Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1053

Last change on this file since 1053 was 1047, checked in by maronga, 12 years ago
last commit documented / added nc2vdf
Property svn:keywords set to `Id Rev`
File size: 53.0 KB

Line
1	#!/bin/ksh
2	#--------------------------------------------------------------------------------#
3	# This file is part of PALM.
4	#
5	# PALM is free software: you can redistribute it and/or modify it under the terms
6	# of the GNU General Public License as published by the Free Software Foundation,
7	# either version 3 of the License, or (at your option) any later version.
8	#
9	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
10	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
11	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
12	#
13	# You should have received a copy of the GNU General Public License along with
14	# PALM. If not, see <http://www.gnu.org/licenses/>.
15	#
16	# Copyright 1997-2012 Leibniz University Hannover
17	#--------------------------------------------------------------------------------#
18	#
19	# Current revisions:
20	# -----------------
21	#
22	# Former revisions:
23	# -----------------
24	# $Id: subjob 1047 2012-11-09 15:32:58Z hoffmann $
25	#
26	# 1046 2012-11-09 14:38:45Z maronga
27	# code put under GPL (PALM 3.9)
28	#
29	# subjob - Plot-Shellskript Version: @(#)SUBJOB 1.0
30	# Prozedur zum automatischen Generieren von Batch-Jobs, die unter NQS
31	# laufen sollen und deren Ergebnis (Dayfile) zum Job-generierenden
32	# Host zurueckgeschickt werden sollen
33
34
35	# letzte Aenderung:
36	# 29/06/94 - Siggi - Beginn mit Erstellung der Originalversion
37	# 08/07/94 - Siggi - Originalversion abgeschlossen (Version 1.0)
38	# 06/02/98 - Siggi - berte validiert
39	# 27/01/01 - Siggi - ground.yonsei.ac.kr validiert, Jobs zur T3E in Korea
40	# moeglich
41	# 08/02/01 - Siggi - alle subjob-Meldungen ins englische uebersetzt
42	# 25/05/02 - Siggi - Unterstuetzung des LoadLeveler
43	# 30/05/02 - Siggi - Validierung fuer ibm-Rechner in Seoul (nobel) sowie
44	# allgemeine Anpassungen fuer ibm-Rechner
45	# 15/10/02 - Siggi - Neue Default-Jobklasse (p_normal) fuer IBM in Seoul
46	# Ruecktransfer des Jobprotokolls fuer diese
47	# Maschine abgeschaltet
48	# 31/10/02 - Siggi - berni validiert
49	# 06/11/02 - Siggi - Neue Jobklassen auf ibmb und ibmh
50	# 08/11/02 - Siggi - quanero validiert
51	# 11/12/02 - Siggi - Notification fuer Transfer-Jobs abgeschaltet
52	# 23/01/03 - Siggi - hostname nobel changed to nobela
53	# 06/02/03 - Siggi - gregale validated
54	# 12/02/03 - Siggi - orkan and maestro validated
55	# 21/02/03 - Siggi - all nobel nodes in Seoul validated
56	# 12/03/03 - Siggi - nec at DKRZ validated
57	# 13/03/03 - Siggi - new nqs resource variable Memory
58	# 07/04/03 - Siggi - processor request option -c on nech needs tasks per
59	# node
60	# 11/04/03 - Siggi - network on ibms has attribute "shared"
61	# 31/07/03 - Siggi - nqs2 on nech implemented (provisional: -h nech2)
62	# cxxl added to ibmh
63	# 29/08/03 - Siggi - changes in job queues and communication system on
64	# ibms
65	# 24/10/03 - Siggi - using alternate hanni address 130.75.4.2
66	# 30/10/03 - Siggi - nech is not supported any more
67	# 10/11/03 - Siggi - nech2 renamed to nech
68	# 20/11/03 - Siggi - submit command on nech changed from qsub.test to qsub
69	# 29/03/04 - Siggi - ground not supported any more, gfdl3 validated
70	# 31/03/04 - Siggi - new option -N for node usage
71	# 12/04/04 - Siggi - scp2 instead of scp used for transfer from decalpha
72	# due to error in ssh installation (otherwise a prompt
73	# for the password appears)
74	# 23/07/04 - Siggi - changes due to the new berni configuration
75	# (federation switch)
76	# 01/09/04 - Gerald new job-classes on hanni
77	# 08/09/04 - Siggi - hanni IP address changed to 130.75.4.10
78	# 23/11/04 - Siggi - new job class cdata on hanni and berni
79	# 03/12/04 - Siggi - notification on ibm switched of in case of
80	# delete_dayfile = true, node usage in cdev set to
81	# shared
82	# 16/02/05 - Gerald hababai validated
83	# 29/03/05 - Micha - new job class channi on hanni
84	# 11/05/05 - Siggi - ConsumableMemory is now required as resource keyword
85	# on ibms
86	# 24/05/05 - Siggi - Default queue on ibms changed from p_normal_1.3 to
87	# p_normal
88	# 30/06/05 - Siggi - network changed for queue cdev from "us" to "ip"
89	# 12/07/05 - Siggi - in network.mpi on ibmh/ibmb "csss" changed to
90	# "sn_all", new job class cexp
91	# 08/09/05 - Siggi - IP-address of gfdl3 changed
92	# 31/10/05 - Siggi - new job class pp on hurricane, serial jobs on
93	# hurricane (with -X0)
94	# 01/11/05 - Siggi - missing queue for jobs submitted on nech (for nech)
95	# added
96	# 30/12/05 - Siggi - change of IP adresses in subnet 130.75.105
97	# 09/02/06 - Siggi - ibmy admitted
98	# 10/02/06 - Siggi - scp2 changed to /bin/scp on decalpha
99	# 13/04/06 - Siggi - ostria admitted
100	# 18/04/06 - Siggi - new option -O for OpenMP usage
101	# 24/05/06 - Siggi - lctit admitted, ftpjob renamed scpjob
102	# 25/07/06 - Siggi - gfdl5 (ibmy) admitted for submitting jobs
103	# 27/09/06 - Siggi - breg/hreg extended with berni/hanni
104	# 25/10/06 - Siggi - data_limit set to 1.76 GByte on hanni and berni
105	# 28/11/06 - Siggi - levanto admitted
106	# 13/02/07 - Siggi - hpmuk releated code removed
107	# 01/03/07 - Siggi - adjustments for RIAM machines gate and NEC-SX8 (n-sx)
108	# 12/04/07 - Siggi - option -f (filetransfer protocol) removed, scp only
109	# 27/07/07 - Siggi - autan admitted
110	# 03/08/07 - Marcus- lcfimm admitted
111	# 08/10/07 - Siggi - further job classes added for hanni (csoloh...)
