unforum

[domroberts]

This won't help us one jot

[JebJoya]

Hey all!

Just to mention something I had said earlier in this thread - although the throttle can be used, it is often better to have full throttle and low priority settings. This is easy to set up in Windows. Basically you have 2 options:

1) Set priority to Low in Task Manager:

- Go to config.txt in your install directory, and set the number on the second line to 0
- Run 13th Labour program
- Press ctrl-alt-del
- Click processes
- Find Labour13.exe
- Right Click
- Choose Set Priority... -> Low

2) Make a shortcut which automatically loads it in low priority mode (this is the good one!):

- Right click Labour13.exe and choose Create a Shortcut...
- Make your shortcut, find it, right click and choose properties
- Set target to:

[Rifflesby]

Is there any way to do that in Mac OSX?

[duckiemonster]

[xnera]

Yeah, changing the priority in the task manager was one of the first things I tried. It did help some, but not as much as changing the throttle, alas.

[Juxta]

Throttle controls how fast the program will attempt to process the WU's.

Process priority controls how much of the CPU is used to run it - by setting it to "Below Normal" just means that your CPU will give other programs precedence if they want to use the CPU. However, once those programs stop using it, then the 13th Labour client will immediately grab as much as possible back again, until something comes along with a higher priority.

Moral of the Story:
1974 Minis won't run at 130 MPH
2006 VW Golf GTI will run at 130 MPH

If you attempt to get the former to go that fast, it will go bang. Test your machine out to find an "optimum" level that the client will run at, without exploding.



J

[Skizz]

Running the client at full throttle should not be a problem for any PC. The CPU is designed to cope. If you are experiencing problems then either:

a) your CPU is overheating due to poor cooling or
b) you've overclocked the CPU.

For (a) make sure the fans are clean, the heat sink on the CPU is secure and that there is good air flow around the base unit. If it's under your desk then try it on your desk or put a big fan near it to get the air moving around.

For (b) don't overclock it. When overclocking, you're pushing at the limits the CPU was designed to be used at. When it gets hot, the CPU can start to misbehave (flip flops not flipping or flopping quick enough).

Skizz

[poozle]

[Juxta]

[chimera245]

There has been some discussion in places that we MAY not be performing the correct RC5 decryption process. In fact this morning I had my own doubts.

SO, I talked to the organ grinder (not the monkey) as it were, and referred to the original paper - to confirm once and for all in my own mind that all was well:

http://theory.lcs.mit.edu/~rivest/Rivest-rc5.pdf

I would like to answer this one and for all now.

To quote from the paper:

As noted above RC is wordoriented all of the basic computational oper
ations have wbit words as inputs and outputs
RC is a blockcipher with a
twoword input plaintext block size and a twoword ciphertext output block
size


The 64/12/8 refers to the following

w/r/b

w == Word Size (64 bits)
r == Number of Rounds
b == Number of Bytes in Key

The RCCRYPT package, which I used to base my code on, and is the subject of Von's hint, uses EXACTLY this pattern.

I know a few of you have reviewed the code privately, but just to quell any further speculation, here is, size by side, the code from RCCRYPT and my code for open comparison:

RCCRYPT:

/*now we have sixteen bytes ready to work on*/
/*split it into two ULLONGs, left and right*/

L = (ULLONG *)buffer;
R = (ULLONG *)(buffer+;

/*Put the data into host byte order*/
*L = ntohll (*L);
*R = ntohll (*R);

/*decrypt the data...*/
for (i=rounds;i>=1;i--)
{
*R = rrotate(*R - S[2*i+1] , *L ) ^ *L;
*L = rrotate(*L - S[2*i], *R) ^ *R;
}
*L -= S[0];
*R -= S[1];

/*Convert buffer back to network order*/
*L = htonll(*L);
*R = htonll(*R);

And my DLL:

L = System.BitConverter.ToUInt64(buffer,0);
R = System.BitConverter.ToUInt64(buffer,;

if (this.m_clientType == 1)
{
L = NToHLL(L);
R = NToHLL(R);
}

//decrypt the data
for (i=this.m_rounds;i>=1;i--)
{
R = RRotate(R - this.m_S[2*i+1] , L ) ^ L;
L = RRotate(L - this.m_S[2*i], R) ^ R;
}

L -= this.m_S[0];
R -= this.m_S[1];

if (this.m_clientType == 1)
{
L = HToNLL(L);
R = HToNLL(R);
}


Hopefully this can now move on.

[jazzychad]

Whew! Thanks for posting that, chimera. I was wondering, myself, if we were processing 128 bits at a time or not. I feel much better now! We'll get this cracked in no time now.

One other question though... on the stats page, the number of active clients seemed to be increasing linearly over the past week or so, which should mean that the number of WUs done should go up exponentially....while the graph shows again a linear increase. How are the active clients accounted for on that graph? Just curious...

[chimera245]

I'm not actively expiring clients at the moment, so any churn (new clients in as old clients go out) appears as an increase in client number, when in fact the client number is fairly stable. I'll most likely do an expiry later on in the week - which will reset the numbers again.

The actual number of clients (and associated load) follows a fairly complex daily and weekly cycle, overlaid over a more erratic cycle of bursts of new clients (following some publicity or information sharing to a new location) followed by a slow drop off (as people get bored).

As long as we continue to exceed my much revised target (now standing at 75000 WUs per day) - I will be very happy.

[themandotcom]

Chimera, I'm just wondering, how long would it take to decrypt while doing 100k wu per day? Thanks, man! Keep on Rockin'

[Joe]

[arnezami]

Just thinking out loud here.

We are currently searching all 8 letter passwords containing a-z,A-Z,0-9. And for each block the server determines the first four letters. Right?

My question is: would it be a good idea (if practical of course) to let the server first try out all 4 letters with either only uppercase or only lowercase characters? I believe that might raise our chance of finding the password sooner rather than later.


Or alternatively: alongside this project try out only combinations with either only a-z or only A-Z (with only a couple of dozen computers or so).

From my calculations: if you try only a-z or only A-Z (for all 8 characters that is) the search would be 1045 times faster for that! And there is a chance MC expected us to use rccrypt without speed improvement so they might have made it easier for us than we think.

Just an idea.

arnezami