[A51] test Kraken

Jan Hrach jenda at yakumo.hrach.eu
Tue Feb 23 12:52:47 CET 2016


I have measured it with my client (napalmex from gsmtk) and it was really only 10 kfrag/s.

Regenerated kernel to use only 4x32bit vectors (genkernel32).

Now it computes 2 bursts per second with 60s lag (could be probably fine-tuned), which is ~expected (we have 3 per second on 3x7970). (and now the connection died)

On 23.2.2016 12:16, Milinko Isakovic wrote:
> Jan thank you for assistance, in setup and,
> 
> 
> Currently we are facing issue with deka setup. so far we ware able to crack keys with HD7900 series GPU. but issue is time required to crack one burst or 100 bursts is approx 3 min
> how can we reduce this time.
> 
> just for example if 1 gpu is used per burst then cant we use 4 GPU per burst  to decrease lookup time. also currently I have 2 GPU with me but time taken to crack is same.
> here is my procedure
> start paplon.py
> start oclvankus.py
> choose one gpu per instance and run 2 instance.
> start one instance of delta_client.py
> start cracking
> 
> now here are some example result of my crack
> 
> crack #0 took 110463 msec
> crack #5 took 111005 msec
> crack #2 took 113481 msec
> crack #6 took 114020 msec
> crack #18 took 115290 msec
> crack #7 took 116799 msec
> crack #9 took 117100 msec
> crack #3 took 118306 msec
> crack #4 took 118606 msec
> crack #15 took 118906 msec
> crack #13 took 119809 msec
> crack #8 took 120348 msec
> crack #16 took 120649 msec
> crack #10 took 123359 msec
> crack #19 took 124863 msec
> crack #11 took 126129 msec
> crack #14 took 127872 msec
> crack #17 took 130940 msec
> crack #12 took 137195 msec
> 
> if you can see the time is increasing as it moves with next burst.
> 
> 
> i also checked in  bursts.samplesession.tar.gz it start with 45000ms and ends with 65000 ms
> 
> now i will result of one gpu instance
> 
> crack #5 took 135398 msec
> crack #0 took 135700 msec
> crack #2 took 138767 msec
> crack #15 took 140602 msec
> crack #9 took 140905 msec
> crack #6 took 141208 msec
> crack #3 took 144273 msec
> crack #4 took 144574 msec
> crack #7 took 144874 msec
> crack #16 took 146093 msec
> crack #13 took 146395 msec
> crack #8 took 147616 msec
> crack #21 took 150978 msec
> crack #14 took 152812 msec
> crack #10 took 153114 msec
> crack #19 took 153413 msec
> 
> there is very less difference in it
> 
> 
> it will be good if one please share actual steps to take while  using multi gpu.
> 
> In case of  adding for example 14 GPU, and speed will have point only in rising burst in parallel: for example from 200 to 1000, or we cracking speed for 200 burst in parallel will be decreasing, and how much it will decrease?
> 
> Also I am wondering, with CPU rising what will happed?  In case I use multiprocessor MB (4cpu or 8cpu), will speed for one burst will go down, from current 20 second to 5-8 second  ?  Or only will be able to fish 8 bursts in 20 seconds ? 
> 
> We are  trying to to get speed of 1 second for each burst... Is this possible anyhow with DEKA ?  On what part we need to focus ?
> 
> Later we will have questions regarding cluster, but firstly we would like to get maximum from one computer...
> 
> 
> 
> 
> On Sun, Feb 21, 2016 at 3:03 PM, Jan Hrach <jenda at yakumo.hrach.eu <mailto:jenda at yakumo.hrach.eu>> wrote:
> 
>     I think it looks OK now. I have raised QSIZE to 100. Looks like the queue management in Deka sucks :-/
> 
>     PS: your cracker did not find "21D142736909C3A5 @ 35 #1 (table:108)" from https://brmlab.cz/project/gsm/deka/deka-test (crack.txt in the linked tarball), the others are OK.
> 
> 
>     On 21.2.2016 22:51, Milinko Isakovic wrote:
>     > If you are still there /dev/sdg sould be now ok. I think it was cable issue...
>     >
>     > On Sun, Feb 21, 2016 at 9:29 PM, Jan Hrach <jenda at yakumo.hrach.eu <mailto:jenda at yakumo.hrach.eu> <mailto:jenda at yakumo.hrach.eu <mailto:jenda at yakumo.hrach.eu>>> wrote:
>     >
>     >     (ssh'ed there, installed htop and iotop)
>     >
>     >     This is because you try to crack only one burst at once. The kernel is designed in a way that it loads num_kernels * vector_size fragments at once, i.e., 4096*32 = 131072 fragments (one burst is 16320 fragments) and then time for one round is almost constant. So you need to crack as many bursts in parallel as to keep it fully saturated. If you are cracking only one burst, the CPU *is* faster. But if you submit for example 20 bursts at once (I would recommend even more, say up to 100), you will make the advantage of 2048 cores of your GPU.
