Hashcat Brain bottleneck bypass
#1
1st of all I wanna to say a hashcat brain function is amazing. Thanks to atom and all the contributors. I`ve tried to operate it with different configurations and watched the online speed differences from offline run without network sockets. Network is an element which is hard to improve in regard of hashcat. I`m having a notion, isn`t there a way how to rapidly enlarge a LAN bottleneck via using IPFC and RFC 2625 over fibre channel Brocade optical switches ?  I have several Brocades equipped by 16Gbit Gbic`s and theoretically there could be a very fast response time between brain and clients, far beyond any ethernet swich possibilities.  Just an idea, but maybe there is a way.  Thanks for any opinion or maybe explaining me I haven`t considered an unknown obstacle or any other bottleneck reason.
Reply
#2
maybe explain what type of hash you are trying to crack and why the (ethernet) speed in your opinion is really that important ?

brain was never meant to be used with that much of data anyways, it was developed for very slow salted hashes like bcrypt and scrypt
Reply
#3
(10-08-2020, 08:54 PM)philsmd Wrote: maybe explain what type of hash you are trying to crack and why the (ethernet) speed in your opinion is really that important ?

brain was never meant to be used with that much of data anyways, it was developed for very slow salted hashes like bcrypt and scrypt

I`ve tried many algos, but even mentioned slow hashes bcrypt and scrypt are rapidly slowing down via brain session with 5+ clients, in contrast with a single brain client, which is  always perfect on slow hashes.   I see it on client`s performance monitor. In offline mode without brain  there is 10-20% better utilisation of the client`s hardware and hashing speed on slow hashes ( measured -D1 and -D2 extra each ) measured  from 5+ clients. A brain hardware probably is not a bottleneck, as I saw , except NIC of c. ( Never run brain on W2012 with multiple faces, it is a performance tragedy, 40% hashing speed lost even on slow algos with 5+ clients, only linux version is usable with 10+ clients  ) I also noticed, it is a bad idea to connect clients with poor performance to a brain together with powerful clients , it somehow slows down the powerful ones. I tried a lot of HW combinations. The last thing which gave me a slightly  better result in clients utilisation decrease  was 10Gbit ethetnet. Not because of the speed, but probably because of the latency.  But fiber channel has much more power to be unleashed.
Reply