Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 481

Last change on this file since 481 was 475, checked in by raasch, 15 years ago
make options can be set in configuration file; bugfix for generating serial jobs; bugfix for masked output
Property svn:keywords set to `Id Rev`
File size: 47.5 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 475 2010-02-04 02:26:16Z 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	# 03/02/10 - Siggi - bug in serial jobs removed
115
116
117
118	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
119	delete_dayfile=false
120	email_notification=none
121	group_number=none
122	locat=normal
123	no_default_queue=none
124	no_submit=false
125	job_catalog="~/job_queue"
126	job_name=none
127	local_user=$LOGNAME
128	node_usage=shared
129	numprocs=0
130	punkte="..........................................................."
131	submcom=qsub
132	queue=default
133	remote_host=none
134	remote_user=""
135	verify=true
136
137	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
138	typeset -i inumprocs nodes=0 tasks_per_node=0 threads_per_task=1
139	typeset -L20 spalte1
140	typeset -R40 spalte2
141	typeset -L60 spalte3
142
143
144
145	# FEHLERBEHANDLUNG
146	# BEI EXIT:
147	trap 'if [[ $locat != normal ]]
148	then
149	case $locat in
150	(option) printf "\n --> available optios can be displayed"
151	printf " by typing:"
152	printf "\n \"subjob ?\" \n";;
153	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
154	(*) printf "\n +++ unknown error"
155	printf "\n please inform S. Raasch!\n"
156	esac
157	[[ -f $job_to_send ]] && rm $job_to_send
158	printf "\n\n+++ SUBJOB killed \n\n"
159	fi' exit
160
161
162	# BEI TERMINAL-BREAK:
163	trap '[[ -f $job_to_send ]] && rm $job_to_send
164	printf "\n\n+++ SUBJOB killed \n\n"
165	exit
166	' 2
167
168
169
170
171	# LOKALEN HOSTNAMEN FESTSTELLEN
172	local_host=$(hostname)
173
174
175
176	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
177	# UEBERHAUPT ZULAESSIG IST
178	case $local_host in
179	(atmos) local_addres=172.20.25.35; local_host=lcide;;
180	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
181	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
182	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
183	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
184	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
185	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
186	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
187	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
188	(breg-en0\|berni-en0) local_addres=130.73.230.10; local_host=ibmb;;
189	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
190	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
191	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
192	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
193	(cs*) local_addres=136.172.44.131; local_host=nech;;
194	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
195	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
196	(node*) local_addres=165.132.26.61 local_host=lck;;
197	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
198	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
199	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
200	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
201	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
202	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
203	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
204	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
205	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
206	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
207	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
208	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
209	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
210	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
211	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
212	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
213	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
214	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
215	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
216	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
217	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
218	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
219	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
220	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
221	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
222	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
223	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
224	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
225	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
226	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
227	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
228	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
229	(*) printf "\n +++ \"$local_host\" unknown";
230	printf "\n please inform S. Raasch!";
231	locat=parameter; exit;;
232	esac
233
234
235
236	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
237	remote_host=$local_host
238
239
240
241
242	# PROZEDUROPTIONEN EINLESEN
243	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
244	do
245	case $option in
246	(c) job_catalog=$OPTARG;;
247	(d) delete_dayfile=true;;
248	(D) no_submit=true;;
249	(e) email_notification=$OPTARG;;
250	(g) group_number=$OPTARG;;
251	(h) remote_host=$OPTARG;;
252	(m) memory=$OPTARG;;
253	(n) job_name=$OPTARG;;
254	(N) node_usage=$OPTARG;;
255	(O) threads_per_task=$OPTARG;;
256	(q) no_default_queue=$OPTARG;;
257	(t) cputime=$OPTARG;;
258	(T) tasks_per_node=$OPTARG;;
259	(u) remote_user=$OPTARG;;
260	(v) verify=false;;
261	(X) numprocs=$OPTARG;;
262	(\?) printf "\n +++ Option $OPTARG unknown \n";
263	locat=option; exit;;
264	esac
265	done
266
267
268	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
269	shift OPTIND-1; file_to_send=$1
270
271
272	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
273	if [ "$1" = "?" ]
274	then
275	(printf "\n *** subjob can be called as follows:\n"
276	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
277	printf "\n Description of available options:\n"
278	printf "\n Option Description Default-Value"
279	printf "\n -c job-input- and output-catalog ~/job_queue"
280	printf "\n -d no job-protocol will be created ---"
281	printf "\n -D only the job-file will be created ---"
282	printf "\n -h execution host, available hosts: $remote_host"
283	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
284	printf "\n lctit, nech, necriam, unics"
285	printf "\n -m memory demand per process in MByte ---"
286	printf "\n -n jobname <jobdatei>"
287	printf "\n -O threads per task (for OpenMP usage) 1"
288	printf "\n -q job-queue to be used default"
289	printf "\n -t allowed cpu-time in seconds ---"
290	printf "\n -T tasks per node (on parallel hosts) ---"
291	printf "\n -u username on execution host from .netrc"
292	printf "\n -v no prompt for confirmation ---"
293	printf "\n -X # of processors (on parallel hosts) 1"
294	printf "\n "
295	printf "\n The only possible positional parameter is <jobfile>:"
296	printf "\n The complete NQS-job must be provided here."
