Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 412

Last change on this file since 412 was 407, checked in by maronga, 15 years ago
humidity for non-flat topography implemented, re-adjustments for lcxt4
Property svn:keywords set to `Id Rev`
File size: 45.4 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 407 2009-12-01 15:01:15Z 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
113	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
114	delete_dayfile=false
115	email_notification=none
116	group_number=none
117	locat=normal
118	no_default_queue=none
119	no_submit=false
120	job_catalog="~/job_queue"
121	job_name=none
122	local_user=$LOGNAME
123	node_usage=shared
124	numprocs=1
125	punkte="..........................................................."
126	submcom=qsub
127	queue=default
128	remote_host=none
129	remote_user=""
130	verify=true
131
132	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
133	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
134	typeset -L20 spalte1
135	typeset -R40 spalte2
136	typeset -L60 spalte3
137
138
139
140	# FEHLERBEHANDLUNG
141	# BEI EXIT:
142	trap 'if [[ $locat != normal ]]
143	then
144	case $locat in
145	(option) printf "\n --> available optios can be displayed"
146	printf " by typing:"
147	printf "\n \"subjob ?\" \n";;
148	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
149	(*) printf "\n +++ unknown error"
150	printf "\n please inform S. Raasch!\n"
151	esac
152	[[ -f $job_to_send ]] && rm $job_to_send
153	printf "\n\n+++ SUBJOB killed \n\n"
154	fi' exit
155
156
157	# BEI TERMINAL-BREAK:
158	trap '[[ -f $job_to_send ]] && rm $job_to_send
159	printf "\n\n+++ SUBJOB killed \n\n"
160	exit
161	' 2
162
163
164
165
166	# LOKALEN HOSTNAMEN FESTSTELLEN
167	local_host=$(hostname)
168
169
170
171	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
172	# UEBERHAUPT ZULAESSIG IST
173	case $local_host in
174	(atmos) local_addres=172.20.25.35; local_host=lcide;;
175	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
176	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
177	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
178	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
179	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
180	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
181	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
182	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
183	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
184	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
185	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
186	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
187	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
188	(cs*) local_addres=136.172.44.131; local_host=nech;;
189	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
190	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
191	(node*) local_addres=165.132.26.61 local_host=lck;;
192	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
193	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
194	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
195	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
196	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
197	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
198	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
199	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
200	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
201	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
202	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
203	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
204	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
205	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
206	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
207	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
208	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
209	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
210	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
211	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
212	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
213	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
214	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
215	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
216	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
217	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
218	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
219	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
220	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
221	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
222	(*) printf "\n +++ \"$local_host\" unknown";
223	printf "\n please inform S. Raasch!";
224	locat=parameter; exit;;
225	esac
226
227
228
229	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
230	remote_host=$local_host
231
232
233
234
235	# PROZEDUROPTIONEN EINLESEN
236	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
237	do
238	case $option in
239	(c) job_catalog=$OPTARG;;
240	(d) delete_dayfile=true;;
241	(D) no_submit=true;;
242	(e) email_notification=$OPTARG;;
243	(g) group_number=$OPTARG;;
244	(h) remote_host=$OPTARG;;
245	(m) memory=$OPTARG;;
246	(n) job_name=$OPTARG;;
247	(N) node_usage=$OPTARG;;
248	(O) threads_per_task=$OPTARG;;
249	(q) no_default_queue=$OPTARG;;
250	(t) cputime=$OPTARG;;
251	(T) tasks_per_node=$OPTARG;;
252	(u) remote_user=$OPTARG;;
253	(v) verify=false;;
254	(X) numprocs=$OPTARG;;
255	(\?) printf "\n +++ Option $OPTARG unknown \n";
256	locat=option; exit;;
257	esac
258	done
259
260
261	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
262	shift OPTIND-1; file_to_send=$1
263
264
265	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
266	if [ "$1" = "?" ]
267	then
268	(printf "\n *** subjob can be called as follows:\n"
269	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
270	printf "\n Description of available options:\n"
271	printf "\n Option Description Default-Value"
272	printf "\n -c job-input- and output-catalog ~/job_queue"
273	printf "\n -d no job-protocol will be created ---"
274	printf "\n -D only the job-file will be created ---"
275	printf "\n -h execution host, available hosts: $remote_host"
276	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
277	printf "\n lctit, nech, necriam, unics"
278	printf "\n -m memory demand per process in MByte ---"
279	printf "\n -n jobname <jobdatei>"
280	printf "\n -O threads per task (for OpenMP usage) 1"
281	printf "\n -q job-queue to be used default"
282	printf "\n -t allowed cpu-time in seconds ---"
283	printf "\n -T tasks per node (on parallel hosts) ---"
284	printf "\n -u username on execution host from .netrc"
285	printf "\n -v no prompt for confirmation ---"
286	printf "\n -X # of processors (on parallel hosts) 1"
287	printf "\n "
288	printf "\n The only possible positional parameter is <jobfile>:"
289	printf "\n The complete NQS-job must be provided here."