112	# 15/10/07 - Siggi - preliminary adjustments for lctit based on Jin's
113	# suggestions
114	# 19/10/07 - Marcus- add new optional argument -g group_number
115	# 19/10/07 - Siggi - a ";" was missing in the last change done by Marcus
116	# 30/10/07 - Marcus- further adjustments for queues on lctit
117	# 15/05/08 - Siggi - adjustments for lcxt4 (Bergen Center for Computational
118	# Science)
119	# 14/07/08 - Siggi - adjustments for lcsgih
120	# 23/09/08 - Gerald- paesano admitted
121	# 02/10/08 - Siggi - PBS adjustments for lcxt4
122	# 02/03/09 - Siggi - Adjustments for new NEC-SX9 at RIAM
123	# 16/04/09 - Marcus- Adjustments for lcsgib and lcsgih
124	# 21/04/09 - Siggi - adjustments for new IBM at DKRZ, which is now ibmh
125	# 18/05/09 - Siggi - Settings for serial jobs on lcsgi changed
126	# 24/06/09 - BjornM- adjustments for lcxt4 (loading modules manually)
127	# 08/07/09 - Siggi - option -e added (email notification on lcsgih/b)
128	# 20/07/09 - Siggi - On lcsgi, jobs for returning the job protocol are
129	# now run on the data nodes (feature=data)
130	# 25/08/09 - BjornM- adapted for lck
131	# 26/08/09 - Marcus- caurus admitted; optional qos feature hiprio on lcsgi
132	# 03/09/09 - Siggi - PBS sgi feature directive only used if explicitly
133	# set in the config file by the user
134	# 16/10/09 - Carolin-adjustments for archiving on SGI-ICE of binary files;
135	# adding special1q
136	# 01/12/09 - BjornM- re-adjustments for lcxt4, added sno (130.75.105.113)
137	# 16/10/09 - Carolin-adding permq
138	# 01/02/10 - Siggi - adapted for lcxt5m and lckyoto (Fujitsu HX600)
139	# 03/02/10 - Siggi - bug in serial jobs removed
140	# 26/02/10 - BjornM- re-adjustments for lcxt4 (new modules, email
141	# notification)
142	# 01/03/10 - Siggi - loading of modules controlled by environment variable
143	# module_calls
144	# 17/08/10 - BjornM- re-adjustments for lcxt4 (location of qsub)
145	# 25/08/10 - BjornM- account geofysisk replaced by guest for lcxt4
146	# 25/08/10 - Siggi - new variable project_account in pbs-statements for
147	# lcxt4
148	# 08/12/10 - Siggi - initialization of the module command changed for
149	# SGI-ICE/lcsgi
150	# adjustments for Kyushu Univ. (lcrte, ibmku)
151	# 14/12/10 - Siggi - adjustments for new Tsubame system at Tokyo
152	# institute of technology (lctit)
153	# 02/02/11 - Siggi - further asjustments for Tsubame concerning openMP
154	# 06/03/11 - Siggi - adjustments for ibmkisti
155	# 17/03/11 - Siggi - adjustments for openmp usage on ibmkisti
156	# 03/04/11 - Micha - added lckordi
157	# 17/08/11 - Siggi - hicegate0 added
158	# 18/08/11 - Siggi - workaround on lcsgi in order to avoid appends to file
159	# last_job_transfer_protocol
160	# 21/08/11 - Siggi - inferno admitted
161	# 29/11/11 - Siggi - adjustments for lcsgih/lcsgib queues, pingui admitted
162	# 21/12/11 - Theres- solano admitted
163	# 31/01/12 - Matthias - adjustments for ibmh
164	# 02/05/12 - Bjoern - lcxt4 is now lcxe6, adjustments for lcxe6,
165	# bugfixes for the parameter file check, initial
166	# commands on ssh calls sourced out to .mrun.config
167	# 02/10/12 - Siggi - adjustments for lcsb
168	# 01/11/12 - Siggi - adjustments for lckiaps
169
170
171
172	# VARIABLENVEREINBARUNGEN + DEFAULTWERTE
173	delete_dayfile=false
174	email_notification=none
175	group_number=none
176	locat=normal
177	no_default_queue=none
178	no_submit=false
179	job_catalog="~/job_queue"
180	job_name=none
181	local_user=$LOGNAME
182	node_usage=shared
183	numprocs=0
184	punkte="..........................................................."
185	submcom=qsub
186	queue=default
187	remote_host=none
188	remote_user=""
189	verify=true
190
191	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
192	typeset -i inumprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
193	typeset -L20 spalte1
194	typeset -R40 spalte2
195	typeset -L60 spalte3
196
197
198
199	# FEHLERBEHANDLUNG
200	# BEI EXIT:
201	trap 'if [[ $locat != normal ]]
202	then
203	case $locat in
204	(option) printf "\n --> available optios can be displayed"
205	printf " by typing:"
206	printf "\n \"subjob ?\" \n";;
207	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
208	(*) printf "\n +++ unknown error"
209	printf "\n please inform S. Raasch!\n"
210	esac
211	[[ -f $job_to_send ]] && rm $job_to_send
212	printf "\n\n+++ SUBJOB killed \n\n"
213	fi' exit
214
215
216	# BEI TERMINAL-BREAK:
217	trap '[[ -f $job_to_send ]] && rm $job_to_send
218	printf "\n\n+++ SUBJOB killed \n\n"
219	exit
220	' 2
221
222
223
224
225	# LOKALEN HOSTNAMEN FESTSTELLEN
226	local_host=$(hostname)
227
228
229
230	# HOSTSPEZIFISCHE VARIABLEN VEREINBAREN BZW. PRUEFEN, OB LOKALER HOST
231	# UEBERHAUPT ZULAESSIG IST
232	# Note: One of the entries for "lck" or "lckordi" always has to be
233	# comment out, because the hostname (node*) is same for both machines
234	case $local_host in
235	(atmos) local_addres=172.20.25.35; local_host=lcide;;
236	(autan) local_addres=130.75.105.57; local_host=lcmuk;;
237	(bora) local_addres=130.75.105.103; local_host=lcmuk;;
238	(bd1) local_addres=130.73.232.64; local_host=lcsgib;;
239	(bd2) local_addres=130.73.232.65; local_host=lcsgib;;
240	(bd3) local_addres=130.73.232.66; local_host=lcsgib;;
241	(bd4) local_addres=130.73.232.67; local_host=lcsgib;;
242	(b01*\|bicegate1) local_addres=130.73.232.102; local_host=lcsgib;;
243	(bicegate2) local_addres=130.73.232.103; local_host=lcsgib;;