>     >
>     >     PS: your /dev/sdg died, so currently it does not crack at all :-(
>     >
>     >
>     >     On 21.2.2016 19:05, Milinko Isakovic wrote:
>     >     > Hi to all,
>     >     >
>     >     > I tired deka, but seems to be a lot of problems. First of all, I had problem to install drivers (Debian Jessie) amd64,  at the end, finaly I did it somehow.
>     >     >
>     >     > Configuration is:
>     >     >
>     >     > 32 Gb ram
>     >     > 2x 7970
>     >     > 8x SAS disk
>     >     > i7 cpu
>     >     >
>     >     > Main problem for me is that it seems to me that GPU cracking is not working or I do not know how to make an setup.
>     >     >
>     >     > Screenshots I am sending to you describing:
>     >     >
>     >     > 1-5.png  - GUP cracking  ---> 100+ sec
>     >     > cpu1-5.ping CPU cracking ---> 18-19 sec
>     >     >
>     >     > Does anybody know where I am making mistake ?
>     >     >
>     >     >
>     >     > Also I can give ssh to somebody if somebody would like to do tests / checkl where problem is>
>     >     >
>     >     > Regadrs
>     >     > Milinko
>     >     >
>     >
>     >     --
>     >     Jan Hrach | http://jenda.hrach.eu/
>     >     GPG CD98 5440 4372 0C6D 164D A24D F019 2F8E 6527 282E
>     >
>     >
>     >
>     >
>     > --
>     >
>     > Milinko Isakovic, CEO
>     > CS Network Solutions Limited
>     >
>     > 72 Blythe Road, West Kensington, London W14 0HB
>     >
>     >
>     > London Switchboard: +442071933539
>     > Belgrade Operations: +381112620152
>     > Personal mobille:       +381666666666
>     >
>     > Web: www.cs-networks.net <http://www.cs-networks.net> <http://www.cs-networks.net/>
>     > <http://www.facebook.com/smsanywhere> <https://twitter.com/cs_networks> <http://www.linkedin.com/company/cs-networks>  <http://www.youtube.com/csnetworks>
>     > --------------------------
>     >
>     > This message (including any attachments) is confidential and may be privileged. If you have received it by mistake please notify the sender by return e-mail and delete this message from your system. Any unauthorized use or dissemination of this message in whole or in part is strictly prohibited. Please note that e-mails are susceptible to change. CS Networks shall not be liable for the improper or incomplete transmission of the information contained in this communication nor for any delay in its receipt or damage to your system.  CS Networks does not guarantee that the integrity of this communication has been maintained nor that this communication is free of viruses, interceptions or interference.
> 
>     --
>     Jan Hrach | http://jenda.hrach.eu/
>     GPG CD98 5440 4372 0C6D 164D A24D F019 2F8E 6527 282E
> 
> 
> 
> 
> -- 
> 
> Milinko Isakovic, CEO
> CS Network Solutions Limited
> 
> 72 Blythe Road, West Kensington, London W14 0HB
> 
> 
> London Switchboard: +442071933539
> Belgrade Operations: +381112620152
> Personal mobille:       +381666666666
> 
> Web: www.cs-networks.net <http://www.cs-networks.net/>
> <http://www.facebook.com/smsanywhere> <https://twitter.com/cs_networks> <http://www.linkedin.com/company/cs-networks>  <http://www.youtube.com/csnetworks>
> --------------------------
> 
> This message (including any attachments) is confidential and may be privileged. If you have received it by mistake please notify the sender by return e-mail and delete this message from your system. Any unauthorized use or dissemination of this message in whole or in part is strictly prohibited. Please note that e-mails are susceptible to change. CS Networks shall not be liable for the improper or incomplete transmission of the information contained in this communication nor for any delay in its receipt or damage to your system.  CS Networks does not guarantee that the integrity of this communication has been maintained nor that this communication is free of viruses, interceptions or interference.

-- 
Jan Hrach | http://jenda.hrach.eu/
GPG CD98 5440 4372 0C6D 164D A24D F019 2F8E 6527 282E



More information about the A51 mailing list