Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 437

Last change on this file since 437 was 437, checked in by raasch, 15 years ago
script adjustements for lcxt5m
Property svn:keywords set to `Id Rev`
File size: 46.5 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 437 2010-02-01 04:15:58Z 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 - adjustments for lcxt5m
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	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
209	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
210	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
211	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
212	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
213	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
214	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
215	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
216	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
217	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
218	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
219	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
220	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
221	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
222	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
223	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
224	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
225	(tgg*) local_addres=172.17.75.161; local_host=lctit;;
226	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
227	(*) printf "\n +++ \"$local_host\" unknown";
228	printf "\n please inform S. Raasch!";
229	locat=parameter; exit;;
230	esac
231
232
233
234	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
235	remote_host=$local_host
236
237
238
239
240	# PROZEDUROPTIONEN EINLESEN
241	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
242	do
243	case $option in
244	(c) job_catalog=$OPTARG;;
245	(d) delete_dayfile=true;;
246	(D) no_submit=true;;
247	(e) email_notification=$OPTARG;;
248	(g) group_number=$OPTARG;;
249	(h) remote_host=$OPTARG;;
250	(m) memory=$OPTARG;;
251	(n) job_name=$OPTARG;;
252	(N) node_usage=$OPTARG;;
253	(O) threads_per_task=$OPTARG;;
254	(q) no_default_queue=$OPTARG;;
255	(t) cputime=$OPTARG;;
256	(T) tasks_per_node=$OPTARG;;
257	(u) remote_user=$OPTARG;;
258	(v) verify=false;;
259	(X) numprocs=$OPTARG;;
260	(\?) printf "\n +++ Option $OPTARG unknown \n";
261	locat=option; exit;;
262	esac
263	done
264
265
266	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
267	shift OPTIND-1; file_to_send=$1
268
269
270	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
271	if [ "$1" = "?" ]
272	then
273	(printf "\n *** subjob can be called as follows:\n"
274	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
275	printf "\n Description of available options:\n"
276	printf "\n Option Description Default-Value"
277	printf "\n -c job-input- and output-catalog ~/job_queue"
278	printf "\n -d no job-protocol will be created ---"
279	printf "\n -D only the job-file will be created ---"
280	printf "\n -h execution host, available hosts: $remote_host"
281	printf "\n ibm, ibmb, ibmh, ibms, ibmy, lcmuk,"
282	printf "\n lctit, nech, necriam, unics"
283	printf "\n -m memory demand per process in MByte ---"
284	printf "\n -n jobname <jobdatei>"
285	printf "\n -O threads per task (for OpenMP usage) 1"
286	printf "\n -q job-queue to be used default"
287	printf "\n -t allowed cpu-time in seconds ---"
288	printf "\n -T tasks per node (on parallel hosts) ---"
289	printf "\n -u username on execution host from .netrc"
290	printf "\n -v no prompt for confirmation ---"
291	printf "\n -X # of processors (on parallel hosts) 1"
292	printf "\n "
293	printf "\n The only possible positional parameter is <jobfile>:"
294	printf "\n The complete NQS-job must be provided here."
