
1024	#PBS -o $remote_dayfile
1025	#PBS -N $job_name
1026	#PBS -j o
1027	#PBS -v MPIPROGINV=YES
1028
1029	%%END%%
1030
1031	else
1032	cat > $job_to_send << %%END%%
1033	#!/bin/ksh
1034	#PBS -o $remote_dayfile
1035	#PBS -j o
1036
1037	%%END%%
1038
1039	fi
1040
1041	elif [[ $remote_host = lctit ]]
1042	then
1043	cat > $job_to_send << %%END%%
1044	#!/bin/bash
1045	#$ -S /bin/bash
1046	cd $job_catalog
1047	export PATH=$PALM_BIN:\$PATH
1048	echo \$PATH
1049
1050	%%END%%
1051
1052	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1053	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
1054	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
1055
1056	else
1057
1058	cat > $job_to_send << %%END%%
1059	# @\$-q ${queue}
1060	# @\$-l${qsubtime} $timestring
1061	# @\$-l${qsubmem} ${memory}mb
1062	# @\$-o $remote_dayfile
1063	# @\$-eo
1064
1065	%%END%%
1066
1067	fi
1068
1069
1070	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1071	# BEI EXIT VERANLASSEN
1072	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1073	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1074	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1075	then
1076	echo "set +vx" >> $job_to_send
1077	echo "trap '" >> $job_to_send
1078	echo "set +vx" >> $job_to_send
1079	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lctit ]]
1080	then
1081	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
1082	then
1083	return_queue=c1
1084	elif [[ $remote_host = ibms ]]
1085	then
1086	return_queue=p_normal
1087	elif [[ $remote_host = ibmy ]]
1088	then
1089	return_queue=serial
1090	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1091	then
1092	return_queue=serialq
1093	elif [[ $remote_host = lctit ]]
1094	then
1095	return_queue=sla3
1096	elif [[ $remote_host = necriam ]]
1097	then
1098	return_queue=SP
1099	else
1100	return_queue=unknown
1101	fi
1102
1103	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1104	then
1105
1106	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1107	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1108	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1109	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1110	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1111	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1112	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1113	if [[ $host != "ibmh" ]]
1114	then
1115	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1116	fi
1117	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1118	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1119
1120	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1121	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1122
1123	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1124	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1125	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1126
1127	elif [[ $remote_host = nech ]]
1128	then
1129	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1130	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1131	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1132	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1133	echo "#PBS -j o" >> $job_to_send
1134	echo " " >> $job_to_send
1135	echo "set -x" >> $job_to_send
1136	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1137	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1138	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1139	echo "%%END%%" >> $job_to_send
1140
1141	elif [[ $remote_host = necriam ]]
1142	then
1143	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1144	echo "#PBS -q $return_queue" >> $job_to_send
1145	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1146	echo "#PBS -j o" >> $job_to_send
1147	echo " " >> $job_to_send
1148	echo "set -x" >> $job_to_send
1149	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1150	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1151	echo "%%END%%" >> $job_to_send
1152
1153	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1154	then
1155	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1156	echo "#!/bin/bash" >> $job_to_send
1157	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1158	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1159	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1160	echo "#PBS -l feature=data" >> $job_to_send
1161	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1162	echo "#PBS -j oe" >> $job_to_send
1163	echo " " >> $job_to_send
1164	echo ". /usr/share/modules/init/bash" >> $job_to_send
1165	echo "set -x" >> $job_to_send
1166	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1167	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1168	echo "%%END%%" >> $job_to_send
1169
1170	elif [[ $remote_host = lctit ]]
1171	then
1172	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1173	echo "set -x" >> $job_to_send
1174	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1175	echo "cd $job_catalog" >> $job_to_send
1176	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1177	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1178	echo "%%END%%" >> $job_to_send
1179
1180	else
1181
1182	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1183	echo "# @\\\$-q $return_queue" >> $job_to_send
1184	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1185	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1186	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1187	then
1188	echo "# @\$-l mpp_p=0" >> $job_to_send
1189	fi
1190	echo '# @\$-lF 10mb' >> $job_to_send
1191	# echo '# @\$-o /dev/null' >> $job_to_send
1192	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1193	echo '# @\\\$-eo' >> $job_to_send
1194	echo " " >> $job_to_send