290	printf "\n <jobfile>=? creates this outline\n\n") \| more
291	exit
292	fi
293
294
295
296	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
297	if [[ "$file_to_send" = "" ]]
298	then
299	printf "\n +++ job-file missing"
300	locat=parameter; exit
301	else
302	if [[ -f $file_to_send ]]
303	then
304	true
305	else
306	printf "\n +++ job-file: "
307	printf "\n $file_to_send"
308	printf "\n does not exist"
309	locat=parameter; exit
310	fi
311	fi
312
313
314
315	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
316	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
317	if [[ $job_name = none ]]
318	then
319	job_name=$file_to_send
320	fi
321	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
322	then
323	printf "\n +++ job-file name: "
324	printf "\n $job_name"
325	printf "\n must not contain \"/\"-characters"
326	locat=parameter; exit
327	fi
328
329
330
331
332	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
333	# ODER WENN HOST NICHT ANGEGEBEN WURDE
334	if [[ $remote_host = none ]]
335	then
336	printf "\n +++ host missing"
337	locat=option; exit
338	else
339	case $remote_host in
340	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
341	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
342	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
343	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
344	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
345	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
346	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
347	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
348	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
349	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
350	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
351	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
352	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
353	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
354	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
355	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
356	locat=parameter; exit;;
357	esac
358	fi
359
360
361	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
362	if [[ $no_default_queue != none ]]
363	then
364	error=false
365	ndq=$no_default_queue
366	case $remote_host in
367	(ibm) case $ndq in
368	(p690_express\|p690_standard\|p690_long) error=false;;
369	(*) error=true;;
370	esac;;
371	(ibmb) case $ndq in
372	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
373	(*) error=true;;
374	esac;;
375	(ibmh) case $ndq in
376	(no_class) error=false;;
377	(*) error=true;;
378	esac;;
379	(ibms) case $ndq in
380	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
381	(*) error=true;;
382	esac;;
383	(ibmy) case $ndq in
384	(parallel) error=false;;
385	(*) error=true;;
386	esac;;
387	(lcsgib\|lcsgih) case $ndq in
388	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|special1q) error=false;;
389	(*) error=true;;
390	esac;;
391	(lctit) case $ndq in
392	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
393	(*) error=true;;
394	esac;;
395	(t3eb) case $ndq in
396	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
397	(*) error=true;;
398	esac;;
399	(necriam) case $ndq in
400	(SP\|SS\|P6) error=false;;
401	(*) error=true;;
402	esac;;
403	(t3eh) case $ndq in
404	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
405	(*) error=true;;
406	esac;;
407	(t3ej2\|t3ej5) case $ndq in
408	(low\|normal\|high) error=false;;
409	(*) error=true;;
410	esac;;
411	(t3es) case $ndq in
