Home

Context Navigation

source: palm/trunk/SCRIPTS/subjob @ 1918

Last change on this file since 1918 was 1917, checked in by witha, 9 years ago
Corrected documented changes in the file headers
Property svn:keywords set to `Id Rev`
File size: 57.8 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 1917 2016-05-27 14:28:12Z raasch $
30	#
31	# 1866 2016-04-15 06:50:59Z raasch
32	# adjusted for lcocean
33	#
34	# 1841 2016-04-07 19:14:06Z raasch
35	# script now running under bash
36	#
37	# 1701 2015-11-02 07:43:04Z maronga
38	# Bugfix: added missing init_cmds for lccrayh/lccrayb
39	#
40	# 1621 2015-07-17 11:39:33Z heinze
41	# adjustments for Mistral at DKRZ Hamburg (lcbullhh)
42	#
43	# 1575 2015-03-27 09:56:27Z raasch
44	# mpp2-queues added to lccrayh
45	#
46	# 1547 2015-01-29 15:09:12Z witha
47	# adjustments for ForWind computing cluster (lcflow)
48	#
49	# 1545 2015-01-29 06:52:23Z heinze
50	# local host name for blizzard further specified
51	#
52	# 1480 2014-10-17 14:41:49Z raasch
53	# adjustments for 2nd stage of HLRNIII
54	#
55	# 1468 2014-09-24 14:06:57Z maronga
56	# Typo removed (addres->address)
57	# Adjustments for lcxe6
58	#
59	# 1452 2014-08-22 09:41:06Z heinze
60	# local hosts for blizzard added
61	#
62	# 1450 2014-08-21 07:31:51Z heinze
63	# HLRN-III (lccrayb): testq queue adjusted to mpp1testq
64	#
65	# 1442 2014-07-28 07:09:10Z raasch
66	# HLRN-III (lccrayb/lccrayh) queues adjusted
67	#
68	# 1378 2014-04-28 06:04:58Z raasch
69	# -et option added for lctit
70	#
71	# 1350 2014-04-04 13:01:30Z maronga
72	# location of qsub updated for lcxe6
73	#
74	# 1289 2014-03-04 07:12:34Z raasch
75	# German comments translated to English
76	# fimm-, necriam-, scirocco-, ibmy-, and sgi-specific code removed
77	#
78	# 1279 2014-01-28 12:10:14Z raasch
79	# node calculation modified due to changes in mrun (tasks_per_node must not be
80	# an integral divisor of numprocs any more)
81	#
82	# 1274 2014-01-09 13:14:54Z heinze
83	# adjustments for lccrayh
84	#
85	# 1266 2013-12-11 12:07:34Z heinze
86	# further adjustments for lccrayb (use msub instead of qsub)
87	#
88	# 1264 2013-12-09 12:46:09Z fricke
89	# Bugfix: Using number of nodes instead of number of processors (lccrayb)
90	#
91	# 1262 2013-12-09 10:57:20Z fricke
92	# further adjustments for lccrayb
93	#
94	# 1260 2013-12-04 12:48:04Z raasch
95	# jaboticaba admitted
96	#
97	# 1255 2013-11-07 14:43:35Z raasch
98	# further adjustments for lccrayb
99	#
100	# 1224 2013-09-16 07:27:23Z raasch
101	# first adjustments for lccrayb
102	#
103	# 1202 2013-07-10 16:22:07Z witha
104	# adjustments for Forwind cluster (lcflow)
105	#
106	# 1199 2013-07-05 14:52:22Z raasch
107	# adjustments for CSC Helsinki (lccrayf)
108	#
109	# use of cluster/express queue enabled (ibmh)
110	# vinessa added (imuk)
111	#
112	# 1103 2013-02-20 02:15:53Z raasch
113	# bash compatibility adjustments (usage of OPTIND, output formatting with printf
114	# instead typeset -L/R),
115	# further adjustments for lckyuh
116	#
117	# 2013-02-10 01:47:43Z raasch
118	# adjustments for Kyushu-Univeristy computing center (lckyuh - hayaka)
119	# and for Forwind cluster (lcflow)
120	#
121	# 1094 2013-02-03 01:52:12Z raasch
122	# new option -P for explicit setting of ssh/scp port,
123	# decalpha parts (yonsei) removed
124	#
125	# 2013-02-02 07:06:13Z raasch
126	# adjustments for Kyushu-University computing center (lckyut - tatara)
127	# old changelog messages removed
128	#
129	# 1046 2012-11-09 14:38:45Z maronga
130	# code put under GPL (PALM 3.9)
131	#
132	# 08/07/94 - Siggi - first version finished
133	# 29/06/94 - Siggi - script development started
134	#--------------------------------------------------------------------------------#
135	# subjob - script for automatic generation and submission of batch-job files
136	# for various batch queuing systems
137	#--------------------------------------------------------------------------------#
138
139
140	# VARIABLE-DECLARATIONS AND DEFAULT VALUES
141	delete_dayfile=false
142	email_notification=none
143	group_number=none
144	locat=normal
145	no_default_queue=none
146	no_submit=false
147	job_catalog="~/job_queue"
148	job_name=none
149	local_user=$LOGNAME
150	node_usage=shared
151	numprocs=0
152	punkte="..........................................................."
153	submcom=qsub
154	queue=default
155	remote_host=none
156	remote_user=""
157	verify=true
158
159	typeset -i cputime=0 memory=0 Memory=0 minuten resttime sekunden stunden
160	typeset -i numprocs mpi_tasks=0 nodes=0 processes_per_node=0 tasks_per_node=0 threads_per_task=1
161
162
163
164	# ERROR HANDLING
165	# IN CASE OF EXIT:
166	trap 'if [[ $locat != normal ]]
167	then
168	case $locat in
169	(option) printf "\n --> available optios can be displayed"
170	printf " by typing:"
171	printf "\n \"subjob ?\" \n";;
172	(ftpcopy\|parameter\|scp\|verify) printf "\n";;
173	(*) printf "\n +++ unknown error"
174	printf "\n please inform S. Raasch!\n"
175	esac
176	[[ -f $job_to_send ]] && rm $job_to_send
177	printf "\n\n+++ SUBJOB killed \n\n"
178	fi' exit
179
180
181	# IN CASE OF TERMINAL-BREAK:
182	trap '[[ -f $job_to_send ]] && rm $job_to_send
183	printf "\n\n+++ SUBJOB killed \n\n"
184	exit
185	' 2
186
187
188	# DETERMINE NAME OF LOCAL HOST
189	local_host=$(hostname)
190
191	# SET HOST-SPECIFIC VARIABLES VEREINBAREN (CHECK, IF LOCAL HOST
192	# IS ADMITTED AT ALL)
193	# NOTE: ONE OF THE ENTRIES FOR "lck" OR "lckordi" ALWAYS HAS TO BE
194	# COMMENT OUT, BECAUSE THE HOSTNAME (node*) IS SAME FOR BOTH MACHINES
195	case $local_host in
196	(ambiel-lx) local_address=134.106.74.48; local_host=lcfor;;
197	(atmos) local_address=172.20.25.35; local_host=lcide;;
198	(austru) local_address=130.75.105.128; local_host=lcmuk;;
199	(autan) local_address=130.75.105.57; local_host=lcmuk;;
200	(bora) local_address=130.75.105.103; local_host=lcmuk;;