297	printf "\n <jobfile>=? creates this outline\n\n") \| more
298	exit
299	fi
300
301
302
303	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
304	if [[ "$file_to_send" = "" ]]
305	then
306	printf "\n +++ job-file missing"
307	locat=parameter; exit
308	else
309	if [[ -f $file_to_send ]]
310	then
311	true
312	else
313	printf "\n +++ job-file: "
314	printf "\n $file_to_send"
315	printf "\n does not exist"
316	locat=parameter; exit
317	fi
318	fi
319
320
321
322	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
323	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
324	if [[ $job_name = none ]]
325	then
326	job_name=$file_to_send
327	fi
328	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
329	then
330	printf "\n +++ job-file name: "
331	printf "\n $job_name"
332	printf "\n must not contain \"/\"-characters"
333	locat=parameter; exit
334	fi
335
336
337
338
339	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
340	# ODER WENN HOST NICHT ANGEGEBEN WURDE
341	if [[ $remote_host = none ]]
342	then
343	printf "\n +++ host missing"
344	locat=option; exit
345	else
346	case $remote_host in
347	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
348	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
349	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
350	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
351	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
352	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
353	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
354	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
355	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
356	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
357	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
358	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
359	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
360	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
361	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
362	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
363	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
364	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
365	locat=parameter; exit;;
366	esac
367	fi
368
369
370	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
371	if [[ $no_default_queue != none ]]
372	then
373	error=false
374	ndq=$no_default_queue
375	case $remote_host in
376	(ibm) case $ndq in
377	(p690_express\|p690_standard\|p690_long) error=false;;
378	(*) error=true;;
379	esac;;
380	(ibmb) case $ndq in
381	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
382	(*) error=true;;
383	esac;;
384	(ibmh) case $ndq in
385	(no_class) error=false;;
386	(*) error=true;;
387	esac;;
388	(ibms) case $ndq in
389	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
390	(*) error=true;;
391	esac;;
392	(ibmy) case $ndq in
393	(parallel) error=false;;
394	(*) error=true;;
395	esac;;
396	(lckyoto) case $ndq in
397	(eh\|ph) error=false;;
398	(*) error=true;;
399	esac;;
400	(lcsgib\|lcsgih) case $ndq in
401	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
402	(*) error=true;;
403	esac;;
404	(lctit) case $ndq in
405	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
406	(*) error=true;;
407	esac;;
408	(t3eb) case $ndq in
409	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
410	(*) error=true;;
411	esac;;
412	(necriam) case $ndq in
413	(SP\|SS\|P6) error=false;;
414	(*) error=true;;
415	esac;;
416	(t3eh) case $ndq in
417	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
418	(*) error=true;;
419	esac;;
420	(t3ej2\|t3ej5) case $ndq in