412	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
413	(*) error=true;;
414	esac;;
415	(unics) case $ndq in
416	(unics\|ht) error=false;;
417	(*) error=true;;
418	esac;;
419	esac
420	if [[ $error = true ]]
421	then
422	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
423	locat=parameter; exit
424	else
425	queue=$no_default_queue
426	fi
427	fi
428
429
430
431	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
432	if [[ $node_usage != shared && $queue = cdev ]]
433	then
434	node_usage=shared
435	fi
436
437
438
439	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
440	# AUFTEILEN
441	done=false
442	while [[ $done = false ]]
443	do
444	if (( $cputime <= 0 ))
445	then
446	printf "\n +++ wrong cpu-time or cpu-time missing"
447	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
448	printf "\n >>> "
449	read cputime 1>/dev/null 2>&1
450	else
451	done=true
452	fi
453	done
454	if [[ $remote_host = nech ]]
455	then
456	if (( tasks_per_node != 0 ))
457	then
458	(( cputime = cputime * tasks_per_node ))
459	elif [[ $numprocs != 0 ]]
460	then
461	(( cputime = cputime * numprocs ))
462	fi
463	fi
464	(( stunden = cputime / 3600 ))
465	(( resttime = cputime - stunden * 3600 ))
466	(( minuten = resttime / 60 ))
467	(( sekunden = resttime - minuten * 60 ))
468	timestring=${stunden}:${minuten}:${sekunden}
469
470
471
472	# PRUEFEN DER KERNSPEICHERANFORDERUNG
473	done=false
474	while [[ $done = false ]]
475	do
476	if (( memory <= 0 ))
477	then
478	printf "\n +++ wrong memory demand or memory demand missing"
479	printf "\n >>> Please type memory in MByte per process as INTEGER:"
480	printf "\n >>> "
481	read memory 1>/dev/null 2>&1
482	else
483	done=true
484	fi
485	done
486
487	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
488	then
489	if (( tasks_per_node != 0 ))
490	then
491	(( Memory = memory * tasks_per_node / 1000 ))
492	elif [[ $numprocs != 0 ]]
493	then
494	(( Memory = memory * numprocs / 1000 ))
495	else
496	(( Memory = memory / 1000 ))
497	fi
498	fi
499
500
501	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
502	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
503	then
504	(( memory = memory * threads_per_task ))
505	fi
506
507
508	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
509	if (( tasks_per_node != 0 ))
510	then
511	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
512	then
513	(( nodes = numprocs / tasks_per_node ))
514	else
515	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
516	fi
517	fi
518
519
520
521	# HEADER-AUSGABE
522	if [[ $verify = true ]]
523	then
524	printf "\n\n"
525	printf "#--------------------------------------------------------------# \n"
526	spalte1=SUBJOB;spalte2=$(date)
527	printf "\| $spalte1$spalte2 \| \n"
528	printf "\| \| \n"
529	printf "\| values of parameters/options: \| \n"
530	spalte1=local_host$punkte; spalte2=$punkte$local_host
531	printf "\| $spalte1$spalte2 \| \n"
532	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
533	printf "\| $spalte1$spalte2 \| \n"
534	spalte1=queue$punkte; spalte2=$punkte$queue
535	printf "\| $spalte1$spalte2 \| \n"
536	spalte1=memory$punkte; spalte2="$punkte$memory mb"
537	printf "\| $spalte1$spalte2 \| \n"
538	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
539	printf "\| $spalte1$spalte2 \| \n"
540	spalte1=job_name$punkte; spalte2="$punkte$job_name"
541	printf "\| $spalte1$spalte2 \| \n"
542	printf "#--------------------------------------------------------------# \n\n"
543
544
545	# KONTROLLABFRAGE, OB ALLES O.K.