295	printf "\n <jobfile>=? creates this outline\n\n") \| more
296	exit
297	fi
298
299
300
301	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
302	if [[ "$file_to_send" = "" ]]
303	then
304	printf "\n +++ job-file missing"
305	locat=parameter; exit
306	else
307	if [[ -f $file_to_send ]]
308	then
309	true
310	else
311	printf "\n +++ job-file: "
312	printf "\n $file_to_send"
313	printf "\n does not exist"
314	locat=parameter; exit
315	fi
316	fi
317
318
319
320	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
321	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
322	if [[ $job_name = none ]]
323	then
324	job_name=$file_to_send
325	fi
326	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
327	then
328	printf "\n +++ job-file name: "
329	printf "\n $job_name"
330	printf "\n must not contain \"/\"-characters"
331	locat=parameter; exit
332	fi
333
334
335
336
337	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
338	# ODER WENN HOST NICHT ANGEGEBEN WURDE
339	if [[ $remote_host = none ]]
340	then
341	printf "\n +++ host missing"
342	locat=option; exit
343	else
344	case $remote_host in
345	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
346	(ibmb) queue=cpar; remote_addres=130.73.230.10; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
347	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
348	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
349	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
350	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
351	(lcsgib) queue=smallq; remote_addres=130.73.232.102; submcom=/opt/moab/bin/msub;;
352	(lcsgih) queue=smallq; remote_addres=130.75.4.103; submcom=/opt/moab/bin/msub;;
353	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
354	(lctit) queue=lctit; remote_addres=172.17.75.161; submcom=/n1ge/TITECH_GRID/tools/bin/n1ge;;
355	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.3.0/bin/qsub;;
356	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
357	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
358	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
359	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
360	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
361	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
362	locat=parameter; exit;;
363	esac
364	fi
365
366
367	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
368	if [[ $no_default_queue != none ]]
369	then
370	error=false
371	ndq=$no_default_queue
372	case $remote_host in
373	(ibm) case $ndq in
374	(p690_express\|p690_standard\|p690_long) error=false;;
375	(*) error=true;;
376	esac;;
377	(ibmb) case $ndq in
378	(cdata\|cdev\|cexp\|c1\|cshare\|csolo\|cspec) error=false;;
379	(*) error=true;;
380	esac;;
381	(ibmh) case $ndq in
382	(no_class) error=false;;
383	(*) error=true;;
384	esac;;
385	(ibms) case $ndq in
386	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
387	(*) error=true;;
388	esac;;
389	(ibmy) case $ndq in
390	(parallel) error=false;;
391	(*) error=true;;
392	esac;;
393	(lcsgib\|lcsgih) case $ndq in
394	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
395	(*) error=true;;
396	esac;;
397	(lctit) case $ndq in
398	(novice\|sla1\|sla2\|sla3\|RAM64GB\|RAM128GB) error=false;;
399	(*) error=true;;
400	esac;;
401	(t3eb) case $ndq in
402	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
403	(*) error=true;;
404	esac;;
405	(necriam) case $ndq in
406	(SP\|SS\|P6) error=false;;
407	(*) error=true;;
408	esac;;
409	(t3eh) case $ndq in
410	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
411	(*) error=true;;
412	esac;;
413	(t3ej2\|t3ej5) case $ndq in
414	(low\|normal\|high) error=false;;
415	(*) error=true;;
416	esac;;
417	(t3es) case $ndq in
418	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
419	(*) error=true;;
420	esac;;
421	(unics) case $ndq in
422	(unics\|ht) error=false;;
423	(*) error=true;;
424	esac;;
425	esac
426	if [[ $error = true ]]
427	then
428	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
429	locat=parameter; exit
430	else