244	(blizzard1) local_addres=136.172.40.15; local_host=ibmh;;
245	(breva) local_addres=130.75.105.98; local_host=lcmuk;;
246	(caurus) local_addres=130.75.105.19; local_host=lcmuk;;
247	(climate*) local_addres=165.132.26.68; local_host=lcyon;;
248	(compute-*.local) local_addres=172.20.4.2; local_host=lcfimm;;
249	(cs*) local_addres=136.172.44.131; local_host=nech;;
250	(elephanta) local_addres=130.75.105.6; local_host=lcmuk;;
251	(fimm.bccs.uib.no) local_addres=172.20.4.2; local_host=lcfimm;;
252	(node*) local_addres=165.132.26.61 local_host=lck;;
253	# (node*) local_addres=210.219.61.8 local_host=lckordi;;
254	(gaia*) local_addres=150.183.146.24; local_host=ibmkisti;;
255	(gallego) local_addres=130.75.105.10; local_host=lcmuk;;
256	(gfdl5) local_addres=165.132.26.58; local_host=ibmy;;
257	(gfdl3.yonsei.ac.kr) local_addres=165.132.26.56; local_host=decalpha;;
258	(gregale) local_addres=130.75.105.109; local_host=lcmuk;;
259	(hababai) local_addres=130.75.105.108; local_host=lcmuk;;
260	(hexagon.bccs.uib.no) local_addres=129.177.20.113; local_host=lcxe6;;
261	(hd1) local_addres=130.75.4.104; local_host=lcsgih;;
262	(hd2) local_addres=130.75.4.105; local_host=lcsgih;;
263	(hd3) local_addres=130.75.4.106; local_host=lcsgih;;
264	(hd4) local_addres=130.75.4.107; local_host=lcsgih;;
265	(hicegate0) local_addres=130.75.4.101; local_host=lcsgih;;
266	(h01*\|hicegate1) local_addres=130.75.4.102; local_host=lcsgih;;
267	(hicegate2) local_addres=130.75.4.103; local_host=lcsgih;;
268	(hx*) local_addres=133.3.51.11; local_host=lckyoto;;
269	(inferno) local_addres=130.75.105.5; local_host=lcmuk;;
270	(irifi) local_addres=130.75.105.104; local_host=lcmuk;;
271	(sno) local_addres=130.75.105.113; local_host=lcmuk;;
272	(levanto) local_addres=130.75.105.45; local_host=lcmuk;;
273	(login*) local_addres=118.128.66.223; local_host=lckiaps;;
274	(maestro) local_addres=130.75.105.2; local_host=lcmuk;;
275	(meteo-login*) local_addres=193.166.211.144;local_host=lcxt5m;;
276	(hexagon*) local_addres=129.177.20.113; local_host=lcxe6;;
277	(nobel*) local_addres=150.183.5.101; local_host=ibms;;
278	(orkan) local_addres=130.75.105.3; local_host=lcmuk;;
279	(ostria) local_addres=130.75.105.106; local_host=lcmuk;;
280	(paesano) local_addres=130.75.105.46; local_host=lcmuk;;
281	(pingui) local_addres=134.106.74.118; local_host=lcfor;;
282	(quanero) local_addres=130.75.105.107; local_host=lcmuk;;
283	(rte*) local_addres=133.5.185.60; local_host=lcrte;;
284	(scirocco) local_addres=172.20.25.41; local_host=lcmuk;;
285	(solano) local_addres=130.75.105.110; local_host=lcmuk;;
286	(sun1\|sun2) local_addres=130.75.6.1; local_host=unics;;
287	(sx-*) local_addres=172.16.1.131; local_host=necriam;;
288	(r1*) local_addres=130.75.4.102; local_host=lcsgih;;
289	(r2*) local_addres=130.73.232.102; local_host=lcsgib;;
290	(t2a*) local_addres=10.1.6.165; local_host=lctit;;
291	(urban*) local_addres=147.46.30.151 local_host=lcsb;;
292	(vorias) local_addres=172.20.25.43; local_host=lcmuk;;
293	(*.cc.kyushu-u.ac.jp) local_addres=133.5.4.129; local_host=ibmku;;
294	(*) printf "\n +++ \"$local_host\" unknown";
295	printf "\n please inform S. Raasch!";
296	locat=parameter; exit;;
297	esac
298
299
300
301	# REMOTE HOST DEFAULTMAESSIG = LOCAL HOST SETZEN
302	remote_host=$local_host
303
304
305
306
307	# PROZEDUROPTIONEN EINLESEN
308	while getopts :c:dDe:g:h:m:n:N:O:q:t:T:u:vX: option
309	do
310	case $option in
311	(c) job_catalog=$OPTARG;;
312	(d) delete_dayfile=true;;
313	(D) no_submit=true;;
314	(e) email_notification=$OPTARG;;
315	(g) group_number=$OPTARG;;
316	(h) remote_host=$OPTARG;;
317	(m) memory=$OPTARG;;
318	(n) job_name=$OPTARG;;
319	(N) node_usage=$OPTARG;;
320	(O) threads_per_task=$OPTARG;;
321	(q) no_default_queue=$OPTARG;;
322	(t) cputime=$OPTARG;;
323	(T) tasks_per_node=$OPTARG;;
324	(u) remote_user=$OPTARG;;
325	(v) verify=false;;
326	(X) numprocs=$OPTARG;;
327	(\?) printf "\n +++ Option $OPTARG unknown \n";
328	locat=option; exit;;
329	esac
330	done
331
332
333	# JOBDATEINAMEN ALS NAECHSTES ARGUMENT HOLEN
334	shift OPTIND-1; file_to_send=$1
335
336
337	# KURZE AUFRUFBESCHREIBUNG WIRD HIER AUSGEGEBEN
338	if [ "$1" = "?" ]
339	then
340	(printf "\n *** subjob can be called as follows:\n"
341	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
342	printf "\n Description of available options:\n"
343	printf "\n Option Description Default-Value"
344	printf "\n -c job-input- and output-catalog ~/job_queue"
345	printf "\n -d no job-protocol will be created ---"
346	printf "\n -D only the job-file will be created ---"
347	printf "\n -h execution host, available hosts: $remote_host"
348	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, ibmy, lc...,"
349	printf "\n lckiaps, lctit, nech, necriam, unics"
350	printf "\n -m memory demand per process in MByte ---"
351	printf "\n -n jobname <jobdatei>"
352	printf "\n -O threads per task (for OpenMP usage) 1"
353	printf "\n -q job-queue to be used default"
354	printf "\n -t allowed cpu-time in seconds ---"
355	printf "\n -T tasks per node (on parallel hosts) ---"
356	printf "\n -u username on execution host from .netrc"
357	printf "\n -v no prompt for confirmation ---"
358	printf "\n -X # of processors (on parallel hosts) 1"
359	printf "\n "
360	printf "\n The only possible positional parameter is <jobfile>:"
361	printf "\n The complete NQS-job must be provided here."
