Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 676

Last change on this file since 676 was 635, checked in by raasch, 14 years ago
mrun/mbuild/subjob adjustments for new Tsubama system at Tokyo Institute of Techology (lctit)
Property svn:keywords set to `Id Rev`
File size: 46.1 KB

Line
1	#!/bin/ksh
2	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
3	# $Id: subjob 635 2010-12-14 13:33:02Z suehring $
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	# 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email
116	# notification)
117	# 01/03/10 - Siggi - loading of modules controlled by environment variable
118	# module_calls
119	# 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub)
120	# 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4
121	# 25/08/10 - Siggi - new variable project_account in pbs-statements for
122	# lcxt4
123	# 08/12/10 - Siggi - initialization of the module command changed for
124	# SGI-ICE/lcsgi
125	# adjustments for Kyushu Univ. (lcrte, ibmku)
126	# 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
127	# institute of technology (lctit)
128
129
130
131	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
132	delete_dayfile=false
133	email_notification=none
134	group_number=none
135	locat=normal
136	no_default_queue=none
137	no_submit=false
138	job_catalog="~/job_queue"
139	job_name=none
140	local_user=$LOGNAME
141	node_usage=shared
142	numprocs=0
143	punkte="..........................................................."
144	submcom=qsub
145	queue=default
146	remote_host=none
147	remote_user=""
148	verify=true
149
150	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
151	typeset -i inumprocs nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
152	typeset -L20 spalte1
153	typeset -R40 spalte2
154	typeset -L60 spalte3
155
156
157
158	# FEHLERBEHANDLUNG
159	# BEI EXIT:
160	trap 'if [[ $locat != normal ]]
161	then
162	case $locat in
163	(option) printf "\n --> available optios can be displayed"
164	printf " by typing:"
165	printf "\n \"subjob ?\" \n";;
166	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
167	(*) printf "\n +++ unknown error"
168	printf "\n please inform S. Raasch!\n"
169	esac
170	[[ -f $job_to_send ]] && rm $job_to_send
171	printf "\n\n+++ SUBJOB killed \n\n"
172	fi' exit
173
174
175	# BEI TERMINAL-BREAK:
176	trap '[[ -f $job_to_send ]] && rm $job_to_send
177	printf "\n\n+++ SUBJOB killed \n\n"
178	exit
179	' 2
180
181
182
183
184	# LOKALEN HOSTNAMEN FESTSTELLEN
185	local_host=$(hostname)
186
187
188
189	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
190	# UEBERHAUPT ZULAESSIG IST
191	case $local_host in
192	(atmos) local_addres=172.20.25.35; local_host=lcide;;
193	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
194	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
195	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
196	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
197	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
198	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
199	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
200	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
201	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
202	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
203	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
204	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
205	(cs*) local_addres=136.172.44.131; local_host=nech;;
206	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
207	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
208	(node*) local_addres=165.132.26.61 local_host=lck;;
209	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
210	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
211	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
212	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
213	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
214	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxt4;;
215	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
216	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
217	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
218	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
219	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
220	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
221	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
222	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
223	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
224	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
225	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
226	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
227	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
228	(nid*) local_addres=129.177.20.113; local_host=lcxt4;;
229	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
230	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
231	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
232	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
