C question in All Things Unix http://www.dslreports.com/forum/r20964802 en Wed, 10 Feb 2010 10:32:18 EDT Wed, 10 Feb 2010 10:32:18 EDT Re: C question http://www.dslreports.com/forum/remark,21134084 Dr_Death : TSI Gabe, did you have a look on nick breese POC.
»www.security-assessment.com/file···1.tar.gz
it include two files PPU_md5.c & SPU_md5.c
it may help you to reach 1.4 billion calculation a second.

there is a guy who modify john the ripper ppc version to run from spu instead of cpu, but he didnt publish it.
»chamewco.livejournal.com/107802.html

good luck.]]> http://www.dslreports.com/forum/remark,21134084 Thu, 18 Sep 2008 22:02:11 EDT Re: C question http://www.dslreports.com/forum/remark,20972077 TakeTheFifth : Inspiration ?

Phil]]> http://www.dslreports.com/forum/remark,20972077 Mon, 18 Aug 2008 23:47:08 EDT Re: C question http://www.dslreports.com/forum/remark,20969642 TSI Gabe : The problem with the SPUs is that its _extremely_ slow at branching. Any jmp will cause it to stall for at least 20 cycles. So basically the whole md5 function is __inline__'d and does very simple spu_or,xor,and on 4 32-bit vectors.

So I'm trying to code a function that is as much unrolled and pipelined as possible to generate a word list.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20969642 Mon, 18 Aug 2008 15:51:02 EDT Re: C question http://www.dslreports.com/forum/remark,20969612 dave :
said by Steve See Profile :

... that appears to violate the laws of physics.
The PS3 uses logic elements made from resublimated thiotimoline.]]> http://www.dslreports.com/forum/remark,20969612 Mon, 18 Aug 2008 15:46:36 EDT Re: C question http://www.dslreports.com/forum/remark,20969588 TSI Gabe : I'll open it up when I'm satisfied with what I have.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20969588 Mon, 18 Aug 2008 15:42:55 EDT Re: C question http://www.dslreports.com/forum/remark,20969467 GILXA1226 : This code going to be open, or you are going to keep it closed?]]> http://www.dslreports.com/forum/remark,20969467 Mon, 18 Aug 2008 15:23:22 EDT Re: C question http://www.dslreports.com/forum/remark,20969433 BeesTea :
said by Steve See Profile :

For instance, many onlookers will find it hard to reconcile your difficulties with basic C casting with the notion that you've implemented some killer hashing algorithm that appears to violate the laws of physics.
Physics schmysics!
--
Overpower, overcome.]]> http://www.dslreports.com/forum/remark,20969433 Mon, 18 Aug 2008 15:17:48 EDT Re: C question http://www.dslreports.com/forum/remark,20967705 TSI Gabe :
said by dave See Profile :

Just for info, what is the length of the source string over which these hashes are generated?
8 chars. So yes very simple.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20967705 Mon, 18 Aug 2008 09:53:52 EDT Re: C question http://www.dslreports.com/forum/remark,20967437 Steve :
said by TSI Gabe See Profile :

Anyway...I really don't have to prove myself.
No, you don't, but that won't stop us from reaching our own conclusions based on the evidence here.

For instance, many onlookers will find it hard to reconcile your difficulties with basic C casting with the notion that you've implemented some killer hashing algorithm that appears to violate the laws of physics.

But in any case, good luck.

Steve
--
Stephen J. Friedl | Unix Wizard | Microsoft Security MVP | Tustin, California USA | my web site]]> http://www.dslreports.com/forum/remark,20967437 Mon, 18 Aug 2008 08:36:23 EDT Re: C question http://www.dslreports.com/forum/remark,20967424 dave : Just for info, what is the length of the source string over which these hashes are generated?]]> http://www.dslreports.com/forum/remark,20967424 Mon, 18 Aug 2008 08:30:26 EDT Re: C question http://www.dslreports.com/forum/remark,20967388 TSI Gabe : Yeah well the SPUs can do 5 billion hashes per SPU but without storing them back in memory. Of course the PS3 doesn't have enough memory bandwidth to handle it. Keep in mind that the SPUs can do 4 32 bit instructions per operand.

The problem I'm having is sending all that data back to the PPU which is proving to be impossible since it would fill up the PS3's memory in a few seconds, not to mention it couldn't handle it.

So the only viable thing it can do is search for one hash or maybe 4 at a time.