201	(b04*) local_address=133.5.4.33; local_host=lckyuh;;
202	(blizzard1\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.15; local_host=ibmh;;
203	(blizzard2\|p0\|p1\|p2\|p3\|p4\|p5\|p6\|p7\|p8\|p9) local_address=136.172.40.16; local_host=ibmh;;
204	(blogin\|bxc) local_address=130.73.233.1; local_host=lccrayb;;
205	(hlogin\|hxc) local_address=130.75.4.1; local_host=lccrayh;;
206	(breva) local_address=130.75.105.98; local_host=lcmuk;;
207	(buran) local_address=130.75.105.58; local_host=lcmuk;;
208	(caurus) local_address=130.75.105.19; local_host=lcmuk;;
209	(climate*) local_address=165.132.26.68; local_host=lcyon;;
210	(clogin*) local_address=86.50.166.21; local_host=lccrayf;;
211	(cs*) local_address=136.172.44.131; local_host=nech;;
212	(elephanta) local_address=130.75.105.6; local_host=lcmuk;;
213	(flow01) local_address=10.141.255.71; local_host=lcflow;;
214	(flow02) local_address=10.141.255.72; local_host=lcflow;;
215	(node*) local_address=165.132.26.61 local_host=lck;;
216	# (node*) local_address=210.219.61.8 local_host=lckordi;;
217	(gaia*) local_address=150.183.146.24; local_host=ibmkisti;;
218	(gallego) local_address=130.75.105.10; local_host=lcmuk;;
219	(gregale) local_address=130.75.105.109; local_host=lcmuk;;
220	(hababai) local_address=130.75.105.108; local_host=lcmuk;;
221	(hayaka*) local_address=133.5.4.33; local_host=lckyuh;;
222	(hexagon.bccs.uib.no) local_address=129.177.20.113; local_host=lcxe6;;
223	(hx*) local_address=133.3.51.11; local_host=lckyoto;;
224	(inferno) local_address=130.75.105.5; local_host=lcmuk;;
225	(irifi) local_address=130.75.105.104; local_host=lcmuk;;
226	(jaboticaba) local_address=150.163.25.181; local_host=lcbr;;
227	(sno) local_address=130.75.105.113; local_host=lcmuk;;
228	(levanto) local_address=130.75.105.45; local_host=lcmuk;;
229	(login*) local_address=118.128.66.223; local_host=lckiaps;;
230	(maestro) local_address=130.75.105.2; local_host=lcmuk;;
231	(meller) local_address=134.106.74.155; local_host=lcfor;;
232	(meteo-login*) local_address=193.166.211.144;local_host=lcxt5m;;
233	(mlogin1\|m1) local_address=136.172.50.13; local_host=lcbullhh;;
234	(hexagon*) local_address=129.177.20.113; local_host=lcxe6;;
235	(nobel*) local_address=150.183.5.101; local_host=ibms;;
236	(ocean) local_address="ocean"; local_host=lcocean;;
237	(orkan) local_address=130.75.105.3; local_host=lcmuk;;
238	(ostria) local_address=130.75.105.106; local_host=lcmuk;;
239	(paesano) local_address=130.75.105.46; local_host=lcmuk;;
240	(pcj*) local_address=172.31.120.1; local_host=lckyut;;
241	(pingui) local_address=134.106.74.118; local_host=lcfor;;
242	(quanero) local_address=130.75.105.107; local_host=lcmuk;;
243	(rte*) local_address=133.5.185.60; local_host=lcrte;;
244	(schultzl-Latitude-E6540) local_address="schultzl-Latitude-E6540"; local_host=lcsch;;
245	(shiokaze-lx) local_address=134.106.74.123; local_host=lcfor;;
246	(sisu-login*) local_address=86.50.166.21; local_host=lccrayf;;
247	(solano) local_address=130.75.105.110; local_host=lcmuk;;
248	(sugoka*) local_address=172.31.120.1; local_host=lckyut;;
249	(tc*) local_address="ocean"; local_host=lcocean;;
250	(t2a*) local_address=10.1.6.165; local_host=lctit;;
251	(urban*) local_address=147.46.30.151 local_host=lcsb;;
252	(vinessa) local_address=130.75.105.112; local_host=lcmuk;;
253	(vorias) local_address=172.20.25.43; local_host=lcmuk;;
254	(*.cc.kyushu-u.ac.jp) local_address=133.5.4.129; local_host=ibmku;;
255	(*) printf "\n +++ \"$local_host\" unknown";
256	printf "\n please contact the PALM group at IMUK";
257	locat=parameter; exit;;
258	esac
259
260
261
262	# BY DEFAULT, THE REMOTE HOST IS THE LOCAL HOST
263	remote_host=$local_host
264
265
266
267
268	# READ THE SHELLSCRIPT-OPTIONS
269	while getopts :c:dDe:g:h:m:n:N:O:P:q:t:T:u:vX: option
270	do
271	case $option in
272	(c) job_catalog=$OPTARG;;
273	(d) delete_dayfile=true;;
274	(D) no_submit=true;;
275	(e) email_notification=$OPTARG;;
276	(g) group_number=$OPTARG;;
277	(h) remote_host=$OPTARG;;
278	(m) memory=$OPTARG;;
279	(n) job_name=$OPTARG;;
280	(N) node_usage=$OPTARG;;
281	(O) threads_per_task=$OPTARG;;
282	(P) scp_port=$OPTARG;;
283	(q) no_default_queue=$OPTARG;;
284	(t) cputime=$OPTARG;;
285	(T) tasks_per_node=$OPTARG;;
286	(u) remote_user=$OPTARG;;
287	(v) verify=false;;
288	(X) numprocs=$OPTARG;;
289	(\?) printf "\n +++ Option $OPTARG unknown \n";
290	locat=option; exit;;
291	esac
292	done
293
294
295	# GET THE NAME OF THE JOBFILE AS NEXT ARGUMENT
296	(( to_shift = $OPTIND - 1 ))
297	shift $to_shift; file_to_send=$1
298
299
300	# OUTPUT OF SHORT DESCRIPTION OF SCRIPT-OPTIONS
301	if [ "$1" = "?" ]
302	then
303	(printf "\n *** subjob can be called as follows:\n"
304	printf "\n subjob -c.. -d -D -h.. -m.. -q.. -t.. -u.. -v <jobfile>\n"
305	printf "\n Description of available options:\n"
306	printf "\n Option Description Default-Value"
307	printf "\n -c job-input- and output-catalog ~/job_queue"
308	printf "\n -d no job-protocol will be created ---"
309	printf "\n -D only the job-file will be created ---"
310	printf "\n -h execution host, available hosts: $remote_host"
311	printf "\n ibm, ibmh, ibmkisti, ibmku, ibms, lc...,"
312	printf "\n lckiaps, lctit, nech"
313	printf "\n -m memory demand per process in MByte ---"
314	printf "\n -n jobname <jobdatei>"
315	printf "\n -O threads per task (for OpenMP usage) 1"
316	printf "\n -P ssh/scp port default port"
317	printf "\n -q job-queue to be used default"
318	printf "\n -t allowed cpu-time in seconds ---"
319	printf "\n -T tasks per node (on parallel hosts) ---"
320	printf "\n -u username on execution host from .netrc"
321	printf "\n -v no prompt for confirmation ---"
322	printf "\n -X # of processors (on parallel hosts) 1"
323	printf "\n "
324	printf "\n The only possible positional parameter is <jobfile>:"
325	printf "\n The complete NQS-job must be provided here."