233	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
234	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
235	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
236	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
237	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
238	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
239	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
240	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
241	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
242	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
243	(*) printf "\n +++ \"$local_host\" unknown";
244	printf "\n please inform S. Raasch!";
245	locat=parameter; exit;;
246	esac
247
248
249
250	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
251	remote_host=$local_host
252
253
254
255
256	# PROZEDUROPTIONEN EINLESEN
257	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
258	do
259	case $option in
260	(c) job_catalog=$OPTARG;;
261	(d) delete_dayfile=true;;
262	(D) no_submit=true;;
263	(e) email_notification=$OPTARG;;
264	(g) group_number=$OPTARG;;
265	(h) remote_host=$OPTARG;;
266	(m) memory=$OPTARG;;
267	(n) job_name=$OPTARG;;
268	(N) node_usage=$OPTARG;;
269	(O) threads_per_task=$OPTARG;;
270	(q) no_default_queue=$OPTARG;;
271	(t) cputime=$OPTARG;;
272	(T) tasks_per_node=$OPTARG;;
273	(u) remote_user=$OPTARG;;
274	(v) verify=false;;
275	(X) numprocs=$OPTARG;;
276	(\?) printf "\n +++ Option $OPTARG unknown \n";
277	locat=option; exit;;
278	esac
279	done
280
281
282	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
283	shift OPTIND-1; file_to_send=$1
284
285
286	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
287	if [ "$1" = "?" ]
288	then
289	(printf "\n *** subjob can be called as follows:\n"
290	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
291	printf "\n Description of available options:\n"
292	printf "\n Option Description Default-Value"
293	printf "\n -c job-input- and output-catalog ~/job_queue"
294	printf "\n -d no job-protocol will be created ---"
295	printf "\n -D only the job-file will be created ---"
296	printf "\n -h execution host, available hosts: $remote_host"
297	printf "\n ibm, ibmh, ibmku, ibms, ibmy, lc...,"
298	printf "\n lctit, nech, necriam, unics"
299	printf "\n -m memory demand per process in MByte ---"
300	printf "\n -n jobname <jobdatei>"
301	printf "\n -O threads per task (for OpenMP usage) 1"
302	printf "\n -q job-queue to be used default"
303	printf "\n -t allowed cpu-time in seconds ---"
304	printf "\n -T tasks per node (on parallel hosts) ---"
305	printf "\n -u username on execution host from .netrc"
306	printf "\n -v no prompt for confirmation ---"
307	printf "\n -X # of processors (on parallel hosts) 1"
308	printf "\n "
309	printf "\n The only possible positional parameter is <jobfile>:"
310	printf "\n The complete NQS-job must be provided here."
311	printf "\n <jobfile>=? creates this outline\n\n") \| more
312	exit
313	fi
314
315
316
317	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
318	if [[ "$file_to_send" = "" ]]
319	then
320	printf "\n +++ job-file missing"
321	locat=parameter; exit
322	else
323	if [[ -f $file_to_send ]]
324	then
325	true
326	else
327	printf "\n +++ job-file: "
328	printf "\n $file_to_send"
329	printf "\n does not exist"
330	locat=parameter; exit
331	fi
332	fi
333
334
335
336	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
337	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
338	if [[ $job_name = none ]]
339	then
340	job_name=$file_to_send
341	fi
342	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
343	then
344	printf "\n +++ job-file name: "
345	printf "\n $job_name"
346	printf "\n must not contain \"/\"-characters"
347	locat=parameter; exit
348	fi
349
350
351
352
353	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
354	# ODER WENN HOST NICHT ANGEGEBEN WURDE
355	if [[ $remote_host = none ]]
356	then
357	printf "\n +++ host missing"
358	locat=option; exit
359	else
360	case $remote_host in
361	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
362	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
363	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
364	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
365	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
366	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
367	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
368	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
369	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
370	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
371	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
372	(lcxt4) remote_addres=129.177.20.113; submcom=/opt/torque/2.4.9-snap.201005191035/bin/qsub;;
373	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
374	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
375	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
376	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
377	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