362	printf "\n <jobfile>=? creates this outline\n\n") \| more
363	exit
364	fi
365
366
367
368	# PRUEFEN, OB JOBDATEI ANGEGEBEN WURDE UND OB SIE AUCH EXISTIERT
369	if [[ "$file_to_send" = "" ]]
370	then
371	printf "\n +++ job-file missing"
372	locat=parameter; exit
373	else
374	if [[ -f $file_to_send ]]
375	then
376	true
377	else
378	printf "\n +++ job-file: "
379	printf "\n $file_to_send"
380	printf "\n does not exist"
381	locat=parameter; exit
382	fi
383	fi
384
385
386
387	# FALLS KEIN JOBNAME ANGEGEBEN WURDE, WIRD JOBNAME = JOBDATEINAME
388	# GESETZT. VORAUSSETZUNG: JOBDATEINAME BEINHALTET KEINE PFADE
389	if [[ $job_name = none ]]
390	then
391	job_name=$file_to_send
392	fi
393	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
394	then
395	printf "\n +++ job-file name: "
396	printf "\n $job_name"
397	printf "\n must not contain \"/\"-characters"
398	locat=parameter; exit
399	fi
400
401
402
403
404	# HOSTSPEZIFISCHE GROESSEN VEREINBAREN BZW. ABBRUCH BEI UNZULAESSIGEM HOST
405	# ODER WENN HOST NICHT ANGEGEBEN WURDE
406	if [[ $remote_host = none ]]
407	then
408	printf "\n +++ host missing"
409	locat=option; exit
410	else
411	case $remote_host in
412	(ibm) queue=p690_standard; remote_addres=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
413	(ibmh) queue=no_class; remote_addres=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
414	(ibmkisti) queue=class.32plus; remote_addres=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
415	(ibmku) queue=s4; remote_addres=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
416	(ibms) queue=p_normal; remote_addres=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
417	(ibmy) queue=parallel; remote_addres=165.132.26.58; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
418	(lcfimm) remote_addres=172.20.4.2; submcom=/opt/torque/bin/qsub;;
419	(lckyoto) remote_addres=133.3.51.11; submcom=/thin/local/bin/qsub;;
420	(lcsgib) queue=smallq; remote_addres=130.73.232.104; submcom=/opt/moab/bin/msub;;
421	(lcsgih) queue=smallq; remote_addres=130.75.4.101; submcom=/opt/moab/bin/msub;;
422	(lck) remote_addres=165.132.26.61; submcom=/usr/torque/bin/qsub;;
423	(lckiaps) remote_addres=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
424	(lckordi) remote_addres=210.219.61.8; submcom=/usr/torque/bin/qsub;;
425	(lcsb) remote_addres=147.46.30.151; submcom=/usr/torque/bin/qsub;;
426	(lctit) queue=S; remote_addres=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
427	(lcxe6) remote_addres=129.177.20.113; submcom=/opt/torque/2.5.10/bin/qsub;;
428	(lcxt5m) remote_addres=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
429	(lcyon) remote_addres=165.132.26.68; submcom=/usr/torque/bin/qsub;;
430	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=136.172.44.147; submcom="/usr/local/bin/qsub";;
431	(necriam) qsubmem=memsz_job; qsubtime=cputim_job; remote_addres=172.16.1.131; submcom="/usr/bin/nqsII/qsub";;
432	(vpp) qsubmem=m; qsubtime=t; queue=vpp; remote_addres=130.75.4.130;;
433	(unics) qsubmem=d; qsubtime=t; queue=unics; remote_addres=130.75.6.1;;
434	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
435	locat=parameter; exit;;
436	esac
437	fi
438
439
440	# EVTL. PRUEFEN, OB ANGEGEBENE QUEUE ZULAESSIG IST
441	if [[ $no_default_queue != none ]]
442	then
443	error=false
444	ndq=$no_default_queue
445	case $remote_host in
446	(ibm) case $ndq in
447	(p690_express\|p690_standard\|p690_long) error=false;;
448	(*) error=true;;
449	esac;;
450	(ibmh) case $ndq in
451	(no_class) error=false;;
452	(*) error=true;;
453	esac;;
454	(ibmkisti) case $ndq in
455	(class.32plus\|class.1-2\|class.2-32) error=false;;
456	(*) error=true;;
457	esac;;
458	(ibmku) case $ndq in
459	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
460	(*) error=true;;
461	esac;;
462	(ibms) case $ndq in
463	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
464	(*) error=true;;
465	esac;;
466	(ibmy) case $ndq in
467	(parallel) error=false;;
468	(*) error=true;;
469	esac;;
470	(lckiaps) case $ndq in
471	(express\|normal) error=false;;
472	(*) error=true;;
473	esac;;
474	(lckyoto) case $ndq in
475	(eh\|ph) error=false;;
476	(*) error=true;;
477	esac;;
478	(lcsgib\|lcsgih) case $ndq in
479	(testq\|serialq\|smallq\|mediumq\|bigq\|workq\|dataq\|permq\|special1q) error=false;;
480	(*) error=true;;
481	esac;;
482	(lctit) case $ndq in
483	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
484	(*) error=true;;
485	esac;;
486	(t3eb) case $ndq in
487	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
488	(*) error=true;;
489	esac;;
490	(necriam) case $ndq in
491	(SP\|SS\|P6) error=false;;
492	(*) error=true;;
493	esac;;
494	(t3eh) case $ndq in
495	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
496	(*) error=true;;
497	esac;;
498	(t3ej2\|t3ej5) case $ndq in
499	(low\|normal\|high) error=false;;
500	(*) error=true;;
501	esac;;
502	(t3es) case $ndq in
503	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
504	(*) error=true;;
505	esac;;
506	(unics) case $ndq in
507	(unics\|ht) error=false;;
508	(*) error=true;;
509	esac;;
510	esac
511	if [[ $error = true ]]
512	then
513	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
514	locat=parameter; exit
515	else
516	queue=$no_default_queue
517	fi
518	fi
519
520
521
522	# PRUEFEN DER CPU-ZEIT, ZEIT NACH STUNDEN, MINUTEN UND SEKUNDEN
523	# AUFTEILEN
524	done=false
525	while [[ $done = false ]]
526	do
527	if (( $cputime <= 0 ))
528	then
529	printf "\n +++ wrong cpu-time or cpu-time missing"
530	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
531	printf "\n >>> "
532	read cputime 1>/dev/null 2>&1
533	else
534	done=true
535	fi
536	done
537	if [[ $remote_host = nech ]]
538	then
539	if (( tasks_per_node != 0 ))
540	then
541	(( cputime = cputime * tasks_per_node ))
542	elif [[ $numprocs != 0 ]]
543	then
544	(( cputime = cputime * numprocs ))
545	fi
546	fi
547	(( stunden = cputime / 3600 ))
548	(( resttime = cputime - stunden * 3600 ))
549	(( minuten = resttime / 60 ))
550	(( sekunden = resttime - minuten * 60 ))
551	timestring=${stunden}:${minuten}:${sekunden}
552
553
554
555	# PRUEFEN DER KERNSPEICHERANFORDERUNG
556	done=false
557	while [[ $done = false ]]
558	do
559	if (( memory <= 0 ))
560	then
561	printf "\n +++ wrong memory demand or memory demand missing"
562	printf "\n >>> Please type memory in MByte per process as INTEGER:"
563	printf "\n >>> "
564	read memory 1>/dev/null 2>&1
565	else
566	done=true
567	fi
568	done
569
570	if [[ $remote_host = nech \|\| $remote_host = necriam ]]
571	then
572	if (( tasks_per_node != 0 ))
573	then
574	(( Memory = memory * tasks_per_node / 1000 ))
575	elif [[ $numprocs != 0 ]]
576	then
577	(( Memory = memory * numprocs / 1000 ))
578	else
579	(( Memory = memory / 1000 ))
580	fi
581	elif [[ $remote_host = lctit ]]
582	then
583	(( Memory = memory * tasks_per_node / 1000 ))
584	fi
585
586
587	# SPEICHERBERECHNUNG BEI OPENMP-NUTZUNG
588	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
589	then
590	(( memory = memory * threads_per_task ))
591	fi
592
593
594	# BERECHNUNG DER ANZAHL DER ZU VERWENDENDEN KNOTEN
595	if (( tasks_per_node != 0 ))
596	then
597	(( nodes = numprocs / ( tasks_per_node * threads_per_task ) ))
598	fi
599
600	# Calculate number of processes per node
601	(( processes_per_node = tasks_per_node * threads_per_task ))
602
603	# Calculate number of MPI tasks
604	(( mpi_tasks = numprocs / threads_per_task ))
605
606
607	# HEADER-AUSGABE
608	if [[ $verify = true ]]
609	then
610	printf "\n\n"
611	printf "#--------------------------------------------------------------# \n"
612	spalte1=SUBJOB;spalte2=$(date)
613	printf "\| $spalte1$spalte2 \| \n"
614	printf "\| \| \n"
615	printf "\| values of parameters/options: \| \n"
616	spalte1=local_host$punkte; spalte2=$punkte$local_host
617	printf "\| $spalte1$spalte2 \| \n"
618	spalte1=remote_host$punkte; spalte2=$punkte$remote_host
619	printf "\| $spalte1$spalte2 \| \n"
620	spalte1=queue$punkte; spalte2=$punkte$queue
621	printf "\| $spalte1$spalte2 \| \n"
622	spalte1=memory$punkte; spalte2="$punkte$memory mb"
623	printf "\| $spalte1$spalte2 \| \n"
624	spalte1=cputime$punkte; spalte2="$punkte$cputime sec"
625	printf "\| $spalte1$spalte2 \| \n"
626	spalte1=job_name$punkte; spalte2="$punkte$job_name"
627	printf "\| $spalte1$spalte2 \| \n"
628	printf "#--------------------------------------------------------------# \n\n"
629
630
631	# KONTROLLABFRAGE, OB ALLES O.K.