326	printf "\n <jobfile>=? creates this outline\n\n") \| more
327	exit
328	fi
329
330
331
332	# CHECK, IF JOB-FILE HAS BEEN GIVEN AS ARGUMENT AND IF THE FILE ITSELF EXISTS
333	if [[ "$file_to_send" = "" ]]
334	then
335	printf "\n +++ job-file missing"
336	locat=parameter; exit
337	else
338	if [[ -f $file_to_send ]]
339	then
340	true
341	else
342	printf "\n +++ job-file: "
343	printf "\n $file_to_send"
344	printf "\n does not exist"
345	locat=parameter; exit
346	fi
347	fi
348
349
350
351	# IF NO JOBNAME HAS BEEN GIVEN, JOBNAME IS SET TO THE NAME OF THE JOB-FILE,
352	# PROVIDED THAT THE JOB-FILE NAME DOES NOT CONTAIN ANY PATH
353	if [[ $job_name = none ]]
354	then
355	job_name=$file_to_send
356	fi
357	if [[ $(echo $job_name \| grep -c "/") != 0 ]]
358	then
359	printf "\n +++ job-file name: "
360	printf "\n $job_name"
361	printf "\n must not contain \"/\"-characters"
362	locat=parameter; exit
363	fi
364
365
366
367
368	# SET HOST-SPECIFIC QUANTITIES, OR TERMINATE IN CASE OF UNKNOWN HOST,
369	# OR IF NO HOST HAS BEEN GIVEN
370	if [[ $remote_host = none ]]
371	then
372	printf "\n +++ host missing"
373	locat=option; exit
374	else
375	case $remote_host in
376	(ibm) queue=p690_standard; remote_address=134.76.99.81; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
377	(ibmh) queue=cluster; remote_address=136.172.40.15; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
378	(ibmkisti) queue=class.32plus; remote_address=150.183.146.24; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
379	(ibmku) queue=s4; remote_address=133.5.4.129; submcom=/usr/local/bin/llsubmit;;
380	(ibms) queue=p_normal; remote_address=150.183.5.101; submcom=/usr/lpp/LoadL/full/bin/llsubmit;;
381	(lcbullhh) queue=compute; remote_address=136.172.50.13; submcom=/usr/bin/sbatch;;
382	(lccrayb) queue=mpp1testq; remote_address=130.73.233.1; submcom=/opt/moab/default/bin/msub;;
383	(lccrayh) queue=mpp1testq; remote_address=130.75.4.1; submcom=/opt/moab/default/bin/msub;;
384	(lccrayf) queue=small; remote_address=86.50.166.21; submcom=/opt/slurm/default/bin/sbatch;;
385	(lcflow) remote_address=10.140.1.72; submcom=qsub;;
386	(lckyoto) remote_address=133.3.51.11; submcom=/thin/local/bin/qsub;;
387	(lck) remote_address=165.132.26.61; submcom=/usr/torque/bin/qsub;;
388	(lckiaps) remote_address=118.128.66.223; submcom=/cm/shared/apps/pbspro/11.0.2.110766/bin/qsub;;
389	(lckordi) remote_address=210.219.61.8; submcom=/usr/torque/bin/qsub;;
390	(lckyuh) remote_address=133.5.4.33; submcom=/usr/bin/pjsub;;
391	(lckyut) remote_address=133.5.4.37; submcom=/usr/bin/pjsub;;
392	(lcocean) remote_address="ocean"; submcom=qsub;;
393	(lcsb) remote_address=147.46.30.151; submcom=/usr/torque/bin/qsub;;
394	(lctit) queue=S; remote_address=10.1.6.165; submcom=/opt/pbs/tools/bin/t2sub;;
395	(lcxe6) remote_address=129.177.20.113; submcom=/opt/torque/default/bin/qsub;;
396	(lcxt5m) remote_address=193.166.211.144; submcom=/opt/pbs/10.1.0.91350/bin/qsub;;
397	(lcyon) remote_address=165.132.26.68; submcom=/usr/torque/bin/qsub;;
398	(nech) qsubmem=memsz_job; qsubtime=cputim_job; remote_address=136.172.44.147; submcom="/usr/local/bin/qsub";;
399	(*) printf "\n +++ hostname \"$remote_host\" not allowed";
400	locat=parameter; exit;;
401	esac
402	fi
403
404
405	# CHECK, IF A VALID QUEUE HAS BEEN GIVEN
406	if [[ $no_default_queue != none ]]
407	then
408	error=false
409	ndq=$no_default_queue
410	case $remote_host in
411	(ibm) case $ndq in
412	(p690_express\|p690_standard\|p690_long) error=false;;
413	(*) error=true;;
414	esac;;
415	(ibmh) case $ndq in
416	(cluster\|express) error=false;;
417	(*) error=true;;
418	esac;;
419	(ibmkisti) case $ndq in
420	(class.32plus\|class.1-2\|class.2-32) error=false;;
421	(*) error=true;;
422	esac;;
423	(ibmku) case $ndq in
424	(sdbg1\|sdbg2\|sdbg4\|s4\|s16\|s32\|s32-s) error=false;;
425	(*) error=true;;
426	esac;;
427	(ibms) case $ndq in
428	(express\|normal\|p_express\|p_normal\|p_normal_1.3\|p_normal_1.7\|grand) error=false;;
429	(*) error=true;;
430	esac;;
431	(lcbullhh) case $ndq in
432	(compute\|shared) error=false;;
433	(*) error=true;;
434	esac;;
435	(lccrayb) case $ndq in
436	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
437	(*) error=true;;
438	esac;;
439	(lccrayh) case $ndq in
440	(dataq\|mpp1q\|mpp1testq\|mpp2q\|mpp2testq\|smp1q\|smp1testq\|specialm1q) error=false;;
441	(*) error=true;;
442	esac;;
443	(lccrayf) case $ndq in
444	(usup\|test*\|small\|large) error=false;;
445	(*) error=true;;
446	esac;;
447	(lcflow) case $ndq in
448	(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;;
449	(*) error=true;;
450	esac;;
451	(lckiaps) case $ndq in
452	(express\|normal) error=false;;
453	(*) error=true;;
454	esac;;
455	(lckyoto) case $ndq in
456	(eh\|ph) error=false;;
457	(*) error=true;;
458	esac;;
459	(lckyuh) case $ndq in
460	(fx-dbg\|fx-single\|fx-small\|fx-middle\|fx-large) error=false;;
461	(*) error=true;;
462	esac;;
463	(lckyut) case $ndq in
464	(cx-dbg\|cx-single\|cx-small\|cx-middle\|cx-large) error=false;;
465	(*) error=true;;
466	esac;;
467	(lctit) case $ndq in
468	(G\|L128\|L256\|L512H\|S\|S96\|V) error=false;;
469	(*) error=true;;
470	esac;;
471	(t3eb) case $ndq in
472	(berte\|p50\|p100\|p392\|forfree\|p25himem) error=false;;
473	(*) error=true;;
474	esac;;
475	(t3eh) case $ndq in
476	(para_t3e\|em\|k\|l\|lm\|comp_t3e\|c\|p\|ht) error=false;;
477	(*) error=true;;
478	esac;;
479	(t3ej2\|t3ej5) case $ndq in
480	(low\|normal\|high) error=false;;
481	(*) error=true;;
482	esac;;
483	(t3es) case $ndq in
484	(batch\|serial-4\|pe4\|p48\|pe16\|pe32\|pe64\|pe128) error=false;;
485	(*) error=true;;
486	esac;;
487	esac
488	if [[ $error = true ]]
489	then
490	printf "\n +++ queue \"$no_default_queue\" on host \"$remote_host\" not allowed"
491	locat=parameter; exit
492	else
493	queue=$no_default_queue
494	fi
495	fi
496
497
498
499	# CHECK THE CPU-TIME
500	# SPLIT TIME INTO HOURS, MINUTES, AND SECONDS
501	done=false
502	while [[ $done = false ]]
503	do
504	if (( $cputime <= 0 ))
505	then
506	printf "\n +++ wrong cpu-time or cpu-time missing"
507	printf "\n >>> Please type cpu-time in seconds as INTEGER:"
508	printf "\n >>> "
509	read cputime 1>/dev/null 2>&1
510	else
511	done=true
512	fi
513	done
514	if [[ $remote_host = nech ]]
515	then
516	if (( tasks_per_node != 0 ))
517	then
518	(( cputime = cputime * tasks_per_node ))
519	elif [[ $numprocs != 0 ]]
520	then
521	(( cputime = cputime * numprocs ))
522	fi
523	fi
524	(( stunden = cputime / 3600 ))
525	(( resttime = cputime - stunden * 3600 ))
526	(( minuten = resttime / 60 ))
