A Tale of Shift Left, Shift Right, and MEV

Research, Thoughts on Security, , ,

The NIST Cybersecurity Framework’s five Functions are: Identify, Protect, Detect, Respond, and Recover. Within this framework “shift left” means that we shift our focus towards Identify and Protect, i.e. towards fixing issues before our system is deployed into production. Say, if you have a bug in your Apple Store/Google Play app, you’d prefer fixing it before you deploy it to the internet, or, similarly you prefer fixing the bug in your cryptocurrency contract before you deploy it in the wild (i.e. “on the mainnet”). While shift left makes lots of sense (let’s prevent the issue in the first place!), in many cases, preventing all issues is likely impossible.

Shifting Left

Firstly, it is important to state that preventing all issues is often impossible because of what safety practitioners call the “blunt end”. When writing code, practitioners, i.e. developers, are at what safety literature would call the “sharp end”: they are like airplane pilots, directly at the control of their craft. While it’s easy to imagine that pilots are in “full control”, in reality, airplane pilots depend on and are affected by many things: accurate weather reports, quality and amount of training, certification bodies, proper aircraft maintenance, production pressure from corporate, etc. These are what safety literature calls the “blunt end” — things that contribute to both success and failure, but are not directly visible/obvious. When the 10th plane drops out of the sky this same year, and Boeing keeps blaming the (now dead) pilots, you know it’s the blunt end you ought to be looking at.

What this boils down to in terms of shift left is that while developers are in “full control”, in reality, there is a large blunt end that they have little to no control over, and which may set up the conditions such that failure is sometimes inevitable, even with the most careful of developers. An easy example of this is previously unknown vulnerability classes: for example unknown issues with CPU cache timing side-channels in classical IT security, or flash loan attacks for cryptocurrency contracts, which are a relatively new phenomenon, allowing anyone to instantly obtain a large capital for a fraction of time to break assumptions about a contract, which can help attackers. Another slightly less obvious example is production pressure, where competing entities are known to be working on the same idea/system and releasing first is critical for business success. It is one thing to “ask for time”, it’s another thing when you gotta release or it was all for nothing. While techniques like fuzzing, formal verification, coding best practices, etc. can help, they are neither bulletproof, nor are they reasonable to expect to be done fully for all contracts.

Notice that given the nature of cryptocurrencies where “code is the law”, it seems counter-intuitive to think that there are factors at play other than the code itself. On the surface, the rules of the game seem set and all mistakes seem to happen to/with code, hence the culprit is always clear. In other words, it seems as if there is no “blunt end”. However, as with all interesting endeavors, in my view, blunt ends are abound, even if they only manifest themselves in code in subtle ways. For example, before overflow protection was enabled by default on the most used Ethereum compiler, Solidity, overflow issues were possible (e.g. issues where 255+1=0, oops!), and could have lead to attacks. In other words, changing the compiler defaults can prevent a type of attack. Similarly, the SMT Checker, shipped as part of Solidity and Remix, can also warn developers of potential mistakes. These systems change the landscape, preventing the “sharp end”, i.e. developers, from making mistakes.

Shifting Right

The IT security sector has realized for some time now that spending all their effort on shift left, i.e. identifying issues and preventing them pre-deployment, is not adequate. With some effort, time, and money being spent on detection, adequate response, and recovery, i.e. post-deployment and hence on the “right” side, the impact of attacks or unforeseen events can be minimized. This is incredibly important, because as per above, it is often impossible to prevent all attacks. From phishing attacks, novel attack classes, persistent attackers, to production pressures, some attacks will go through, and be successful — at which point the only question is how successful will they be, i.e. will they reach all (or any) of their intended goals, or will they be stopped early, minimizing the damage caused.

