Re: [Cryptography] Thoughts on hardware randomness sources
On 09/13/2013 11:32 PM, Jerry Leichter wrote: On Sep 12, 2013, at 11:06 PM, Marcus D. Leech wrote: There are a class of hyper-cheap USB audio dongles with very uncomplicated mixer models. A small flotilla of those might get you some fault-tolerance. My main thought on such things relates to servers, where power consumption isn't really much of an issue I'm not sure what servers you're talking about here. If by server you mean one of those things in a rack at Amazon or Google or Rackspace - power consumption, and its consequence, cooling - is *the* major issue these days. Also, the servers used in such data centers don't have multiple free USB inputs - they may not have any. If by server you mean some quite low-power box in someone's home ... power is again an issue. People want these things small, fan-free, and dead reliable. And they are increasingly aware of the electric bills always-on devices produce. About the only server for which power is not an issue is one of those extra-large desktops that small businesses use. -- Jerry I was mostly contrasting with mobile systems, where power consumption is at an absolute premium. The USB sound systems I'm thinking of consume 350mW while operating, and about 300uW when idle. A couple or three of those on even a stripped-down server would contribute in only the smallest way to extra power consumption. And the extra computational load? When these servers things are running flat-out serving up secured connections? I would guess the phrase an inconsiderable trifle would apply. ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Thoughts on hardware randomness sources
On Sep 12, 2013, at 11:06 PM, Marcus D. Leech wrote: There are a class of hyper-cheap USB audio dongles with very uncomplicated mixer models. A small flotilla of those might get you some fault-tolerance. My main thought on such things relates to servers, where power consumption isn't really much of an issue I'm not sure what servers you're talking about here. If by server you mean one of those things in a rack at Amazon or Google or Rackspace - power consumption, and its consequence, cooling - is *the* major issue these days. Also, the servers used in such data centers don't have multiple free USB inputs - they may not have any. If by server you mean some quite low-power box in someone's home ... power is again an issue. People want these things small, fan-free, and dead reliable. And they are increasingly aware of the electric bills always-on devices produce. About the only server for which power is not an issue is one of those extra-large desktops that small businesses use. -- Jerry ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
[Cryptography] RSA equivalent key length/strength
Recommendations are given herein as: symmetric_key_length - recommended_equivalent_RSA_key_length, in bits. Looking at Wikipedia, I see: As of 2003 RSA Security claims that 1024-bit RSA keys are equivalent in strength to 80-bit symmetric keys, 2048-bit RSA keys to 112-bit symmetric keys and 3072-bit RSA keys to 128-bit symmetric keys. RSA claims that 1024-bit keys are likely to become crackable some time between 2006 and 2010 and that 2048-bit keys are sufficient until 2030. An RSA key length of 3072 bits should be used if security is required beyond 2030.[6] http://www.emc.com/emc-plus/rsa-labs/standards-initiatives/key-size.htm That page doesn't give any actual recommendations or long-term dates from RSA now. It gives the traditional recommendations 80 - 1024 and 112 - 2048, and a 2000 Lenstra/Verheul minimum commercial recommendation for 2010 of 78 - 1369. NIST key management guidelines further suggest that 15360-bit RSA keys are equivalent in strength to 256-bit symmetric keys.[7] http://csrc.nist.gov/publications/nistpubs/800-57/sp800-57_part1_rev3_general.pdf NIST also give the traditional recommendations, 80 - 1024 and 112 - 2048, plus 128 - 3072, 192 - 7680, 256 - 15360. I get that 1024 bits is about on the edge, about equivalent to 80 bits or a little less, and may be crackable either now or sometime soon. But, I wonder, where do these longer equivalent figures come from? I don't know, I'm just asking - and I chose Wikipedia because that's the general wisdom. Is this an area where NSA have shaped the worldwide cryptography marketplace to make it more tractable to advanced cryptanalytic capabilities being developed by NSA/CSS, by perhaps greatly exaggerating the equivalent lengths? And by emphasising the difficulty of using longer keys? As I said, I do not know. I merely raise the possibility. [ Personally, I recommend 1,536 bit RSA keys and DH primes for security to 2030, 2,048 if 1,536 is unavailable, 4,096 bits if paranoid/high value; and not using RSA at all for longer term security. I don't know whether someone will build that sort of quantum computer one day, but they might. ] -- Peter Fairbrother ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] RSA equivalent key length/strength
Also see RFC 3766 from almost a decade ago; it has stood up fairly well. --Paul Hoffman ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] prism proof email, namespaces, and anonymity