632	antwort="dummy"
633	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
634	do
635	read antwort?" >>> continue (y/n) ? "
636	done
637	if [[ $antwort = n \|\| $antwort = N ]]
638	then
639	locat=verify; exit
640	fi
641	printf "\n"
642	fi
643
644
645
646	# ZUFALLSKENNUNG GENERIEREN UND JOBNAMEN AUF ZIELRECHNER BESTIMMEN
647	kennung=$RANDOM
648	job_on_remhost=${job_name}_${kennung}_$local_host
649	job_to_send=job_to_send_$kennung
650	if [[ $delete_dayfile = false ]]
651	then
652	remote_dayfile=${local_host}_${job_name}_result_$kennung
653	local_dayfile=${remote_host}_${job_name}
654	else
655	remote_dayfile=/dev/null
656	fi
657
658
659
660	# Generate the batch job scripts (qsub/msub/LoadLeveler)
661	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
662	then
663
664	# General LoadLeveler settings
665	execute_in_shell="#!/bin/ksh"
666	use_shell="# @ shell = /bin/ksh"
667	consumable_memory="ConsumableMemory($memory mb)"
668	class="# @ class = $queue"
669	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
670	network_to_use="# @ network.mpi = sn_all,shared,us"
671	data_limit="# @ data_limit = 1.76gb"
672	image_size="# @ image_size = 50"
673	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
674
675	if [[ $email_notification = none ]]
676	then
677	notify_user=""
678	else
679	notify_user="# @ notify_user = $email_notification"
680	if [[ $delete_dayfile = true ]]
681	then
682	notification='# @ notification = never'
683	fi
684	fi
685
686	if [[ $remote_host = ibmh ]]
687	then
688	data_limit=""
689	network_to_use=""
690	class=""
691	environment=""
692	rset="# @ rset = RSET_MCM_AFFINITY"
693	task_affinity="# @ task_affinity = core(1)"
694	elif [[ $remote_host = ibmkisti ]]
695	then
696	network_to_use="# @ network.MPI = sn_all,shared,US"
697	wall_clock_limit="# @ wall_clock_limit = $timestring"
698	if [[ $threads_per_task = 1 ]]
699	then
700	rset="# @ rset = RSET_MCM_AFFINITY"
701	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
702	fi
703	environment=""
704	use_shell=""
705	data_limit=""
706	image_size=""
707	elif [[ $remote_host = ibmku ]]
708	then
709	execute_in_shell="#!/usr/bin/ksh"
710	use_shell="# @ shell = /usr/bin/ksh"
711	consumable_memory=""
712	environment=""
713	network_to_use="# @ network.mpi = sn_all,shared,us"
714	data_limit=""
715	image_size=""
716	elif [[ $remote_host = ibms ]]
717	then
718	network_to_use="# @ network.mpi = csss,shared,us"
719	elif [[ $remote_host = ibmy ]]
720	then
721	consumable_memory=""
722	network_to_use=""
723	fi
724
725	cat > $job_to_send << %%END%%
726	$execute_in_shell
727	$use_shell
728
729	# @ job_type = parallel
730	# @ job_name = $job_name
731	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
732	# @ output = $remote_dayfile
733	# @ error = $remote_dayfile
734	$wall_clock_limit
735	$image_size
736	$class
737	$environment
738	$network_to_use
739	$data_limit
740	$rset
741	$mcm_affinity_options
742	$task_affinity
743	$notification
744	$notify_user
745
746	%%END%%
747
748	if (( nodes > 0 ))
749	then
750
751	if [[ $remote_host != ibmkisti ]]
752	then
753
754	cat >> $job_to_send << %%END%%
755	# @ node = $nodes
756	# @ tasks_per_node = $processes_per_node
757	# @ node_usage = $node_usage
758	# @ queue
759
760	%%END%%
761
762	else
763
764	cat >> $job_to_send << %%END%%
765	# @ total_tasks = $mpi_tasks
766	# @ blocking = unlimited
767	# @ queue
768
769	%%END%%
770
771	fi
772
773	else
774
775	if [[ $remote_host != ibmy ]]
776	then
777
778	cat >> $job_to_send << %%END%%
779	# @ blocking = unlimited
780	# @ total_tasks = $numprocs
781	# @ node_usage = $node_usage
782	# @ queue
783
784	%%END%%
785
786	else
787
788	cat >> $job_to_send << %%END%%
789	# @ node = 1
790	# @ total_tasks = $numprocs
791	# @ queue
792
793	%%END%%
794
795	fi
796
797	fi
798
799	# workaround because of silly job filter on ibmkisti
800	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
801	then
802	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
803	fi
804
805	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
806	then
807
808	cat > $job_to_send << %%END%%
809	#!/bin/ksh
810
811	# @ job_type = serial
812	# @ node_usage = $node_usage
813	# @ job_name = palm
814	# @ wall_clock_limit = ${timestring},$timestring
815	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
816	# @ output = $remote_dayfile
817	# @ error = $remote_dayfile
818	$class
819	$notification
820
821	# @ queue
822
823	%%END%%
824
825	elif [[ $remote_host = lcfimm ]]
826	then
827
828	if [[ $numprocs != 0 ]]
829	then
830	cat > $job_to_send << %%END%%
831	#!/bin/ksh
832	#PBS -N $job_name
833	#PBS -A $project_account
834	#PBS -l walltime=$timestring
835	#PBS -l nodes=${nodes}:ppn=$processes_per_node
836	#PBS -l pmem=${memory}mb
837	#PBS -m abe
838	#PBS -o $remote_dayfile
839	#PBS -j oe
840	mpd &
841
842	%%END%%
843
844	else
845	cat > $job_to_send << %%END%%
846	#!/bin/ksh
847	#PBS -N $job_name
848	#PBS -A $project_account
849	#PBS -l walltime=$timestring
850	#PBS -l ncpus=1
851	#PBS -l pmem=${memory}mb
852	#PBS -m abe
853	#PBS -o $remote_dayfile
854	#PBS -j oe
855
856	%%END%%
857
858	fi
859
860	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
861	then
862
863	if [[ $numprocs != 0 ]]
864	then
865	cat > $job_to_send << %%END%%
866	#!/bin/ksh
867	#PBS -N $job_name
868	#PBS -l walltime=$timestring
869	#PBS -l ncpus=$numprocs
870	#PBS -l pmem=${memory}mb
871	#PBS -o $remote_dayfile
872	#PBS -l nodes=$nodes:ppn=${processes_per_node}
873	#PBS -j oe
874
875	mpd &
876
877	%%END%%
878
879	else
880	cat > $job_to_send << %%END%%
881	#!/bin/ksh
882	#PBS -N $job_name
883	#PBS -l walltime=$timestring
884	#PBS -l ncpus=1
885	#PBS -l pmem=${memory}mb
886	#PBS -o $remote_dayfile
887	#PBS -j oe
888
889	%%END%%
890
891	fi
892
893	elif [[ $remote_host = lckiaps ]]
894	then
895
896	if [[ $numprocs != 0 ]]
897	then
898	cat > $job_to_send << %%END%%
899	#!/bin/ksh
900	#PBS -N $job_name
901	#PBS -l walltime=$timestring
902	#PBS -l select=1:ncpus=$numprocs
903	#PBS -l pmem=${memory}mb
