Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 2114

Last change on this file since 2114 was 1945, checked in by raasch, 9 years ago
last commit documented
Property svn:keywords set to `Id Rev`
File size: 57.9 KB

Line
1	#!/bin/bash
2
3	# subjob - script for automatic generation and submission of batch-job files
4	# for various batch queuing systems
5
6	#--------------------------------------------------------------------------------#
7	# This file is part of PALM.
8	#
9	# PALM is free software: you can redistribute it and/or modify it under the terms
10	# of the GNU General Public License as published by the Free Software Foundation,
11	# either version 3 of the License, or (at your option) any later version.
12	#
13	# PALM is distributed in the hope that it will be useful, but WITHOUT ANY
14	# WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
15	# A PARTICULAR PURPOSE. See the GNU General Public License for more details.
16	#
17	# You should have received a copy of the GNU General Public License along with
18	# PALM. If not, see <http://www.gnu.org/licenses/>.
19	#
20	# Copyright 1997-2014 Leibniz Universitaet Hannover
21	#--------------------------------------------------------------------------------#
22	#
23	# Current revisions:
24	# ------------------
25	#
26	#
27	# Former revisions:
28	# -----------------
29	# $Id: subjob 1945 2016-06-15 07:17:24Z kanani $
30	#
31	# 1944 2016-06-15 06:29:00Z raasch
32	# adjustments for using HLRN ssh-keys
33	#
34	# 1940 2016-06-14 05:15:20Z raasch
35	# adjustments for lckiaps
36	#
37	# 1866 2016-04-15 06:50:59Z raasch
38	# adjusted for lcocean
39	#
40	# 1841 2016-04-07 19:14:06Z raasch
41	# script now running under bash
42	#
43	# 1701 2015-11-02 07:43:04Z maronga
44	# Bugfix: added missing init_cmds for lccrayh/lccrayb
45	#
46	# 1621 2015-07-17 11:39:33Z heinze
47	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
48	#
49	# 1575 2015-03-27 09:56:27Z raasch
50	# mpp2-queues added to lccrayh
51	#
52	# 1547 2015-01-29 15:09:12Z witha
53	# adjustments for ForWind computing cluster (lcflow)
54	#
55	# 1545 2015-01-29 06:52:23Z heinze
56	# local host name for blizzard further specified
57	#
58	# 1480 2014-10-17 14:41:49Z raasch
59	# adjustments for 2nd stage of HLRNIII
60	#
61	# 1468 2014-09-24 14:06:57Z maronga
62	# Typo removed (addres->address)
63	# Adjustments for lcxe6
64	#
65	# 1452 2014-08-22 09:41:06Z heinze
66	# local hosts for blizzard added
67	#
68	# 1450 2014-08-21 07:31:51Z heinze
69	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
70	#
71	# 1442 2014-07-28 07:09:10Z raasch
72	# HLRN-III (lccrayb/lccrayh) queues adjusted
73	#
74	# 1378 2014-04-28 06:04:58Z raasch
75	# -et option added for lctit
76	#
77	# 1350 2014-04-04 13:01:30Z maronga
78	# location of qsub updated for lcxe6
79	#
80	# 1289 2014-03-04 07:12:34Z raasch
81	# German comments translated to English
82	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
83	#
84	# 1279 2014-01-28 12:10:14Z raasch
85	# node calculation modified due to changes in mrun (tasks_per_node must not be
86	# an integral divisor of numprocs any more)
87	#
88	# 1274 2014-01-09 13:14:54Z heinze
89	# adjustments for lccrayh
90	#
91	# 1266 2013-12-11 12:07:34Z heinze
92	# further adjustments for lccrayb (use msub instead of qsub)
93	#
94	# 1264 2013-12-09 12:46:09Z fricke
95	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
96	#
97	# 1262 2013-12-09 10:57:20Z fricke
98	# further adjustments for lccrayb
99	#
100	# 1260 2013-12-04 12:48:04Z raasch
101	# jaboticaba admitted
102	#
103	# 1255 2013-11-07 14:43:35Z raasch
104	# further adjustments for lccrayb
105	#
106	# 1224 2013-09-16 07:27:23Z raasch
107	# first adjustments for lccrayb
108	#
109	# 1202 2013-07-10 16:22:07Z witha
110	# adjustments for Forwind cluster (lcflow)
111	#
112	# 1199 2013-07-05 14:52:22Z raasch
113	# adjustments for CSC Helsinki (lccrayf)
114	#
115	# use of cluster/express queue enabled (ibmh)
116	# vinessa added (imuk)
117	#
118	# 1103 2013-02-20 02:15:53Z raasch
119	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
120	# instead typeset -L/R),
121	# further adjustments for lckyuh
122	#
123	# 2013-02-10 01:47:43Z raasch
124	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
125	# and for Forwind cluster (lcflow)
126	#
127	# 1094 2013-02-03 01:52:12Z raasch
128	# new option -P for explicit setting of ssh/scp port,
129	# decalpha parts (yonsei) removed
130	#
131	# 2013-02-02 07:06:13Z raasch
132	# adjustments for Kyushu-University computing center (lckyut - tatara)
133	# old changelog messages removed
134	#
135	# 1046 2012-11-09 14:38:45Z maronga
136	# code put under GPL (PALM 3.9)
137	#
138	# 08/07/94 - Siggi - first version finished
139	# 29/06/94 - Siggi - script development started
140	#--------------------------------------------------------------------------------#
141	# subjob - script for automatic generation and submission of batch-job files
142	# for various batch queuing systems
143	#--------------------------------------------------------------------------------#
144
145
146	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
147	delete_dayfile=false
148	email_notification=none
149	group_number=none
150	locat=normal
151	no_default_queue=none
152	no_submit=false
153	job_catalog="~/job_queue"
154	job_name=none
155	local_user=$LOGNAME
156	node_usage=shared
157	numprocs=0
158	punkte="..........................................................."