On Fri, Sep 13, 2013 at 10:12 PM, Perry E. Metzger pe...@piermont.comwrote: On Fri, 13 Sep 2013 16:55:05 -0400 John Kelsey crypto@gmail.com wrote: Everyone, The more I think about it, the more important it seems that any anonymous email like communications system *not* include people who don't want to be part of it, and have lots of defenses to prevent its anonymous communications from becoming a nightmare for its participants. If the goal is to make PRISM stop working and make the email part of the internet go dark for spies (which definitely includes a lot more than just US spies!), then this system has to be something that lots of people will want to use. There should be multiple defenses against spam and phishing and other nasty things being sent in this system, with enough designed-in flexibility to deal with changes in attacker behavior over tome. Indeed. As I said in the message I just pointed Nico at: http://www.metzdowd.com/pipermail/cryptography/2013-August/016874.html Quoting myself: Spam might be a terrible, terrible problem in such a network since it could not easily be traced to a sender and thus not easily blocked, but there's an obvious solution to that. I've been using Jabber, Facebook and other services where all or essentially all communications require a bi-directional decision to enable messages for years now, and there is virtually no spam in such systems because of it. So, require such bi-directional friending within our postulated new messaging network -- authentication is handled by the public keys of course. The keys. This to me is the critical point for widespread adoption, key management. How do you do this in a way that doesn't put people off immediately. There are two new efforts I'm aware if trying to solve this in a user friendly way are https://parley.co/#how-it-works and http://mailpile.is. Parley's approach does at least deal with the longevity of the private key although it does boil security down to a password, if I can obtain their packet and the salt I can probably brute force the password. I've exchanged mails with the mailpile.is guys and I think they're still looking at the options. Max ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
[Cryptography] Key management, key storage. (was Re: prism proof email, namespaces, and anonymity)
On Sat, 14 Sep 2013 17:23:40 +0100 Max Kington mking...@webhanger.com wrote: The keys. This to me is the critical point for widespread adoption, key management. How do you do this in a way that doesn't put people off immediately. You don't seem to be entirely talking about key management, given that you talk about mailpile and parley. Parley seems to be simply talking about *key storage* for example, which is a different kettle of fish. However, on the topic of key management itself, my own proposal was described here: http://www.metzdowd.com/pipermail/cryptography/2013-August/016870.html In summary, I proposed a way you can map IDs to keys through pure long term observation/widely witnessed events. The idea is not original given that to some extent things like Certificate Transparency already do this in other domains. Perry -- Perry E. Metzgerpe...@piermont.com ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] RSA equivalent key length/strength
On Sat, 14 Sep 2013 09:31:22 -0700 Paul Hoffman paul.hoff...@vpnc.org wrote: Also see RFC 3766 from almost a decade ago; it has stood up fairly well. For those not aware, the document, by Paul and Hilarie Orman, discusses equivalent key strengths and practical brute force methods, giving extensive detail on how all calculations were done. A URL for the lazy: http://tools.ietf.org/html/rfc3766 It is very well done. I'd like to see an update done but it does feel like the methodology was well laid out and is difficult to argue with in general. The detailed numbers are slightly different from others out there, but not so much as to change the general recommendations that have been floating around. Their table, from April 2004, looked like this: +-+---+--+--+ | System | | | | | requirement | Symmetric | RSA or DH| DSA subgroup | | for attack | key size | modulus size | size | | resistance | (bits)| (bits) | (bits) | | (bits) | | | | +-+---+--+--+ | 70 | 70| 947 | 129 | | 80 | 80| 1228 | 148 | | 90 | 90| 1553 | 167 | |100 |100| 1926 | 186 | |150 |150| 4575 | 284 | |200 |200| 8719 | 383 | |250 |250|14596 | 482 | +-+---+--+--+ They had some caveats, such as the statement that if TWIRL like machines appear, we could presume an 11 bit reduction in strength -- see the RFC itself for details. Perry -- Perry E. Metzgerpe...@piermont.com ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Key management, key storage. (was Re: prism proof email, namespaces, and anonymity)
