Politness Costs Nothing but Can Pay Dividends

[Asher]

No, all the more reason to not use an identical seed on startup.

Okay, am I being ****ing punked or something?

I'm here harping about re-seeding often, you're telling me re-seeding is not desirable, and now you're telling me the dangers of using the same seed (eg, not re-seeding).

The story is as follows: the Keno game in question was shut off every night. On startup, the code declared something like strm = new rand(123523);

Throughout the day, the Keno game drew from strm by incrementing. However, because the stream had been reseeded every day with the same seed, the sequence of numbers drawn was identical. Somebody noticed this, took down the sequence of winning draws and won 3 jackpots in a row....

For the life of me I can't figure out why the hell you think this counters my argument that every user gets a new seed every time? It's incredibly retarded to use a fixed number for the seed, using the timestamp is better but not perfect, and this is precisely why in my program, I derive the seed based on timestamp, data from our IR cameras, and mic input.

[Kuciwalker]

Cryptography is far more than ciphers.

My best friend is doing her PhD in crypto right now, and she's not doing much with ciphers at all. Ciphers are but one part of Cryptography, even if they are an extremely common part right now.

So, basically, you meant a limited section of new research in cryptography that isn't actually used in practical applications doesn't regard ciphers, and therefore maybe isn't related to RNGs, so my generalization was horribly, horribly wrong?

[Asher]

The quote I provided is perfectly clear.

Yes, the quote you provided was perfectly clear in demonstrating I said "new rand" starts a new stream. Thus my confusion here.

[Asher]

Cryptography is far more than ciphers.

My best friend is doing her PhD in crypto right now, and she's not doing much with ciphers at all. Ciphers are but one part of Cryptography, even if they are an extremely common part right now.

So, basically, you meant a limited section of new research in cryptography that isn't actually used in practical applications doesn't regard ciphers, and therefore maybe isn't related to RNGs, so my generalization was horribly, horribly wrong?

Your statement that cryptography = RNG is quite obviously wrong because cryptography is a field of study, and a RNG is not.

If you said cipher, I'd have agreed. Say what you mean.

[KrazyHorse]

Okay, am I being ****ing punked or something?

I'm here harping about re-seeding often, you're telling me re-seeding is not desirable, and now you're telling me the dangers of using the same seed (eg, not re-seeding).

Asher, this is just starting to get sad. If you're going to DECLARE A NEW STREAM then obviously you need a new seed (assuming you start at the beginning of the stream). However, there is NO REASON TO BELIEVE that you need to declare a new stream for every user. In fact, it is far safer to NOT declare a new stream for every user, but to persist the stream. This is because RNGs are designed to provide sequences of numbers with defined behaviors in a single stream, or if they are designed for parallel applications, across a set of numbered streams with identical seeds.

[Asher]

Asher, this is just starting to get sad. If you're going to DECLARE A NEW STREAM then obviously you need a new seed (assuming you start at the beginning of the stream.

Yes, I've never disagreed with this.

However, there is NO REASON TO BELIEVE that you need to declare a new stream for every user. In fact, it is far safer to NOT declare a new stream for every user, but to persist the stream. This is because RNGs are designed to provide sequences of numbers with defined behaviors in a single stream, or if they are designed for parallel applications, across a set of numbered streams with identical seeds.

And this is where we actually disagree.

The very fact that subsequent numbers have defined behaviors in the single stream is precisely WHY it should be re-seeded for each user. The goal is to have random numbers, not "well defined behaviours", for each user. If you don't think they should be re-seeded with each user, under what conditions would you re-seed?

I'm not sure how aware you are how utterly crap RNGs are on those embedded devices. Ever used Shuffle on your iPod?

[KrazyHorse]

Yes, I've never disagreed with this.


And this is where we actually disagree.

The very fact that subsequent numbers have defined behaviors in the single stream is precisely WHY it should be re-seeded for each user. The goal is to have random numbers, not "well defined behaviours", for each user. If you don't think they should be re-seeded with each user, under what conditions would you re-seed?

I'm not sure how aware you are how utterly crap RNGs are on those embedded devices. Ever used Shuffle on your iPod?

I'm beginning to see why RNG screwups are so common. People like you use them.

[Asher]

I'm starting to suspect you're simply drunk at this point.

[KrazyHorse]

I'm more embarrassed for you than anything else...

[Asher]

And why won't you answer the question. If you don't re-seed for each user, under what conditions do you re-seed?

[KrazyHorse]

Ideally, never.

[Asher]

I'm more embarrassed for you than anything else...

No reason to be, this wouldn't be the first time I've had to put PhDs in other fields in their place when they try to design proper software.

[Asher]

Ideally, never.

So you think these things should seed based on the MAC address or something?

You've got a lot more learning to do about this, KH. Talk to that colleague of yours.

[KrazyHorse]

Most reasonably good RNGs are fine for 2^128 or more bits. SPRNG is pretty good for any parallel application I've ever needed, and is good for up to 2^256 bits or so, divided through various numbers of independent available streams, depending on what algo you use.

[KrazyHorse]

So you think these things should seed based on the MAC address or something?

Are you a ******?