527	(( sekunden = resttime - minuten * 60 ))
528	timestring=${stunden}:${minuten}:${sekunden}
529
530
531
532	# CHECK THE MEMORY DEMAND
533	done=false
534	while [[ $done = false ]]
535	do
536	if (( memory <= 0 ))
537	then
538	printf "\n +++ wrong memory demand or memory demand missing"
539	printf "\n >>> Please type memory in MByte per process as INTEGER:"
540	printf "\n >>> "
541	read memory 1>/dev/null 2>&1
542	else
543	done=true
544	fi
545	done
546
547	if [[ $remote_host = nech ]]
548	then
549	if (( tasks_per_node != 0 ))
550	then
551	(( Memory = memory * tasks_per_node / 1000 ))
552	elif [[ $numprocs != 0 ]]
553	then
554	(( Memory = memory * numprocs / 1000 ))
555	else
556	(( Memory = memory / 1000 ))
557	fi
558	elif [[ $remote_host = lctit ]]
559	then
560	(( Memory = memory * tasks_per_node / 1000 ))
561	fi
562
563
564	# MEMORY DEMAND IN CASE OF OPENMP-USAGE ON IBM-SYSTEMS
565	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
566	then
567	(( memory = memory * threads_per_task ))
568	fi
569
570
571	# CALCULATE NUMBER OF REQUIRED NODES
572	if (( tasks_per_node != 0 ))
573	then
574	(( nodes = ( numprocs - 1 ) / ( tasks_per_node * threads_per_task ) + 1 ))
575	fi
576
577
578	# CALCULATE NUMBER OF PROCESSES PER NODE
579	(( processes_per_node = tasks_per_node * threads_per_task ))
580
581
582	# CALCULATE NUMBER OF MPI TASKS
583	(( mpi_tasks = numprocs / threads_per_task ))
584
585
586	# SET PORT NUMBER OPTION FOR CALLS OF ssh/scp, subjob AND batch_scp SCRIPTS
587	if [[ "$scp_port" != "" ]]
588	then
589	PORTOPT="-P $scp_port"
590	SSH_PORTOPT="-p $scp_port"
591	fi
592
593
594	# HEADER-OUTPUT
595	if [[ $verify = true ]]
596	then
597	printf "\n\n"
598	printf "#--------------------------------------------------------------# \n"
599	spalte1=SUBJOB;spalte2=$(date)
600	printf "\| %-20s%40s \| \n" "$spalte1" "$spalte2"
601	printf "\| \| \n"
602	printf "\| values of parameters/options: \| \n"
603	spalte1=$(echo local_host$punkte \| cut -c-20)
604	spalte2=$punkte$local_host
605	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
606	spalte1=$(echo remote_host$punkte \| cut -c-20)
607	spalte2=$punkte$remote_host
608	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
609	spalte1=$(echo queue$punkte \| cut -c-20)
610	spalte2=$punkte$queue
611	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
612	spalte1=$(echo memory$punkte \| cut -c-20)
613	spalte2="$punkte$memory mb"
614	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
615	spalte1=$(echo cputime$punkte \| cut -c-20)
616	spalte2="$punkte$cputime sec"
617	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
618	spalte1=$(echo job_name$punkte \| cut -c-20)
619	spalte2="$punkte$job_name"
620	printf "\| %-20s%40s \| \n" "$spalte1" "${spalte2: -40}"
621	printf "#--------------------------------------------------------------# \n\n"
622
623
624	# QUERY CHECK
625	antwort="dummy"
626	while [[ $antwort != y && $antwort != Y && $antwort != n && $antwort != N ]]
627	do
628	read antwort?" >>> continue (y/n) ? "
629	done
630	if [[ $antwort = n \|\| $antwort = N ]]
631	then
632	locat=verify; exit
633	fi
634	printf "\n"
635	fi
636
637	# GENERATE RANDOM IDENTIFIER, AND DETERMINE THE JOBNAME ON THE TARGET HOST
638	identifier=$RANDOM
639	job_on_remhost=${job_name}_${identifier}_$local_host
640	job_to_send=job_to_send_$identifier
641	if [[ $delete_dayfile = false ]]
642	then
643	remote_dayfile=${local_host}_${job_name}_result_$identifier
644	local_dayfile=${remote_host}_${job_name}
645	else
646	remote_dayfile=/dev/null
647	fi
648
649
650	# GENERATE THE BATCH-JOB SCRIPTS (FOR QUEUEING-SYSTEMS qsub/msub/LoadLeveler)
651	if [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs != 0 ]]
652	then
653
654	# GENERAL LOADLEVELER SETTINGS
655	execute_in_shell="#!/bin/ksh"
656	use_shell="# @ shell = /bin/ksh"
657	consumable_memory="ConsumableMemory($memory mb)"
658	class="# @ class = $queue"
659	environment="# @ environment = OMP_NUM_THREADS=$threads_per_task; MP_SHARED_MEMORY=yes"
660	network_to_use="# @ network.mpi = sn_all,shared,us"
661	data_limit="# @ data_limit = 1.76gb"
662	image_size="# @ image_size = 50"
663	wall_clock_limit="# @ wall_clock_limit = ${timestring},$timestring"
664
665	if [[ $email_notification = none ]]
666	then
667	notify_user=""
668	else
669	notify_user="# @ notify_user = $email_notification"
670	if [[ $delete_dayfile = true ]]
671	then
672	notification='# @ notification = never'
673	fi
674	fi
675
676	if [[ $remote_host = ibmh ]]
677	then
678	data_limit=""
679	network_to_use=""
680	class="# @ class = $queue"
681	environment=""
682	rset="# @ rset = RSET_MCM_AFFINITY"
683	task_affinity="# @ task_affinity = core(1)"
684	elif [[ $remote_host = ibmkisti ]]
685	then
686	network_to_use="# @ network.MPI = sn_all,shared,US"
687	wall_clock_limit="# @ wall_clock_limit = $timestring"
688	if [[ $threads_per_task = 1 ]]
689	then
690	rset="# @ rset = RSET_MCM_AFFINITY"
691	mcm_affinity_options="# @ mcm_affinity_options = mcm_mem_pref mcm_sni_none mcm_distribute"
692	fi
693	environment=""
694	use_shell=""
695	data_limit=""
696	image_size=""
697	elif [[ $remote_host = ibmku ]]
698	then
699	execute_in_shell="#!/usr/bin/ksh"
700	use_shell="# @ shell = /usr/bin/ksh"
701	consumable_memory=""
702	environment=""
703	network_to_use="# @ network.mpi = sn_all,shared,us"
704	data_limit=""
705	image_size=""
706	elif [[ $remote_host = ibms ]]
707	then
708	network_to_use="# @ network.mpi = csss,shared,us"
709	fi
710
711	cat > $job_to_send << %%END%%
712	$execute_in_shell
713	$use_shell
714
715	# @ job_type = parallel
716	# @ job_name = $job_name
717	# @ resources = ConsumableCpus($threads_per_task) $consumable_memory
718	# @ output = $remote_dayfile
719	# @ error = $remote_dayfile
720	$wall_clock_limit
721	$image_size
722	$class
723	$environment
724	$network_to_use
725	$data_limit
726	$rset
727	$mcm_affinity_options
728	$task_affinity
729	$notification
730	$notify_user
731
732	%%END%%
733
734	if (( nodes > 0 ))
735	then
736
737	if [[ $remote_host != ibmkisti ]]
738	then
739
740	cat >> $job_to_send << %%END%%
741	# @ node = $nodes
742	# @ tasks_per_node = $processes_per_node
743	# @ node_usage = $node_usage
744	# @ queue
745
746	%%END%%
747
748	else
749
750	cat >> $job_to_send << %%END%%
751	# @ total_tasks = $mpi_tasks
752	# @ blocking = unlimited
753	# @ queue
754
755	%%END%%
756
757	fi
758
759	else
760
761	cat >> $job_to_send << %%END%%
762	# @ blocking = unlimited
763	# @ total_tasks = $numprocs
764	# @ node_usage = $node_usage
765	# @ queue
766
767	%%END%%
768
769	fi
770