On Sat, Sep 14, 2013 at 9:46 AM, Perry E. Metzger pe...@piermont.com wrote: However, on the topic of key management itself, my own proposal was described here: http://www.metzdowd.com/pipermail/cryptography/2013-August/016870.html In summary, I proposed a way you can map IDs to keys through pure long term observation/widely witnessed events. The idea is not original given that to some extent things like Certificate Transparency already do this in other domains. Hi Perry, What you're proposing is multipath probing of email users' public keys. Certificate Transparency isn't the right comparison, but this has certainly been discussed in other domains: Public Spaces Key Infrastructure / SecSpider (Osterweil et al, 2006, 2007) [1] Perspectives (for HTTPS - Wendlant et al, 2008) [3] Convergence (for HTTPS - Marlinspike, 2011) [4] Vantages (for DNSSSEC - Osterweil et al, 2013) [5] Probing servers is easier than probing email users, and publishing a servername - key directory is also easier as server names don't have the same privacy concerns as email names. Still, it's an interesting idea. Key changes are a challenge to this approach, which people tend to overlook. One approach is to have the probed party declare a commitment to maintaining its public key constant for some period of time, and have this commitment be detected by the probing parties. This provides some timing guarantees so that the rest of the system can probe and download new results at regular intervals, without having sudden key changes cause glitches. Things like HPKP [6] and TACK [7] explore this option. Trevor [1] http://irl.cs.ucla.edu/papers/pski.pdf [2] http://secspider.cs.ucla.edu/docs.html [3] http://perspectives-project.org/ [4] http://convergence.io/ [5] http://irl.cs.ucla.edu/~eoster/doc/pubdata-tpds13.pdf [6] http://tools.ietf.org/html/draft-ietf-websec-key-pinning-08 [7] http://tack.io ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] RSA equivalent key length/strength
On 14/09/13 17:14, Perry E. Metzger wrote: On Sat, 14 Sep 2013 16:53:38 +0100 Peter Fairbrother zenadsl6...@zen.co.uk wrote: NIST also give the traditional recommendations, 80 - 1024 and 112 - 2048, plus 128 - 3072, 192 - 7680, 256 - 15360. [...] But, I wonder, where do these longer equivalent figures come from? I don't know, I'm just asking - and I chose Wikipedia because that's the general wisdom. [...] [ Personally, I recommend 1,536 bit RSA keys and DH primes for security to 2030, 2,048 if 1,536 is unavailable, 4,096 bits if paranoid/high value; and not using RSA at all for longer term security. I don't know whether someone will build that sort of quantum computer one day, but they might. ] On what basis do you select your numbers? Have you done calculations on the time it takes to factor numbers using modern algorithms to produce them? Yes, some - but I don't believe that's enough. Historically, it would not have been (and wasn't) - it doesn't take account of algorithm development. I actually based the 1,536-bit figure on the old RSA factoring challenges, and how long it took to break them. We are publicly at 768 bits now, and that's very expensive http://eprint.iacr.org/2010/006.pdf - and, over the last twenty years the rate of public advance has been about 256 bits per decade. So at that rate 1,536 bits would become possible but very expensive in 2043, and would still be impossible in 2030. If 1,024 is possible but very expensive for NSA now, and 256 bits per decade is right, then 1,536 may just be on the verge of edging into possibility in 2030 - but I think progress is going to slow (unless they develop quantum computers). We have already found many of the easy-to-find advances in theory. -- Peter Fairbrother ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] RSA equivalent key length/strength
On 14/09/13 18:53 PM, Peter Fairbrother wrote: But, I wonder, where do these longer equivalent figures come from? http://keylength.com/ (is a better repository to answer your question.) iang ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] RSA equivalent key length/strength
On Sat, Sep 14, 2013 at 12:56:02PM -0400, Perry E. Metzger wrote: http://tools.ietf.org/html/rfc3766 | requirement | Symmetric | RSA or DH| DSA subgroup | | for attack | key size | modulus size | size | +-+---+--+--+ |100 |100| 1926 | 186 | if TWIRL like machines appear, we could presume an 11 bit reduction in strength 100-11 = 89 bits. Bitcoin is pushing 75 bits/year right now with GPUs and 65nm ASICs (not sure what balance). Does that place ~2000 bit modulus around the safety margin of 56-bit DES when that was being argued about (the previous generation NSA key-strength sabotage)? Anyone have some projections for the cost of a TWIRL to crack 2048 bit RSA? Projecting 2048 out to a 2030 doesnt seem like a hugely conservative estimate. Bear in mind NSA would probably be willing to drop $1b one-off to be able to crack public key crypto for the next decade. There have been cost and performance, power, density improvements since TWIRL was proposed. Maybe the single largest employer of mathematicians can squeeze a few incremetal optimizations of the TWIRL algorithm or implementation strategy. Tin foil or not: maybe its time for 3072 RSA/DH and 384/512 ECC? Adam ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Perfection versus Forward Secrecy