Anyway...I really don't have to prove myself.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20967388 Mon, 18 Aug 2008 08:14:37 EDT Re: C question http://www.dslreports.com/forum/remark,20966693 Tony_B : Getting ~5 billion hashes a second per core is next to impossible with a PS3 cell processor. Generating *one* MD5 hash requires many CPU instructions (each taking many clock cycles). I will have to concur with Steve See Profile on this one. You must have meant millions... ]]> http://www.dslreports.com/forum/remark,20966693 Mon, 18 Aug 2008 00:23:15 EDT Re: C question http://www.dslreports.com/forum/remark,20966650 Steve :
said by TSI Gabe See Profile :

I've been able to reach arround 30 billion MD5 hashes per second using the 6 SPU cores on the PS3...
I call bullshit.

30B MD5 hashes per second over 6 cores means 5B hashes per second per core. I suppose I could believe that the PS3 could do 5B CPU instructions per second per core, but I don't buy that it can do 5B MD5 hashes (even simple ones) per second. I doubt even the memory bandwidth is sufficient for that amount of computation.

Did you mean million?

Steve
--
Stephen J. Friedl | Unix Wizard | Microsoft Security MVP | Tustin, California USA | my web site]]> http://www.dslreports.com/forum/remark,20966650 Mon, 18 Aug 2008 00:09:38 EDT Re: C question http://www.dslreports.com/forum/remark,20966550 BeesTea :
said by TSI Gabe See Profile :

I guess might as well get it out there.

I've been able to reach arround 30 billion MD5 hashes per second using the 6 SPU cores on the PS3...I'm trying to make it even better still..
Thank ${diety} (or math) for salted hashes then.

What's your practical application? Cracking apache htpasswd files?
--
Overpower, overcome.]]> http://www.dslreports.com/forum/remark,20966550 Sun, 17 Aug 2008 23:37:00 EDT Re: C question http://www.dslreports.com/forum/remark,20966139 TSI Gabe : No you do have access to the SPUs (only 6 out of 8 though). The hypervisor controls access to the RSX video card and hard drive which isn't great I agree.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20966139 Sun, 17 Aug 2008 22:00:20 EDT Re: C question http://www.dslreports.com/forum/remark,20965332 netjunk : You are not even using cell processors as Sony made sure users don't have direct access to ps3 hardware. So you shouldn't really expect too much out it. What you really need to do is to fix its firmware and then try to find a way to harness real ps3 power. :D]]> http://www.dslreports.com/forum/remark,20965332 Sun, 17 Aug 2008 18:51:50 EDT Re: C question http://www.dslreports.com/forum/remark,20964935 TSI Gabe :
said by dave See Profile :

So this is "cracking security for fun and profit with household toys" ?
I guess you can put it that way.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20964935 Sun, 17 Aug 2008 17:19:55 EDT Re: C question http://www.dslreports.com/forum/remark,20964932 dave : So this is "cracking security for fun and profit with household toys" ?]]> http://www.dslreports.com/forum/remark,20964932 Sun, 17 Aug 2008 17:19:18 EDT Re: C question http://www.dslreports.com/forum/remark,20964911 TSI Gabe : I guess might as well get it out there.

I've been able to reach arround 30 billion MD5 hashes per second using the 6 SPU cores on the PS3...I'm trying to make it even better still..
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20964911 Sun, 17 Aug 2008 17:13:49 EDT Re: C question http://www.dslreports.com/forum/remark,20964878 TSI Gabe : I'm actually programming for one arch (PS3 Cell Processor to be exact) so that doesn't bother me.

It does appear to be working so thanks
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20964878 Sun, 17 Aug 2008 17:07:02 EDT Re: C question http://www.dslreports.com/forum/remark,20964872 nwrickert : "abcd" could then be 0x61626364 or 0x64636261 depending on the byte order of the machine. This can get tricky if you want to avoid such problems. With copying allowed, there are library functions to avoid that problem ("htonl()" and friends).
--
AT&T dsl; Westell 327w modem/router; openSuSE 11.0; firefox 3.0.1]]> http://www.dslreports.com/forum/remark,20964872 Sun, 17 Aug 2008 17:05:05 EDT Re: C question http://www.dslreports.com/forum/remark,20964843 dave : This'll probably work:

x = *(unsigned int*)array

This is non-portable, dependent on the integer size, dependent on endianness, and may fail on architectures that require aligned integers. There is no solution without data movement, otherwise.]]> http://www.dslreports.com/forum/remark,20964843 Sun, 17 Aug 2008 16:59:16 EDT C question http://www.dslreports.com/forum/remark,20964802 TSI Gabe : This may be unrelated to this forum but it's the closest thing I've found.

Basically what I'm trying to do is casting an array of char into an array of unsigned integers with their ascii value. for example "aaaa" would cast to 0x61616161. And this without using memcpy and would be as much as possible O(1).

I know the answer might be simple enough...but I can't seem to figure this one out.
--
TSI Gabe - TekSavvy Solutions Inc.]]> http://www.dslreports.com/forum/remark,20964802 Sun, 17 Aug 2008 16:46:20 EDT