There are some interesting properties of spending proper time dealing with post-production incidents. Firstly, it allows the system designers and developers to understand what attack patterns are typical against systems. In classical IT security, e.g. honeypots are often used to understand attacker behavior. These are (often under-protected) systems deployed either on the public internet, or more interestingly, on the intranet(s) of organizations, to see if (and how) attackers try to attack it. These systems can help spot (novel) attack patterns, and attackers in general. Other information-sharing systems are also in-place in the traditional IT security world: in Germany the BSI is responsible to warn large enterprises of (novel/successful) attack patterns in the wild. Within the space of cryptocurrencies, attacks tend to be on the public chain, and can be observed directly both as they happen (unless they are in a so-called private mempool, more on this later), and retrospectively, thanks to the public ledger. There is plenty to learn from how others succeed, or fail.

MEV, and On-Chain Security-as-a-Service

With all the above out of the way, let’s get to the meat of this article: how behavior patterns in the cryptocurrency world that are often cited as undesirable are actually doing a form of shift right. This inadvertent shift right is something I believe we could exploit to improve the safety of the chain overall.

Generalized frontrunning is a technique that consists of people listening to the set of sent, but unconfirmed transactions on the chain, and seeing if copying the same transaction themselves would make them money. They then submit the same transaction with a higher fee, giving more fee to the miners who confirm transactions. This leads to both transactions being executed, but theirs being executed first (and the 2nd likely failing), thereby giving them the money. Clearly, miners (who confirm transactions) could always do such frontrunning, and could do it much easier, since they decide who is going to be first, so they can make themselves first without paying a higher (or any) fee. Transaction frontrunning is slightly more complicated: it tries to both front- and back-run a transaction (also called sandwiching), manipulating the price before the transaction is executed (the “front” part) and then selling the excess (the “back” part). Generalized and transaction frontrunning attacks all fit under the umbrella of transaction reordering attacks and are also called Miner Extractable Value, or MEV for short. MEV is big business, with 8 million USD extracted in the past 30 days on the Ethereum chain, as of writing.

When you perform generalized frontrunning, you do two things: (1) you constantly check for potential value to be extracted, such an attack would do, and (2) you make sure your transaction runs first (hence the name “front-run”) that extracts the value to get the value for yourself. This clearly maps to the post-deployment part of the Cybersecurity Framework of NIST: detect & respond. In other words, it would actually make sense to constantly run a bot on-chain to detect and front run attacks against your own contracts. In a sense, the frontrunners are operating what in the traditional IT security world would be called a SOC — a Security Operations Center that is constantly on the lookout for attacks.

In this sense, MEV is not undesired, but a price to pay to have a non-stop security system. Of course, it is not a given that MEV bot operators extracting MEV will give back your money for the attack they front-ran. However, I posit that it’s much more likely that you’ll get your money back from them than the attackers. For example, at the MEV day for Devconnect, there was a discussion I participated in where a large MEV bot operator mentioned that they were willing to give back the funds of attacks they front-ran for a 10% fee. I expect that large organizations will in the future run their own MEV bots, and medium-sized ones will either simply pay the fee (10% or whatever it will be) when an attack happens, or perhaps pay a monthly fee for the protection service provided.

(Footnote: there is not only bad MEV. For example, backrunning (i.e. running after) liquidation events are incentivized to happen, and are considered normal for the system to run as designed.)

Current Solutions to MEV, and their Impact

There are two obvious ways to deal with the MEV issue: creating economic incentives to reduce their impact, and hiding pending transactions. Let’s to to get into both.

One way to deal with MEV is by not making pending transaction visible. So-called private mempools are operated by some miners (i.e. confirmers of transactions) that don’t allow the public to see the not-yet confirmed transactions. This is advantageous because it means that attackers cannot run clients to frontrun/sandwich/etc. pending transactions sent to these private mempools. In a way, private mempools are kind of a trusted environment, where you can be safe from people trying to extract MEV. However, it also means that malicious behavior submitted to these private pools cannot be front-run. Since these private pools are trust-based, it would be possible to allow some trusted parties to observe these private pools and frontrun attacks. Rather than being purely trust-based, it is possible that some sort of incentives-based system could also be set up, which seems like an open question/potential for improvement. This would be similar to what Flashbots does. Flashbots tries to use economic incentives through an open marketplace to make MEV extraction less disruptive to the ecosystem (among other benefits). However, it does not work for private mempools, since, well, they are private, and Flashbots requires publicly visible unconfirmed transactions to be able to work, due to its open marketplace concept.