On Thu, Sep 12, 2013 at 11:08 PM, Eugen Leitl eu...@leitl.org wrote: I do not think that the spooks are too far away from open research in QC hardware. It does not seem likely that we'll be getting real QC any time soon, if ever. I don't think the spooks are ahead of the public either, and I really doubt the NSA has a large quantum computer. We still haven't seen quantum computers built yet which can truly rival their conventional electronic brethren, especially if you look at it from a cost perspective. DWave computers are interesting from a novelty perspective, but not really ready to replace existing computers, even for highly specialized tasks like running Shor's algorithm. Nevertheless, if you've been following the trends in quantum computers over the last few years, they are getting larger, and DWave is an example of them moving out of the labs and turning into something you can buy. I wouldn't be surprised to see a large quantum computer built in the next two decades. -- Tony Arcieri ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
[Cryptography] Quantum Computers for Shor's Algorithm (was Re: Perfection versus Forward Secrecy)
On Sat, 14 Sep 2013 11:49:50 -0700 Tony Arcieri basc...@gmail.com wrote: We still haven't seen quantum computers built yet which can truly rival their conventional electronic brethren, especially if you look at it from a cost perspective. DWave computers are interesting from a novelty perspective, but not really ready to replace existing computers, even for highly specialized tasks like running Shor's algorithm. Nevertheless, if you've been following the trends in quantum computers over the last few years, they are getting larger, and DWave is an example of them moving out of the labs and turning into something you can buy. I wouldn't be surprised to see a large quantum computer built in the next two decades. DWave has never unambiguously shown their machine actually is a quantum computer, and even if it is, given its design it very specifically cannot run Shor's algorithm or anything like it. I'm unaware of a quantum computer of more than five qbits that has been demonstrated that can run Shor's algorithm, and that specific method, using a molecule with five distinct NMR peaks, cannot really be extended further. If you can find a reference to quantum computer with more qbits that can run Shor's algorithm that has been demonstrated in public, I would be very interested. (And yes, I'm aware of the two photon device that factored the number 21, though I believe the team used tricks to make that work -- opinions on whether that work could scale would be welcome of course.) Perry -- Perry E. Metzgerpe...@piermont.com ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Quantum Computers for Shor's Algorithm (was Re: Perfection versus Forward Secrecy)
On Sat, Sep 14, 2013 at 12:12 PM, Perry E. Metzger pe...@piermont.comwrote: DWave has never unambiguously shown their machine actually is a quantum computer There was some controversy about that a few months ago. In the end, my understanding is it netted out that it *is* a real (albeit limited) quantum computer: http://www.wired.com/wiredenterprise/2013/06/d-wave-quantum-computer-usc/ and even if it is, given its design it very specifically cannot run Shor's algorithm or anything like it. Sure, I never said it could ;) I also said that conventional computers can still outpace it. I'm certainly NOT saying, that in their present capacity, that DWave computers are any sort of threat to modern cryptography. But still, it goes to show that quantum computers are happening. Now it's just a question of whether a large computer capable of running Shor's algorithm is actually on the horizon, or if it falls into a category like nuclear fusion where work on it drags on indefinitely. -- Tony Arcieri ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Quantum Computers for Shor's Algorithm (was Re: Perfection versus Forward Secrecy)
On Sat, 14 Sep 2013 12:42:22 -0700 Tony Arcieri basc...@gmail.com wrote: Sure, I never said it could ;) I also said that conventional computers can still outpace it. I'm certainly NOT saying, that in their present capacity, that DWave computers are any sort of threat to modern cryptography. But still, it goes to show that quantum computers are happening. Given that the DWave design is totally unsuitable for Shor's algorithm, it seems to have no real bearing on the situation in either direction. To break 1024 bit keys (a minimum capability for a useful Shor machine, I'd say), you need several thousand qbits. I've not heard of a demonstration of more than a half dozen, and I've seen no progress on the topic in a while. It isn't like last year we could do six and the year before five and this year someone announced fifteen -- there have been no incremental improvements. It is of course possible that there's been secret research on this at NSA which has gotten far further, but I would expect that the manufacturing technology needed to do that would require a huge number of people to pull off, too many to keep quiet indefinitely. Perry -- Perry E. Metzgerpe...@piermont.com ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] Thoughts on hardware randomness sources
