cluster distribution with 24PC

[GoogleCodeExporter]

I write my thesis about wifi security. This include what is the breaking time 
on wpa. Now I hire classroom from the school and I would like to ask to help 
somebody.

What steps will reproduce the problem?

1. Approximatly I will get 24 core2quad q6600 and nvidia gtx260 PC with gigabit 
network. Can I use more than one cpu core? (I read about that, one CPU-core eat 
one GPU, but what about the network serve? more cpu core help the network 
distribution.)

2. How can you control the huge data flow, when you have got 24 clients? In my 
oppinion I will get more than 200000PMK/s. Does my winchester keep up the 
loading? (500G, 7200rpm, sata2)
2b.If I chose database, I will get higher performance? 

3. I noticed that, the synchronization on the server eat enormous cpu power. I 
really interested in that,Does the synchronization loading on 1 core or 
distributing all core? (will I get higher power if I enable HyperThreading?)

4. What do you think about? Use core i5-650 3,2ghz (2core, 4thread) or use 
core2 quad 6600 only 2,4ghz (4core, 4thread) with or without gtx260.
4b.How important the memory bandwith? (

What version of the product are you using? On what operating system?
svn280 backtrack4 r2 fresh

Please provide any additional information below.
What does it mean RTT and TTS?
Serving 0 active clients; 10283 PMKs/s; 0.0 TTS
#3: 'Network-Clients': 6380.4 PMKs/s (RTT 7.0)

Sorry my bad english

Original issue reported on code.google.com by andor...@gmail.com on 16 Jan 2011 at 11:54

[GoogleCodeExporter]

I speak in function of my experience and what I read on the issues in these 
months, so it means I don't have the TRUE on my tongue, ok?

As far as I know, pyrit has problem to scale (see issue 173) and there are some 
bottlenecks around (threads, moltiple GPU managements, etc) and there are also 
other bottleneck (DB performace issues, Network I/O, hard disk I/O, etc).

If you want to follow the "cluster way", my suggestion is to set a few number 
of nodes (lets start with one master and one slave) and then test. Then add a 
new node and ther again till you reach bottleneck.

Anyway, I think you hard disk will be not able to feed 24 nodes: I think best 
solution (under performances viewpoint) is manage 24 nodes indipendently, 
charge 1/24th of passwords on each of them and then control from remote using 
ssh and/or scripts.
In this case, you will have PSK/s that will grove in linear mode in function of 
number of node, otherwise I am afraid you will suffer a lot of bottlenck than 
will "eat" so many power.... c'est la vie!

PS: i don't think "issue" page is the best way to manage this topic, it should 
be bettor to post in group discussion.

Original comment by pyrit.lo...@gmail.com on 17 Jan 2011 at 12:20

[GoogleCodeExporter]

would be nice to see unified clustering platform for worlds top cracking 
programs

John the ripper has MPI support, if Pyrit has MPI support that would be nice 
because it would allow both programs to run on the same distributed computing 
platform

Original comment by thefixe...@gmail.com on 1 Feb 2011 at 8:30

[GoogleCodeExporter]

you are very welcome to write one :-)

Original comment by lukas.l...@gmail.com on 2 Feb 2011 at 7:35

[GoogleCodeExporter]

I have done the test with 20pc (19client + 1 server)

I will write the results as soon as possible

Original comment by andor...@gmail.com on 2 Feb 2011 at 2:35

[GoogleCodeExporter]

hey lucas, your awesome dude!! I wish I knew how to code I would help you but 
im a n00b, I bow to thee, and thank you for making your work available.

Original comment by thefixe...@gmail.com on 5 Feb 2011 at 5:05

[GoogleCodeExporter]

Original comment by lukas.l...@gmail.com on 15 Feb 2011 at 4:17

• Changed state: Invalid