421	(low\|normal\|high) error=false;;
422	(*) error=true;;
423	esac;;
424	(t3es) case $ndq in
425	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
426	(*) error=true;;
427	esac;;
428	(unics) case $ndq in
429	(unics\|ht) error=false;;
430	(*) error=true;;
431	esac;;
432	esac
433	if [[ $error = true ]]
434	then
435	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
436	locat=parameter; exit
437	else
438	queue=$no_default_queue
439	fi
440	fi
441
442
443
444	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
445	if [[ $node_usage != shared && $queue = cdev ]]
446	then
447	node_usage=shared
448	fi
449
450
451
452	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
453	# AUFTEILEN
454	done=false
455	while [[ $done = false ]]
456	do
457	if (( $cputime <= 0 ))
458	then
459	printf "\n +++ wrong cpu-time or cpu-time missing"
460	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
461	printf "\n >>> "
462	read cputime 1>/dev/null 2>&1
463	else
464	done=true
465	fi
466	done
467	if [[ $remote_host = nech ]]
468	then
469	if (( tasks_per_node != 0 ))
470	then
471	(( cputime = cputime * tasks_per_node ))
472	elif [[ $numprocs != 0 ]]
473	then
474	(( cputime = cputime * numprocs ))
475	fi
476	fi
477	(( stunden = cputime / 3600 ))
478	(( resttime = cputime - stunden * 3600 ))
479	(( minuten = resttime / 60 ))
480	(( sekunden = resttime - minuten * 60 ))
481	timestring=${stunden}:${minuten}:${sekunden}
482
483
484
485	# PRUEFEN DER KERNSPEICHERANFORDERUNG
486	done=false
487	while [[ $done = false ]]
488	do
489	if (( memory <= 0 ))
490	then
491	printf "\n +++ wrong memory demand or memory demand missing"
492	printf "\n >>> Please type memory in MByte per process as INTEGER:"
493	printf "\n >>> "
494	read memory 1>/dev/null 2>&1
495	else
496	done=true
497	fi
498	done
499
500	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
501	then
502	if (( tasks_per_node != 0 ))
503	then
504	(( Memory = memory * tasks_per_node / 1000 ))
505	elif [[ $numprocs != 0 ]]
506	then
507	(( Memory = memory * numprocs / 1000 ))
508	else
509	(( Memory = memory / 1000 ))
510	fi
511	fi
512
513
514	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
515	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
516	then
517	(( memory = memory * threads_per_task ))
518	fi
519
520
521	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
522	if (( tasks_per_node != 0 ))
523	then
524	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
525	then
526	(( nodes = numprocs / tasks_per_node ))
527	else
528	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
529	fi
530	fi
531
532
533
534	# HEADER-AUSGABE
535	if [[ $verify = true ]]
536	then
537	printf "\n\n"
538	printf "#--------------------------------------------------------------# \n"
539	spalte1=SUBJOB;spalte2=$(date)
540	printf "\| $spalte1$spalte2 \| \n"
541	printf "\| \| \n"
542	printf "\| values of parameters/options: \| \n"
543	spalte1=local_host$punkte; spalte2=$punkte$local_host
544	printf "\| $spalte1$spalte2 \| \n"
545	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
546	printf "\| $spalte1$spalte2 \| \n"
547	spalte1=queue$punkte; spalte2=$punkte$queue
548	printf "\| $spalte1$spalte2 \| \n"
549	spalte1=memory$punkte; spalte2="$punkte$memory mb"
550	printf "\| $spalte1$spalte2 \| \n"
551	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
552	printf "\| $spalte1$spalte2 \| \n"
553	spalte1=job_name$punkte; spalte2="$punkte$job_name"
554	printf "\| $spalte1$spalte2 \| \n"
555	printf "#--------------------------------------------------------------# \n\n"
556
557
558	# KONTROLLABFRAGE, OB ALLES O.K.