159	submcom=qsub
160	queue=default
161	remote_host=none
162	remote_user=""
163	verify=true
164
165	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
166	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
167
168
169
170	# ERROR HANDLING
171	# IN CASE OF EXIT:
172	trap 'if [[ $locat != normal ]]
173	then
174	case $locat in
175	(option) printf "\n --> available optios can be displayed"
176	printf " by typing:"
177	printf "\n \"subjob ?\" \n";;
178	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
179	(*) printf "\n +++ unknown error"
180	printf "\n please inform S. Raasch!\n"
181	esac
182	[[ -f $job_to_send ]] && rm $job_to_send
183	printf "\n\n+++ SUBJOB killed \n\n"
184	fi' exit
185
186
187	# IN CASE OF TERMINAL-BREAK:
188	trap '[[ -f $job_to_send ]] && rm $job_to_send
189	printf "\n\n+++ SUBJOB killed \n\n"
190	exit
191	' 2
192
193
194	# DETERMINE NAME OF LOCAL HOST
195	local_host=$(hostname)
196
197	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
198	# IS ADMITTED AT ALL)
199	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
200	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
201	case $local_host in
202	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
203	(atmos) local_address=172.20.25.35; local_host=lcide;;
204	(austru) local_address=130.75.105.128; local_host=lcmuk;;
205	(autan) local_address=130.75.105.57; local_host=lcmuk;;
206	(bora) local_address=130.75.105.103; local_host=lcmuk;;
207	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
208	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
209	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
210	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
211	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
212	(breva) local_address=130.75.105.98; local_host=lcmuk;;
213	(buran) local_address=130.75.105.58; local_host=lcmuk;;
214	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
215	(climate*) local_address=165.132.26.68; local_host=lcyon;;
216	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
217	(cs*) local_address=136.172.44.131; local_host=nech;;
218	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
219	(flow01) local_address=10.141.255.71; local_host=lcflow;;
220	(flow02) local_address=10.141.255.72; local_host=lcflow;;
221	(node*) local_address=165.132.26.61 local_host=lck;;
222	# (node*) local_address=210.219.61.8 local_host=lckordi;;
223	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
224	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
225	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
226	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
227	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
228	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
229	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
230	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
231	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
232	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
233	(sno) local_address=130.75.105.113; local_host=lcmuk;;
234	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
235	(login*) local_address=118.128.66.201; local_host=lckiaps;;
236	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
237	(meller) local_address=134.106.74.155; local_host=lcfor;;
238	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
239	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
240	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
241	(nobel*) local_address=150.183.5.101; local_host=ibms;;
242	(ocean) local_address="ocean"; local_host=lcocean;;
243	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
244	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
245	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
246	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
247	(pingui) local_address=134.106.74.118; local_host=lcfor;;
248	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
249	(rte*) local_address=133.5.185.60; local_host=lcrte;;
250	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
251	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
252	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
253	(solano) local_address=130.75.105.110; local_host=lcmuk;;
254	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
255	(tc*) local_address="ocean"; local_host=lcocean;;
256	(t2a*) local_address=10.1.6.165; local_host=lctit;;
257	(urban*) local_address=147.46.30.151 local_host=lcsb;;
258	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
259	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
260	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
261	(*) printf "\n +++ \"$local_host\" unknown";
262	printf "\n please contact the PALM group at IMUK";
263	locat=parameter; exit;;
264	esac
265
266
267
268	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
269	remote_host=$local_host
270
271
272
273
274	# READ THE SHELLSCRIPT-OPTIONS
275	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
276	do
277	case $option in
278	(c) job_catalog=$OPTARG;;
279	(d) delete_dayfile=true;;
280	(D) no_submit=true;;
281	(e) email_notification=$OPTARG;;
282	(g) group_number=$OPTARG;;
283	(h) remote_host=$OPTARG;;
284	(m) memory=$OPTARG;;
285	(n) job_name=$OPTARG;;
286	(N) node_usage=$OPTARG;;
287	(O) threads_per_task=$OPTARG;;
288	(P) scp_port=$OPTARG;;
289	(q) no_default_queue=$OPTARG;;
290	(t) cputime=$OPTARG;;
291	(T) tasks_per_node=$OPTARG;;
292	(u) remote_user=$OPTARG;;
293	(v) verify=false;;
294	(X) numprocs=$OPTARG;;
295	(\?) printf "\n +++ Option $OPTARG unknown \n";
296	locat=option; exit;;
297	esac
298	done
299
300
301	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
302	(( to_shift = $OPTIND - 1 ))
303	shift $to_shift; file_to_send=$1
304
305
306	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
307	if [ "$1" = "?" ]
308	then
309	(printf "\n *** subjob can be called as follows:\n"
310	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
311	printf "\n Description of available options:\n"
312	printf "\n Option Description Default-Value"
313	printf "\n -c job-input- and output-catalog ~/job_queue"
314	printf "\n -d no job-protocol will be created ---"
315	printf "\n -D only the job-file will be created ---"
316	printf "\n -h execution host, available hosts: $remote_host"
317	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
318	printf "\n lckiaps, lctit, nech"
319	printf "\n -m memory demand per process in MByte ---"
320	printf "\n -n jobname <jobdatei>"
321	printf "\n -O threads per task (for OpenMP usage) 1"
322	printf "\n -P ssh/scp port default port"
323	printf "\n -q job-queue to be used default"
324	printf "\n -t allowed cpu-time in seconds ---"
325	printf "\n -T tasks per node (on parallel hosts) ---"
326	printf "\n -u username on execution host from .netrc"
327	printf "\n -v no prompt for confirmation ---"
328	printf "\n -X # of processors (on parallel hosts) 1"
329	printf "\n "
330	printf "\n The only possible positional parameter is <jobfile>:"
331	printf "\n The complete NQS-job must be provided here."