431	queue=$no_default_queue
432	fi
433	fi
434
435
436
437	# KNOTENNUTZUNG IN ENTWICKLERQUEUE MUSS SHARED SEIN
438	if [[ $node_usage != shared && $queue = cdev ]]
439	then
440	node_usage=shared
441	fi
442
443
444
445	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
446	# AUFTEILEN
447	done=false
448	while [[ $done = false ]]
449	do
450	if (( $cputime <= 0 ))
451	then
452	printf "\n +++ wrong cpu-time or cpu-time missing"
453	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
454	printf "\n >>> "
455	read cputime 1>/dev/null 2>&1
456	else
457	done=true
458	fi
459	done
460	if [[ $remote_host = nech ]]
461	then
462	if (( tasks_per_node != 0 ))
463	then
464	(( cputime = cputime * tasks_per_node ))
465	elif [[ $numprocs != 0 ]]
466	then
467	(( cputime = cputime * numprocs ))
468	fi
469	fi
470	(( stunden = cputime / 3600 ))
471	(( resttime = cputime - stunden * 3600 ))
472	(( minuten = resttime / 60 ))
473	(( sekunden = resttime - minuten * 60 ))
474	timestring=${stunden}:${minuten}:${sekunden}
475
476
477
478	# PRUEFEN DER KERNSPEICHERANFORDERUNG
479	done=false
480	while [[ $done = false ]]
481	do
482	if (( memory <= 0 ))
483	then
484	printf "\n +++ wrong memory demand or memory demand missing"
485	printf "\n >>> Please type memory in MByte per process as INTEGER:"
486	printf "\n >>> "
487	read memory 1>/dev/null 2>&1
488	else
489	done=true
490	fi
491	done
492
493	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
494	then
495	if (( tasks_per_node != 0 ))
496	then
497	(( Memory = memory * tasks_per_node / 1000 ))
498	elif [[ $numprocs != 0 ]]
499	then
500	(( Memory = memory * numprocs / 1000 ))
501	else
502	(( Memory = memory / 1000 ))
503	fi
504	fi
505
506
507	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
508	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
509	then
510	(( memory = memory * threads_per_task ))
511	fi
512
513
514	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
515	if (( tasks_per_node != 0 ))
516	then
517	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
518	then
519	(( nodes = numprocs / tasks_per_node ))
520	else
521	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
522	fi
523	fi
524
525
526
527	# HEADER-AUSGABE
528	if [[ $verify = true ]]
529	then
530	printf "\n\n"
531	printf "#--------------------------------------------------------------# \n"
532	spalte1=SUBJOB;spalte2=$(date)
533	printf "\| $spalte1$spalte2 \| \n"
534	printf "\| \| \n"
535	printf "\| values of parameters/options: \| \n"
536	spalte1=local_host$punkte; spalte2=$punkte$local_host
537	printf "\| $spalte1$spalte2 \| \n"
538	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
539	printf "\| $spalte1$spalte2 \| \n"
540	spalte1=queue$punkte; spalte2=$punkte$queue
541	printf "\| $spalte1$spalte2 \| \n"
542	spalte1=memory$punkte; spalte2="$punkte$memory mb"
543	printf "\| $spalte1$spalte2 \| \n"
544	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
545	printf "\| $spalte1$spalte2 \| \n"
546	spalte1=job_name$punkte; spalte2="$punkte$job_name"
547	printf "\| $spalte1$spalte2 \| \n"
548	printf "#--------------------------------------------------------------# \n\n"
549
550
551	# KONTROLLABFRAGE, OB ALLES O.K.
552	antwort="dummy"
553	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
554	do
555	read antwort?" >>> continue (y/n) ? "
556	done
557	if [[ $antwort = n \|\| $antwort = N ]]
558	then
559	locat=verify; exit
560	fi
561	printf "\n"
562	fi
563
564
565
566	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
567	kennung=$RANDOM
568	job_on_remhost=${job_name}_${kennung}_$local_host
569	job_to_send=job_to_send_$kennung
570	if [[ $delete_dayfile = false ]]
571	then
572	remote_dayfile=${local_host}_${job_name}_result_$kennung
573	local_dayfile=${remote_host}_${job_name}
574	else
575	remote_dayfile=/dev/null
576	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
577	then
578	notification='# @ notification = never'
579	fi
580	fi
581
582
583
584	# QSUB- ODER LL-KOMMANDOS BZW. SKRIPTE GENERIEREN
585	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