559	antwort="dummy"
560	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
561	do
562	read antwort?" >>> continue (y/n) ? "
563	done
564	if [[ $antwort = n \|\| $antwort = N ]]
565	then
566	locat=verify; exit
567	fi
568	printf "\n"
569	fi
570
571
572
573	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
574	kennung=$RANDOM
575	job_on_remhost=${job_name}_${kennung}_$local_host
576	job_to_send=job_to_send_$kennung
577	if [[ $delete_dayfile = false ]]
578	then
579	remote_dayfile=${local_host}_${job_name}_result_$kennung
580	local_dayfile=${remote_host}_${job_name}
581	else
582	remote_dayfile=/dev/null
583	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
584	then
585	notification='# @ notification = never'
586	fi
587	fi
588
589
590
591	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
592	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
593	then
594
595	if [[ $remote_host = ibmy ]]
596	then
597	consumable_memory=""
598	else
599	consumable_memory="ConsumableMemory($memory mb)"
600	fi
601
602	if [[ $remote_host = ibmh ]]
603	then
604	data_limit=""
605	network_to_use=""
606	class=""
607	environment=""
608	else
609	class="# @ class = $queue"
610	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
611	if [[ $queue = cdev ]]
612	then
613	data_limit="# @ data_limit = 1.76gb"
614	network_to_use="# @ network.mpi = sn_all,shared,ip"
615	else
616	if [[ $remote_host = ibms ]]
617	then
618	network_to_use="# @ network.mpi = csss,shared,us"
619	elif [[ $remote_host = ibmy ]]
620	then
621	network_to_use=""
622	else
623	network_to_use="# @ network.mpi = sn_all,shared,us"
624	data_limit="# @ data_limit = 1.76gb"
625	fi
626	fi
627	fi
628
629	cat > $job_to_send << %%END%%
630	#!/bin/ksh
631	# @ shell = /bin/ksh
632
633	# @ job_type = parallel
634	# @ job_name = palm
635	# @ wall_clock_limit = ${timestring},$timestring
636	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
637	# @ output = $remote_dayfile
638	# @ error = $remote_dayfile
639	# @ image_size = 50
640	$class
641	$environment
642	$network_to_use
643	$data_limit
644	$notification
645
646	%%END%%
647
648	if (( nodes > 0 ))
649	then
650
651	cat >> $job_to_send << %%END%%
652	# @ node = $nodes
653	# @ tasks_per_node = $tasks_per_node
654	# @ node_usage = $node_usage
655	# @ queue
656
657	%%END%%
658
659	else
660
661	if [[ $remote_host != ibmy ]]
662	then
663
664	cat >> $job_to_send << %%END%%
665	# @ blocking = unlimited
666	# @ total_tasks = $numprocs
667	# @ node_usage = $node_usage
668	# @ queue
669
670	%%END%%
671
672	else
673
674	cat >> $job_to_send << %%END%%
675	# @ node = 1
676	# @ total_tasks = $numprocs
677	# @ queue
678
679	%%END%%
680
681	fi
682
683	fi
684
685	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
686	then
687
688	cat > $job_to_send << %%END%%
689	#!/bin/ksh
690
691	# @ job_type = serial
692	# @ node_usage = $node_usage
693	# @ job_name = palm
694	# @ wall_clock_limit = ${timestring},$timestring
695	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
696	# @ output = $remote_dayfile
697	# @ error = $remote_dayfile
698	$class
699	$notification
700
701	# @ queue
702
703	%%END%%
704
705	elif [[ $remote_host = lcfimm ]]
706	then
707
708	if [[ $numprocs != 0 ]]
709	then
710	cat > $job_to_send << %%END%%
711	#!/bin/ksh
712	#PBS -N $job_name
713	#PBS -A geofysisk
714	#PBS -l walltime=$timestring
715	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
716	#PBS -l pmem=${memory}mb
717	#PBS -m abe
718	#PBS -M maronga@muk.uni-hannover.de
719	#PBS -o $remote_dayfile
720	#PBS -j oe
721	mpd &
722
723	%%END%%
724
725	else
726	cat > $job_to_send << %%END%%
727	#!/bin/ksh
728	#PBS -N $job_name
729	#PBS -A geofysisk
730	#PBS -l walltime=$timestring
731	#PBS -l ncpus=1
732	#PBS -l pmem=${memory}mb
733	#PBS -m abe
734	#PBS -M maronga@muk.uni-hannover.de
735	#PBS -o $remote_dayfile
736	#PBS -j oe
737
738	%%END%%
739
740	fi
741
742	elif [[ $remote_host = lck ]]
743	then
744
745	if [[ $numprocs != 0 ]]
746	then
747	cat > $job_to_send << %%END%%
748	#!/bin/ksh
749	#PBS -N $job_name
750	#PBS -l walltime=$timestring
751	#PBS -l ncpus=$numprocs
752	#PBS -l pmem=${memory}mb
753	#PBS -o $remote_dayfile
754	#PBS -j oe
755
756	mpd &
757
758	%%END%%
759
760	else
761	cat > $job_to_send << %%END%%
762	#!/bin/ksh