546	antwort="dummy"
547	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
548	do
549	read antwort?" >>> continue (y/n) ? "
550	done
551	if [[ $antwort = n \|\| $antwort = N ]]
552	then
553	locat=verify; exit
554	fi
555	printf "\n"
556	fi
557
558
559
560	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
561	kennung=$RANDOM
562	job_on_remhost=${job_name}_${kennung}_$local_host
563	job_to_send=job_to_send_$kennung
564	if [[ $delete_dayfile = false ]]
565	then
566	remote_dayfile=${local_host}_${job_name}_result_$kennung
567	local_dayfile=${remote_host}_${job_name}
568	else
569	remote_dayfile=/dev/null
570	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
571	then
572	notification='# @ notification = never'
573	fi
574	fi
575
576
577
578	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
579	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
580	then
581
582	if [[ $remote_host = ibmy ]]
583	then
584	consumable_memory=""
585	else
586	consumable_memory="ConsumableMemory($memory mb)"
587	fi
588
589	if [[ $remote_host = ibmh ]]
590	then
591	data_limit=""
592	network_to_use=""
593	class=""
594	environment=""
595	else
596	class="# @ class = $queue"
597	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
598	if [[ $queue = cdev ]]
599	then
600	data_limit="# @ data_limit = 1.76gb"
601	network_to_use="# @ network.mpi = sn_all,shared,ip"
602	else
603	if [[ $remote_host = ibms ]]
604	then
605	network_to_use="# @ network.mpi = csss,shared,us"
606	elif [[ $remote_host = ibmy ]]
607	then
608	network_to_use=""
609	else
610	network_to_use="# @ network.mpi = sn_all,shared,us"
611	data_limit="# @ data_limit = 1.76gb"
612	fi
613	fi
614	fi
615
616	cat > $job_to_send << %%END%%
617	#!/bin/ksh
618	# @ shell = /bin/ksh
619
620	# @ job_type = parallel
621	# @ job_name = palm
622	# @ wall_clock_limit = ${timestring},$timestring
623	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
624	# @ output = $remote_dayfile
625	# @ error = $remote_dayfile
626	# @ image_size = 50
627	$class
628	$environment
629	$network_to_use
630	$data_limit
631	$notification
632
633	%%END%%
634
635	if (( nodes > 0 ))
636	then
637
638	cat >> $job_to_send << %%END%%
639	# @ node = $nodes
640	# @ tasks_per_node = $tasks_per_node
641	# @ node_usage = $node_usage
642	# @ queue
643
644	%%END%%
645
646	else
647
648	if [[ $remote_host != ibmy ]]
649	then
650
651	cat >> $job_to_send << %%END%%
652	# @ blocking = unlimited
653	# @ total_tasks = $numprocs
654	# @ node_usage = $node_usage
655	# @ queue
656
657	%%END%%
658
659	else
660
661	cat >> $job_to_send << %%END%%
662	# @ node = 1
663	# @ total_tasks = $numprocs
664	# @ queue
665
666	%%END%%
667
668	fi
669
670	fi
671
672	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
673	then
674
675	cat > $job_to_send << %%END%%
676	#!/bin/ksh
677
678	# @ job_type = serial
679	# @ node_usage = $node_usage
680	# @ job_name = palm
681	# @ wall_clock_limit = ${timestring},$timestring
682	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
683	# @ output = $remote_dayfile
684	# @ error = $remote_dayfile
685	$class
686	$notification
687
688	# @ queue
689
690	%%END%%
691
692	elif [[ $remote_host = lcfimm ]]
693	then
694
695	if [[ $numprocs != 0 ]]
696	then
697	cat > $job_to_send << %%END%%
698	#!/bin/ksh
699	#PBS -N $job_name
700	#PBS -A geofysisk
701	#PBS -l walltime=$timestring
702	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
703	#PBS -l pmem=${memory}mb
704	#PBS -m abe
705	#PBS -M maronga@muk.uni-hannover.de
706	#PBS -o $remote_dayfile
707	#PBS -j oe
708	mpd &
709
710	%%END%%
711
712	else
713	cat > $job_to_send << %%END%%
714	#!/bin/ksh
715	#PBS -N $job_name
716	#PBS -A geofysisk
717	#PBS -l walltime=$timestring
718	#PBS -l ncpus=1
719	#PBS -l pmem=${memory}mb
720	#PBS -m abe
721	#PBS -M maronga@muk.uni-hannover.de
722	#PBS -o $remote_dayfile
723	#PBS -j oe
724
725	%%END%%
726
727	fi
728
729	elif [[ $remote_host = lck ]]
730	then