378	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
379	locat=parameter; exit;;
380	esac
381	fi
382
383
384	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
385	if [[ $no_default_queue != none ]]
386	then
387	error=false
388	ndq=$no_default_queue
389	case $remote_host in
390	(ibm) case $ndq in
391	(p690_express\|p690_standard\|p690_long) error=false;;
392	(*) error=true;;
393	esac;;
394	(ibmh) case $ndq in
395	(no_class) error=false;;
396	(*) error=true;;
397	esac;;
398	(ibmku) case $ndq in
399	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
400	(*) error=true;;
401	esac;;
402	(ibms) case $ndq in
403	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
404	(*) error=true;;
405	esac;;
406	(ibmy) case $ndq in
407	(parallel) error=false;;
408	(*) error=true;;
409	esac;;
410	(lckyoto) case $ndq in
411	(eh\|ph) error=false;;
412	(*) error=true;;
413	esac;;
414	(lcsgib\|lcsgih) case $ndq in
415	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
416	(*) error=true;;
417	esac;;
418	(lctit) case $ndq in
419	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
420	(*) error=true;;
421	esac;;
422	(t3eb) case $ndq in
423	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
424	(*) error=true;;
425	esac;;
426	(necriam) case $ndq in
427	(SP\|SS\|P6) error=false;;
428	(*) error=true;;
429	esac;;
430	(t3eh) case $ndq in
431	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
432	(*) error=true;;
433	esac;;
434	(t3ej2\|t3ej5) case $ndq in
435	(low\|normal\|high) error=false;;
436	(*) error=true;;
437	esac;;
438	(t3es) case $ndq in
439	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
440	(*) error=true;;
441	esac;;
442	(unics) case $ndq in
443	(unics\|ht) error=false;;
444	(*) error=true;;
445	esac;;
446	esac
447	if [[ $error = true ]]
448	then
449	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
450	locat=parameter; exit
451	else
452	queue=$no_default_queue
453	fi
454	fi
455
456
457
458	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
459	# AUFTEILEN
460	done=false
461	while [[ $done = false ]]
462	do
463	if (( $cputime <= 0 ))
464	then
465	printf "\n +++ wrong cpu-time or cpu-time missing"
466	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
467	printf "\n >>> "
468	read cputime 1>/dev/null 2>&1
469	else
470	done=true
471	fi
472	done
473	if [[ $remote_host = nech ]]
474	then
475	if (( tasks_per_node != 0 ))
476	then
477	(( cputime = cputime * tasks_per_node ))
478	elif [[ $numprocs != 0 ]]
479	then
480	(( cputime = cputime * numprocs ))
481	fi
482	fi
483	(( stunden = cputime / 3600 ))
484	(( resttime = cputime - stunden * 3600 ))
485	(( minuten = resttime / 60 ))
486	(( sekunden = resttime - minuten * 60 ))
487	timestring=${stunden}:${minuten}:${sekunden}
488
489
490
491	# PRUEFEN DER KERNSPEICHERANFORDERUNG
492	done=false
493	while [[ $done = false ]]
494	do
495	if (( memory <= 0 ))
496	then
497	printf "\n +++ wrong memory demand or memory demand missing"
498	printf "\n >>> Please type memory in MByte per process as INTEGER:"
499	printf "\n >>> "
500	read memory 1>/dev/null 2>&1
501	else
502	done=true
503	fi
504	done
505
506	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
507	then
508	if (( tasks_per_node != 0 ))
509	then
510	(( Memory = memory * tasks_per_node / 1000 ))
511	elif [[ $numprocs != 0 ]]
512	then
513	(( Memory = memory * numprocs / 1000 ))
514	else
515	(( Memory = memory / 1000 ))
516	fi
517	elif [[ $remote_host = lctit ]]
518	then
519	(( Memory = memory * tasks_per_node / 1000 ))
520	fi
521
522
523	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
524	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
525	then
526	(( memory = memory * threads_per_task ))
527	fi
528
529
530	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
531	if (( tasks_per_node != 0 ))
532	then
533	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
534	fi
535
536	# Calculate number of processes per node
537	(( processes_per_node = tasks_per_node * threads_per_task ))
538
539
540	# HEADER-AUSGABE
541	if [[ $verify = true ]]
542	then
543	printf "\n\n"
544	printf "#--------------------------------------------------------------# \n"
545	spalte1=SUBJOB;spalte2=$(date)
546	printf "\| $spalte1$spalte2 \| \n"
547	printf "\| \| \n"
548	printf "\| values of parameters/options: \| \n"
549	spalte1=local_host$punkte; spalte2=$punkte$local_host
550	printf "\| $spalte1$spalte2 \| \n"
551	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
552	printf "\| $spalte1$spalte2 \| \n"
553	spalte1=queue$punkte; spalte2=$punkte$queue
554	printf "\| $spalte1$spalte2 \| \n"
555	spalte1=memory$punkte; spalte2="$punkte$memory mb"
556	printf "\| $spalte1$spalte2 \| \n"
557	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
558	printf "\| $spalte1$spalte2 \| \n"
559	spalte1=job_name$punkte; spalte2="$punkte$job_name"
560	printf "\| $spalte1$spalte2 \| \n"
561	printf "#--------------------------------------------------------------# \n\n"
562
563
564	# KONTROLLABFRAGE, OB ALLES O.K.