763	#PBS -N $job_name
764	#PBS -l walltime=$timestring
765	#PBS -l ncpus=1
766	#PBS -l pmem=${memory}mb
767	#PBS -o $remote_dayfile
768	#PBS -j oe
769
770	%%END%%
771
772	fi
773
774	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
775	then
776
777	if [[ "$sgi_feature" != "" ]]
778	then
779	feature_directive="#PBS -l feature=$sgi_feature"
780	else
781	feature_directive=""
782	fi
783
784	if [[ $queue = dataq \|\| $queue = permq ]]
785	then
786	feature_directive="#PBS -l feature=data"
787	fi
788
789	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
790	then
791	queue_directive="#PBS -q $queue"
792	else
793	queue_directive=""
794	fi
795
796	if [[ $email_notification = none ]]
797	then
798	email_directive=""
799	else
800	email_directive="#PBS -M $email_notification"
801	fi
802
803	if [[ $numprocs != 0 ]]
804	then
805	cat > $job_to_send << %%END%%
806	#!/bin/bash
807	#PBS -N $job_name
808	#PBS -l walltime=$timestring
809	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
810	#PBS -l naccesspolicy=$node_usage
811	#PBS -o $remote_dayfile
812	#PBS -j oe
813	$feature_directive
814	$queue_directive
815	$email_directive
816
817	. /usr/share/modules/init/bash
818	module load ifort/11.0.069
819	module load $mpilib
820	module load netcdf
821
822	echo ld_library_path=\$LD_LIBRARY_PATH
823
824	%%END%%
825
826	else
827	cat > $job_to_send << %%END%%
828	#PBS -S /bin/bash
829	#PBS -N $job_name
830	#PBS -l walltime=$timestring
831	#PBS -l ncpus=1
832	#PBS -o $remote_dayfile
833	#PBS -j oe
834	$queue_directive
835	$email_directive
836
837	. /usr/share/modules/init/bash
838	module load ifort/11.0.069
839	module load netcdf
840
841	%%END%%
842
843	fi
844
845	elif [[ $remote_host = lcxt4 ]]
846	then
847
848	if [[ $numprocs != 0 ]]
849	then
850	cat > $job_to_send << %%END%%
851	#!/bin/ksh
852	#PBS -S /bin/ksh
853	#PBS -N $job_name
854	#PBS -A geofysisk
855	#PBS -j oe
856	#PBS -l walltime=$timestring
857	#PBS -l mppwidth=${numprocs}
858	#PBS -l mppnppn=${tasks_per_node}
859	#PBS -m abe
860	#PBS -M maronga@muk.uni-hannover.de
861	#PBS -o $remote_dayfile
862
863	. /opt/modules/default/init/ksh
864	module load modules/3.1.6
865	module load PrgEnv-pgi/2.1.56HDA
866	module load Base-opts/2.1.56HDA
867	module load moab/5.2.5-s13570
868	module load torque/2.3.4-snap.200809221601
869
870
871	%%END%%
872
873	else
874	cat > $job_to_send << %%END%%
875	#!/bin/ksh
876	#PBS -S /bin/ksh
877	#PBS -N $job_name
878	#PBS -A geofysisk
879	#PBS -j oe
880	#PBS -l walltime=$timestring
881	#PBS -l ncpus=1
882	#PBS -l pmem=${memory}mb
883	#PBS -m abe
884	#PBS -M maronga@muk.uni-hannover.de
885	#PBS -o $remote_dayfile
886
887	. /opt/modules/default/init/ksh
888	module load modules/3.1.6
889	module load PrgEnv-pgi/2.1.56HD
890	module load Base-opts/2.1.56HD
891	module load moab/5.2.5-s13441
892	module load torque/2.3.4-snap.200809221601
893
894	%%END%%
895
896	fi
897
898	elif [[ $remote_host = lckyoto ]]
899	then
900
901	cat > $job_to_send << %%END%%
902	#!/bin/ksh
903	# @\$-o $remote_dayfile
904	# @\$-eo -oi
905	# @\$-lP 16
906	# @\$-lp 1
907	# @\$-lm 28gb -llm unlimited -ls unlimited
908	# @\$-q $queue
909	# @\$-Pvn abs_pack
910	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
911	#. /thin/local/etc/setprofile/intel-11.0.sh
912	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
913	. ~/.myprofile
914	#. /home2/t/t51254/palm/current_version/myprofile
915	#. /thin/apps/pgi/mpi.sh
916	#
917	env
918	#
919	set -x
920
921	%%END%%
922
923	elif [[ $remote_host = lcxt5m ]]
924	then
925
926	if [[ $numprocs != 0 ]]
927	then
928	cat > $job_to_send << %%END%%
929	#!/bin/ksh
930	#PBS -S /bin/ksh
931	#PBS -N $job_name
932	#PBS -j oe
933	#PBS -l walltime=$timestring
934	#PBS -l mppwidth=${numprocs}
935	#PBS -l mppnppn=${tasks_per_node}
936	#PBS -m abe
937	#PBS -o $remote_dayfile
938
939	. /opt/modules/default/init/ksh
940	module load modules/3.1.6
941	module load netcdf/3.6.2
942	module load PrgEnv-pgi/2.2.41
943	module load Base-opts/2.2.41
944
945	%%END%%
946
947	else
948	cat > $job_to_send << %%END%%
949	#!/bin/ksh
950	#PBS -S /bin/ksh
951	#PBS -N $job_name
952	#PBS -j oe
953	#PBS -l walltime=$timestring
954	#PBS -l ncpus=1
955	#PBS -l pmem=${memory}mb
956	#PBS -m abe
957	#PBS -o $remote_dayfile
958