586	then
587
588	if [[ $remote_host = ibmy ]]
589	then
590	consumable_memory=""
591	else
592	consumable_memory="ConsumableMemory($memory mb)"
593	fi
594
595	if [[ $remote_host = ibmh ]]
596	then
597	data_limit=""
598	network_to_use=""
599	class=""
600	environment=""
601	else
602	class="# @ class = $queue"
603	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
604	if [[ $queue = cdev ]]
605	then
606	data_limit="# @ data_limit = 1.76gb"
607	network_to_use="# @ network.mpi = sn_all,shared,ip"
608	else
609	if [[ $remote_host = ibms ]]
610	then
611	network_to_use="# @ network.mpi = csss,shared,us"
612	elif [[ $remote_host = ibmy ]]
613	then
614	network_to_use=""
615	else
616	network_to_use="# @ network.mpi = sn_all,shared,us"
617	data_limit="# @ data_limit = 1.76gb"
618	fi
619	fi
620	fi
621
622	cat > $job_to_send << %%END%%
623	#!/bin/ksh
624	# @ shell = /bin/ksh
625
626	# @ job_type = parallel
627	# @ job_name = palm
628	# @ wall_clock_limit = ${timestring},$timestring
629	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
630	# @ output = $remote_dayfile
631	# @ error = $remote_dayfile
632	# @ image_size = 50
633	$class
634	$environment
635	$network_to_use
636	$data_limit
637	$notification
638
639	%%END%%
640
641	if (( nodes > 0 ))
642	then
643
644	cat >> $job_to_send << %%END%%
645	# @ node = $nodes
646	# @ tasks_per_node = $tasks_per_node
647	# @ node_usage = $node_usage
648	# @ queue
649
650	%%END%%
651
652	else
653
654	if [[ $remote_host != ibmy ]]
655	then
656
657	cat >> $job_to_send << %%END%%
658	# @ blocking = unlimited
659	# @ total_tasks = $numprocs
660	# @ node_usage = $node_usage
661	# @ queue
662
663	%%END%%
664
665	else
666
667	cat >> $job_to_send << %%END%%
668	# @ node = 1
669	# @ total_tasks = $numprocs
670	# @ queue
671
672	%%END%%
673
674	fi
675
676	fi
677
678	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
679	then
680
681	cat > $job_to_send << %%END%%
682	#!/bin/ksh
683
684	# @ job_type = serial
685	# @ node_usage = $node_usage
686	# @ job_name = palm
687	# @ wall_clock_limit = ${timestring},$timestring
688	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
689	# @ output = $remote_dayfile
690	# @ error = $remote_dayfile
691	$class
692	$notification
693
694	# @ queue
695
696	%%END%%
697
698	elif [[ $remote_host = lcfimm ]]
699	then
700
701	if [[ $numprocs != 0 ]]
702	then
703	cat > $job_to_send << %%END%%
704	#!/bin/ksh
705	#PBS -N $job_name
706	#PBS -A geofysisk
707	#PBS -l walltime=$timestring
708	#PBS -l nodes=${nodes}:ppn=$tasks_per_node
709	#PBS -l pmem=${memory}mb
710	#PBS -m abe
711	#PBS -M maronga@muk.uni-hannover.de
712	#PBS -o $remote_dayfile
713	#PBS -j oe
714	mpd &
715
716	%%END%%
717
718	else
719	cat > $job_to_send << %%END%%
720	#!/bin/ksh
721	#PBS -N $job_name
722	#PBS -A geofysisk
723	#PBS -l walltime=$timestring
724	#PBS -l ncpus=1
725	#PBS -l pmem=${memory}mb
726	#PBS -m abe
727	#PBS -M maronga@muk.uni-hannover.de
728	#PBS -o $remote_dayfile
729	#PBS -j oe
730
731	%%END%%
732
733	fi
734
735	elif [[ $remote_host = lck ]]
736	then
737
738	if [[ $numprocs != 0 ]]
739	then
740	cat > $job_to_send << %%END%%
741	#!/bin/ksh
742	#PBS -N $job_name
743	#PBS -l walltime=$timestring
744	#PBS -l ncpus=$numprocs
745	#PBS -l pmem=${memory}mb
746	#PBS -o $remote_dayfile
747	#PBS -j oe
748
749	mpd &
750
751	%%END%%
752
753	else
754	cat > $job_to_send << %%END%%
755	#!/bin/ksh
756	#PBS -N $job_name
757	#PBS -l walltime=$timestring
758	#PBS -l ncpus=1
759	#PBS -l pmem=${memory}mb
760	#PBS -o $remote_dayfile
761	#PBS -j oe
762
763	%%END%%
764
765	fi
766
767	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
768	then
769
770	if [[ "$sgi_feature" != "" ]]
771	then
772	feature_directive="#PBS -l feature=$sgi_feature"
773	else
774	feature_directive=""
775	fi
776
777	if [[ $queue = dataq \|\| $queue = permq ]]
778	then
779	feature_directive="#PBS -l feature=data"
780	fi
781
782	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
783	then
784	queue_directive="#PBS -q $queue"
785	else
786	queue_directive=""