771	# WORKAROUND BECAUSE OF SILLY JOB FILTER ON ibmkisti
772	if [[ $remote_host = ibmkisti && $threads_per_task != 1 ]]
773	then
774	echo "export OMP_NUM_THREADS=$threads_per_task" >> $job_to_send
775	fi
776
777	elif [[ $(echo $remote_host \| cut -c1-3) = ibm && $numprocs = 0 ]]
778	then
779
780	cat > $job_to_send << %%END%%
781	#!/bin/ksh
782
783	# @ job_type = serial
784	# @ node_usage = $node_usage
785	# @ job_name = palm
786	# @ wall_clock_limit = ${timestring},$timestring
787	# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)
788	# @ output = $remote_dayfile
789	# @ error = $remote_dayfile
790	$class
791	$notification
792
793	# @ queue
794
795	%%END%%
796
797	elif [[ $remote_host = lcbullhh ]]
798	then
799	if [[ $numprocs != 0 ]]
800	then
801	cat > $job_to_send << %%END%%
802	#!/bin/bash -l
803	#SBATCH -J $job_name
804	#SBATCH -t $timestring
805	#SBATCH -N $nodes
806	#SBATCH --ntasks-per-node=$processes_per_node
807	#SBATCH -p $queue
808	#SBATCH -o $remote_dayfile
809	#SBATCH -e $remote_dayfile
810	#SBATCH -A $project_account
811
812	$init_cmds
813	$module_calls
814
815	%%END%%
816
817	else
818	cat > $job_to_send << %%END%%
819	#!/bin/bash -l
820	#SBATCH -J $job_name
821	#SBATCH -t $timestring
822	#SBATCH -l ncpus=1
823	#SBATCH -l pmem=${memory}mb
824	#SBATCH -m abe
825	#SBATCH -o $remote_dayfile
826	#SBATCH -e $remote_dayfile
827	#SBATCH -A $project_account
828
829	$init_cmds
830	$module_calls
831
832	%%END%%
833
834	fi
835
836	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
837	then
838
839	if [[ "$feature" != "" ]]
840	then
841	featuredir="#PBS -l feature=$feature"
842	fi
843
844	if [[ $numprocs != 0 ]]
845	then
846	cat > $job_to_send << %%END%%
847	#!/bin/bash
848	#PBS -N $job_name
849	#PBS -l walltime=$timestring
850	#PBS -l nodes=$nodes:ppn=$processes_per_node
851	#PBS -o $remote_dayfile
852	#PBS -j oe
853	#PBS -q $queue
854	$featuredir
855
856	$init_cmds
857	$module_calls
858
859	%%END%%
860
861	else
862
863	continue
864
865	fi
866
867	elif [[ $remote_host = lccrayf ]]
868	then
869
870	if [[ $numprocs != 0 ]]
871	then
872	cat > $job_to_send << %%END%%
873	#!/bin/bash -l
874	#SBATCH -J $job_name
875	#SBATCH -t $timestring
876	#SBATCH -N $nodes
877	#SBATCH --ntasks-per-node=$processes_per_node
878	#SBATCH -p $queue
879	#SBATCH -o $remote_dayfile
880	#SBATCH -e $remote_dayfile
881
882	$init_cmds
883	$module_calls
884
885	%%END%%
886
887	else
888	cat > $job_to_send << %%END%%
889	#!/bin/bash -l
890	#SBATCH -J $job_name
891	#SBATCH -t $timestring
892	#SBATCH -l ncpus=1
893	#SBATCH -l pmem=${memory}mb
894	#SBATCH -m abe
895	#SBATCH -o $remote_dayfile
896	#SBATCH -e $remote_dayfile
897
898	$init_cmds
899	$module_calls
900
901	%%END%%
902
903	fi
904
905	elif [[ $remote_host = lcflow ]]
906	then
907	if [[ $numprocs != 0 ]]
908	then
909	pe_set="#$ -pe impi $numprocs"
910	else
911	pe_set="#$ -pe impi 1"
912	fi
913	if [[ $queue = default ]]
914	then
915	queue_set=""
916	else
917	queue_set="#$ -q $queue"
918	fi
919	[[ "$disc_space" = "" ]] && disc_space=50
920
921	cat > $job_to_send << %%END%%
922	#!/bin/bash
923	#$ -S /bin/bash
924	#$ -N $job_name
925	#$ -cwd
926	#$ -l h_rt=$timestring
927	#$ -l h_vmem=${memory}M
928	#$ -o $remote_dayfile
929	#$ -j y
930	$pe_set
931	#$ -R y
932	#$ -l h_fsize=${disc_space}G
933	$queue_set
934
935	%%END%%
936
937	elif [[ $remote_host = lck \|\| $remote_host = lckordi \|\| $remote_host = lcsb ]]
938	then
939
940	if [[ $numprocs != 0 ]]
941	then
942	cat > $job_to_send << %%END%%
943	#!/bin/ksh
944	#PBS -N $job_name
945	#PBS -l walltime=$timestring
946	#PBS -l ncpus=$numprocs
947	#PBS -l pmem=${memory}mb
948	#PBS -o $remote_dayfile
949	#PBS -l nodes=$nodes:ppn=${processes_per_node}
950	#PBS -j oe
951
952	mpd &
953
954	%%END%%
955
956	else
957	cat > $job_to_send << %%END%%
958	#!/bin/ksh
959	#PBS -N $job_name
960	#PBS -l walltime=$timestring
961	#PBS -l ncpus=1
962	#PBS -l pmem=${memory}mb
963	#PBS -o $remote_dayfile
964	#PBS -j oe
965
966	%%END%%
967
968	fi
969
970	elif [[ $remote_host = lckiaps ]]
971	then
972
973	if [[ $numprocs != 0 ]]
974	then
975	cat > $job_to_send << %%END%%
976	#!/bin/ksh
977	#PBS -N $job_name
978	#PBS -l walltime=$timestring
979	#PBS -l select=1:ncpus=$numprocs
980	#PBS -l pmem=${memory}mb
981	#PBS -q $queue
982	#PBS -o $remote_dayfile
983	#PBS -j oe
984	#PBS -V
985
986	%%END%%
987
988	else
989	cat > $job_to_send << %%END%%
990	#!/bin/ksh
991	#PBS -N $job_name
992	#PBS -l walltime=$timestring
993	#PBS -l ncpus=1
994	#PBS -l pmem=${memory}mb
995	#PBS -o $remote_dayfile
996	#PBS -j oe
997
998	%%END%%
999
1000	fi
1001
1002	elif [[ $remote_host = lcyon ]]
1003	then
1004
1005	if [[ $numprocs != 0 ]]
1006	then
1007	cat > $job_to_send << %%END%%
1008	#!/bin/ksh
1009	#PBS -N $job_name
1010	#PBS -l walltime=$timestring
1011	#PBS -l ncpus=$numprocs
1012	#PBS -l pmem=${memory}mb
1013	#PBS -o $remote_dayfile
1014	#PBS -j oe
1015
1016	%%END%%
1017
1018	else
1019	cat > $job_to_send << %%END%%
1020	#!/bin/ksh
1021	#PBS -N $job_name
1022	#PBS -l walltime=$timestring
1023	#PBS -l ncpus=1
1024	#PBS -l pmem=${memory}mb
1025	#PBS -o $remote_dayfile
1026	#PBS -j oe
1027
1028	%%END%%
1029
1030	fi
1031
1032	elif [[ $remote_host = lcxe6 ]]
1033	then
1034
1035	if [[ $numprocs != 0 ]]
1036	then
1037	cat > $job_to_send << %%END%%
1038	#!/bin/ksh
1039	#PBS -S /bin/ksh
1040	#PBS -N $job_name
1041	#PBS -A $project_account
1042	#PBS -j oe
1043	#PBS -l walltime=$timestring
1044	#PBS -l mppwidth=${numprocs}
1045	#PBS -l mppnppn=${processes_per_node}
1046	#PBS -m abe
1047	#PBS -o $remote_dayfile
1048	$email_directive
1049
1050	$init_cmds
1051	$module_calls
1052
1053	%%END%%
1054
1055	else
1056	cat > $job_to_send << %%END%%
1057	#!/bin/ksh
1058	#PBS -S /bin/ksh
1059	#PBS -N $job_name
1060	#PBS -A $project_account
1061	#PBS -j oe
1062	#PBS -l walltime=$timestring
1063	#PBS -l ncpus=1
1064	#PBS -l pmem=${memory}mb
1065	#PBS -m abe
1066	$email_directive
1067	#PBS -o $remote_dayfile
1068
1069	$init_cmds
1070	$module_calls
1071
1072	%%END%%
1073
1074	fi
1075
1076	elif [[ $remote_host = lckyoto ]]
1077	then
1078
1079	cat > $job_to_send << %%END%%
1080	#!/bin/ksh
1081	# @\$-o $remote_dayfile
1082	# @\$-eo -oi
1083	# @\$-lP 16
1084	# @\$-lp 1
1085	# @\$-lm 28gb -llm unlimited -ls unlimited
1086	# @\$-q $queue
1087	# @\$-Pvn abs_pack
1088	##for intel? @\$-Pvn abs_unpack -Pvs unpack -Pvc unpack
1089	#. /thin/local/etc/setprofile/intel-11.0.sh
1090	#. /thin/local/etc/setprofile/mvapich2-1.4+intel-11.0.sh
1091	. ~/.myprofile
1092	#. /home2/t/t51254/palm/current_version/myprofile
