Not true. Password hashing algorithms should be resource intensive enough to prevent brute force calculation from being a viable route. This is why bcrypt stores a salt, a hash, and the current number of rounds. That number of rounds should increase as CPUs get faster to prevent older hashes from existing in the wild which can be more effectively broken by newer CPUs.
I was incorrect about the goal being minimal resources. I should have written that that goal was to have controlled resource usage. The salt does not increase the expense of the the hash function. Key stretching techniques like adding rounds increase the expense to reach the final hash output but does not increase the expense of the hash function. High password length allowances of several thousand characters should not lead to a denial of service attack but they don’t materially increase security after a certain length either.
They’re designed to be resource intensive to calculate to make them harder to brute force, and impossible to reverse.
Some literally have a parameter which acts as a sliding scale for how difficult they are to calculate, so that you can increase security as hardware power advances.
I was incorrect but I still disagree with you. The hashing function is not designed to be resource intensive but to have a controlled cost. Key stretching by adding rounds repeats the controlled cost to make computing the final hash more expensive but the message length passed to the function isn’t really an issue. After the first round it doesn’t matter if the message length was 10, 128, or 1024 bytes because each round after is only getting exactly the number of bytes the one way hash outputs.
Yes, a hashing function is designed to be resource intensive, since that’s what makes it hard to brute force. No, a hashing function isn’t designed to be infinitely expensive, because that would be insane. Yes, it’s still a bad thing to provide somebody with a force multiplier like that if they want to run a denial-of-service.
I’m a bit behind on password specific hashing techniques. Thanks for the education.
My background more in general purpose one way hashing functions where we want to be able to calculate hashes quickly, without collisions, and using a consistent amount of resources.
If the goal is to be resource intensive why don’t modern hashing functions designed to use more resources? What’s the technical problem keeping Argon2 from being designed to eat even more cycles?
Argon2 has parameters that allow you to specify the execution time, the memory required, and the degree of parallelism.
But at a certain point you get diminishing returns and you’re just wasting resources. It seems like a similar question to why not just use massive encryption keys.
you have to limit it somewhere or you’re opening yourself up for a DoS attack
password hashing algorithms are literally designed to be resource intensive
Edited to remove untrue information. Thanks for the corrections everyone.
See “Password Hashing” here: https://en.m.wikipedia.org/wiki/Key_derivation_function
It is actually important to have a controlled cost to calculate in the forward direction too.
Totally true. I stand corrected. Thank you.
Hashes are one way functions. You can’t get from hash back to input
True. I was all kinds of incorrect in my hasty typing. I’ll update it to be less wrong.
Only if the hash function is designed well
Not true. Password hashing algorithms should be resource intensive enough to prevent brute force calculation from being a viable route. This is why bcrypt stores a salt, a hash, and the current number of rounds. That number of rounds should increase as CPUs get faster to prevent older hashes from existing in the wild which can be more effectively broken by newer CPUs.
I was incorrect about the goal being minimal resources. I should have written that that goal was to have controlled resource usage. The salt does not increase the expense of the the hash function. Key stretching techniques like adding rounds increase the expense to reach the final hash output but does not increase the expense of the hash function. High password length allowances of several thousand characters should not lead to a denial of service attack but they don’t materially increase security after a certain length either.
Incorrect.
They’re designed to be resource intensive to calculate to make them harder to brute force, and impossible to reverse.
Some literally have a parameter which acts as a sliding scale for how difficult they are to calculate, so that you can increase security as hardware power advances.
I was incorrect but I still disagree with you. The hashing function is not designed to be resource intensive but to have a controlled cost. Key stretching by adding rounds repeats the controlled cost to make computing the final hash more expensive but the message length passed to the function isn’t really an issue. After the first round it doesn’t matter if the message length was 10, 128, or 1024 bytes because each round after is only getting exactly the number of bytes the one way hash outputs.
It depends on the hash. E.g., OWASP only recommends 2 iterations of Argon2id as a minimum.
Yes, a hashing function is designed to be resource intensive, since that’s what makes it hard to brute force. No, a hashing function isn’t designed to be infinitely expensive, because that would be insane. Yes, it’s still a bad thing to provide somebody with a force multiplier like that if they want to run a denial-of-service.
I’m a bit behind on password specific hashing techniques. Thanks for the education.
My background more in general purpose one way hashing functions where we want to be able to calculate hashes quickly, without collisions, and using a consistent amount of resources.
If the goal is to be resource intensive why don’t modern hashing functions designed to use more resources? What’s the technical problem keeping Argon2 from being designed to eat even more cycles?
Argon2 has parameters that allow you to specify the execution time, the memory required, and the degree of parallelism.
But at a certain point you get diminishing returns and you’re just wasting resources. It seems like a similar question to why not just use massive encryption keys.