731
732	if [[ $numprocs != 0 ]]
733	then
734	cat > $job_to_send << %%END%%
735	#!/bin/ksh
736	#PBS -N $job_name
737	#PBS -l walltime=$timestring
738	#PBS -l ncpus=$numprocs
739	#PBS -l pmem=${memory}mb
740	#PBS -o $remote_dayfile
741	#PBS -j oe
742
743	mpd &
744
745	%%END%%
746
747	else
748	cat > $job_to_send << %%END%%
749	#!/bin/ksh
750	#PBS -N $job_name
751	#PBS -l walltime=$timestring
752	#PBS -l ncpus=1
753	#PBS -l pmem=${memory}mb
754	#PBS -o $remote_dayfile
755	#PBS -j oe
756
757	%%END%%
758
759	fi
760
761	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
762	then
763
764	if [[ "$sgi_feature" != "" ]]
765	then
766	feature_directive="#PBS -l feature=$sgi_feature"
767	else
768	feature_directive=""
769	fi
770
771	if [[ $queue = dataq ]]
772	then
773	feature_directive="#PBS -l feature=data"
774	fi
775
776	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = serialq \|\| $queue = special1q ]]
777	then
778	queue_directive="#PBS -q $queue"
779	else
780	queue_directive=""
781	fi
782
783	if [[ $email_notification = none ]]
784	then
785	email_directive=""
786	else
787	email_directive="#PBS -M $email_notification"
788	fi
789
790	if [[ $numprocs != 0 ]]
791	then
792	cat > $job_to_send << %%END%%
793	#!/bin/bash
794	#PBS -N $job_name
795	#PBS -l walltime=$timestring
796	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
797	#PBS -l naccesspolicy=$node_usage
798	#PBS -o $remote_dayfile
799	#PBS -j oe
800	$feature_directive
801	$queue_directive
802	$email_directive
803
804	. /usr/share/modules/init/bash
805	module load ifort/11.0.069
806	module load $mpilib
807	module load netcdf
808
809	echo ld_library_path=\$LD_LIBRARY_PATH
810
811	%%END%%
812
813	else
814	cat > $job_to_send << %%END%%
815	#PBS -S /bin/bash
816	#PBS -N $job_name
817	#PBS -l walltime=$timestring
818	#PBS -l ncpus=1
819	#PBS -o $remote_dayfile
820	#PBS -e $remote_dayfile
821	$queue_directive
822	$email_directive
823
824	%%END%%
825
826	fi
827
828	elif [[ $remote_host = lcxt4 ]]
829	then
830
831	if [[ $numprocs != 0 ]]
832	then
833	cat > $job_to_send << %%END%%
834	#!/bin/ksh
835	#PBS -S /bin/ksh
836	#PBS -N $job_name
837	#PBS -A geofysisk
838	#PBS -j oe
839	#PBS -l walltime=$timestring
840	#PBS -l mppwidth=${numprocs}
841	#PBS -l mppnppn=${tasks_per_node}
842	#PBS -m abe
843	#PBS -M maronga@muk.uni-hannover.de
844	#PBS -o $remote_dayfile
845
846	. /opt/modules/default/init/ksh
847	module load modules/3.1.6
848	module load PrgEnv-pgi/2.1.56HDA
849	module load Base-opts/2.1.56HDA
850	module load moab/5.2.5-s13570
851	module load torque/2.3.4-snap.200809221601
852
853
854	%%END%%
855
856	else
857	cat > $job_to_send << %%END%%
858	#!/bin/ksh
859	#PBS -S /bin/ksh
860	#PBS -N $job_name
861	#PBS -A geofysisk
862	#PBS -j oe
863	#PBS -l walltime=$timestring
864	#PBS -l ncpus=1
865	#PBS -l pmem=${memory}mb
866	#PBS -m abe
867	#PBS -M maronga@muk.uni-hannover.de
868	#PBS -o $remote_dayfile
869
870	. /opt/modules/default/init/ksh
871	module load modules/3.1.6
872	module load PrgEnv-pgi/2.1.56HD
873	module load Base-opts/2.1.56HD
874	module load moab/5.2.5-s13441
875	module load torque/2.3.4-snap.200809221601
876
877	%%END%%
878
879	fi
880
881	elif [[ $remote_host = nech ]]
882	then
883
884	if (( nodes > 1 ))
885	then
886	cat > $job_to_send << %%END%%
887	#!/bin/ksh
888	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
889	#PBS -l ${qsubmem}=${Memory}gb
890	#PBS -b $nodes
891	#PBS -o $remote_dayfile
892	#PBS -N palm
893	#PBS -j o
894	#PBS -T mpisx
895
896	%%END%%
897
898	elif [[ $numprocs != 0 ]]
899	then
900	cat > $job_to_send << %%END%%
901	#!/bin/ksh
902	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
903	#PBS -l ${qsubmem}=${Memory}gb
904	#PBS -o $remote_dayfile
905	#PBS -N palm
906	#PBS -j o
907
908	%%END%%
909
910	else
911	cat > $job_to_send << %%END%%
912	#!/bin/ksh
913	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
914	#PBS -o $remote_dayfile
915	#PBS -j o