1093	#. /thin/apps/pgi/mpi.sh
1094	#
1095	env
1096	#
1097	set -x
1098
1099	%%END%%
1100
1101	elif [[ $remote_host = lcxt5m ]]
1102	then
1103
1104	if [[ $numprocs != 0 ]]
1105	then
1106	cat > $job_to_send << %%END%%
1107	#!/bin/ksh
1108	#PBS -S /bin/ksh
1109	#PBS -N $job_name
1110	#PBS -j oe
1111	#PBS -l walltime=$timestring
1112	#PBS -l mppwidth=${numprocs}
1113	#PBS -l mppnppn=${processes_per_node}
1114	#PBS -m abe
1115	#PBS -o $remote_dayfile
1116
1117	$init_cmds
1118	$module_calls
1119
1120	%%END%%
1121
1122	else
1123	cat > $job_to_send << %%END%%
1124	#!/bin/ksh
1125	#PBS -S /bin/ksh
1126	#PBS -N $job_name
1127	#PBS -j oe
1128	#PBS -l walltime=$timestring
1129	#PBS -l ncpus=1
1130	#PBS -l pmem=${memory}mb
1131	#PBS -m abe
1132	#PBS -o $remote_dayfile
1133
1134	$init_cmds
1135	$module_calls
1136
1137	%%END%%
1138
1139	fi
1140
1141	elif [[ $remote_host = lckyuh ]]
1142	then
1143	cat > $job_to_send << %%END%%
1144	#!/bin/bash
1145	#PJM -L "rscgrp=$queue"
1146	#PJM -L "node=$nodes"
1147	#PJM --mpi "proc=$numprocs"
1148	#PJM -L "elapse=$timestring"
1149	#PJM -o $remote_dayfile
1150	#PJM -j
1151	#PJM -X
1152	#PJM --no-stging
1153
1154	export LANG=en_US.UTF-8
1155	%%END%%
1156
1157	elif [[ $remote_host = lckyut ]]
1158	then
1159	cat > $job_to_send << %%END%%
1160	#!/bin/bash
1161	#PJM -L "rscgrp=$queue"
1162	#PJM -L "vnode=$numprocs"
1163	#PJM -L "vnode-core=1"
1164	#PJM -L "elapse=$timestring"
1165	#PJM --mpi proc=$numprocs
1166	#PJM -o $remote_dayfile
1167	#PJM -j
1168	#PJM -X
1169	#PJM --no-stging
1170
1171	export LANG=en_US.UTF-8
1172	%%END%%
1173
1174	elif [[ $remote_host = lcocean ]]
1175	then
1176	cat > $job_to_send << %%END%%
1177	#!/bin/bash
1178	#$ -cwd
1179	#$ -V
1180	#$ -N $job_name
1181	#$ -pe orte $numprocs
1182	#$ -o $remote_dayfile
1183	#$ -j y
1184	#$ -R y
1185	$init_cmds
1186	$module_calls
1187
1188	%%END%%
1189
1190	elif [[ $remote_host = nech ]]
1191	then
1192
1193	if (( nodes > 1 ))
1194	then
1195	cat > $job_to_send << %%END%%
1196	#!/bin/ksh
1197	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1198	#PBS -l ${qsubmem}=${Memory}gb
1199	#PBS -b $nodes
1200	#PBS -o $remote_dayfile
1201	#PBS -N palm
1202	#PBS -j o
1203	#PBS -T mpisx
1204
1205	%%END%%
1206
1207	elif [[ $numprocs != 0 ]]
1208	then
1209	cat > $job_to_send << %%END%%
1210	#!/bin/ksh
1211	#PBS -l cpunum_prc=$processes_per_node,cputim_job=$cputime
1212	#PBS -l ${qsubmem}=${Memory}gb
1213	#PBS -o $remote_dayfile
1214	#PBS -N palm
1215	#PBS -j o
1216
1217	%%END%%
1218
1219	else
1220	cat > $job_to_send << %%END%%
1221	#!/bin/ksh
1222	#PBS -l ${qsubmem}=${Memory}gb,${qsubtime}=$cputime
1223	#PBS -o $remote_dayfile
1224	#PBS -j o
1225
1226	%%END%%
1227
1228	fi
1229
1230	elif [[ $remote_host = lctit ]]
1231	then
1232	cat > $job_to_send << %%END%%
1233	#!/bin/ksh
1234	$init_cmds
1235	$module_calls
1236
1237	%%END%%
1238
1239	# SET OPTIONS FOR SUBMIT-COMMAND
1240	if [[ $tasks_per_node != $processes_per_node ]]
1241	then
1242	submcom="$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 $queue "
1243	else
1244	submcom="$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 "
1245	fi
1246
1247	else
1248
1249	cat > $job_to_send << %%END%%
1250	# @\$-q ${queue}
1251	# @\$-l${qsubtime} $timestring
1252	# @\$-l${qsubmem} ${memory}mb
1253	# @\$-o $remote_dayfile
1254	# @\$-eo
1255
1256	%%END%%
1257
1258	fi
1259
1260
1261	# IN CASE OF JOBS EXECUTING ON REMOTE-HOSTS, THE TRANSFER OF THE DAYFILES
1262	# TO THE LOCAL HOSTS WILL BE INITIATED BY TRAP ON EXIT
1263	# NO TRANSFER POSSIBLE ON IBM IN SEOUL
1264	if [[ $delete_dayfile = false && $remote_host != $local_host ]]
1265	then
1266	echo "set +vx" >> $job_to_send
1267	echo "trap '" >> $job_to_send
1268	echo "set +vx" >> $job_to_send
1269	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 ]]
1270	then
1271	if [[ $remote_host = ibmh ]]
1272	then
1273	return_queue=c1
1274	elif [[ $remote_host = ibmkisti ]]
1275	then
1276	return_queue=class.1-2
1277	elif [[ $remote_host = ibmku ]]
1278	then
1279	return_queue=sdbg2
1280	elif [[ $remote_host = ibms ]]
1281	then
1282	return_queue=p_normal
1283	elif [[ $remote_host = lcbullhh ]]
1284	then
1285	return_queue=shared
1286	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1287	then
1288	return_queue=dataq
1289	elif [[ $remote_host = lcxe6 ]]
1290	then
1291	return_queue=debug
1292	elif [[ $remote_host = lckiaps ]]
1293	then
1294	return_queue=express
1295	elif [[ $remote_host = lckyuh ]]
1296	then
1297	return_queue=cx-single
1298	elif [[ $remote_host = lckyut ]]
1299	then
1300	return_queue=cx-single
1301	else
1302	return_queue=unknown
1303	fi
1304
1305	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1306	then
1307
1308	if [[ $remote_host = ibmku ]]
1309	then
1310	echo "echo \"#!/usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1311	echo "echo \"# @ shell = /usr/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1312	else
1313	echo "echo \"#!/bin/ksh\" >> scpjob.$identifier" >> $job_to_send
1314	fi
1315	echo "echo \"# @ job_type = serial\" >> scpjob.$identifier" >> $job_to_send
1316	echo "echo \"# @ job_name = transfer\" >> scpjob.$identifier" >> $job_to_send
1317	echo "echo \"# @ resources = ConsumableCpus(1) ConsumableMemory(1 gb)\" >> scpjob.$identifier" >> $job_to_send
1318	echo "echo \"# @ wall_clock_limit = 00:10:00,00:10:00\" >> scpjob.$identifier " >> $job_to_send
1319	echo "echo \"# @ output = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1320	echo "echo \"# @ error = job_queue/last_job_transfer_protocol\" >> scpjob.$identifier" >> $job_to_send
1321	if [[ $host != "ibmh" ]]
1322	then
1323	echo "echo \"# @ class = $return_queue\" >> scpjob.$identifier" >> $job_to_send
1324	fi
1325	echo "echo \"# @ image_size = 10\" >> scpjob.$identifier" >> $job_to_send
1326	echo "echo \"# @ notification = never\" >> scpjob.$identifier" >> $job_to_send
1327
1328	echo "echo \"# @ queue\" >> scpjob.$identifier" >> $job_to_send
1329	echo "echo \" \" >> scpjob.$identifier" >> $job_to_send
1330
1331	echo "echo \"set -x\" >> scpjob.$identifier" >> $job_to_send
1332	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
1333	if [[ $remote_host = ibmku ]]
1334	then
1335	echo "echo \"rm scpjob.$identifier\" >> scpjob.$identifier" >> $job_to_send
1336	fi
1337	echo "echo \"exit\" >> scpjob.$identifier" >> $job_to_send
1338
1339	elif [[ $remote_host = nech ]]
1340	then
1341	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1342	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1343	echo "#PBS -l ${qsubmem}=1GB,${qsubtime}=100" >> $job_to_send