959	. /opt/modules/default/init/ksh
960	module load modules/3.1.6
961	module load netcdf/3.6.2
962	module load PrgEnv-pgi/2.2.41
963	module load Base-opts/2.2.41
964
965	%%END%%
966
967	fi
968
969	elif [[ $remote_host = nech ]]
970	then
971
972	if (( nodes > 1 ))
973	then
974	cat > $job_to_send << %%END%%
975	#!/bin/ksh
976	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
977	#PBS -l ${qsubmem}=${Memory}gb
978	#PBS -b $nodes
979	#PBS -o $remote_dayfile
980	#PBS -N palm
981	#PBS -j o
982	#PBS -T mpisx
983
984	%%END%%
985
986	elif [[ $numprocs != 0 ]]
987	then
988	cat > $job_to_send << %%END%%
989	#!/bin/ksh
990	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
991	#PBS -l ${qsubmem}=${Memory}gb
992	#PBS -o $remote_dayfile
993	#PBS -N palm
994	#PBS -j o
995
996	%%END%%
997
998	else
999	cat > $job_to_send << %%END%%
1000	#!/bin/ksh
1001	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1002	#PBS -o $remote_dayfile
1003	#PBS -j o
1004
1005	%%END%%
1006
1007	fi
1008
1009	elif [[ $remote_host = necriam ]]
1010	then
1011
1012	if (( nodes > 1 ))
1013	then
1014	cat > $job_to_send << %%END%%
1015	#!/bin/ksh
1016	#PBS -b $nodes
1017	#PBS -o $remote_dayfile
1018	#PBS -N $job_name
1019	#PBS -j o
1020	#PBS -v MPIPROGINV=YES
1021
1022	%%END%%
1023
1024	elif [[ $numprocs != 0 ]]
1025	then
1026	cat > $job_to_send << %%END%%
1027	#!/bin/ksh
1028	#PBS -q ${queue}
1029	#PBS -o $remote_dayfile
1030	#PBS -N $job_name
1031	#PBS -j o
1032	#PBS -v MPIPROGINV=YES
1033
1034	%%END%%
1035
1036	else
1037	cat > $job_to_send << %%END%%
1038	#!/bin/ksh
1039	#PBS -o $remote_dayfile
1040	#PBS -j o
1041
1042	%%END%%
1043
1044	fi
1045
1046	elif [[ $remote_host = lctit ]]
1047	then
1048	cat > $job_to_send << %%END%%
1049	#!/bin/bash
1050	#$ -S /bin/bash
1051	cd $job_catalog
1052	export PATH=$PALM_BIN:\$PATH
1053	echo \$PATH
1054
1055	%%END%%
1056
1057	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1058	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
1059	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
1060
1061	else
1062
1063	cat > $job_to_send << %%END%%
1064	# @\$-q ${queue}
1065	# @\$-l${qsubtime} $timestring
1066	# @\$-l${qsubmem} ${memory}mb
1067	# @\$-o $remote_dayfile
1068	# @\$-eo
1069
1070	%%END%%
1071
1072	fi
1073
1074
1075	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1076	# BEI EXIT VERANLASSEN
1077	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1078	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1079	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1080	then
1081	echo "set +vx" >> $job_to_send
1082	echo "trap '" >> $job_to_send
1083	echo "set +vx" >> $job_to_send
1084	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 ]]
1085	then
1086	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
1087	then
1088	return_queue=c1
1089	elif [[ $remote_host = ibms ]]
1090	then
1091	return_queue=p_normal
1092	elif [[ $remote_host = ibmy ]]
1093	then
1094	return_queue=serial
1095	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1096	then
1097	return_queue=serialq
1098	elif [[ $remote_host = lctit ]]
1099	then
1100	return_queue=sla3
1101	elif [[ $remote_host = necriam ]]
1102	then
1103	return_queue=SP
1104	else
1105	return_queue=unknown
1106	fi
1107
1108	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1109	then
1110
1111	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1112	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1113	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1114	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1115	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1116	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1117	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1118	if [[ $host != "ibmh" ]]
1119	then
1120	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1121	fi
1122	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1123	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1124
1125	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1126	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1127
1128	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1129	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1130	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1131
1132	elif [[ $remote_host = nech ]]