904	#PBS -q $queue
905	#PBS -o $remote_dayfile
906	#PBS -j oe
907	#PBS -V
908
909	%%END%%
910
911	else
912	cat > $job_to_send << %%END%%
913	#!/bin/ksh
914	#PBS -N $job_name
915	#PBS -l walltime=$timestring
916	#PBS -l ncpus=1
917	#PBS -l pmem=${memory}mb
918	#PBS -o $remote_dayfile
919	#PBS -j oe
920
921	%%END%%
922
923	fi
924
925	elif [[ $remote_host = lcyon ]]
926	then
927
928	if [[ $numprocs != 0 ]]
929	then
930	cat > $job_to_send << %%END%%
931	#!/bin/ksh
932	#PBS -N $job_name
933	#PBS -l walltime=$timestring
934	#PBS -l ncpus=$numprocs
935	#PBS -l pmem=${memory}mb
936	#PBS -o $remote_dayfile
937	#PBS -j oe
938
939	%%END%%
940
941	else
942	cat > $job_to_send << %%END%%
943	#!/bin/ksh
944	#PBS -N $job_name
945	#PBS -l walltime=$timestring
946	#PBS -l ncpus=1
947	#PBS -l pmem=${memory}mb
948	#PBS -o $remote_dayfile
949	#PBS -j oe
950
951	%%END%%
952
953	fi
954
955	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
956	then
957
958	if [[ "$sgi_feature" != "" ]]
959	then
960	feature_directive="#PBS -l feature=$sgi_feature"
961	else
962	feature_directive=""
963	fi
964
965	if [[ $queue = dataq \|\| $queue = permq ]]
966	then
967	feature_directive="#PBS -l feature=data"
968	fi
969
970	if [[ $queue = testq \|\| $queue = mediumq \|\| $queue = bigq \|\| $queue = workq \|\| $queue = dataq \|\| $queue = permq \|\| $queue = serialq \|\| $queue = special1q ]]
971	then
972	queue_directive="#PBS -q $queue"
973	else
974	queue_directive=""
975	fi
976
977	if [[ $email_notification = none ]]
978	then
979	email_directive=""
980	else
981	email_directive="#PBS -M $email_notification"
982	fi
983
984	if [[ $numprocs != 0 ]]
985	then
986	cat > $job_to_send << %%END%%
987	#!/bin/bash
988	#PBS -N $job_name
989	#PBS -l walltime=$timestring
990	#PBS -l nodes=$nodes:ppn=${processes_per_node}
991	#PBS -l naccesspolicy=$node_usage
992	#PBS -o $remote_dayfile
993	#PBS -j oe
994	$feature_directive
995	$queue_directive
996	$email_directive
997
998	eval \`/sw/swdist/bin/modulesinit\`
999	#. /usr/share/modules/init/bash
1000	$init_cmds
1001	$module_calls
1002
1003	echo ld_library_path=\$LD_LIBRARY_PATH
1004
1005	%%END%%
1006
1007	else
1008	cat > $job_to_send << %%END%%
1009	#PBS -S /bin/bash
1010	#PBS -N $job_name
1011	#PBS -l walltime=$timestring
1012	#PBS -l ncpus=1
1013	#PBS -o $remote_dayfile
1014	#PBS -j oe
1015	$queue_directive
1016	$email_directive
1017
1018	eval \`/sw/swdist/bin/modulesinit\`
1019	#. /usr/share/modules/init/bash
1020	$init_cmds
1021	$module_calls
1022
1023	%%END%%
1024
1025	fi
1026
1027	elif [[ $remote_host = lcxe6 ]]
1028	then
1029
1030	if [[ $numprocs != 0 ]]
1031	then
1032	cat > $job_to_send << %%END%%
1033	#!/bin/ksh
1034	#PBS -S /bin/ksh
1035	#PBS -N $job_name
1036	#PBS -A $project_account
1037	#PBS -j oe
1038	#PBS -l walltime=$timestring
1039	#PBS -l mppwidth=${numprocs}
1040	#PBS -l mppnppn=${processes_per_node}
1041	#PBS -m abe
1042	#PBS -o $remote_dayfile
1043	$email_directive
1044
1045	$init_cmds
1046	$module_calls
1047
1048	%%END%%
1049
1050	else
1051	cat > $job_to_send << %%END%%
1052	#!/bin/ksh
1053	#PBS -S /bin/ksh
1054	#PBS -N $job_name
1055	#PBS -A $project_account
1056	#PBS -j oe
1057	#PBS -l walltime=$timestring
1058	#PBS -l ncpus=1
1059	#PBS -l pmem=${memory}mb
1060	#PBS -m abe
1061	$email_directive
1062	#PBS -o $remote_dayfile
1063
1064	$init_cmds
1065	$module_calls
1066
1067	%%END%%
1068
1069	fi
1070
1071	elif [[ $remote_host = lckyoto ]]
1072	then
1073
1074	cat > $job_to_send << %%END%%
1075	#!/bin/ksh
1076	# @\$-o $remote_dayfile
1077	# @\$-eo -oi
1078	# @\$-lP 16
1079	# @\$-lp 1
1080	# @\$-lm 28gb -llm unlimited -ls unlimited
1081	# @\$-q $queue
1082	# @\$-Pvn abs_pack
1083	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1084	#. /thin/local/etc/setprofile/intel-11.0.sh
1085	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1086	. ~/.myprofile
1087	#. /home2/t/t51254/palm/current_version/myprofile
1088	#. /thin/apps/pgi/mpi.sh
1089	#
1090	env
1091	#
1092	set -x
1093
1094	%%END%%
1095
1096	elif [[ $remote_host = lcxt5m ]]
1097	then
1098
1099	if [[ $numprocs != 0 ]]
1100	then
1101	cat > $job_to_send << %%END%%
1102	#!/bin/ksh
1103	#PBS -S /bin/ksh
1104	#PBS -N $job_name
1105	#PBS -j oe
1106	#PBS -l walltime=$timestring
1107	#PBS -l mppwidth=${numprocs}
1108	#PBS -l mppnppn=${processes_per_node}
1109	#PBS -m abe
1110	#PBS -o $remote_dayfile
1111
1112	$init_cmds
1113	$module_calls
1114
1115	%%END%%
1116
1117	else
1118	cat > $job_to_send << %%END%%
1119	#!/bin/ksh
1120	#PBS -S /bin/ksh
1121	#PBS -N $job_name
1122	#PBS -j oe
1123	#PBS -l walltime=$timestring
1124	#PBS -l ncpus=1
1125	#PBS -l pmem=${memory}mb
1126	#PBS -m abe
1127	#PBS -o $remote_dayfile
1128
1129	$init_cmds
1130	$module_calls
1131
1132	%%END%%
1133
1134	fi
1135
1136	elif [[ $remote_host = nech ]]
1137	then
1138
1139	if (( nodes > 1 ))
1140	then
1141	cat > $job_to_send << %%END%%
1142	#!/bin/ksh
1143	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1144	#PBS -l ${qsubmem}=${Memory}gb
1145	#PBS -b $nodes
1146	#PBS -o $remote_dayfile
1147	#PBS -N palm
1148	#PBS -j o
1149	#PBS -T mpisx
1150
1151	%%END%%
1152
1153	elif [[ $numprocs != 0 ]]
1154	then
1155	cat > $job_to_send << %%END%%
1156	#!/bin/ksh
1157	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1158	#PBS -l ${qsubmem}=${Memory}gb
1159	#PBS -o $remote_dayfile
1160	#PBS -N palm
1161	#PBS -j o
1162
1163	%%END%%
1164
1165	else
1166	cat > $job_to_send << %%END%%
1167	#!/bin/ksh
1168	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1169	#PBS -o $remote_dayfile
1170	#PBS -j o
1171
1172	%%END%%
1173
1174	fi
1175
1176	elif [[ $remote_host = necriam ]]
1177	then
1178
1179	if (( nodes > 1 ))
1180	then
1181	cat > $job_to_send << %%END%%
1182	#!/bin/ksh
1183	#PBS -b $nodes
1184	#PBS -o $remote_dayfile
1185	#PBS -N $job_name
1186	#PBS -j o
1187	#PBS -v MPIPROGINV=YES
1188
1189	%%END%%
1190
1191	elif [[ $numprocs != 0 ]]
1192	then
1193	cat > $job_to_send << %%END%%
1194	#!/bin/ksh
