Posts by TJM

21) Message boards : News : Enigma Cuda 1.07 (Message 6353)
Posted 4 Sep 2017 by Profile TJM
Post:
I think that both the 540M and 460M are too slow to run the GPU app, which is optimized for faster cards.
22) Message boards : Number crunching : A GPU version of Enigma software (Message 6284)
Posted 16 Aug 2017 by Profile TJM
Post:
I had to reset GPU performance stats listed here GPU speed because the script did not sum points correctly when the app runtime was very short and the app used less than 100s of CPU time. Mostly AMD cards running OpenCL apps were affected and I finally found the bug when I noticed RX Vega at the bottom with speed index of 0.
23) Message boards : Number crunching : KLDIO broken (Message 6118)
Posted 4 Aug 2017 by Profile TJM
Post:
This is just the way the server handles workunits.

When running a single work set which can have one or multiple batches, if the server detects something that needs review (typically happens everytime the top score changes), it will stop work unit generator completely and there will be work outage.

Now if there are more work sets (which basically means more than one message loaded and enabled), the server will fall back to the one with lower priority. If two batches have the same priority, the server alternates between them sharing the resources roughly 50/50.
24) Message boards : Number crunching : KLDIO broken (Message 6093)
Posted 3 Aug 2017 by Profile TJM
Post:
nice that we broke the message.
i thinke, it would be nice, if we can read in the forum, which message we crunch with which application. so we all know, what we are doing here. i believe, there is a lot of confusion...


There are two workunit sets right now, ALQFI at low priority and WEUWY.1 (the one that's still listed as unbroken) at normal priority, both from Cryptocellar.
WEUWY at 48 letters might be very hard to break using hillclimb but the overall speed we get from running on multiple GPUs (doing more than 1 full walk per minute) certainly boost our chances a bit.
25) Message boards : Number crunching : KLDIO broken (Message 6091)
Posted 3 Aug 2017 by Profile TJM
Post:
I've got private message from Dan Girard, who was unable to post on the forums due to the spam filter, with a possible explanation of the mysterious LSI:


I think the "Lsi" refers to the village "Bol'shiye L'zi" (Greater L'zi) and a smaller village "Malyye L'zi" (Small L'zi) adjacent to it on its west side. At the bottom of cryptocellar.org's page on Frode Weierud and Olaf Ostwald's paper "Modern Breaking of Enigma Ciphertexts," there is a link to the University of Texas Libraries' Eastern European map collection, where I found this map:

http://www.lib.utexas.edu/maps/ams/eastern_europe/txu-oclc-6519747-no-35-6.jpg

In the middle of the lower right quadrant of the map is an intersection of two roads marked in red, with one coming in from the east to meet the other running from southwest to northeast. By clicking at the point of intersection to zoom in, you'll find the village of Nikolajewo (spelled "Nikolayevo" on the map) at the intersection of those roads, with Bol'shiye L'zi (spelled "Bol'shiyye L'zi" on the map) about 10 km to the northeast. Malyye L'zi is not marked, however, presumably because it was too small.

I see that a few of the commenters seem to think that KLDIO is a naval message; but, like the other messages from 1941 on Frode's site, it is an Army message, from the early stages of Operation Barbarossa, the German invasion of the Soviet Union.


I've just edited Dan's account and now he should be able to join us on the forums.
26) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 6078)
Posted 3 Aug 2017 by Profile TJM
Post:
The plan for this weekend is that 1.03 OpenCL Windows app will be moved from Enigma Test and merged with CUDA app as Enigma GPU.
CUDA & OpenCL Linux v1.00 apps will be added to Enigma Test, just to verify that they're working.
27) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 6055)
Posted 3 Aug 2017 by Profile TJM
Post:
I received my first task, at last !! Great but...
Impossible to complete, my low end AMD AM1 (154GFLOPS) would spend 10-12 hours to complete but if I switch off or suspend computer, the task is restarting from 0% , no memory in disk about % completed every moment. Sad


Actually, the task does not start from 0%.
The app writes it's own checkpoints but the wrapper and BOINC manager do not know about this, so they report a fresh start. Eventually the task will finish, as long as the app has enough time to checkpoint between restarts.
There are plans to update the wrapper to read the current progress, or at some point the BOINC API will be built in into the app.
28) Message boards : Number crunching : KLDIO broken (Message 6054)
Posted 3 Aug 2017 by Profile TJM
Post:
Here is the translation from Frode Weierud:

Quartermaster unit at 20:00 [hours] (LSI) 10 km northeast [of] Nikolajewo.