1344	echo "#PBS -o last_job_transfer_protocol" >> $job_to_send
1345	echo "#PBS -j o" >> $job_to_send
1346	echo " " >> $job_to_send
1347	echo "set -x" >> $job_to_send
1348	echo "cd /pf/b/${remote_user}/job_queue" >> $job_to_send
1349	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1350	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1351	echo "%%END%%" >> $job_to_send
1352
1353	elif [[ $remote_host = lcbullhh ]]
1354	then
1355	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1356	echo "#!/bin/bash" >> $job_to_send
1357	echo "#SBATCH --job-name=job_protocol_transfer" >> $job_to_send
1358	echo "#SBATCH -t 00:20:00" >> $job_to_send
1359	echo "#SBATCH -N 1" >> $job_to_send
1360	echo "#SBATCH -n 1" >> $job_to_send
1361	echo "#SBATCH -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1362	echo "#SBATCH -o $remote_dayfile" >> $job_to_send
1363	echo "#SBATCH -e $remote_dayfile" >> $job_to_send
1364	echo "#SBATCH -A $project_account" >> $job_to_send
1365	echo "#SBATCH -p $return_queue" >> $job_to_send
1366	echo " " >> $job_to_send
1367	echo "set -x" >> $job_to_send
1368	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1369	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1370	echo "%%END%%" >> $job_to_send
1371
1372	elif [[ $remote_host = lckyuh ]]
1373	then
1374	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1375	echo "#!/bin/bash" >> $job_to_send
1376	echo "#PJM -L \"node=1\"" >> $job_to_send
1377	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1378	echo "#PJM --no-stging" >> $job_to_send
1379	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1380	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1381	echo "#PJM -j" >> $job_to_send
1382	echo " " >> $job_to_send
1383	echo "export LANG=en_US.UTF-8" >> $job_to_send
1384	echo "set -x" >> $job_to_send
1385	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1386	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1387	echo "%%END%%" >> $job_to_send
1388
1389	elif [[ $remote_host = lckyut ]]
1390	then
1391	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1392	echo "#!/bin/bash" >> $job_to_send
1393	echo "#PJM -L \"vnode=1\"" >> $job_to_send
1394	echo "#PJM -L \"rscgrp=$return_queue\"" >> $job_to_send
1395	echo "#PJM --no-stging" >> $job_to_send
1396	echo "#PJM -L \"elapse=30:00\"" >> $job_to_send
1397	echo "#PJM -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1398	echo "#PJM -j" >> $job_to_send
1399	echo " " >> $job_to_send
1400	echo "export LANG=en_US.UTF-8" >> $job_to_send
1401	echo "set -x" >> $job_to_send
1402	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address $remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1403	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1404	echo "%%END%%" >> $job_to_send
1405
1406	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1407	then
1408	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1409	echo "#!/bin/bash" >> $job_to_send
1410	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1411	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1412	echo "#PBS -l nodes=1:ppn=1" >> $job_to_send
1413	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1414	echo "#PBS -j oe" >> $job_to_send
1415	echo " " >> $job_to_send
1416	echo "set -x" >> $job_to_send
1417	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1418	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1419	echo "%%END%%" >> $job_to_send
1420
1421	elif [[ $remote_host = lcocean ]]
1422	then
1423	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1424	echo "#!/bin/bash" >> $job_to_send
1425	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1426	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1427	echo "SGEPREFIX -cwd" >> $job_to_send
1428	echo "SGEPREFIX -j y" >> $job_to_send
1429	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1430	echo " " >> $job_to_send
1431	echo "set -x" >> $job_to_send
1432	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1433	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1434	echo "" >> $job_to_send
1435	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1436	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1437	echo "rm -f scpjob.${identifier}" >> $job_to_send
1438	echo "%%END%%" >> $job_to_send
1439	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1440	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1441
1442	elif [[ $remote_host = lcflow ]]
1443	then
1444	echo "cat > scpjob.${identifier}.tmp << %%END%%" >> $job_to_send
1445	echo "#!/bin/bash" >> $job_to_send
1446	echo "SGEPREFIX -S /bin/bash" >> $job_to_send
1447	echo "SGEPREFIX -N transfer_$job_name" >> $job_to_send
1448	echo "SGEPREFIX -cwd" >> $job_to_send
1449	echo "SGEPREFIX -l h_rt=01:00:00" >> $job_to_send
1450	echo "SGEPREFIX -l h_vmem=500M" >> $job_to_send
1451	echo "SGEPREFIX -j y" >> $job_to_send
1452	echo "SGEPREFIX -o ${local_host}_${job_name}_scpjob_$identifier" >> $job_to_send
1453	echo " " >> $job_to_send
1454	echo "set -x" >> $job_to_send
1455	echo "export PALM_BIN=$PALM_BIN" \| sed -e 's:'$HOME':$HOME:' >> $job_to_send
1456	echo "export PATH=\$PATH:\$PALM_BIN" >> $job_to_send
1457	echo "" >> $job_to_send
1458	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1459	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1460	echo "rm -f scpjob.${identifier}" >> $job_to_send
1461	echo "%%END%%" >> $job_to_send
1462	echo "sed -e 's/SGEPREFIX/#$/g' scpjob.${identifier}.tmp > scpjob.${identifier}" >> $job_to_send
1463	echo "rm -f scpjob.${identifier}.tmp" >> $job_to_send
1464	elif [[ $remote_host = lcxe6 ]]
1465	then
1466	echo "cat > scpjob.${identifier} << %%END%%" >> $job_to_send
1467	echo "#!/bin/ksh" >> $job_to_send
1468	echo "#PBS -N job_protocol_transfer" >> $job_to_send
1469	echo "#PBS -l walltime=00:30:00" >> $job_to_send
1470	echo "#PBS -A $project_account" >> $job_to_send
1471	echo "#PBS -l mppwidth=1" >> $job_to_send
1472	echo "#PBS -l mppnppn=1" >> $job_to_send
1473	echo "#PBS -o \$HOME/job_queue/last_job_transfer_protocol" >> $job_to_send
1474	echo "#PBS -j oe" >> $job_to_send
1475	echo " " >> $job_to_send
1476	echo "set -x" >> $job_to_send
1477	echo "batch_scp $PORTOPT -d -w 10 -u $local_user $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1478	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1479	echo "%%END%%" >> $job_to_send
1480	else
1481
1482	echo "cat > scpjob.$identifier << %%END%%" >> $job_to_send
1483	echo "# @\\\$-q $return_queue" >> $job_to_send
1484	echo "# @\\\$-l${qsubtime} 10" >> $job_to_send
1485	echo "# @\\\$-l${qsubmem} 10mb" >> $job_to_send