1195	#PBS -q ${queue}
1196	#PBS -o $remote_dayfile
1197	#PBS -N $job_name
1198	#PBS -j o
1199	#PBS -v MPIPROGINV=YES
1200
1201	%%END%%
1202
1203	else
1204	cat > $job_to_send << %%END%%
1205	#!/bin/ksh
1206	#PBS -o $remote_dayfile
1207	#PBS -j o
1208
1209	%%END%%
1210
1211	fi
1212
1213	elif [[ $remote_host = lctit ]]
1214	then
1215	cat > $job_to_send << %%END%%
1216	#!/bin/ksh
1217	$init_cmds
1218	$module_calls
1219
1220	%%END%%
1221
1222	# OPTIONEN FUER SUBMIT-KOMMANDO ZUSAMMENSTELLEN
1223	if [[ $tasks_per_node != $processes_per_node ]]
1224	then
1225	submcom="$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 -o $remote_dayfile -j oe -q $queue "
1226	else
1227	submcom="$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 "
1228	fi
1229
1230	else
1231
1232	cat > $job_to_send << %%END%%
1233	# @\$-q ${queue}
1234	# @\$-l${qsubtime} $timestring
1235	# @\$-l${qsubmem} ${memory}mb
1236	# @\$-o $remote_dayfile
1237	# @\$-eo
1238
1239	%%END%%
1240
1241	fi
1242
1243
1244	# BEI RECHNUNG AUF REMOTE-MASCHINEN RUECKTRANSFER DES DAYFILES PER TRAP
1245	# BEI EXIT VERANLASSEN
1246	# VEKTORRECHNER MUSS EIGENEN JOB STARTEN, DA DORT NOHUP NICHT FUNKTIONIERT
1247	# AUF IBM IN SEOUL IST RUECKTRANSFER ZUR ZEIT GENERELL NICHT MOEGLICH
1248	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1249	then
1250	echo "set +vx" >> $job_to_send
1251	echo "trap '" >> $job_to_send
1252	echo "set +vx" >> $job_to_send
1253	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $(echo $remote_host \| cut -c1-5) = lcsgi \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lckiaps ]]
1254	then
1255	if [[ $remote_host = ibmh ]]
1256	then
1257	return_queue=c1
1258	elif [[ $remote_host = ibmkisti ]]
1259	then
1260	return_queue=class.1-2
1261	elif [[ $remote_host = ibmku ]]
1262	then
1263	return_queue=sdbg2
1264	elif [[ $remote_host = ibms ]]
1265	then
1266	return_queue=p_normal
1267	elif [[ $remote_host = ibmy ]]
1268	then
1269	return_queue=serial
1270	elif [[ $remote_host = lcsgih \|\| $remote_host = lcsgib ]]
1271	then
1272	return_queue=serialq
1273	elif [[ $remote_host = necriam ]]
1274	then
1275	return_queue=SP
1276	elif [[ $remote_host = lckiaps ]]
1277	then
1278	return_queue=express
1279	else
1280	return_queue=unknown
1281	fi
1282
1283	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1284	then
1285
1286	if [[ $remote_host = ibmku ]]
1287	then
1288	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1289	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1290	else
1291	echo "echo \"#!/bin/ksh\" >> scpjob.$kennung" >> $job_to_send
1292	fi
1293	echo "echo \"# @ job_type = serial\" >> scpjob.$kennung" >> $job_to_send
1294	echo "echo \"# @ job_name = transfer\" >> scpjob.$kennung" >> $job_to_send
1295	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$kennung" >> $job_to_send
1296	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$kennung " >> $job_to_send
1297	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1298	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$kennung" >> $job_to_send
1299	if [[ $host != "ibmh" ]]
1300	then
1301	echo "echo \"# @ class = $return_queue\" >> scpjob.$kennung" >> $job_to_send
1302	fi
1303	echo "echo \"# @ image_size = 10\" >> scpjob.$kennung" >> $job_to_send
1304	echo "echo \"# @ notification = never\" >> scpjob.$kennung" >> $job_to_send
1305
1306	echo "echo \"# @ queue\" >> scpjob.$kennung" >> $job_to_send
1307	echo "echo \" \" >> scpjob.$kennung" >> $job_to_send
1308
1309	echo "echo \"set -x\" >> scpjob.$kennung" >> $job_to_send
1310	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
1311	if [[ $remote_host = ibmku ]]
1312	then
1313	echo "echo \"rm scpjob.$kennung\" >> scpjob.$kennung" >> $job_to_send
1314	fi
1315	echo "echo \"exit\" >> scpjob.$kennung" >> $job_to_send
1316
1317	elif [[ $remote_host = nech ]]
1318	then
1319	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1320	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1321	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1322	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1323	echo "#PBS -j o" >> $job_to_send
1324	echo " " >> $job_to_send
1325	echo "set -x" >> $job_to_send
1326	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1327	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1328	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1329	echo "%%END%%" >> $job_to_send
1330
1331	elif [[ $remote_host = necriam ]]
1332	then
1333	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1334	echo "#PBS -q $return_queue" >> $job_to_send
1335	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1336	echo "#PBS -j o" >> $job_to_send
1337	echo " " >> $job_to_send
1338	echo "set -x" >> $job_to_send
1339	echo "batch_scp -d -w 10 -u $local_user $local_addres $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1340	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1341	echo "%%END%%" >> $job_to_send
1342
1343	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1344	then
1345	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1346	echo "#!/bin/bash" >> $job_to_send
1347	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1348	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1349	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1350	echo "#PBS -l feature=data" >> $job_to_send
1351	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1352	echo "#PBS -j oe" >> $job_to_send
1353	echo " " >> $job_to_send
1354	echo ". /usr/share/modules/init/bash" >> $job_to_send
1355	echo "set -x" >> $job_to_send
1356	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1357	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1358	echo "%%END%%" >> $job_to_send
1359
1360	else
1361
1362	echo "cat > scpjob.$kennung << %%END%%" >> $job_to_send
1363	echo "# @\\\$-q $return_queue" >> $job_to_send
1364	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1365	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1366	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1367	then