At 08:32 PM 9/13/2013, Jerry Leichter wrote: If by server you mean one of those things in a rack at Amazon or Google or Rackspace - power consumption, and its consequence, cooling - is *the* major issue these days. Also, the servers used in such data centers don't have multiple free USB inputs - they may not have any. More to the point, the servers in the data centers aren't going to let you plug things in to them, especially if you're just renting a virtual machine or cloud minutes and don't get to connect to the real hardware at all (which also means you're not going to be able to use disk drive timing.) A tablet computer has lots of sensors in it; even turning the cameras on at boot time and hashing the raw pixels should give you a reasonable chunk of entropy; you're not going to turn your virtual machine upside down and shake it like an Etch-A-Sketch. I realize it's possible for somebody to try to manipulate this, but I've always assumed that ethernet packet timing ought to give you some entropy even so, and even though with virtual machines you may only get quantized versions of interrupt times. Startup processes are probably going to include pinging a router and a name server, or at least they could if you wanted. ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] real random numbers
On 09/14/2013 03:29 PM, John Denker wrote: Things like clock skew are usually nothing but squish ... not reliably predictable, but also not reliably unpredictable. I'm not interested in squish, and I'm not interested in speculation about things that might be random. I see theoretical the enemy of the good here. The term squish is entertaining, but be careful that once you paint away with your broad brush that you don't dismiss engineering realities that matter. I can see there is an appeal to entropy sources that you can work back to some quantum origin, but even they will fail horribly if you don't build a larger system that is secure, and secure at some non-trivial radius. (How much Tempest-hardening are you going to do?) And once we have built such vaguely secure systems, why reject entropy sources within those systems, merely because they you think they look like squish? If there is a random component, why toss it out? You seem to respect using hashing to condition and stretch entropy--though why any existing hash shouldn't also fall to your squish generalization, I don't know. It seems that you would reject using a coin toss as a source of entropy because coins are not perfectly fair and there are biases in their results. So? You respect hashing, why not clean the output with a good hash? You dismiss things like clock skew, but when I start to imagine ways to defeat interrupt timing as an entropy source, your Johnson noise source also fails: by the time the adversary has enough information about what is going on inside the GHz-plus box to infer precise clock phase, precise interrupt timing, and how fast the CPU responds...they have also tapped into the code that is counting your Johnson. There are a lot of installed machines that can get useful entropy from existing sources, and it seems you would have the man who is dying of thirst die, because the water isn't pure enough. Certainly, if hardware manufacturers want to put dedicated entropy sources in machines, I approve, and I am even going to use rdrand as *part* of my random numbers, but in the mean time, give the poor servers a sip of entropy. (And bravo to Linux distributions that overruled the purist Linux maintainer who thought no entropy was better than poorly audited entropy, we are a lot more secure because of them.) -kb ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography
Re: [Cryptography] real random numbers
Your first two categories are talking about the distribution of entropy--we assume some unpredictability exists, and we want to quantify it in terms of bits of entropy per bit of output. That's a useful distinction to make, and as you said, if you can get even a little entropy per bit and know how much you're getting, you can get something very close to ideal random bits out. Your second two categories are talking about different kinds of sources--completely deterministic, or things that can have randomness but don't always. That leaves out sources that always have a particular amount of entropy (or at least are always expected to!). I'd say even the squish category can be useful in two ways: a. If you have sensible mechanisms for collecting entropy, they can't hurt and sometimes help. For example, if you sample an external clock, most of the time, the answer may be deterministic, but once in awhile, you may get some actual entropy, in the sense that the clock drift is sufficient that the sampled value could have one of two values, and an attacker can't know which. b. If you sample enough squishes, you may accumulate a lot of entropy. Some ring oscillator designs are built like this, hoping to occasionally sample the transition in value on one of the oscillators. The idea is that the rest of the behavior of the oscillators might possibly be predicted by an attacker, but what value gets read when you sample a value that's transitioning between a 0 and a 1 is really random, changed by thermal noise. I think the big problem with (b) is in quantifying the entropy you get. I also think that (b) describes a lot of what commonly gets collected by the OS and put into the entropy pool. --John ___ The cryptography mailing list cryptography@metzdowd.com http://www.metzdowd.com/mailman/listinfo/cryptography