1486	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 \|\| $remote_host = t3es ]]
1487	then
1488	echo "# @\$-l mpp_p=0" >> $job_to_send
1489	fi
1490	echo '# @\$-lF 10mb' >> $job_to_send
1491	echo '# @\$-o job_queue/last_job_transfer_protocol' >> $job_to_send
1492	echo '# @\\\$-eo' >> $job_to_send
1493	echo " " >> $job_to_send
1494	if [[ $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1495	then
1496	echo "set +vx" >> $job_to_send
1497	echo ". .profile" >> $job_to_send
1498	fi
1499	echo "set -x" >> $job_to_send
1500	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
1501	echo "[[ \"\$for_subjob_to_do\" != \"\" ]] && eval \$for_subjob_to_do" >> $job_to_send
1502	echo "%%END%%" >> $job_to_send
1503
1504	fi
1505
1506	if [[ $(echo $remote_host \| cut -c1-3) = ibm ]]
1507	then
1508	echo "llsubmit scpjob.$identifier" >> $job_to_send
1509	elif [[ $remote_host = lcbullhh ]]
1510	then
1511	echo "sbatch scpjob.$identifier" >> $job_to_send
1512	elif [[ $remote_host = lccrayb \|\| $remote_host = lccrayh ]]
1513	then
1514	echo "msub -q $return_queue scpjob.$identifier" >> $job_to_send
1515	elif [[ $remote_host = t3eb \|\| $remote_host = t3eh \|\| $remote_host = t3ej2 \|\| $remote_host = t3ej5 ]]
1516	then
1517	echo "qsub -J n scpjob.$identifier" >> $job_to_send
1518	elif [[ $remote_host = t3es ]]
1519	then
1520	echo "qsub -J n -s /bin/ksh scpjob.$identifier" >> $job_to_send
1521	elif [[ $remote_host = lckiaps ]]
1522	then
1523	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1524	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"$submcom ${job_catalog}/scpjob.$identifier\" " >> $job_to_send
1525	echo "rm ${job_catalog}/scpjob.$identifier" >> $job_to_send
1526	elif [[ $remote_host = lckyu* ]]
1527	then
1528	echo "scp $PORTOPT scpjob.$identifier ${remote_username}@${remote_address}:job_queue" >> $job_to_send
1529	echo "ssh $SSH_PORTOPT ${remote_username}@${remote_address} \"cd job_queue; $submcom scpjob.$identifier; rm scpjob.$identifier\" " >> $job_to_send
1530	elif [[ $remote_host = lcflow \|\| $remote_host = lcocean ]]
1531	then
1532	echo "mv scpjob.$identifier $job_catalog" >> $job_to_send
1533	echo "/usr/bin/ssh ${remote_username}@${remote_address} \"$init_cmds $module_calls cd $job_catalog; $submcom scpjob.$identifier\" " >> $job_to_send
1534	else
1535	echo "$submcom scpjob.$identifier" >> $job_to_send
1536	fi
1537	if [[ $remote_host != ibmku && $remote_host != lckiaps ]]
1538	then
1539	echo "rm scpjob.$identifier" >> $job_to_send
1540	fi
1541	if [[ $remote_host = nech ]]
1542	then
1543	echo "cd -" >> $job_to_send
1544	fi
1545	else
1546	# echo "ftpcopy -d $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile" >> $job_to_send
1547	# ??? funktioniert das ÃŒberhaupt noch ???
1548	echo "nohup ftpcopy -d -w 15 $local_address ${job_catalog}/$remote_dayfile \"$job_catalog\" $local_dayfile > /dev/null &" >> $job_to_send
1549	fi
1550	echo "set -x" >> $job_to_send
1551	echo " ' exit" >> $job_to_send
1552	echo "set -x" >> $job_to_send
1553	fi
1554
1555
1556	# APPEND THE JOB-FILE (CREATE BY mrun) TO THE JOB-DIRECTIVES GENERATED ABOVE
1557	cat $file_to_send >> $job_to_send
1558
1559	if [[ $remote_host = ibm ]]
1560	then
1561	echo " " >> $job_to_send
1562	echo "exit" >> $job_to_send
1563	fi
1564
1565	# REMOVE JOB-FILE
1566	if [[ $remote_host = lctit \|\| $remote_host = ibmku \|\| $remote_host = lcflow ]]
1567	then
1568	echo " " >> $job_to_send
1569	echo "rm ~/job_queue/$job_on_remhost" >> $job_to_send
1570	fi
1571
1572
1573	# TRANSFER JOB TO THE TARGET HOST (JOB-DIRECTORY)
1574	if [[ $no_submit = false ]]
1575	then
1576	if [[ $remote_host != $local_host ]]
1577	then
1578	[[ $verify = true ]] && printf "\n >>> transfering job to \"$remote_host\"..."
1579	if [[ $remote_host = ibms ]] # ssh on ibms cannot handle "~/"
1580	then
1581	job_catalog_save=$job_catalog
1582	job_catalog=job_queue
1583	elif [[ $remote_host = nech ]]
1584	then
1585	job_catalog_save=$job_catalog
1586	job_catalog=/hpf/b/${remote_user}/job_queue
1587	fi
1588	if [[ $remote_host = nech ]]
1589	then
1590	# FILES CAN ONLY BE TRANSFERED VIA DKRZ'S ARCHIVE-SERVER
1591	scp $PORTOPT $job_to_send ${remote_user}@136.172.44.205:${job_catalog}/$job_on_remhost
1592	else
1593	scp $PORTOPT $job_to_send ${remote_user}@${remote_address}:${job_catalog}/$job_on_remhost
1594	fi
1595	if [[ $? = 1 ]]
1596	then
1597	locat=scp; exit
1598	fi
1599	if [[ $remote_host = ibms ]]
1600	then
1601	job_catalog=$job_catalog_save
1602	fi
1603	[[ $verify = true ]] && printf "\n >>> finished\n"
1604	else
1605	eval job_catalog=$job_catalog
1606	cp $job_to_send ${job_catalog}/$job_on_remhost
1607	fi
1608
1609
1610
1611	# START NQS- / LOADLEVELER-JOB
1612	if [[ $remote_host != $local_host ]]
1613	then
1614	[[ $verify = true ]] && printf "\n >>> submitting job using \"qsub\"...\n"
1615
1616	if [[ $remote_host = ibmku ]]
1617	then
1618	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost"
1619	elif [[ $remote_host = lcflow ]]
1620	then
1621	/usr/bin/ssh $SSH_PORTOPT $remote_address -l $remote_user "$init_cmds $module_calls cd $job_catalog; $submcom $job_on_remhost"
1622	else
1623	ssh $SSH_PORTOPT $remote_address -l $remote_user "cd $job_catalog; $submcom $job_on_remhost; rm $job_on_remhost"
1624	fi
1625
1626	[[ $verify = true ]] && printf " >>> o.k.\n"
1627	else
1628	cd $job_catalog
1629	if [[ $(echo $local_host \| cut -c1-3) = ibm \|\| $(echo $local_host \| cut -c1-6) = lccray ]]
1630	then
1631	eval $submcom $job_on_remhost
1632	elif [[ $local_host = lctit \|\| $localhost = lcxe6 \|\| $localhost = lck \|\| $localhost = lckordi \|\| $localhost = lcyon \|\| $localhost = lcsb \|\| $localhost = lckyu* ]]
1633	then
1634	chmod u+x $job_on_remhost
1635	eval $submcom $job_on_remhost
1636	elif [[ $local_host = nech ]]
1637	then
1638	if [[ $queue = default ]]
1639	then
1640	eval $submcom $job_on_remhost
1641	else
1642	eval $submcom -q $queue $job_on_remhost
1643	fi
1644	elif [[ $local_host = lcbullhh ]]
1645	then
1646	if [[ $queue = default ]]
1647	then
1648	eval $submcom $job_on_remhost
1649	fi
1650	else
1651	qsub $job_on_remhost
1652	fi
1653
1654	# JOBFILE MUST NOT BE DELETED ON lctit/ibmku/lcflow. THIS WILL BE DONE
1655	# AT THE END OF THE JOB
1656	if [[ $local_host != lctit && $local_host != ibmku && $local_host != lcflow ]]
1657	then
1658	rm $job_on_remhost
1659	fi
1660	cd - > /dev/null
1661	fi
1662	fi
1663
1664	# FINAL ACTIONS
1665	if [[ $no_submit = false ]]
1666	then
1667	rm -f $job_to_send
1668	fi
1669	[[ $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 |