scyther/gui/notes-brutus-mpa.txt

run test-MPA with --pickle to some file FILE.

Choose a STEP integer: how many verifications are batched into a single job.

Then:

./make-bsub.py FILE STEP -W 1:00 [OTHER BSUB OPTIONS] > tests.sh

Then

bash tests.sh


When all is done, rerun the original thing without pickle.



This invokes then:
json-scyther.py
in different batches





Test run for real


Fri Dec 31 16:33:20 CET 2010

Login & screen on brutus3 node.

bsub -W 2:00 ./test-mpa.py --pickle mpa-tests.json -A Protocols/MultiProtocolAttacks/*.spdl

Fri Dec 31 18:48:29 CET 2010

Given the 6 minutes timeout, decided to batch into the 1h queues. Thus 9
verifications can safely go in a batch.

./make-bsub.py mpa-tests.json 9 -W 1:00 >mpa-tests.sh

bash mpa-tests.sh


Hmm. For the 1h queue on Brutus, there is a 10.000 pending jobs limit. Thus my
40.000+ jobs get stuck here.

So I could have done the division such that the jobs can be pended at onces
but it would have meant putting the jobs in the 8h or more queues.

For the batching thing, it would be nice to print a counter every 10 bsubs so
if it gets stuck, you can see where it is (or better: how much is left).

The lsf.o* output files clog up the directory. Find a way to disable them!

Woops, we get mail once in a while. Not good. Unclear under which conditions
this occurs, it seems to be errors only. (Probably stale file pointers from
the old watch & rm solution.)

Sun Jan  2 10:54:23 CET 2011

All jobs have been submitted, now only 3000 pending.

There may be a limit for me of about 128 active jobs at the same time.

Sun Jan  2 11:30:30 CET 2011

2200 pending.

Sun Jan  2 12:38:48 CET 2011

1155 pending.
(bjobs -p | grep PEND | wc -l)

Sun Jan  2 13:59:04 CET 2011

0 jobs pending, 32 jobs active.

Sun Jan  2 14:18:11 CET 2011

Done. Recomp started (without --pickle FILE above)
Takes too long on login node. Killed at 14:40.

Instead, rerunning with:

bsub -I -N ./test-mpa.py  -A Protocols/MultiProtocolAttacks/*.spdl

-I for interactive, -N for mail at end.

Sun Jan  2 14:45:04 CET 2011

Above job is running. It also seems faster.

Sun Jan  2 20:07:58 CET 2011

Sigh. It got killed after one hour because no time limit was set.
Rerunning with -W 6:00






Sun Jan  2 14:30:19 CET 2011

In parallel, starting new huge job; biggest possible using current script options.

bsub -W 7:00 ./test-mpa.py --pickle test-full-mpa.json --self-communication -A Protocols/MultiProtocolAttacks/*.spdl 

Actually, these big jobs should be started with finishing e-mail notification
or the switch that makes the bsub command only return after the jobs has
finished, otherwise we end up watching bjobs all the time, which is boring.

Sun Jan  2 14:40:08 CET 2011

The above test generation is now running.

Sun Jan  2 20:09:42 CET 2011

The test generation seems to have finished at 15:31.

./make-bsub.py test-full-mpa.json 10 -W 1:00 >test-full-mpa.sh

This finished at 20:11.
So now running

nice bash test-full-mpa.sh

G



Sun Jan  2 15:07:13 CET 2011

A third parallel test:

batcher.sh OPTIONS_AND_FILES_FOR_TEST_MPA_SCRIPT

Running with -L5. This should automate all of the previous stuff.




Wed Jan  5 15:37:11 CET 2011

Running for cryptrec (with new Scyther version and new batches of 5 things)
./batcher.sh ~/papers/iso/*.spdl



Tue Jan 18 17:10:49 CET 2011

./batcher.sh -m 1 --all-types --self-communication ~/papers/iso/*.spdl
The batcher has jobid 930582

(error, reverting to os.makedirs(path))

Tue Jan 18 23:45:15 CET 2011

./test-iso-combo.sh

Tue Jan 18 23:49:15 CET 2011

./batcher.sh -m 2 --all-types --self-communication ~/papers/iso/*.spdl

Solved: do "watch -n 10 ./WIPER.sh 11"
(wiper.sh finds lsf files accessed longer ago than 11 minutes and wipes them)

./test-mpa-alltypes.sh 



Mon Jan 24 14:55:23 CET 2011

./batcher.sh -m 2 --all-types Protocols/MultiProtocolAttacks/*.spdl



Sat Jan 29 13:35:22 CET 2011

./batcher.sh -m 2 -A --self-communication Protocols/MultiProtocolAttacks/*.spdl