916
917	%%END%%
918
919	fi
920
921	elif [[ $remote_host = necriam ]]
922	then
923
924	if (( nodes > 1 ))
925	then
926	cat > $job_to_send << %%END%%
927	#!/bin/ksh
928	#PBS -b $nodes
929	#PBS -o $remote_dayfile
930	#PBS -N $job_name
931	#PBS -j o
932	#PBS -v MPIPROGINV=YES
933
934	%%END%%
935
936	elif [[ $numprocs != 0 ]]
937	then
938	cat > $job_to_send << %%END%%
939	#!/bin/ksh
940	#PBS -q ${queue}
941	#PBS -o $remote_dayfile
942	#PBS -N $job_name
943	#PBS -j o
944	#PBS -v MPIPROGINV=YES
945
946	%%END%%
947
948	else
949	cat > $job_to_send << %%END%%
950	#!/bin/ksh
951	#PBS -o $remote_dayfile
952	#PBS -j o
953
954	%%END%%
955
956	fi
957
958	elif [[ $remote_host = lctit ]]
959	then
960	cat > $job_to_send << %%END%%
961	#!/bin/bash
962	#$ -S /bin/bash
963	cd $job_catalog
964	export PATH=$PALM_BIN:\$PATH
965	echo \$PATH
966
967	%%END%%
968
969	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
970	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
971	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
972
973	else
974
975	cat > $job_to_send << %%END%%
976	# @\$-q ${queue}
977	# @\$-l${qsubtime} $timestring
978	# @\$-l${qsubmem} ${memory}mb
979	# @\$-o $remote_dayfile
980	# @\$-eo
981
982	%%END%%
983
984	fi
985
986
987	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
988	# BEI EXIT VERANLASSEN
989	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
990	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
991	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
992	then
993	echo "set +vx" >> $job_to_send
994	echo "trap '" >> $job_to_send
995	echo "set +vx" >> $job_to_send
996	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 ]]
997	then
998	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
999	then
1000	return_queue=c1
1001	elif [[ $remote_host = ibms ]]
1002	then
1003	return_queue=p_normal
1004	elif [[ $remote_host = ibmy ]]
1005	then
1006	return_queue=serial
1007	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1008	then
1009	return_queue=serialq
1010	elif [[ $remote_host = lctit ]]
1011	then
1012	return_queue=sla3
1013	elif [[ $remote_host = necriam ]]
1014	then
1015	return_queue=SP
1016	else
1017	return_queue=unknown
1018	fi
1019
1020	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1021	then
1022
1023	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1024	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1025	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1026	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1027	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1028	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1029	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1030	if [[ $host != "ibmh" ]]
1031	then
1032	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1033	fi
1034	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1035	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1036
1037	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1038	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1039
1040	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1041	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1042	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1043
1044	elif [[ $remote_host = nech ]]
1045	then
1046	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1047	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1048	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1049	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1050	echo "#PBS -j o" >> $job_to_send
1051	echo " " >> $job_to_send
1052	echo "set -x" >> $job_to_send
1053	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1054	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1055	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1056	echo "%%END%%" >> $job_to_send
1057
1058	elif [[ $remote_host = necriam ]]
1059	then
1060	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1061	echo "#PBS -q $return_queue" >> $job_to_send