332	printf "\n <jobfile>=? creates this outline\n\n") \| more
333	exit
334	fi
335
336
337
338	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
339	if [[ "$file_to_send" = "" ]]
340	then
341	printf "\n +++ job-file missing"
342	locat=parameter; exit
343	else
344	if [[ -f $file_to_send ]]
345	then
346	true
347	else
348	printf "\n +++ job-file: "
349	printf "\n $file_to_send"
350	printf "\n does not exist"
351	locat=parameter; exit
352	fi
353	fi
354
355
356
357	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
358	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
359	if [[ $job_name = none ]]
360	then
361	job_name=$file_to_send
362	fi
363	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
364	then
365	printf "\n +++ job-file name: "
366	printf "\n $job_name"
367	printf "\n must not contain \"/\"-characters"
368	locat=parameter; exit
369	fi
370
371
372
373
374	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
375	# OR IF NO HOST HAS BEEN GIVEN
376	if [[ $remote_host = none ]]
377	then
378	printf "\n +++ host missing"
379	locat=option; exit
380	else
381	case $remote_host in
382	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
383	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
384	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
385	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
386	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
387	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
388	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
389	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
390	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
391	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
392	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
393	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
394	(lckiaps) remote_address=118.128.66.201; submcom=/opt/pbs/default/bin/qsub;;
395	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
396	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
397	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
398	(lcocean) remote_address="ocean"; submcom=qsub;;
399	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
400	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
401	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
402	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
403	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
404	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
405	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
406	locat=parameter; exit;;
407	esac
408	fi
409
410
411	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
412	if [[ $no_default_queue != none ]]
413	then
414	error=false
415	ndq=$no_default_queue
416	case $remote_host in
417	(ibm) case $ndq in
418	(p690_express\|p690_standard\|p690_long) error=false;;
419	(*) error=true;;
420	esac;;
421	(ibmh) case $ndq in
422	(cluster\|express) error=false;;
423	(*) error=true;;
424	esac;;
425	(ibmkisti) case $ndq in
426	(class.32plus\|class.1-2\|class.2-32) error=false;;
427	(*) error=true;;
428	esac;;
429	(ibmku) case $ndq in
430	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
431	(*) error=true;;
432	esac;;
433	(ibms) case $ndq in
434	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
435	(*) error=true;;
436	esac;;
437	(lcbullhh) case $ndq in
438	(compute\|shared) error=false;;
439	(*) error=true;;
440	esac;;
441	(lccrayb) case $ndq in
442	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
443	(*) error=true;;
444	esac;;
445	(lccrayh) case $ndq in
446	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
447	(*) error=true;;
448	esac;;
449	(lccrayf) case $ndq in
450	(usup\|test*\|small\|large) error=false;;
451	(*) error=true;;
452	esac;;
453	(lcflow) case $ndq in
454	(cfd_lom_long.q\|cfd_him_long.q\|cfd_lom_serl.q\|cfd_lom_shrt.q\|cfd_him_shrt.q\|cfd_ivy_shrt.q) error=false;;
455	(*) error=true;;
456	esac;;
457	(lckiaps) case $ndq in
458	(express\|normal\|normal20\|quickq) error=false;;
459	(*) error=true;;
460	esac;;
461	(lckyoto) case $ndq in
462	(eh\|ph) error=false;;
463	(*) error=true;;
464	esac;;
465	(lckyuh) case $ndq in
466	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
467	(*) error=true;;
468	esac;;
469	(lckyut) case $ndq in
470	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
471	(*) error=true;;
472	esac;;
473	(lctit) case $ndq in
474	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
475	(*) error=true;;
476	esac;;
477	(t3eb) case $ndq in
478	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
479	(*) error=true;;
480	esac;;
481	(t3eh) case $ndq in
482	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
483	(*) error=true;;
484	esac;;
485	(t3ej2\|t3ej5) case $ndq in
486	(low\|normal\|high) error=false;;
487	(*) error=true;;
488	esac;;
489	(t3es) case $ndq in
490	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
491	(*) error=true;;
492	esac;;
493	esac
494	if [[ $error = true ]]
495	then
496	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
497	locat=parameter; exit
498	else
499	queue=$no_default_queue
500	fi
501	fi
502
503
504
505	# CHECK THE CPU-TIME
506	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
507	done=false
508	while [[ $done = false ]]
509	do
510	if (( $cputime <= 0 ))
511	then
512	printf "\n +++ wrong cpu-time or cpu-time missing"
513	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
514	printf "\n >>> "
515	read cputime 1>/dev/null 2>&1
516	else
517	done=true
518	fi
519	done
520	if [[ $remote_host = nech ]]
521	then
522	if (( tasks_per_node != 0 ))
523	then
524	(( cputime = cputime * tasks_per_node ))
525	elif [[ $numprocs != 0 ]]
526	then
527	(( cputime = cputime * numprocs ))
528	fi
529	fi
530	(( stunden = cputime / 3600 ))
531	(( resttime = cputime - stunden * 3600 ))
532	(( minuten = resttime / 60 ))
533	(( sekunden = resttime - minuten * 60 ))
534	timestring=${stunden}:${minuten}:${sekunden}
535
536
537
538	# CHECK THE MEMORY DEMAND
539	done=false
540	while [[ $done = false ]]
541	do
542	if (( memory <= 0 ))
543	then
544	printf "\n +++ wrong memory demand or memory demand missing"