1133	then
1134	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1135	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1136	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1137	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1138	echo "#PBS -j o" >> $job_to_send
1139	echo " " >> $job_to_send
1140	echo "set -x" >> $job_to_send
1141	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1142	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1143	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1144	echo "%%END%%" >> $job_to_send
1145
1146	elif [[ $remote_host = necriam ]]
1147	then
1148	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1149	echo "#PBS -q $return_queue" >> $job_to_send
1150	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1151	echo "#PBS -j o" >> $job_to_send
1152	echo " " >> $job_to_send
1153	echo "set -x" >> $job_to_send
1154	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1155	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1156	echo "%%END%%" >> $job_to_send
1157
1158	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1159	then
1160	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1161	echo "#!/bin/bash" >> $job_to_send
1162	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1163	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1164	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1165	echo "#PBS -l feature=data" >> $job_to_send
1166	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1167	echo "#PBS -j oe" >> $job_to_send
1168	echo " " >> $job_to_send
1169	echo ". /usr/share/modules/init/bash" >> $job_to_send
1170	echo "set -x" >> $job_to_send
1171	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1172	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1173	echo "%%END%%" >> $job_to_send
1174
1175	elif [[ $remote_host = lctit ]]
1176	then
1177	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1178	echo "set -x" >> $job_to_send
1179	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1180	echo "cd $job_catalog" >> $job_to_send
1181	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1182	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1183	echo "%%END%%" >> $job_to_send
1184
1185	else
1186
1187	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1188	echo "# @\\\$-q $return_queue" >> $job_to_send
1189	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1190	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1191	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1192	then
1193	echo "# @\$-l mpp_p=0" >> $job_to_send
1194	fi
1195	echo '# @\$-lF 10mb' >> $job_to_send
1196	# echo '# @\$-o /dev/null' >> $job_to_send
1197	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1198	echo '# @\\\$-eo' >> $job_to_send
1199	echo " " >> $job_to_send
1200	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1201	then
1202	echo "set +vx" >> $job_to_send
1203	echo ". .profile" >> $job_to_send
1204	fi
1205	echo "set -x" >> $job_to_send
1206	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1207	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1208	echo "%%END%%" >> $job_to_send
1209	fi
1210
1211	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1212	then
1213	echo "llsubmit scpjob.$kennung" >> $job_to_send
1214	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1215	then
1216	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1217	echo "msub scpjob.$kennung" >> $job_to_send
1218	elif [[ $remote_host = lctit ]]
1219	then
1220	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1221	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1222	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1223	echo "sleep 10" >> $job_to_send
1224	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1225	then
1226	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1227	elif [[ $remote_host = t3es ]]
1228	then
1229	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1230	else
1231	echo "qsub scpjob.$kennung" >> $job_to_send
1232	fi
1233	echo "rm scpjob.$kennung" >> $job_to_send
1234	if [[ $remote_host = nech ]]
1235	then
1236	echo "cd -" >> $job_to_send
1237	fi
1238	else
1239	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1240	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1241	fi