1062	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1063	echo "#PBS -j o" >> $job_to_send
1064	echo " " >> $job_to_send
1065	echo "set -x" >> $job_to_send
1066	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1067	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1068	echo "%%END%%" >> $job_to_send
1069
1070	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1071	then
1072	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1073	echo "#!/bin/bash" >> $job_to_send
1074	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1075	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1076	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1077	echo "#PBS -l feature=data" >> $job_to_send
1078	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1079	echo "#PBS -j oe" >> $job_to_send
1080	echo " " >> $job_to_send
1081	echo ". /usr/share/modules/init/bash" >> $job_to_send
1082	echo "set -x" >> $job_to_send
1083	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1084	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1085	echo "%%END%%" >> $job_to_send
1086
1087	elif [[ $remote_host = lctit ]]
1088	then
1089	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1090	echo "set -x" >> $job_to_send
1091	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1092	echo "cd $job_catalog" >> $job_to_send
1093	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1094	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1095	echo "%%END%%" >> $job_to_send
1096
1097	else
1098
1099	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1100	echo "# @\\\$-q $return_queue" >> $job_to_send
1101	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1102	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1103	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1104	then
1105	echo "# @\$-l mpp_p=0" >> $job_to_send
1106	fi
1107	echo '# @\$-lF 10mb' >> $job_to_send
1108	# echo '# @\$-o /dev/null' >> $job_to_send
1109	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1110	echo '# @\\\$-eo' >> $job_to_send
1111	echo " " >> $job_to_send
1112	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1113	then
1114	echo "set +vx" >> $job_to_send
1115	echo ". .profile" >> $job_to_send
1116	fi
1117	echo "set -x" >> $job_to_send
1118	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1119	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1120	echo "%%END%%" >> $job_to_send
1121	fi
1122
1123	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1124	then
1125	echo "llsubmit scpjob.$kennung" >> $job_to_send
1126	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1127	then
1128	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1129	echo "msub scpjob.$kennung" >> $job_to_send
1130	elif [[ $remote_host = lctit ]]
1131	then
1132	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1133	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1134	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1135	echo "sleep 10" >> $job_to_send
1136	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1137	then
1138	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1139	elif [[ $remote_host = t3es ]]
1140	then
1141	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1142	else
1143	echo "qsub scpjob.$kennung" >> $job_to_send
1144	fi
1145	echo "rm scpjob.$kennung" >> $job_to_send
1146	if [[ $remote_host = nech ]]
1147	then
1148	echo "cd -" >> $job_to_send
1149	fi
1150	else
1151	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1152	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1153	fi
1154	echo "set -x" >> $job_to_send
1155	echo " ' exit" >> $job_to_send
1156	echo "set -x" >> $job_to_send
1157	fi
1158
1159
1160
1161	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1162	cat $file_to_send >> $job_to_send
1163
1164	if [[ $remote_host = ibm ]]
1165	then
1166	echo " " >> $job_to_send
1167	echo "exit" >> $job_to_send