545	printf "\n >>> Please type memory in MByte per process as INTEGER:"
546	printf "\n >>> "
547	read memory 1>/dev/null 2>&1
548	else
549	done=true
550	fi
551	done
552
553	if [[ $remote_host = nech ]]
554	then
555	if (( tasks_per_node != 0 ))
556	then
557	(( Memory = memory * tasks_per_node / 1000 ))
558	elif [[ $numprocs != 0 ]]
559	then
560	(( Memory = memory * numprocs / 1000 ))
561	else
562	(( Memory = memory / 1000 ))
563	fi
564	elif [[ $remote_host = lctit ]]
565	then
566	(( Memory = memory * tasks_per_node / 1000 ))
567	fi
568
569
570	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
571	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
572	then
573	(( memory = memory * threads_per_task ))
574	fi
575
576
577	# CALCULATE NUMBER OF REQUIRED NODES
578	if (( tasks_per_node != 0 ))
579	then
580	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
581	fi
582
583
584	# CALCULATE NUMBER OF PROCESSES PER NODE
585	(( processes_per_node = tasks_per_node * threads_per_task ))
586
587
588	# CALCULATE NUMBER OF MPI TASKS
589	(( mpi_tasks = numprocs / threads_per_task ))
590
591
592	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
593	if [[ "$scp_port" != "" ]]
594	then
595	PORTOPT="-P $scp_port"
596	SSH_PORTOPT="-p $scp_port"
597	fi
598
599
600	# HEADER-OUTPUT
601	if [[ $verify = true ]]
602	then
603	printf "\n\n"
604	printf "#--------------------------------------------------------------# \n"
605	spalte1=SUBJOB;spalte2=$(date)
606	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
607	printf "\| \| \n"
608	printf "\| values of parameters/options: \| \n"
609	spalte1=$(echo local_host$punkte \| cut -c-20)
610	spalte2=$punkte$local_host
611	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
612	spalte1=$(echo remote_host$punkte \| cut -c-20)
613	spalte2=$punkte$remote_host
614	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
615	spalte1=$(echo queue$punkte \| cut -c-20)
616	spalte2=$punkte$queue
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	spalte1=$(echo memory$punkte \| cut -c-20)
619	spalte2="$punkte$memory mb"
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	spalte1=$(echo cputime$punkte \| cut -c-20)
622	spalte2="$punkte$cputime sec"
623	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
624	spalte1=$(echo job_name$punkte \| cut -c-20)
625	spalte2="$punkte$job_name"
626	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
627	printf "#--------------------------------------------------------------# \n\n"
628
629
630	# QUERY CHECK
631	antwort="dummy"
632	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
633	do
634	read antwort?" >>> continue (y/n) ? "
635	done
636	if [[ $antwort = n \|\| $antwort = N ]]
637	then
638	locat=verify; exit
639	fi
640	printf "\n"
641	fi
642
643	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
644	identifier=$RANDOM
645	job_on_remhost=${job_name}_${identifier}_$local_host
646	job_to_send=job_to_send_$identifier
647	if [[ $delete_dayfile = false ]]
648	then
649	remote_dayfile=${local_host}_${job_name}_result_$identifier
650	local_dayfile=${remote_host}_${job_name}
651	else
652	remote_dayfile=/dev/null
653	fi
654
655
656	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
657	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
658	then
659
660	# GENERAL LOADLEVELER SETTINGS
661	execute_in_shell="#!/bin/ksh"
662	use_shell="# @ shell = /bin/ksh"
663	consumable_memory="ConsumableMemory($memory mb)"
664	class="# @ class = $queue"
665	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
666	network_to_use="# @ network.mpi = sn_all,shared,us"
667	data_limit="# @ data_limit = 1.76gb"
668	image_size="# @ image_size = 50"
669	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
670
671	if [[ $email_notification = none ]]
672	then
673	notify_user=""
674	else
675	notify_user="# @ notify_user = $email_notification"
676	if [[ $delete_dayfile = true ]]
677	then
678	notification='# @ notification = never'
679	fi
680	fi
681
682	if [[ $remote_host = ibmh ]]
683	then
684	data_limit=""
685	network_to_use=""
686	class="# @ class = $queue"
687	environment=""
688	rset="# @ rset = RSET_MCM_AFFINITY"
689	task_affinity="# @ task_affinity = core(1)"
690	elif [[ $remote_host = ibmkisti ]]
691	then
692	network_to_use="# @ network.MPI = sn_all,shared,US"
693	wall_clock_limit="# @ wall_clock_limit = $timestring"
694	if [[ $threads_per_task = 1 ]]
695	then
696	rset="# @ rset = RSET_MCM_AFFINITY"
697	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
698	fi
699	environment=""
700	use_shell=""
701	data_limit=""
702	image_size=""
703	elif [[ $remote_host = ibmku ]]
704	then
705	execute_in_shell="#!/usr/bin/ksh"
706	use_shell="# @ shell = /usr/bin/ksh"
707	consumable_memory=""
708	environment=""
709	network_to_use="# @ network.mpi = sn_all,shared,us"
710	data_limit=""
711	image_size=""
712	elif [[ $remote_host = ibms ]]
713	then
714	network_to_use="# @ network.mpi = csss,shared,us"
715	fi
716
717	cat > $job_to_send << %%END%%
718	$execute_in_shell
719	$use_shell
720
721	# @ job_type = parallel
722	# @ job_name = $job_name
723	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
724	# @ output = $remote_dayfile
725	# @ error = $remote_dayfile
726	$wall_clock_limit
727	$image_size
728	$class
729	$environment
730	$network_to_use
731	$data_limit
732	$rset
733	$mcm_affinity_options
734	$task_affinity
735	$notification
736	$notify_user
737
738	%%END%%
739
740	if (( nodes > 0 ))
741	then
742
743	if [[ $remote_host != ibmkisti ]]
744	then
745
746	cat >> $job_to_send << %%END%%
747	# @ node = $nodes
748	# @ tasks_per_node = $processes_per_node
749	# @ node_usage = $node_usage
750	# @ queue
751
752	%%END%%
753
754	else
755
756	cat >> $job_to_send << %%END%%
757	# @ total_tasks = $mpi_tasks
758	# @ blocking = unlimited
759	# @ queue
760
761	%%END%%
762
763	fi
764
765	else
766
767	cat >> $job_to_send << %%END%%
768	# @ blocking = unlimited
769	# @ total_tasks = $numprocs
770	# @ node_usage = $node_usage
771	# @ queue
772
773	%%END%%
774
775	fi
776
777	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
778	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
779	then
780	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