565	antwort="dummy"
566	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
567	do
568	read antwort?" >>> continue (y/n) ? "
569	done
570	if [[ $antwort = n \|\| $antwort = N ]]
571	then
572	locat=verify; exit
573	fi
574	printf "\n"
575	fi
576
577
578
579	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
580	kennung=$RANDOM
581	job_on_remhost=${job_name}_${kennung}_$local_host
582	job_to_send=job_to_send_$kennung
583	if [[ $delete_dayfile = false ]]
584	then
585	remote_dayfile=${local_host}_${job_name}_result_$kennung
586	local_dayfile=${remote_host}_${job_name}
587	else
588	remote_dayfile=/dev/null
589	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
590	then
591	notification='# @ notification = never'
592	fi
593	fi
594
595
596
597	# Generate the batch job scripts (qsub/msub/LoadLeveler)
598	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
599	then
600
601	# General LoadLeveler settings
602	execute_in_shell="#!/bin/ksh"
603	use_shell="# @ shell = /bin/ksh"
604	consumable_memory="ConsumableMemory($memory mb)"
605	class="# @ class = $queue"
606	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
607	network_to_use="# @ network.mpi = sn_all,shared,us"
608	data_limit="# @ data_limit = 1.76gb"
609	image_size="# @ image_size = 50"
610
611
612	if [[ $remote_host = ibmh ]]
613	then
614	data_limit=""
615	network_to_use=""
616	class=""
617	environment=""
618	elif [[ $remote_host = ibmku ]]
619	then
620	execute_in_shell="#!/usr/bin/ksh"
621	use_shell="# @ shell = /usr/bin/ksh"
622	consumable_memory=""
623	environment=""
624	network_to_use="# @ network.mpi = sn_all,shared,us"
625	data_limit=""
626	image_size=""
627	elif [[ $remote_host = ibms ]]
628	then
629	network_to_use="# @ network.mpi = csss,shared,us"
630	elif [[ $remote_host = ibmy ]]
631	then
632	consumable_memory=""
633	network_to_use=""
634	fi
635
636	cat > $job_to_send << %%END%%
637	$execute_in_shell
638	$use_shell
639
640	# @ job_type = parallel
641	# @ job_name = palm
642	# @ wall_clock_limit = ${timestring},$timestring
643	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
644	# @ output = $remote_dayfile
645	# @ error = $remote_dayfile
646	$image_size
647	$class
648	$environment
649	$network_to_use
650	$data_limit
651	$notification
652
653	%%END%%
654
655	if (( nodes > 0 ))
656	then
657
658	cat >> $job_to_send << %%END%%
659	# @ node = $nodes
660	# @ tasks_per_node = $processes_per_node
661	# @ node_usage = $node_usage
662	# @ queue
663
664	%%END%%
665
666	else
667
668	if [[ $remote_host != ibmy ]]
669	then
670
671	cat >> $job_to_send << %%END%%
672	# @ blocking = unlimited
673	# @ total_tasks = $numprocs
674	# @ node_usage = $node_usage
675	# @ queue
676
677	%%END%%
678
679	else
680
681	cat >> $job_to_send << %%END%%
682	# @ node = 1
683	# @ total_tasks = $numprocs
684	# @ queue
685
686	%%END%%
687
688	fi
689
690	fi
691
692	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
693	then
694
695	cat > $job_to_send << %%END%%
696	#!/bin/ksh
697
698	# @ job_type = serial
699	# @ node_usage = $node_usage
700	# @ job_name = palm
701	# @ wall_clock_limit = ${timestring},$timestring
702	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
703	# @ output = $remote_dayfile
704	# @ error = $remote_dayfile
705	$class
706	$notification
707
708	# @ queue
709
710	%%END%%
711
712	elif [[ $remote_host = lcfimm ]]
713	then
714
715	if [[ $numprocs != 0 ]]
716	then
717	cat > $job_to_send << %%END%%
718	#!/bin/ksh
719	#PBS -N $job_name
720	#PBS -A $project_account
721	#PBS -l walltime=$timestring
722	#PBS -l nodes=${nodes}:ppn=$processes_per_node
723	#PBS -l pmem=${memory}mb
724	#PBS -m abe
725	#PBS -o $remote_dayfile
726	#PBS -j oe
727	mpd &
728
729	%%END%%
730
731	else
732	cat > $job_to_send << %%END%%
733	#!/bin/ksh
734	#PBS -N $job_name
735	#PBS -A $project_account
736	#PBS -l walltime=$timestring
737	#PBS -l ncpus=1
738	#PBS -l pmem=${memory}mb
739	#PBS -m abe
740	#PBS -o $remote_dayfile
741	#PBS -j oe