787	fi
788
789	if [[ $email_notification = none ]]
790	then
791	email_directive=""
792	else
793	email_directive="#PBS -M $email_notification"
794	fi
795
796	if [[ $numprocs != 0 ]]
797	then
798	cat > $job_to_send << %%END%%
799	#!/bin/bash
800	#PBS -N $job_name
801	#PBS -l walltime=$timestring
802	#PBS -l nodes=$nodes:ppn=${tasks_per_node}
803	#PBS -l naccesspolicy=$node_usage
804	#PBS -o $remote_dayfile
805	#PBS -j oe
806	$feature_directive
807	$queue_directive
808	$email_directive
809
810	. /usr/share/modules/init/bash
811	module load ifort/11.0.069
812	module load $mpilib
813	module load netcdf
814
815	echo ld_library_path=\$LD_LIBRARY_PATH
816
817	%%END%%
818
819	else
820	cat > $job_to_send << %%END%%
821	#PBS -S /bin/bash
822	#PBS -N $job_name
823	#PBS -l walltime=$timestring
824	#PBS -l ncpus=1
825	#PBS -o $remote_dayfile
826	#PBS -e $remote_dayfile
827	$queue_directive
828	$email_directive
829
830	%%END%%
831
832	fi
833
834	elif [[ $remote_host = lcxt4 ]]
835	then
836
837	if [[ $numprocs != 0 ]]
838	then
839	cat > $job_to_send << %%END%%
840	#!/bin/ksh
841	#PBS -S /bin/ksh
842	#PBS -N $job_name
843	#PBS -A geofysisk
844	#PBS -j oe
845	#PBS -l walltime=$timestring
846	#PBS -l mppwidth=${numprocs}
847	#PBS -l mppnppn=${tasks_per_node}
848	#PBS -m abe
849	#PBS -M maronga@muk.uni-hannover.de
850	#PBS -o $remote_dayfile
851
852	. /opt/modules/default/init/ksh
853	module load modules/3.1.6
854	module load PrgEnv-pgi/2.1.56HDA
855	module load Base-opts/2.1.56HDA
856	module load moab/5.2.5-s13570
857	module load torque/2.3.4-snap.200809221601
858
859
860	%%END%%
861
862	else
863	cat > $job_to_send << %%END%%
864	#!/bin/ksh
865	#PBS -S /bin/ksh
866	#PBS -N $job_name
867	#PBS -A geofysisk
868	#PBS -j oe
869	#PBS -l walltime=$timestring
870	#PBS -l ncpus=1
871	#PBS -l pmem=${memory}mb
872	#PBS -m abe
873	#PBS -M maronga@muk.uni-hannover.de
874	#PBS -o $remote_dayfile
875
876	. /opt/modules/default/init/ksh
877	module load modules/3.1.6
878	module load PrgEnv-pgi/2.1.56HD
879	module load Base-opts/2.1.56HD
880	module load moab/5.2.5-s13441
881	module load torque/2.3.4-snap.200809221601
882
883	%%END%%
884
885	fi
886
887	elif [[ $remote_host = lcxt5m ]]
888	then
889
890	if [[ $numprocs != 0 ]]
891	then
892	cat > $job_to_send << %%END%%
893	#!/bin/ksh
894	#PBS -S /bin/ksh
895	#PBS -N $job_name
896	#PBS -j oe
897	#PBS -l walltime=$timestring
898	#PBS -l mppwidth=${numprocs}
899	#PBS -l mppnppn=${tasks_per_node}
900	#PBS -m abe
901	#PBS -o $remote_dayfile
902
903	. /opt/modules/default/init/ksh
904	module load modules/3.1.6
905	module load netcdf/3.6.2
906	module load PrgEnv-pgi/2.2.41
907	module load Base-opts/2.2.41
908
909	%%END%%
910
911	else
912	cat > $job_to_send << %%END%%
913	#!/bin/ksh
914	#PBS -S /bin/ksh
915	#PBS -N $job_name
916	#PBS -j oe
917	#PBS -l walltime=$timestring
918	#PBS -l ncpus=1
919	#PBS -l pmem=${memory}mb
920	#PBS -m abe
921	#PBS -o $remote_dayfile
922
923	. /opt/modules/default/init/ksh
924	module load modules/3.1.6
925	module load netcdf/3.6.2
926	module load PrgEnv-pgi/2.2.41
927	module load Base-opts/2.2.41
928
929	%%END%%
930
931	fi
932
933	elif [[ $remote_host = nech ]]
934	then
935
936	if (( nodes > 1 ))
937	then
938	cat > $job_to_send << %%END%%
939	#!/bin/ksh
940	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
941	#PBS -l ${qsubmem}=${Memory}gb
942	#PBS -b $nodes
943	#PBS -o $remote_dayfile
944	#PBS -N palm
945	#PBS -j o
946	#PBS -T mpisx
947
948	%%END%%
949
950	elif [[ $numprocs != 0 ]]
951	then
952	cat > $job_to_send << %%END%%
953	#!/bin/ksh
954	#PBS -l cpunum_prc=$tasks_per_node,cputim_job=$cputime
955	#PBS -l ${qsubmem}=${Memory}gb
956	#PBS -o $remote_dayfile
957	#PBS -N palm
958	#PBS -j o
959
960	%%END%%
961
962	else
963	cat > $job_to_send << %%END%%
964	#!/bin/ksh
965	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
966	#PBS -o $remote_dayfile
967	#PBS -j o
968
969	%%END%%
970
971	fi
972
973	elif [[ $remote_host = necriam ]]
974	then
975
976	if (( nodes > 1 ))
977	then