781	fi
782
783	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
784	then
785
786	cat > $job_to_send << %%END%%
787	#!/bin/ksh
788
789	# @ job_type = serial
790	# @ node_usage = $node_usage
791	# @ job_name = palm
792	# @ wall_clock_limit = ${timestring},$timestring
793	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
794	# @ output = $remote_dayfile
795	# @ error = $remote_dayfile
796	$class
797	$notification
798
799	# @ queue
800
801	%%END%%
802
803	elif [[ $remote_host = lcbullhh ]]
804	then
805	if [[ $numprocs != 0 ]]
806	then
807	cat > $job_to_send << %%END%%
808	#!/bin/bash -l
809	#SBATCH -J $job_name
810	#SBATCH -t $timestring
811	#SBATCH -N $nodes
812	#SBATCH --ntasks-per-node=$processes_per_node
813	#SBATCH -p $queue
814	#SBATCH -o $remote_dayfile
815	#SBATCH -e $remote_dayfile
816	#SBATCH -A $project_account
817
818	$init_cmds
819	$module_calls
820
821	%%END%%
822
823	else
824	cat > $job_to_send << %%END%%
825	#!/bin/bash -l
826	#SBATCH -J $job_name
827	#SBATCH -t $timestring
828	#SBATCH -l ncpus=1
829	#SBATCH -l pmem=${memory}mb
830	#SBATCH -m abe
831	#SBATCH -o $remote_dayfile
832	#SBATCH -e $remote_dayfile
833	#SBATCH -A $project_account
834
835	$init_cmds
836	$module_calls
837
838	%%END%%
839
840	fi
841
842	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
843	then
844
845	if [[ "$feature" != "" ]]
846	then
847	featuredir="#PBS -l feature=$feature"
848	fi
849
850	if [[ $numprocs != 0 ]]
851	then
852	cat > $job_to_send << %%END%%
853	#!/bin/bash
854	#PBS -N $job_name
855	#PBS -l walltime=$timestring
856	#PBS -l nodes=$nodes:ppn=$processes_per_node
857	#PBS -o $remote_dayfile
858	#PBS -j oe
859	#PBS -q $queue
860	$featuredir
861
862	$init_cmds
863	$module_calls
864
865	%%END%%
866
867	else
868
869	continue
870
871	fi
872
873	elif [[ $remote_host = lccrayf ]]
874	then
875
876	if [[ $numprocs != 0 ]]
877	then
878	cat > $job_to_send << %%END%%
879	#!/bin/bash -l
880	#SBATCH -J $job_name
881	#SBATCH -t $timestring
882	#SBATCH -N $nodes
883	#SBATCH --ntasks-per-node=$processes_per_node
884	#SBATCH -p $queue
885	#SBATCH -o $remote_dayfile
886	#SBATCH -e $remote_dayfile
887
888	$init_cmds
889	$module_calls
890
891	%%END%%
892
893	else
894	cat > $job_to_send << %%END%%
895	#!/bin/bash -l
896	#SBATCH -J $job_name
897	#SBATCH -t $timestring
898	#SBATCH -l ncpus=1
899	#SBATCH -l pmem=${memory}mb
900	#SBATCH -m abe
901	#SBATCH -o $remote_dayfile
902	#SBATCH -e $remote_dayfile
903
904	$init_cmds
905	$module_calls
906
907	%%END%%
908
909	fi
910
911	elif [[ $remote_host = lcflow ]]
912	then
913	if [[ $numprocs != 0 ]]
914	then
915	pe_set="#$ -pe impi $numprocs"
916	else
917	pe_set="#$ -pe impi 1"
918	fi
919	if [[ $queue = default ]]
920	then
921	queue_set=""
922	else
923	queue_set="#$ -q $queue"
924	fi
925	[[ "$disc_space" = "" ]] && disc_space=50
926
927	cat > $job_to_send << %%END%%
928	#!/bin/bash
929	#$ -S /bin/bash
930	#$ -N $job_name
931	#$ -cwd
932	#$ -l h_rt=$timestring
933	#$ -l h_vmem=${memory}M
934	#$ -o $remote_dayfile
935	#$ -j y
936	$pe_set
937	#$ -R y
938	#$ -l h_fsize=${disc_space}G
939	$queue_set
940
941	%%END%%
942
943	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
944	then
945
946	if [[ $numprocs != 0 ]]
947	then
948	cat > $job_to_send << %%END%%
949	#!/bin/ksh
950	#PBS -N $job_name
951	#PBS -l walltime=$timestring
952	#PBS -l ncpus=$numprocs
953	#PBS -l pmem=${memory}mb
954	#PBS -o $remote_dayfile
955	#PBS -l nodes=$nodes:ppn=${processes_per_node}
956	#PBS -j oe
957
958	mpd &
959
960	%%END%%
961
962	else
963	cat > $job_to_send << %%END%%
964	#!/bin/ksh
965	#PBS -N $job_name
966	#PBS -l walltime=$timestring
967	#PBS -l ncpus=1
968	#PBS -l pmem=${memory}mb
969	#PBS -o $remote_dayfile
970	#PBS -j oe
971
972	%%END%%
973
974	fi
975
976	elif [[ $remote_host = lckiaps ]]
977	then
978
979	if [[ $numprocs != 0 ]]
980	then
981	cat > $job_to_send << %%END%%
982	#!/bin/bash
983	#PBS -N $job_name
984	#PBS -l walltime=$timestring
985	#PBS -l select=1:ncpus=$numprocs
986	#PBS -l pmem=${memory}mb
987	#PBS -q $queue
988	#PBS -o $remote_dayfile
989	#PBS -j oe
990	#PBS -V
991
992	%%END%%
993
994	else
995	cat > $job_to_send << %%END%%
996	#!/bin/bash
997	#PBS -N $job_name
998	#PBS -l walltime=$timestring
999	#PBS -l ncpus=1
1000	#PBS -l pmem=${memory}mb
1001	#PBS -o $remote_dayfile
1002	#PBS -j oe
1003
1004	%%END%%
1005
1006	fi
1007
1008	elif [[ $remote_host = lcyon ]]
1009	then
1010
1011	if [[ $numprocs != 0 ]]
1012	then
1013	cat > $job_to_send << %%END%%
1014	#!/bin/ksh
1015	#PBS -N $job_name
1016	#PBS -l walltime=$timestring
1017	#PBS -l ncpus=$numprocs
1018	#PBS -l pmem=${memory}mb
1019	#PBS -o $remote_dayfile
1020	#PBS -j oe
1021
1022	%%END%%
1023
1024	else
1025	cat > $job_to_send << %%END%%
1026	#!/bin/ksh
1027	#PBS -N $job_name
1028	#PBS -l walltime=$timestring
1029	#PBS -l ncpus=1
1030	#PBS -l pmem=${memory}mb
1031	#PBS -o $remote_dayfile
1032	#PBS -j oe
1033
1034	%%END%%
1035
1036	fi
1037
1038	elif [[ $remote_host = lcxe6 ]]
1039	then
1040
1041	if [[ $numprocs != 0 ]]
1042	then
1043	cat > $job_to_send << %%END%%
1044	#!/bin/ksh
1045	#PBS -S /bin/ksh
1046	#PBS -N $job_name
1047	#PBS -A $project_account
1048	#PBS -j oe
1049	#PBS -l walltime=$timestring
1050	#PBS -l mppwidth=${numprocs}
1051	#PBS -l mppnppn=${processes_per_node}
1052	#PBS -m abe
1053	#PBS -o $remote_dayfile
1054	$email_directive
1055
1056	$init_cmds
1057	$module_calls
1058
1059	%%END%%
1060
1061	else
1062	cat > $job_to_send << %%END%%
1063	#!/bin/ksh
1064	#PBS -S /bin/ksh
1065	#PBS -N $job_name
1066	#PBS -A $project_account
1067	#PBS -j oe
1068	#PBS -l walltime=$timestring
1069	#PBS -l ncpus=1
1070	#PBS -l pmem=${memory}mb
1071	#PBS -m abe
1072	$email_directive
1073	#PBS -o $remote_dayfile
1074
1075	$init_cmds
1076	$module_calls
1077
1078	%%END%%
1079
1080	fi
1081
1082	elif [[ $remote_host = lckyoto ]]
1083	then
1084
1085	cat > $job_to_send << %%END%%
1086	#!/bin/ksh
1087	# @\$-o $remote_dayfile
1088	# @\$-eo -oi
1089	# @\$-lP 16
1090	# @\$-lp 1
1091	# @\$-lm 28gb -llm unlimited -ls unlimited
1092	# @\$-q $queue
1093	# @\$-Pvn abs_pack