“Quelle-Abteilung” also often seen as “Qu Abt” or “Quelle Abt” is an abbreviation for Quartiermeisterabteilung which means Quartermaster unit. The meaning of LSI I still do not know.



Is this the "fifth and final message" we had been working on before we switched to AV and GPU?

Nope this one was taken from cryptocellar.org, and there are some more unbroken messages left.
29) Message boards : Number crunching : KLDIO broken (Message 6041)
Posted 2 Aug 2017 by Profile TJM
Post:
I have sent an email to Frode Weierud and in return I got more information about the message, with some corrections:

With the given indicator VOF ZVZ it turns out that the complete Ringstellung must be MJW, which gives as start PCL.
It is a left-hand wheel turnover at the 33rd letter, which might explain why we have failed to break this previously. If the rings are not correct it will mean that the message is broken into to parts one with 32 letters and the other 58 letters long. If treated as one message one or the other part will then ruin the statistics of the other part and the hill climbing will fail. Alex’s code which takes into account also the turnover of the left-hand wheel will therefore succeed where we failed.

The plaintext has some possible errors. I have corrected one:
quelle striq abteilung ab zwo null null null x
lsi x lsi x flam eins null km nordostl x
nikolajewo x nikolajewo

The original “nordostw x” should be “nordostl x”, nordostlich. I am not sure what LSI means and nor do I see what “flam” means. However, if I look at the original message form then I see that the 10th cipher group, VAWKQ, in reality should be VOWKQ. The letter “o” had been corrected and looked like an “a” and this is how it was transcribed. Changing this cipher group to VOWKQ turns “flam” into “klam” which is an abbreviation for Klammer, in English bracket or parenthesis. Hence the emended plaintext then becomes:

quelle striq abteilung ab zwo null null null x
lsi x lsi x klam eins null km nordostl x
nikolajewo x nikolajewo

Usually one would expect two brackets but perhaps this messages was written and enciphered in a hurry. If I would guess I think the other bracket might have been intended before the mysterious “lsi x lsi x”.

Best wishes,
Frode
30) Message boards : Number crunching : KLDIO broken (Message 6040)
Posted 2 Aug 2017 by Profile TJM
Post:
Sure, new batch will be online in just few hours for GPU apps and day or two later for CPU - I have to upgrade the CPU work generator because after all the server patches it is now outdated and does not work with current database schema.
31) Message boards : Number crunching : KLDIO broken (Message 6037)
Posted 2 Aug 2017 by Profile TJM
Post:
I think it would be even faster with one of the other dictionary sets, but I've started this group of batches randomly with the '1941'.
It might be better to just add all the batches at once and let the server iterate through them, instead of adding one by one. Perhaps I'll try this method with another message.
32) Message boards : News : KLDIO broken (Message 6035)
Posted 2 Aug 2017 by Profile TJM
Post:
(this is a quick repost from Number Crunching)

A few hours ago Dave GPU returned the plaintext of KLDIO 90.
More info was posted in this forum thread

Please abort all the g3_kldio* workunits that haven't started yet, as they're not needed anymore.
33) Message boards : Number crunching : KLDIO broken (Message 6034)
Posted 2 Aug 2017 by Profile TJM
Post:
About an hour ago Dave GPU (uid 68703) returned a plaintext of KLDIO 90:
quellestriqabteilungabzwonullnullnullxlsixlsixflameinsnullkmnordostwxnikolajewoxnikolajewo

with the following machine settings:
(H) B:JW:DCL:521:AWCSDREYFOKULZNVPXQT

this was done on walk #2120 for the wheels order 521 and UKW B (2190 restarts/walks on average for the entire batch), running Alex's GPU app. All this work was done in less than two days.