1168	fi
1169	if [[ $remote_host = lctit ]]
1170	then
1171	echo " " >> $job_to_send
1172	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1173	fi
1174
1175
1176
1177	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1178	if [[ -z $remote_user ]]
1179	then
1180	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1181	then
1182	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1183	fi
1184	fi
1185
1186
1187
1188	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1189	if [[ $no_submit = false ]]
1190	then
1191	if [[ $remote_host != $local_host ]]
1192	then
1193	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1194	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1195	then
1196	job_catalog_save=$job_catalog
1197	job_catalog=job_queue
1198	elif [[ $remote_host = nech ]]
1199	then
1200	job_catalog_save=$job_catalog
1201	job_catalog=/hpf/b/${remote_user}/job_queue
1202	fi
1203	if [[ $local_host = decalpha ]]
1204	then
1205	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1206	# PASSWORT AB
1207	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1208	elif [[ $remote_host = nech ]]
1209	then
1210	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1211	# TRANSFERIERT WERDEN
1212	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1213	else
1214	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1215	fi
1216	if [[ $? = 1 ]]
1217	then
1218	locat=scp; exit
1219	fi
1220	if [[ $remote_host = ibms ]]
1221	then
1222	job_catalog=$job_catalog_save
1223	fi
1224	[[ $verify = true ]] && printf "\n >>> finished\n"
1225	else
1226	eval job_catalog=$job_catalog
1227	cp $job_to_send ${job_catalog}/$job_on_remhost
1228	fi
1229
1230
1231
1232	# NQS- BZW. LOADLEVELER-JOB STARTEN
1233	if [[ $remote_host != $local_host ]]
1234	then
1235	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1236	if [[ $remote_host != lctit ]]
1237	then
1238	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1239	then
1240	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1241	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1242	else
1243	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1244	fi
1245	else
1246	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1247	# MIT SSH, DESHALB AUFRUF PER PIPE
1248	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1249	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1250	echo "first try" > send_protocol
1251	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1252	do
1253	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1254	then
1255	printf "\n trying to submit job to TIT cluster..."
1256	else
1257	printf "\n +++ failed ... trying to submit again ..."
1258	fi
1259	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1260	cat send_protocol
1261	done
1262	sleep 10
1263	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1264	echo "$submcom $job_on_remhost"
1265	rm send_protocol
1266	fi
1267
1268	[[ $verify = true ]] && printf " >>> o.k.\n"
1269	else
1270	cd $job_catalog
1271	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1272	then
1273	eval $submcom $job_on_remhost
1274	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1275	then
1276	eval $submcom $job_on_remhost
1277	echo "$submcom $job_on_remhost"
1278	chmod u+x $job_on_remhost
1279
1280	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1281	then
1282	eval $submcom $job_on_remhost
1283	elif [[ $local_host = nech ]]
1284	then
1285	if [[ $queue = default ]]
1286	then
1287	eval $submcom $job_on_remhost
1288	else
1289	eval $submcom -q $queue $job_on_remhost
1290	fi
1291	else
1292	qsub $job_on_remhost
1293	fi
1294	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1295	[[ $local_host != lctit ]] && rm $job_on_remhost
1296	cd - > /dev/null
1297	fi
1298	fi
1299
1300
1301
1302	# ABSCHLUSSARBEITEN
1303
1304	if [[ $no_submit = false ]]
1305	then
1306	rm $job_to_send
1307	fi
1308	[[ $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 |