1094	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1095	#. /thin/local/etc/setprofile/intel-11.0.sh
1096	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1097	. ~/.myprofile
1098	#. /home2/t/t51254/palm/current_version/myprofile
1099	#. /thin/apps/pgi/mpi.sh
1100	#
1101	env
1102	#
1103	set -x
1104
1105	%%END%%
1106
1107	elif [[ $remote_host = lcxt5m ]]
1108	then
1109
1110	if [[ $numprocs != 0 ]]
1111	then
1112	cat > $job_to_send << %%END%%
1113	#!/bin/ksh
1114	#PBS -S /bin/ksh
1115	#PBS -N $job_name
1116	#PBS -j oe
1117	#PBS -l walltime=$timestring
1118	#PBS -l mppwidth=${numprocs}
1119	#PBS -l mppnppn=${processes_per_node}
1120	#PBS -m abe
1121	#PBS -o $remote_dayfile
1122
1123	$init_cmds
1124	$module_calls
1125
1126	%%END%%
1127
1128	else
1129	cat > $job_to_send << %%END%%
1130	#!/bin/ksh
1131	#PBS -S /bin/ksh
1132	#PBS -N $job_name
1133	#PBS -j oe
1134	#PBS -l walltime=$timestring
1135	#PBS -l ncpus=1
1136	#PBS -l pmem=${memory}mb
1137	#PBS -m abe
1138	#PBS -o $remote_dayfile
1139
1140	$init_cmds
1141	$module_calls
1142
1143	%%END%%
1144
1145	fi
1146
1147	elif [[ $remote_host = lckyuh ]]
1148	then
1149	cat > $job_to_send << %%END%%
1150	#!/bin/bash
1151	#PJM -L "rscgrp=$queue"
1152	#PJM -L "node=$nodes"
1153	#PJM --mpi "proc=$numprocs"
1154	#PJM -L "elapse=$timestring"
1155	#PJM -o $remote_dayfile
1156	#PJM -j
1157	#PJM -X
1158	#PJM --no-stging
1159
1160	export LANG=en_US.UTF-8
1161	%%END%%
1162
1163	elif [[ $remote_host = lckyut ]]
1164	then
1165	cat > $job_to_send << %%END%%
1166	#!/bin/bash
1167	#PJM -L "rscgrp=$queue"
1168	#PJM -L "vnode=$numprocs"
1169	#PJM -L "vnode-core=1"
1170	#PJM -L "elapse=$timestring"
1171	#PJM --mpi proc=$numprocs
1172	#PJM -o $remote_dayfile
1173	#PJM -j
1174	#PJM -X
1175	#PJM --no-stging
1176
1177	export LANG=en_US.UTF-8
1178	%%END%%
1179
1180	elif [[ $remote_host = lcocean ]]
1181	then
1182	cat > $job_to_send << %%END%%
1183	#!/bin/bash
1184	#$ -cwd
1185	#$ -V
1186	#$ -N $job_name
1187	#$ -pe orte $numprocs
1188	#$ -o $remote_dayfile
1189	#$ -j y
1190	#$ -R y
1191	$init_cmds
1192	$module_calls
1193
1194	%%END%%
1195
1196	elif [[ $remote_host = nech ]]
1197	then
1198
1199	if (( nodes > 1 ))
1200	then
1201	cat > $job_to_send << %%END%%
1202	#!/bin/ksh
1203	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1204	#PBS -l ${qsubmem}=${Memory}gb
1205	#PBS -b $nodes
1206	#PBS -o $remote_dayfile
1207	#PBS -N palm
1208	#PBS -j o
1209	#PBS -T mpisx
1210
1211	%%END%%
1212
1213	elif [[ $numprocs != 0 ]]
1214	then
1215	cat > $job_to_send << %%END%%
1216	#!/bin/ksh
1217	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1218	#PBS -l ${qsubmem}=${Memory}gb
1219	#PBS -o $remote_dayfile
1220	#PBS -N palm
1221	#PBS -j o
1222
1223	%%END%%
1224
1225	else
1226	cat > $job_to_send << %%END%%
1227	#!/bin/ksh
1228	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1229	#PBS -o $remote_dayfile
1230	#PBS -j o
1231
1232	%%END%%
1233
1234	fi
1235
1236	elif [[ $remote_host = lctit ]]
1237	then
1238	cat > $job_to_send << %%END%%
1239	#!/bin/ksh
1240	$init_cmds
1241	$module_calls
1242
1243	%%END%%
1244
1245	# SET OPTIONS FOR SUBMIT-COMMAND
1246	if [[ $tasks_per_node != $processes_per_node ]]
1247	then
1248	submcom="$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 -et 1 -q $queue "
1249	else
1250	submcom="$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 -et 1 -q $queue "
1251	fi
1252
1253	else
1254
1255	cat > $job_to_send << %%END%%
1256	# @\$-q ${queue}
1257	# @\$-l${qsubtime} $timestring
1258	# @\$-l${qsubmem} ${memory}mb
1259	# @\$-o $remote_dayfile
1260	# @\$-eo
1261
1262	%%END%%
1263
1264	fi
1265
1266
1267	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1268	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1269	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1270	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1271	then
1272	echo "set +vx" >> $job_to_send
1273	echo "trap '" >> $job_to_send
1274	echo "set +vx" >> $job_to_send
1275	if [[ $(echo $remote_host \| cut -c1-3) = ibm \|\| $remote_host = lcbullhh \|\| $remote_host = lccrayb \|\| $remote_host = lccrayh \|\| $(echo $remote_host \| cut -c1-3) = nec \|\| $remote_host = lcflow \|\| $remote_host = lckiaps \|\| $remote_host = lckyu* \|\| $remote_host = lcxe6 \|\| $remote_host = lcocean ]]
1276	then
1277	if [[ $remote_host = ibmh ]]
1278	then
1279	return_queue=c1
1280	elif [[ $remote_host = ibmkisti ]]
1281	then
1282	return_queue=class.1-2
1283	elif [[ $remote_host = ibmku ]]
1284	then
1285	return_queue=sdbg2
1286	elif [[ $remote_host = ibms ]]
1287	then
1288	return_queue=p_normal
1289	elif [[ $remote_host = lcbullhh ]]
1290	then
1291	return_queue=shared
1292	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1293	then
1294	return_queue=dataq
1295	elif [[ $remote_host = lcxe6 ]]
1296	then
1297	return_queue=debug
1298	elif [[ $remote_host = lckiaps ]]
1299	then
1300	return_queue=express
1301	elif [[ $remote_host = lckyuh ]]
1302	then
1303	return_queue=cx-single
1304	elif [[ $remote_host = lckyut ]]
1305	then
1306	return_queue=cx-single
1307	else
1308	return_queue=unknown
1309	fi
1310
1311	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1312	then
1313
1314	if [[ $remote_host = ibmku ]]
1315	then
1316	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1317	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1318	else
1319	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1320	fi
1321	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1322	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1323	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1324	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1325	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1326	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1327	if [[ $host != "ibmh" ]]
1328	then
1329	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1330	fi
1331	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1332	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1333
1334	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1335	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1336