978	cat > $job_to_send << %%END%%
979	#!/bin/ksh
980	#PBS -b $nodes
981	#PBS -o $remote_dayfile
982	#PBS -N $job_name
983	#PBS -j o
984	#PBS -v MPIPROGINV=YES
985
986	%%END%%
987
988	elif [[ $numprocs != 0 ]]
989	then
990	cat > $job_to_send << %%END%%
991	#!/bin/ksh
992	#PBS -q ${queue}
993	#PBS -o $remote_dayfile
994	#PBS -N $job_name
995	#PBS -j o
996	#PBS -v MPIPROGINV=YES
997
998	%%END%%
999
1000	else
1001	cat > $job_to_send << %%END%%
1002	#!/bin/ksh
1003	#PBS -o $remote_dayfile
1004	#PBS -j o
1005
1006	%%END%%
1007
1008	fi
1009
1010	elif [[ $remote_host = lctit ]]
1011	then
1012	cat > $job_to_send << %%END%%
1013	#!/bin/bash
1014	#$ -S /bin/bash
1015	cd $job_catalog
1016	export PATH=$PALM_BIN:\$PATH
1017	echo \$PATH
1018
1019	%%END%%
1020
1021	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1022	# submcom="$submcom -N $job_name -sgeout $remote_dayfile -q default"
1023	submcom="$submcom -g $group_number -N $job_name -rt 4320 -sgeout $remote_dayfile -q $node_usage "
1024
1025	else
1026
1027	cat > $job_to_send << %%END%%
1028	# @\$-q ${queue}
1029	# @\$-l${qsubtime} $timestring
1030	# @\$-l${qsubmem} ${memory}mb
1031	# @\$-o $remote_dayfile
1032	# @\$-eo
1033
1034	%%END%%
1035
1036	fi
1037
1038
1039	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1040	# BEI EXIT VERANLASSEN
1041	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1042	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1043	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1044	then
1045	echo "set +vx" >> $job_to_send
1046	echo "trap '" >> $job_to_send
1047	echo "set +vx" >> $job_to_send
1048	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 ]]
1049	then
1050	if [[ $remote_host = ibmb \|\| $remote_host = ibmh ]]
1051	then
1052	return_queue=c1
1053	elif [[ $remote_host = ibms ]]
1054	then
1055	return_queue=p_normal
1056	elif [[ $remote_host = ibmy ]]
1057	then
1058	return_queue=serial
1059	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1060	then
1061	return_queue=serialq
1062	elif [[ $remote_host = lctit ]]
1063	then
1064	return_queue=sla3
1065	elif [[ $remote_host = necriam ]]
1066	then
1067	return_queue=SP
1068	else
1069	return_queue=unknown
1070	fi
1071
1072	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1073	then
1074
1075	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1076	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1077	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1078	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1079	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1080	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1081	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1082	if [[ $host != "ibmh" ]]
1083	then
1084	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1085	fi
1086	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1087	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1088
1089	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1090	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1091
1092	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1093	echo "echo \"batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$kennung" >> $job_to_send
1094	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1095
1096	elif [[ $remote_host = nech ]]
1097	then
1098	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1099	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1100	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1101	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1102	echo "#PBS -j o" >> $job_to_send
1103	echo " " >> $job_to_send
1104	echo "set -x" >> $job_to_send
1105	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1106	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1107	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1108	echo "%%END%%" >> $job_to_send
1109
1110	elif [[ $remote_host = necriam ]]
1111	then
1112	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1113	echo "#PBS -q $return_queue" >> $job_to_send