The plaintext score is nowhere near the top scores returned for this batch.
It hardly even makes it to top 50%: -> KLDIO results page

but it was detected by the server, which cross-checks the returned plaintexts with all available dictionary sets and marks all results that have high score in more than one dictionary. It was marked as possible decrypt because it has a decent score when checked against "U534" and "AVv1" (dictionaries provided by Alex) tridicts -> KLDIO cross check

Feel free to abort all the remaining KLDIO workunits as they're not needed anymore.
I have already aborted all that were in "unsent" state, the rest I left untouched, so those who won't read this in time or have the workunits already processed, will still get credit for their CPU/GPU time.
34) Message boards : Number crunching : ALQFI batch (Message 6019)
Posted 30 Jul 2017 by Profile TJM
Post:
For last few weeks the GPU apps were trying to break ALQFI taken from here: http://cryptocellar.org/bgac/GArmy_messages.html

The GPU app is very fast and at the current rate, on average it takes just under a minute to do a full walk over all possible wheel settings of an 'army' enigma (UKW B, wheels 1-5)

So far we did:

- nearly 8000 walks running mode 2 (-a), with scores "023" (icscore - bigram - trigram) using "1941" dictionary set (batch g3_alqfi87_9)
- nearly 6000 walks running mode 2 (-a), with scores "023" (icscore - bigram - trigram) using dictionaries provided by Alex (the author), codenamed AVv1 (batch g3_alqfi87_8)
- nearly 6000 walks running mode 2 (-a), scores "013" (icscore - unigram - trigram) using dictionaries provided by Alex (batch g3_alqfi87_7)
- nearly 10000 walks mode 2 (-a), scores "023" using 86v1 dict set (this is a newer revision of 1941 dictionary) & nearly 7000 walks with the same set running scores "013".

Combined thats nearly 40,000 walks over all the possible wheels settings, and the message is still unbroken (unless the solution is already found but for some reason has a score lower than top scores, but that's unlikely). I'm going to write a script that will dump all the unique results for all batches for review, there used to be a live view but that's unusable now due to database size.

At 87 letters length the message should be breakable quite easily if there is nothing wrong with it (well obviously it's not an easy one because it was listed for quite a long time and remains unbroken).
The next step I'd like to try before giving up on ALQFI is to trim the first or last {n} letters. At around 70 letters it should still be breakable with some luck and lots of restarts, and shortening may help if there is any kind of damage near the beginning or end of the message.

Or, if anyone has better idea, please let me know.
35) Message boards : Number crunching : No work for new AMD work (Message 6000)
Posted 27 Jul 2017 by Profile TJM
Post:
Most of my hosts periodically run Enigma AV - NONE has EVER gotten an Enigma CUDA task. I do have "yes" checked for


You have to enable beta work in your account preferences. Both GPU apps are marked as beta for now (I've checked your preferences and there is an entry with allow_beta_work set to 0).
36) Message boards : Number crunching : Top results history page (Message 5975)
Posted 21 Jul 2017 by Profile TJM
Post:
This is a well known text used as a test workunit to check basic functionality of a new app.

Any workunit starting with m3_test, m4_test, g3_test or g4_test[1] most likely runs on a message for which settings are known already. It is the only way to check if the application can actually find the solution, which is very important because we could run broken app for years without knowing that something is wrong. As a side effect test runs that do not fail are used to gather statistical information.
[1] the first character in WU name is always 'm' for batches of CPU workunits and 'g' for anything assigned to GPU. This is used internally by the server so it does know where to look for workunit description. The second character is enigma version, 4 for M4 and 3 currently used for M3 and Heer (which runs as M3 with settings limited to Heer compatible).
37) Message boards : Number crunching : A GPU version of Enigma software (Message 5970)
Posted 20 Jul 2017 by Profile TJM
Post:
Please wait a minute, that fix is not final.

I don't think there is a need to re-check anything because each candidate key is tried many times, by doing multiple restarts.


Let me know when the app is ready to update, I've prepared release 1.10 that's just waiting for new executables.
38) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 5968)
Posted 20 Jul 2017 by Profile TJM
Post:
There was a problem with translating --device to -d and parsing it inside the app.
I think it is fixed in 1.02 release because I've just seen a first completed and validated task.
39) Message boards : News : New app: "Enigma Test", OpenCL and Linux tests (Message 5965)
Posted 20 Jul 2017 by Profile TJM
Post:
I had to change app weights in the server configuration. With the weight set too low the feeder didn't even touch ATI tasks for some reason. Now it started sending them out as I see numbers raising in the server_status.

For the next few days there will be limited number of opencl test tasks as I'd like to check the app for any possible errors.
40) Message boards : Number crunching : A GPU version of Enigma software (Message 5959)
Posted 20 Jul 2017 by Profile TJM
Post:
It's fixed already, I did a stupid mistake patching the work generator to support test app and it placed all the workunits in the wrong queue.


Previous 20 · Next 20



Copyright © 2024 TJM