1337	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1338	echo "echo \"batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \\\"$job_catalog\\\" $local_dayfile\" >> scpjob.$identifier" >> $job_to_send
1339	if [[ $remote_host = ibmku ]]
1340	then
1341	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1342	fi
1343	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1344
1345	elif [[ $remote_host = nech ]]
1346	then
1347	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1348	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1349	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1350	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1351	echo "#PBS -j o" >> $job_to_send
1352	echo " " >> $job_to_send
1353	echo "set -x" >> $job_to_send
1354	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1355	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1356	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1357	echo "%%END%%" >> $job_to_send
1358
1359	elif [[ $remote_host = lcbullhh ]]
1360	then
1361	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1362	echo "#!/bin/bash" >> $job_to_send
1363	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1364	echo "#SBATCH -t 00:20:00" >> $job_to_send
1365	echo "#SBATCH -N 1" >> $job_to_send
1366	echo "#SBATCH -n 1" >> $job_to_send
1367	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1368	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1369	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1370	echo "#SBATCH -A $project_account" >> $job_to_send
1371	echo "#SBATCH -p $return_queue" >> $job_to_send
1372	echo " " >> $job_to_send
1373	echo "set -x" >> $job_to_send
1374	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1375	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1376	echo "%%END%%" >> $job_to_send
1377
1378	elif [[ $remote_host = lckyuh ]]
1379	then
1380	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1381	echo "#!/bin/bash" >> $job_to_send
1382	echo "#PJM -L \"node=1\"" >> $job_to_send
1383	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1384	echo "#PJM --no-stging" >> $job_to_send
1385	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1386	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1387	echo "#PJM -j" >> $job_to_send
1388	echo " " >> $job_to_send
1389	echo "export LANG=en_US.UTF-8" >> $job_to_send
1390	echo "set -x" >> $job_to_send
1391	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1392	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1393	echo "%%END%%" >> $job_to_send
1394
1395	elif [[ $remote_host = lckyut ]]
1396	then
1397	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1398	echo "#!/bin/bash" >> $job_to_send
1399	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1400	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1401	echo "#PJM --no-stging" >> $job_to_send
1402	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1403	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1404	echo "#PJM -j" >> $job_to_send
1405	echo " " >> $job_to_send
1406	echo "export LANG=en_US.UTF-8" >> $job_to_send
1407	echo "set -x" >> $job_to_send
1408	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1409	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1410	echo "%%END%%" >> $job_to_send
1411
1412	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1413	then
1414	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1415	echo "#!/bin/bash" >> $job_to_send
1416	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1417	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1418	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1419	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1420	echo "#PBS -j oe" >> $job_to_send
1421	echo " " >> $job_to_send
1422	echo "set -x" >> $job_to_send
1423	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1424	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1425	echo "%%END%%" >> $job_to_send
1426
1427	elif [[ $remote_host = lcocean ]]
1428	then
1429	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1430	echo "#!/bin/bash" >> $job_to_send
1431	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1432	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1433	echo "SGEPREFIX -cwd" >> $job_to_send
1434	echo "SGEPREFIX -j y" >> $job_to_send
1435	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1436	echo " " >> $job_to_send
1437	echo "set -x" >> $job_to_send
1438	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1439	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1440	echo "" >> $job_to_send
1441	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1442	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1443	echo "rm -f scpjob.${identifier}" >> $job_to_send
1444	echo "%%END%%" >> $job_to_send
1445	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1446	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1447
1448	elif [[ $remote_host = lcflow ]]
1449	then
1450	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1451	echo "#!/bin/bash" >> $job_to_send
1452	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1453	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1454	echo "SGEPREFIX -cwd" >> $job_to_send
1455	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1456	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1457	echo "SGEPREFIX -j y" >> $job_to_send
1458	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1459	echo " " >> $job_to_send
1460	echo "set -x" >> $job_to_send
1461	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1462	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1463	echo "" >> $job_to_send
1464	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1465	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1466	echo "rm -f scpjob.${identifier}" >> $job_to_send
1467	echo "%%END%%" >> $job_to_send
1468	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1469	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1470	elif [[ $remote_host = lcxe6 ]]
1471	then
1472	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1473	echo "#!/bin/ksh" >> $job_to_send
1474	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1475	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1476	echo "#PBS -A $project_account" >> $job_to_send