1368	echo "# @\$-l mpp_p=0" >> $job_to_send
1369	fi
1370	echo '# @\$-lF 10mb' >> $job_to_send
1371	# echo '# @\$-o /dev/null' >> $job_to_send
1372	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1373	echo '# @\\\$-eo' >> $job_to_send
1374	echo " " >> $job_to_send
1375	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1376	then
1377	echo "set +vx" >> $job_to_send
1378	echo ". .profile" >> $job_to_send
1379	fi
1380	echo "set -x" >> $job_to_send
1381	echo "batch_scp -d -w 10 -u $local_user $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1382	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1383	echo "%%END%%" >> $job_to_send
1384	fi
1385
1386	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1387	then
1388	echo "llsubmit scpjob.$kennung" >> $job_to_send
1389	elif [[ $(echo $remote_host \| cut -c1-5) = lcsgi ]]
1390	then
1391	echo "rm -rf \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1392	echo "chmod u+x scpjob.$kennung" >> $job_to_send
1393	echo "msub scpjob.$kennung" >> $job_to_send
1394	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1395	then
1396	echo "qsub -J n scpjob.$kennung" >> $job_to_send
1397	elif [[ $remote_host = t3es ]]
1398	then
1399	echo "qsub -J n -s /bin/ksh scpjob.$kennung" >> $job_to_send
1400	elif [[ $remote_host = lckiaps ]]
1401	then
1402	echo "mv scpjob.$kennung $job_catalog" >> $job_to_send
1403	echo "ssh ${remote_username}@${remote_addres} \"$submcom ${job_catalog}/scpjob.$kennung\" " >> $job_to_send
1404	echo "rm ${job_catalog}/scpjob.$kennung" >> $job_to_send
1405	else
1406	echo "$submcom scpjob.$kennung" >> $job_to_send
1407	fi
1408	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1409	then
1410	echo "rm scpjob.$kennung" >> $job_to_send
1411	fi
1412	if [[ $remote_host = nech ]]
1413	then
1414	echo "cd -" >> $job_to_send
1415	fi
1416	else
1417	# echo "ftpcopy -d $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1418	echo "nohup ftpcopy -d -w 15 $local_addres ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1419	fi
1420	echo "set -x" >> $job_to_send
1421	echo " ' exit" >> $job_to_send
1422	echo "set -x" >> $job_to_send
1423	fi
1424
1425
1426
1427	# EIGENTLICHE JOB-DATEI AN QSUB-KOMMANDOS ANHAENGEN
1428	cat $file_to_send >> $job_to_send
1429
1430	if [[ $remote_host = ibm ]]
1431	then
1432	echo " " >> $job_to_send
1433	echo "exit" >> $job_to_send
1434	fi
1435
1436	# remove job file
1437	if [[ $remote_host = lctit \|\| $remote_host = ibmku ]]
1438	then
1439	echo " " >> $job_to_send
1440	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1441	fi
1442
1443
1444
1445	# USER-NAME AUF ZIELRECHNER AUS .NETRC-DATEI ERMITTELN
1446	if [[ -z $remote_user ]]
1447	then
1448	if [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es \|\| $remote_host = vpp ]]
1449	then
1450	grep $remote_addres ~/.netrc \| read dum dum dum remote_user dum dum
1451	fi
1452	fi
1453
1454
1455
1456	# JOB AUF ZIELRECHNER TRANSFERIEREN BZW. INS JOBVERZEICHNIS KOPIEREN
1457	if [[ $no_submit = false ]]
1458	then
1459	if [[ $remote_host != $local_host ]]
1460	then
1461	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1462	if [[ $remote_host = ibms \|\| $remote_host = ibmy ]] # ssh on ibms cannot handle "~/"
1463	then
1464	job_catalog_save=$job_catalog
1465	job_catalog=job_queue
1466	elif [[ $remote_host = nech ]]
1467	then
1468	job_catalog_save=$job_catalog
1469	job_catalog=/hpf/b/${remote_user}/job_queue
1470	fi
1471	if [[ $local_host = decalpha ]]
1472	then
1473	# VERWENDUNG VON SCP AUF DECALPHA FRAGT FEHLERHAFTERWEISE
1474	# PASSWORT AB
1475	/bin/scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1476	elif [[ $remote_host = nech ]]
1477	then
1478	# DATEIEN KOENNEN NUR UEBER DEN ARCHIVE-SERVER DES DKRZ
1479	# TRANSFERIERT WERDEN
1480	scp $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1481	else
1482	scp $job_to_send ${remote_user}@${remote_addres}:${job_catalog}/$job_on_remhost
1483	fi
1484	if [[ $? = 1 ]]
1485	then
1486	locat=scp; exit
1487	fi
1488	if [[ $remote_host = ibms ]]
1489	then
1490	job_catalog=$job_catalog_save
1491	fi
1492	[[ $verify = true ]] && printf "\n >>> finished\n"
1493	else
1494	eval job_catalog=$job_catalog
1495	cp $job_to_send ${job_catalog}/$job_on_remhost
1496	fi
1497
1498
1499
1500	# NQS- BZW. LOADLEVELER-JOB STARTEN
1501	if [[ $remote_host != $local_host ]]
1502	then
1503	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1504
1505	if [[ $(echo $remote_host \| cut -c1-5) = lcsgi && $prio = true ]]
1506	then
1507	printf "\n >>> submit with HLRN qos-feature hiprio...\n"
1508	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom -l qos=hiprio $job_on_remhost; rm $job_on_remhost"
1509	elif [[ $remote_host = ibmku ]]
1510	then
1511	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1512	else
1513	ssh $remote_addres -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1514	fi
1515
1516	[[ $verify = true ]] && printf " >>> o.k.\n"
1517	else
1518	cd $job_catalog
1519	if [[ $(echo $local_host \| cut -c1-5) = lcsgi \|\| $(echo $local_host \| cut -c1-3) = ibm ]]
1520	then
1521	eval $submcom $job_on_remhost
1522	elif [[ $local_host = lcfimm \|\| $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb ]]
1523	then
1524	chmod u+x $job_on_remhost
1525	echo "$submcom $job_on_remhost"
1526	eval $submcom $job_on_remhost
1527	elif [[ $local_host = nech ]]
1528	then
1529	if [[ $queue = default ]]
1530	then
1531	eval $submcom $job_on_remhost
1532	else
1533	eval $submcom -q $queue $job_on_remhost
1534	fi
1535	else
1536	qsub $job_on_remhost
1537	fi
1538
1539	# Jobfile must not be deleted on lctit/ibmku!! This will be done
1540	# only at the end of the job.
1541	if [[ $local_host != lctit && $local_host != ibmku ]]
1542	then
1543	rm $job_on_remhost
1544	fi
1545	cd - > /dev/null
1546	fi
1547	fi
1548
1549
1550
1551	# ABSCHLUSSARBEITEN
1552
1553	if [[ $no_submit = false ]]
1554	then
1555	rm $job_to_send
1556	fi
1557	[[ $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 |