1195	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1196	then
1197	echo "set +vx" >> $job_to_send
1198	echo ". .profile" >> $job_to_send
1199	fi
1200	echo "set -x" >> $job_to_send
1201	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1202	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1203	echo "%%END%%" >> $job_to_send
1204	fi
1205
1206	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1207	then
1208	echo "llsubmit scpjob.$kennung" >> $job_to_send
1209	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1210	then
1211	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1212	echo "msub scpjob.$kennung" >> $job_to_send
1213	elif [[ $remote_host = lctit ]]
1214	then
1215	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1216	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1217	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1218	echo "sleep 10" >> $job_to_send
1219	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1220	then
1221	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1222	elif [[ $remote_host = t3es ]]
1223	then
1224	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1225	else
1226	echo "qsub scpjob.$kennung" >> $job_to_send
1227	fi
1228	echo "rm scpjob.$kennung" >> $job_to_send
1229	if [[ $remote_host = nech ]]
1230	then
1231	echo "cd -" >> $job_to_send
1232	fi
1233	else
1234	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1235	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1236	fi
1237	echo "set -x" >> $job_to_send
1238	echo " ' exit" >> $job_to_send
1239	echo "set -x" >> $job_to_send
1240	fi
1241
1242
1243
1244	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1245	cat $file_to_send >> $job_to_send
1246
1247	if [[ $remote_host = ibm ]]
1248	then
1249	echo " " >> $job_to_send
1250	echo "exit" >> $job_to_send
1251	fi
1252	if [[ $remote_host = lctit ]]
1253	then
1254	echo " " >> $job_to_send
1255	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1256	fi
1257
1258
1259
1260	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1261	if [[ -z $remote_user ]]
1262	then
1263	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1264	then
1265	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1266	fi
1267	fi
1268
1269
1270
1271	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1272	if [[ $no_submit = false ]]
1273	then
1274	if [[ $remote_host != $local_host ]]
1275	then
1276	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1277	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1278	then
1279	job_catalog_save=$job_catalog
1280	job_catalog=job_queue
1281	elif [[ $remote_host = nech ]]
1282	then
1283	job_catalog_save=$job_catalog
1284	job_catalog=/hpf/b/${remote_user}/job_queue
1285	fi
1286	if [[ $local_host = decalpha ]]
1287	then
1288	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1289	# PASSWORT AB
1290	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1291	elif [[ $remote_host = nech ]]
1292	then
1293	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1294	# TRANSFERIERT WERDEN
1295	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1296	else
1297	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1298	fi
1299	if [[ $? = 1 ]]
1300	then
1301	locat=scp; exit
1302	fi
1303	if [[ $remote_host = ibms ]]
1304	then
1305	job_catalog=$job_catalog_save
1306	fi
1307	[[ $verify = true ]] && printf "\n >>> finished\n"
1308	else
1309	eval job_catalog=$job_catalog
1310	cp $job_to_send ${job_catalog}/$job_on_remhost
1311	fi
1312
1313
1314
1315	# NQS- BZW. LOADLEVELER-JOB STARTEN
1316	if [[ $remote_host != $local_host ]]
1317	then
1318	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1319	if [[ $remote_host != lctit ]]
1320	then
1321	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1322	then
1323	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1324	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1325	else
1326	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1327	fi
1328	else
1329	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1330	# MIT SSH, DESHALB AUFRUF PER PIPE
1331	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1332	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1333	echo "first try" > send_protocol
1334	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1335	do
1336	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1337	then
1338	printf "\n trying to submit job to TIT cluster..."
1339	else
1340	printf "\n +++ failed ... trying to submit again ..."
1341	fi
1342	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1343	cat send_protocol
1344	done
1345	sleep 10
1346	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1347	echo "$submcom $job_on_remhost"
1348	rm send_protocol
1349	fi
1350
1351	[[ $verify = true ]] && printf " >>> o.k.\n"
1352	else
1353	cd $job_catalog
1354	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1355	then
1356	eval $submcom $job_on_remhost
1357	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1358	then
1359	eval $submcom $job_on_remhost
1360	echo "$submcom $job_on_remhost"
1361	chmod u+x $job_on_remhost
1362
1363	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1364	then
1365	eval $submcom $job_on_remhost
1366	elif [[ $local_host = nech ]]
1367	then
1368	if [[ $queue = default ]]
1369	then
1370	eval $submcom $job_on_remhost
1371	else
1372	eval $submcom -q $queue $job_on_remhost
1373	fi
1374	else
1375	qsub $job_on_remhost
1376	fi
1377	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1378	[[ $local_host != lctit ]] && rm $job_on_remhost
1379	cd - > /dev/null
1380	fi
1381	fi
1382
1383
1384
1385	# ABSCHLUSSARBEITEN
1386
1387	if [[ $no_submit = false ]]
1388	then
1389	rm $job_to_send
1390	fi
1391	[[ $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 |