1114	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1115	echo "#PBS -j o" >> $job_to_send
1116	echo " " >> $job_to_send
1117	echo "set -x" >> $job_to_send
1118	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1119	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1120	echo "%%END%%" >> $job_to_send
1121
1122	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1123	then
1124	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1125	echo "#!/bin/bash" >> $job_to_send
1126	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1127	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1128	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1129	echo "#PBS -l feature=data" >> $job_to_send
1130	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1131	echo "#PBS -j oe" >> $job_to_send
1132	echo " " >> $job_to_send
1133	echo ". /usr/share/modules/init/bash" >> $job_to_send
1134	echo "set -x" >> $job_to_send
1135	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1136	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1137	echo "%%END%%" >> $job_to_send
1138
1139	elif [[ $remote_host = lctit ]]
1140	then
1141	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1142	echo "set -x" >> $job_to_send
1143	echo "PATH=\$PATH:$job_catalog/../pub" >> $job_to_send
1144	echo "cd $job_catalog" >> $job_to_send
1145	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1146	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1147	echo "%%END%%" >> $job_to_send
1148
1149	else
1150
1151	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1152	echo "# @\\\$-q $return_queue" >> $job_to_send
1153	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1154	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1155	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1156	then
1157	echo "# @\$-l mpp_p=0" >> $job_to_send
1158	fi
1159	echo '# @\$-lF 10mb' >> $job_to_send
1160	# echo '# @\$-o /dev/null' >> $job_to_send
1161	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1162	echo '# @\\\$-eo' >> $job_to_send
1163	echo " " >> $job_to_send
1164	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1165	then
1166	echo "set +vx" >> $job_to_send
1167	echo ". .profile" >> $job_to_send
1168	fi
1169	echo "set -x" >> $job_to_send
1170	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1171	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1172	echo "%%END%%" >> $job_to_send
1173	fi
1174
1175	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1176	then
1177	echo "llsubmit scpjob.$kennung" >> $job_to_send
1178	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1179	then
1180	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1181	echo "msub scpjob.$kennung" >> $job_to_send
1182	elif [[ $remote_host = lctit ]]
1183	then
1184	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1185	echo "rm -rf ${job_catalog}/last_job_transfer_protocol" >> $job_to_send
1186	echo "n1ge -N jobtransfer -q $return_queue -sgeout ${job_catalog}/last_job_transfer_protocol scpjob.$kennung" >> $job_to_send
1187	echo "sleep 10" >> $job_to_send
1188	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1189	then
1190	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1191	elif [[ $remote_host = t3es ]]
1192	then
1193	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1194	else
1195	echo "qsub scpjob.$kennung" >> $job_to_send
1196	fi
1197	echo "rm scpjob.$kennung" >> $job_to_send
1198	if [[ $remote_host = nech ]]
1199	then
1200	echo "cd -" >> $job_to_send
1201	fi
1202	else
1203	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1204	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1205	fi
1206	echo "set -x" >> $job_to_send
1207	echo " ' exit" >> $job_to_send
1208	echo "set -x" >> $job_to_send
1209	fi
1210
1211
1212
1213	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1214	cat $file_to_send >> $job_to_send
1215
1216	if [[ $remote_host = ibm ]]
1217	then
1218	echo " " >> $job_to_send
1219	echo "exit" >> $job_to_send
1220	fi
1221	if [[ $remote_host = lctit ]]
1222	then
1223	echo " " >> $job_to_send