742
743	%%END%%
744
745	fi
746
747	elif [[ $remote_host = lck ]]
748	then
749
750	if [[ $numprocs != 0 ]]
751	then
752	cat > $job_to_send << %%END%%
753	#!/bin/ksh
754	#PBS -N $job_name
755	#PBS -l walltime=$timestring
756	#PBS -l ncpus=$numprocs
757	#PBS -l pmem=${memory}mb
758	#PBS -o $remote_dayfile
759	#PBS -j oe
760
761	mpd &
762
763	%%END%%
764
765	else
766	cat > $job_to_send << %%END%%
767	#!/bin/ksh
768	#PBS -N $job_name
769	#PBS -l walltime=$timestring
770	#PBS -l ncpus=1
771	#PBS -l pmem=${memory}mb
772	#PBS -o $remote_dayfile
773	#PBS -j oe
774
775	%%END%%
776
777	fi
778
779	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
780	then
781
782	if [[ "$sgi_feature" != "" ]]
783	then
784	feature_directive="#PBS -l feature=$sgi_feature"
785	else
786	feature_directive=""
787	fi
788
789	if [[ $queue = dataq \|\| $queue = permq ]]
790	then
791	feature_directive="#PBS -l feature=data"
792	fi
793
794	if [[ $queue = testq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
795	then
796	queue_directive="#PBS -q $queue"
797	else
798	queue_directive=""
799	fi
800
801	if [[ $email_notification = none ]]
802	then
803	email_directive=""
804	else
805	email_directive="#PBS -M $email_notification"
806	fi
807
808	if [[ $numprocs != 0 ]]
809	then
810	cat > $job_to_send << %%END%%
811	#!/bin/bash
812	#PBS -N $job_name
813	#PBS -l walltime=$timestring
814	#PBS -l nodes=$nodes:ppn=${processes_per_node}
815	#PBS -l naccesspolicy=$node_usage
816	#PBS -o $remote_dayfile
817	#PBS -j oe
818	$feature_directive
819	$queue_directive
820	$email_directive
821
822	eval \`/sw/swdist/bin/modulesinit\`
823	#. /usr/share/modules/init/bash
824	$module_calls
825
826	echo ld_library_path=\$LD_LIBRARY_PATH
827
828	%%END%%
829
830	else
831	cat > $job_to_send << %%END%%
832	#PBS -S /bin/bash
833	#PBS -N $job_name
834	#PBS -l walltime=$timestring
835	#PBS -l ncpus=1
836	#PBS -o $remote_dayfile
837	#PBS -j oe
838	$queue_directive
839	$email_directive
840
841	eval \`/sw/swdist/bin/modulesinit\`
842	#. /usr/share/modules/init/bash
843	$module_calls
844
845	%%END%%
846
847	fi
848
849	elif [[ $remote_host = lcxt4 ]]
850	then
851
852	if [[ $numprocs != 0 ]]
853	then
854	cat > $job_to_send << %%END%%
855	#!/bin/ksh
856	#PBS -S /bin/ksh
857	#PBS -N $job_name
858	#PBS -A $project_account
859	#PBS -j oe
860	#PBS -l walltime=$timestring
861	#PBS -l mppwidth=${numprocs}
862	#PBS -l mppnppn=${processes_per_node}
863	#PBS -m abe
864	#PBS -o $remote_dayfile
865	$email_directive
866
867	. /opt/modules/default/init/ksh
868	$module_calls
869
870	%%END%%
871
872	else
873	cat > $job_to_send << %%END%%
874	#!/bin/ksh
875	#PBS -S /bin/ksh
876	#PBS -N $job_name
877	#PBS -A $project_account
878	#PBS -j oe
879	#PBS -l walltime=$timestring
880	#PBS -l ncpus=1
881	#PBS -l pmem=${memory}mb
882	#PBS -m abe
883	$email_directive
884	#PBS -o $remote_dayfile
885
886	. /opt/modules/default/init/ksh
887	$module_calls
888
889	%%END%%
890
891	fi
892
893	elif [[ $remote_host = lckyoto ]]
894	then
895
896	cat > $job_to_send << %%END%%
897	#!/bin/ksh
898	# @\$-o $remote_dayfile
899	# @\$-eo -oi
900	# @\$-lP 16
901	# @\$-lp 1
902	# @\$-lm 28gb -llm unlimited -ls unlimited
903	# @\$-q $queue
904	# @\$-Pvn abs_pack
905	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
906	#. /thin/local/etc/setprofile/intel-11.0.sh
907	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
908	. ~/.myprofile
909	#. /home2/t/t51254/palm/current_version/myprofile
910	#. /thin/apps/pgi/mpi.sh
911	#
912	env
913	#
914	set -x
915
916	%%END%%
917
918	elif [[ $remote_host = lcxt5m ]]
919	then
920
921	if [[ $numprocs != 0 ]]
922	then
923	cat > $job_to_send << %%END%%
924	#!/bin/ksh
925	#PBS -S /bin/ksh
926	#PBS -N $job_name
927	#PBS -j oe
928	#PBS -l walltime=$timestring
929	#PBS -l mppwidth=${numprocs}
930	#PBS -l mppnppn=${processes_per_node}
931	#PBS -m abe
932	#PBS -o $remote_dayfile
933
934	. /opt/modules/default/init/ksh
935	$module_calls