1477	echo "#PBS -l mppwidth=1" >> $job_to_send
1478	echo "#PBS -l mppnppn=1" >> $job_to_send
1479	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1480	echo "#PBS -j oe" >> $job_to_send
1481	echo " " >> $job_to_send
1482	echo "set -x" >> $job_to_send
1483	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1484	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1485	echo "%%END%%" >> $job_to_send
1486	else
1487
1488	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1489	echo "# @\\\$-q $return_queue" >> $job_to_send
1490	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1491	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1492	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1493	then
1494	echo "# @\$-l mpp_p=0" >> $job_to_send
1495	fi
1496	echo '# @\$-lF 10mb' >> $job_to_send
1497	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1498	echo '# @\\\$-eo' >> $job_to_send
1499	echo " " >> $job_to_send
1500	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1501	then
1502	echo "set +vx" >> $job_to_send
1503	echo ". .profile" >> $job_to_send
1504	fi
1505	echo "set -x" >> $job_to_send
1506	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null" >> $job_to_send
1507	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1508	echo "%%END%%" >> $job_to_send
1509
1510	fi
1511
1512	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1513	then
1514	echo "llsubmit scpjob.$identifier" >> $job_to_send
1515	elif [[ $remote_host = lcbullhh ]]
1516	then
1517	echo "sbatch scpjob.$identifier" >> $job_to_send
1518	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1519	then
1520	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1521	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1522	then
1523	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1524	elif [[ $remote_host = t3es ]]
1525	then
1526	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1527	elif [[ $remote_host = lckiaps ]]
1528	then
1529	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1530	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1531	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1532	elif [[ $remote_host = lckyu* ]]
1533	then
1534	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1535	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1536	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1537	then
1538	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1539	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1540	else
1541	echo "$submcom scpjob.$identifier" >> $job_to_send
1542	fi
1543	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1544	then
1545	echo "rm scpjob.$identifier" >> $job_to_send
1546	fi
1547	if [[ $remote_host = nech ]]
1548	then
1549	echo "cd -" >> $job_to_send
1550	fi
1551	else
1552	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1553	# ??? funktioniert das ÃŒberhaupt noch ???
1554	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1555	fi
1556	echo "set -x" >> $job_to_send
1557	echo " ' exit" >> $job_to_send
1558	echo "set -x" >> $job_to_send
1559	fi
1560
1561
1562	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1563	cat $file_to_send >> $job_to_send
1564
1565	if [[ $remote_host = ibm ]]
1566	then
1567	echo " " >> $job_to_send
1568	echo "exit" >> $job_to_send
1569	fi
1570
1571	# REMOVE JOB-FILE
1572	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1573	then
1574	echo " " >> $job_to_send
1575	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1576	fi
1577
1578
1579	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1580	if [[ $no_submit = false ]]
1581	then
1582	if [[ $remote_host != $local_host ]]
1583	then
1584	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1585	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1586	then
1587	job_catalog_save=$job_catalog
1588	job_catalog=job_queue
1589	elif [[ $remote_host = nech ]]
1590	then
1591	job_catalog_save=$job_catalog
1592	job_catalog=/hpf/b/${remote_user}/job_queue
1593	fi
1594	if [[ $remote_host = nech ]]
1595	then
1596	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1597	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1598	else
1599	scp $ssh_key $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1600	fi
1601	if [[ $? = 1 ]]
1602	then
1603	locat=scp; exit
1604	fi
1605	if [[ $remote_host = ibms ]]
1606	then
1607	job_catalog=$job_catalog_save
1608	fi
1609	[[ $verify = true ]] && printf "\n >>> finished\n"
1610	else
1611	eval job_catalog=$job_catalog
1612	cp $job_to_send ${job_catalog}/$job_on_remhost
1613	fi
1614
1615
1616
1617	# START NQS- / LOADLEVELER-JOB
1618	if [[ $remote_host != $local_host ]]
1619	then
1620	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1621
1622	if [[ $remote_host = ibmku ]]
1623	then
1624	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1625	elif [[ $remote_host = lcflow ]]
1626	then
1627	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1628	else
1629	ssh $ssh_key $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1630	fi
1631
1632	[[ $verify = true ]] && printf " >>> o.k.\n"
1633	else
1634	cd $job_catalog
1635	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1636	then
1637	eval $submcom $job_on_remhost
1638	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1639	then
1640	chmod u+x $job_on_remhost
1641	eval $submcom $job_on_remhost
1642	elif [[ $local_host = nech ]]
1643	then
1644	if [[ $queue = default ]]
1645	then
1646	eval $submcom $job_on_remhost
1647	else
1648	eval $submcom -q $queue $job_on_remhost
1649	fi
1650	elif [[ $local_host = lcbullhh ]]
1651	then
1652	if [[ $queue = default ]]
1653	then
1654	eval $submcom $job_on_remhost
1655	fi
1656	else
1657	qsub $job_on_remhost
1658	fi
1659
1660	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1661	# AT THE END OF THE JOB
1662	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1663	then
1664	rm $job_on_remhost
1665	fi
1666	cd - > /dev/null
1667	fi
1668	fi
1669
1670	# FINAL ACTIONS
1671	if [[ $no_submit = false ]]
1672	then
1673	rm -f $job_to_send
1674	fi
1675	[[ $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 |