1224	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1225	fi
1226
1227
1228
1229	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1230	if [[ -z $remote_user ]]
1231	then
1232	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1233	then
1234	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1235	fi
1236	fi
1237
1238
1239
1240	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1241	if [[ $no_submit = false ]]
1242	then
1243	if [[ $remote_host != $local_host ]]
1244	then
1245	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1246	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1247	then
1248	job_catalog_save=$job_catalog
1249	job_catalog=job_queue
1250	elif [[ $remote_host = nech ]]
1251	then
1252	job_catalog_save=$job_catalog
1253	job_catalog=/hpf/b/${remote_user}/job_queue
1254	fi
1255	if [[ $local_host = decalpha ]]
1256	then
1257	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1258	# PASSWORT AB
1259	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1260	elif [[ $remote_host = nech ]]
1261	then
1262	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1263	# TRANSFERIERT WERDEN
1264	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1265	else
1266	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1267	fi
1268	if [[ $? = 1 ]]
1269	then
1270	locat=scp; exit
1271	fi
1272	if [[ $remote_host = ibms ]]
1273	then
1274	job_catalog=$job_catalog_save
1275	fi
1276	[[ $verify = true ]] && printf "\n >>> finished\n"
1277	else
1278	eval job_catalog=$job_catalog
1279	cp $job_to_send ${job_catalog}/$job_on_remhost
1280	fi
1281
1282
1283
1284	# NQS- BZW. LOADLEVELER-JOB STARTEN
1285	if [[ $remote_host != $local_host ]]
1286	then
1287	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1288	if [[ $remote_host != lctit ]]
1289	then
1290	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1291	then
1292	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1293	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1294	else
1295	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1296	fi
1297	else
1298	# TIT ERLAUBT NUR DIE AUSFï¿œHRUNG GANZ BESTIMMTER KOMMANDOS
1299	# MIT SSH, DESHALB AUFRUF PER PIPE
1300	# UEBERGANGSWEISE CHECK, OB N1GE ENVIRONMENT WIRKLICH VERFUEGBAR
1301	print "cd $job_catalog; chmod u+x $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1302	echo "first try" > send_protocol
1303	while [[ $(cat send_protocol \| grep -c "Forwarding to N1GE") = 0 ]]
1304	do
1305	if [[ $(cat send_protocol \| grep -c "first try") = 1 ]]
1306	then
1307	printf "\n trying to submit job to TIT cluster..."
1308	else
1309	printf "\n +++ failed ... trying to submit again ..."
1310	fi
1311	print "cd $job_catalog; $submcom $job_on_remhost" \| ssh $remote_addres -l $remote_user > send_protocol 2>&1
1312	cat send_protocol
1313	done
1314	sleep 10
1315	print "cd $job_catalog; rm $job_on_remhost" \| ssh $remote_addres -l $remote_user > /dev/null 2>&1
1316	echo "$submcom $job_on_remhost"
1317	rm send_protocol
1318	fi
1319
1320	[[ $verify = true ]] && printf " >>> o.k.\n"
1321	else
1322	cd $job_catalog
1323	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1324	then
1325	eval $submcom $job_on_remhost
1326	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1327	then
1328	eval $submcom $job_on_remhost
1329	echo "$submcom $job_on_remhost"
1330	chmod u+x $job_on_remhost
1331
1332	elif [[ $(echo $local_host \| cut -c1-5) = lcsgi ]]
1333	then
1334	eval $submcom $job_on_remhost
1335	elif [[ $local_host = nech ]]
1336	then
1337	if [[ $queue = default ]]
1338	then
1339	eval $submcom $job_on_remhost
1340	else
1341	eval $submcom -q $queue $job_on_remhost
1342	fi
1343	else
1344	qsub $job_on_remhost
1345	fi
1346	# JOBFILE DARF AUF LCTIT NICHT GELOESCHT WERDEN!! GESCHIEHT ERST AM JOBENDE
1347	[[ $local_host != lctit ]] && rm $job_on_remhost
1348	cd - > /dev/null
1349	fi
1350	fi
1351
1352
1353
1354	# ABSCHLUSSARBEITEN
1355
1356	if [[ $no_submit = false ]]
1357	then
1358	rm $job_to_send
1359	fi
1360	[[ $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 |