936
937	%%END%%
938
939	else
940	cat > $job_to_send << %%END%%
941	#!/bin/ksh
942	#PBS -S /bin/ksh
943	#PBS -N $job_name
944	#PBS -j oe
945	#PBS -l walltime=$timestring
946	#PBS -l ncpus=1
947	#PBS -l pmem=${memory}mb
948	#PBS -m abe
949	#PBS -o $remote_dayfile
950
951	. /opt/modules/default/init/ksh
952	$module_calls
953
954	%%END%%
955
956	fi
957
958	elif [[ $remote_host = nech ]]
959	then
960
961	if (( nodes > 1 ))
962	then
963	cat > $job_to_send << %%END%%
964	#!/bin/ksh
965	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
966	#PBS -l ${qsubmem}=${Memory}gb
967	#PBS -b $nodes
968	#PBS -o $remote_dayfile
969	#PBS -N palm
970	#PBS -j o
971	#PBS -T mpisx
972
973	%%END%%
974
975	elif [[ $numprocs != 0 ]]
976	then
977	cat > $job_to_send << %%END%%
978	#!/bin/ksh
979	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
980	#PBS -l ${qsubmem}=${Memory}gb
981	#PBS -o $remote_dayfile
982	#PBS -N palm
983	#PBS -j o
984
985	%%END%%
986
987	else
988	cat > $job_to_send << %%END%%
989	#!/bin/ksh
990	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
991	#PBS -o $remote_dayfile
992	#PBS -j o
993
994	%%END%%
995
996	fi
997
998	elif [[ $remote_host = necriam ]]
999	then
1000
1001	if (( nodes > 1 ))
1002	then
1003	cat > $job_to_send << %%END%%
1004	#!/bin/ksh
1005	#PBS -b $nodes
1006	#PBS -o $remote_dayfile
1007	#PBS -N $job_name
1008	#PBS -j o
1009	#PBS -v MPIPROGINV=YES
1010
1011	%%END%%
1012
1013	elif [[ $numprocs != 0 ]]
1014	then
1015	cat > $job_to_send << %%END%%
1016	#!/bin/ksh
1017	#PBS -q ${queue}
1018	#PBS -o $remote_dayfile
1019	#PBS -N $job_name
1020	#PBS -j o
1021	#PBS -v MPIPROGINV=YES
1022
1023	%%END%%
1024
1025	else
1026	cat > $job_to_send << %%END%%
1027	#!/bin/ksh
1028	#PBS -o $remote_dayfile
1029	#PBS -j o
1030
1031	%%END%%
1032
1033	fi
1034
1035	elif [[ $remote_host = lctit ]]
1036	then
1037	cat > $job_to_send << %%END%%
1038	#!/bin/ksh
1039
1040	%%END%%
1041
1042	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1043	submcom="$submcom -W group_list=$group_number -N $job_name -l walltime=$timestring -l select=$nodes:ncpus=$processes_per_node:mpiprocs=$tasks_per_node:mem=${Memory}gb -l place=scatter -o $remote_dayfile -j oe -q $queue "
1044
1045	else
1046
1047	cat > $job_to_send << %%END%%
1048	# @\$-q ${queue}
1049	# @\$-l${qsubtime} $timestring
1050	# @\$-l${qsubmem} ${memory}mb
1051	# @\$-o $remote_dayfile
1052	# @\$-eo
1053
1054	%%END%%
1055
1056	fi
1057
1058
1059	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1060	# BEI EXIT VERANLASSEN
1061	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1062	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1063	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1064	then
1065	echo "set +vx" >> $job_to_send
1066	echo "trap '" >> $job_to_send
1067	echo "set +vx" >> $job_to_send
1068	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec ]]
1069	then
1070	if [[ $remote_host = ibmh ]]
1071	then
1072	return_queue=c1
1073	elif [[ $remote_host = ibmku ]]
1074	then
1075	return_queue=sdbg2
1076	elif [[ $remote_host = ibms ]]
1077	then
1078	return_queue=p_normal
1079	elif [[ $remote_host = ibmy ]]
1080	then
1081	return_queue=serial
1082	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1083	then
1084	return_queue=serialq
1085	elif [[ $remote_host = necriam ]]
1086	then
1087	return_queue=SP
1088	else
1089	return_queue=unknown
1090	fi
1091
1092	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1093	then
1094
1095	if [[ $remote_host = ibmku ]]
1096	then
1097	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1098	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1099	else
1100	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1101	fi
1102	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1103	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1104	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1105	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1106	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1107	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1108	if [[ $host != "ibmh" ]]