1242	echo "set -x" >> $job_to_send
1243	echo " ' exit" >> $job_to_send
1244	echo "set -x" >> $job_to_send
1245	fi
1246
1247
1248
1249	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1250	cat $file_to_send >> $job_to_send
1251
1252	if [[ $remote_host = ibm ]]
1253	then
1254	echo " " >> $job_to_send
1255	echo "exit" >> $job_to_send
1256	fi
1257	if [[ $remote_host = lctit ]]
1258	then
1259	echo " " >> $job_to_send
1260	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1261	fi
1262
1263
1264
1265	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1266	if [[ -z $remote_user ]]
1267	then
1268	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1269	then
1270	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1271	fi
1272	fi
1273
1274
1275
1276	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1277	if [[ $no_submit = false ]]
1278	then
1279	if [[ $remote_host != $local_host ]]
1280	then
1281	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1282	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1283	then
1284	job_catalog_save=$job_catalog
1285	job_catalog=job_queue
1286	elif [[ $remote_host = nech ]]
1287	then
1288	job_catalog_save=$job_catalog
1289	job_catalog=/hpf/b/${remote_user}/job_queue
1290	fi
1291	if [[ $local_host = decalpha ]]
1292	then
1293	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1294	# PASSWORT AB
1295	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1296	elif [[ $remote_host = nech ]]
1297	then
1298	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1299	# TRANSFERIERT WERDEN
1300	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1301	else
1302	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1303	fi
1304	if [[ $? = 1 ]]
1305	then
1306	locat=scp; exit
1307	fi
1308	if [[ $remote_host = ibms ]]
1309	then
1310	job_catalog=$job_catalog_save
1311	fi
1312	[[ $verify = true ]] && printf "\n >>> finished\n"
1313	else
1314	eval job_catalog=$job_catalog
1315	cp $job_to_send ${job_catalog}/$job_on_remhost
1316	fi
1317
1318
1319
1320	# NQS- BZW. LOADLEVELER-JOB STARTEN
1321	if [[ $remote_host != $local_host ]]
1322	then
1323	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1324	if [[ $remote_host != lctit ]]
1325	then
1326	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1327	then
1328	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1329	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1330	else
1331	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1332	fi
1333	else
1334	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1335	# MIT SSH, DESHALB AUFRUF PER PIPE
1336	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1337	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1338	echo "first try" > send_protocol
1339	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1340	do
1341	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1342	then
1343	printf "\n trying to submit job to TIT cluster..."
1344	else
1345	printf "\n +++ failed ... trying to submit again ..."
1346	fi
1347	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1348	cat send_protocol
1349	done
1350	sleep 10
1351	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1352	echo "$submcom $job_on_remhost"
1353	rm send_protocol
1354	fi
1355
1356	[[ $verify = true ]] && printf " >>> o.k.\n"
1357	else
1358	cd $job_catalog
1359	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1360	then
1361	eval $submcom $job_on_remhost
1362	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1363	then
1364	eval $submcom $job_on_remhost
1365	echo "$submcom $job_on_remhost"
1366	chmod u+x $job_on_remhost
1367
1368	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1369	then
1370	eval $submcom $job_on_remhost
1371	elif [[ $local_host = nech ]]
1372	then
1373	if [[ $queue = default ]]
1374	then
1375	eval $submcom $job_on_remhost
1376	else
1377	eval $submcom -q $queue $job_on_remhost
1378	fi
1379	else
1380	qsub $job_on_remhost
1381	fi
1382	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1383	[[ $local_host != lctit ]] && rm $job_on_remhost
1384	cd - > /dev/null
1385	fi
1386	fi
1387
1388
1389
1390	# ABSCHLUSSARBEITEN
1391
1392	if [[ $no_submit = false ]]
1393	then
1394	rm $job_to_send
1395	fi
1396	[[ $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 |