1109	then
1110	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1111	fi
1112	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1113	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1114
1115	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1116	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1117
1118	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1119	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
1120	if [[ $remote_host = ibmku ]]
1121	then
1122	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1123	fi
1124	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1125
1126	elif [[ $remote_host = nech ]]
1127	then
1128	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1129	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1130	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1131	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1132	echo "#PBS -j o" >> $job_to_send
1133	echo " " >> $job_to_send
1134	echo "set -x" >> $job_to_send
1135	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1136	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1137	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1138	echo "%%END%%" >> $job_to_send
1139
1140	elif [[ $remote_host = necriam ]]
1141	then
1142	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1143	echo "#PBS -q $return_queue" >> $job_to_send
1144	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1145	echo "#PBS -j o" >> $job_to_send
1146	echo " " >> $job_to_send
1147	echo "set -x" >> $job_to_send
1148	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1149	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1150	echo "%%END%%" >> $job_to_send
1151
1152	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1153	then
1154	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1155	echo "#!/bin/bash" >> $job_to_send
1156	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1157	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1158	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1159	echo "#PBS -l feature=data" >> $job_to_send
1160	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1161	echo "#PBS -j oe" >> $job_to_send
1162	echo " " >> $job_to_send
1163	echo ". /usr/share/modules/init/bash" >> $job_to_send
1164	echo "set -x" >> $job_to_send
1165	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1166	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1167	echo "%%END%%" >> $job_to_send
1168
1169	else
1170
1171	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1172	echo "# @\\\$-q $return_queue" >> $job_to_send
1173	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1174	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1175	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1176	then
1177	echo "# @\$-l mpp_p=0" >> $job_to_send
1178	fi
1179	echo '# @\$-lF 10mb' >> $job_to_send
1180	# echo '# @\$-o /dev/null' >> $job_to_send
1181	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1182	echo '# @\\\$-eo' >> $job_to_send
1183	echo " " >> $job_to_send
1184	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1185	then
1186	echo "set +vx" >> $job_to_send
1187	echo ". .profile" >> $job_to_send
1188	fi
1189	echo "set -x" >> $job_to_send
1190	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1191	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1192	echo "%%END%%" >> $job_to_send
1193	fi
1194
1195	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1196	then
1197	echo "llsubmit scpjob.$kennung" >> $job_to_send
1198	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1199	then
1200	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1201	echo "msub scpjob.$kennung" >> $job_to_send
1202	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1203	then
1204	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1205	elif [[ $remote_host = t3es ]]
1206	then
1207	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1208	else
1209	echo "qsub scpjob.$kennung" >> $job_to_send
1210	fi
1211	if [[ $remote_host != ibmku ]]
1212	then
1213	echo "rm scpjob.$kennung" >> $job_to_send
1214	fi
1215	if [[ $remote_host = nech ]]
1216	then
1217	echo "cd -" >> $job_to_send
1218	fi
1219	else
1220	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1221	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1222	fi
1223	echo "set -x" >> $job_to_send
1224	echo " ' exit" >> $job_to_send
1225	echo "set -x" >> $job_to_send
1226	fi
1227
1228
1229
1230	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1231	cat $file_to_send >> $job_to_send
1232
1233	if [[ $remote_host = ibm ]]
1234	then
1235	echo " " >> $job_to_send
1236	echo "exit" >> $job_to_send
1237	fi
1238
1239	# remove job file
1240	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1241	then
1242	echo " " >> $job_to_send
1243	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1244	fi
1245
1246
1247
1248	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1249	if [[ -z $remote_user ]]
1250	then
1251	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1252	then
1253	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1254	fi
1255	fi
1256
1257
1258
1259	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1260	if [[ $no_submit = false ]]
1261	then
1262	if [[ $remote_host != $local_host ]]
1263	then
1264	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1265	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1266	then
1267	job_catalog_save=$job_catalog
1268	job_catalog=job_queue
1269	elif [[ $remote_host = nech ]]
1270	then
1271	job_catalog_save=$job_catalog
1272	job_catalog=/hpf/b/${remote_user}/job_queue
1273	fi
1274	if [[ $local_host = decalpha ]]
1275	then
1276	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1277	# PASSWORT AB
1278	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1279	elif [[ $remote_host = nech ]]
1280	then
1281	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1282	# TRANSFERIERT WERDEN
1283	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1284	else
1285	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1286	fi
1287	if [[ $? = 1 ]]
1288	then
1289	locat=scp; exit
1290	fi
1291	if [[ $remote_host = ibms ]]
1292	then
1293	job_catalog=$job_catalog_save
1294	fi
1295	[[ $verify = true ]] && printf "\n >>> finished\n"
1296	else
1297	eval job_catalog=$job_catalog
1298	cp $job_to_send ${job_catalog}/$job_on_remhost
1299	fi
1300
1301
1302
1303	# NQS- BZW. LOADLEVELER-JOB STARTEN
1304	if [[ $remote_host != $local_host ]]
1305	then
1306	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1307
1308	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1309	then
1310	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1311	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1312	elif [[ $remote_host = ibmku ]]
1313	then
1314	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1315	else
1316	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1317	fi
1318
1319	[[ $verify = true ]] && printf " >>> o.k.\n"
1320	else
1321	cd $job_catalog
1322	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1323	then
1324	eval $submcom $job_on_remhost
1325	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxt4 \|\| $localhost = lck ]]
1326	then
1327	chmod u+x $job_on_remhost
1328	echo "$submcom $job_on_remhost"
1329	eval $submcom $job_on_remhost
1330	elif [[ $local_host = nech ]]
1331	then
1332	if [[ $queue = default ]]
1333	then
1334	eval $submcom $job_on_remhost
1335	else
1336	eval $submcom -q $queue $job_on_remhost
1337	fi
1338	else
1339	qsub $job_on_remhost
1340	fi
1341
1342	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1343	# only at the end of the job.
1344	if [[ $local_host != lctit && $local_host != ibmku ]]
1345	then
1346	rm $job_on_remhost
1347	fi
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 |