MEV is not only big business, it’s also wasteful: often, the original transaction will fail, but still execute, consuming computational resources and burning money. MEV systems trying to outbid one another for the prize are effectively participating in an all-pay auction, i.e. everyone pays for the good but only one gets it. Thankfully Flashbots solves this by running an off-chain auction that is not all-pay. However, once we move from proof-of-work to proof-of-stake (i.e. to ETH 2.0), the attack transaction can be directly proposed to the next validator, skipping the public pool entirely. Essentially, the attacker can choose be in the private mempool by default (with some caveats, see Proposer-Builder Separation, and MEV-Boost), rather than by exception.

Conclusions

While Miner Extractable Value (MEV) seems on the surface to be an undesirable side-effect of the combination of current cryptocurrency protocols and the public nature of digital ledgers, it can also be seen as a feature. This feature could allow one to perform detection & response to attacks, allowing legitimate entities to mitigate the impact of attacks on contracts post-deployment. There is some evidence that some people running MEV extraction bots are already effectively providing a form of post-deployment attack-mitigation-as-a-service for a fee. However, as usual, preventative measures are both more reliable and cheaper, and should be the focus of most efforts against attacks.

Proof Traces for SAT solvers

Uncategorized

If you have a look at a modern SAT solver, say kissat or CryptoMiniSat, you’ll see that they have well over 20k lines of code. Reviewing that for bugs is basically impossible. So how can you trust that the solver is giving a correct result? Well, if the result is that the input formula is satisfiable, it’s quite easy — we can simply substitute the solution (all modern SAT solvers provide a satisfying assignment) to the formula, and it should satisfy each constraint. This is really fast to do, linear in the size of the formula.

However, if the solver gives a result that the formula is unsatisfiable (UNSAT), what do we do? Well, we need some form of assurance. One would be to check the code, but again, that’s just too much work. We could run it with another SAT solver, but there have been cases of the same bug being in more than one solver. Actually, this bug seems to have been a conceptual mistake copy-pasted, quite weird. In fact, there is a well-known GCC version that will miscompile most MiniSat-based SAT solvers such that they report wrong results(!). We could also do some fuzzing, but do you want to sit on an airplane knowing its flight control software was fuzz-tested for a full 10 minutes to “meet” compliance requirements? No. We need strong assurances. So, let’s ask the SAT solver to produce a proof for the UNSAT result! This is what proof traces are all about.

The Binary Resolution Operator

At its core, a proof trace essentially records the set of operations that must be performed by a system to arrive at the 0=1 equation (i.e. at obvious nonsense), starting with the input equations. Let’s see a simple example. Let’s have our input formula (so-called CNF) be:

 x1 V  x2 = TRUE
 x1 V -x2 = TRUE
-x1 V  x2 = TRUE
-x1 V -x2 = TRUE

Where x1 and x2 are boolean variables. Actually, let’s make this a bit easier to read by a computer:

 1  2 0
 1 -2 0
-1  2 0
-1 -2 0

This the same as the above, but closer to the standard DIMACS format. Each line is called a clause, each line is made up of literals like 1, -1, 2, and -2, and each clause is terminated by a “0”, marking the end of the clause.

The above CNF has no solution to its set of clauses: no matter what we set x1 and x2, at least one of the clauses is not satisfied. OK, so how do I prove that there is no solution? Well, the easiest way to do it is through the binary resolution operator. The binary resolution basically says that if two clauses have one literal that’s inverted, such as “x1” and “-x1”, then we can do this: (A OR x1) RESOLVE (B OR -x1) => A OR B, where “A” and “B” are any set of literals. So, e.g. “x1 V x2 V -x3” resolved with “x3 V x4” becomes “x1 V x2 V x4”. Nice, so let’s try to prove that the above problem is unsatisfiable:

 1  2 0 RESOLVE  1 -2 0 =>  1 0
-1  2 0 RESOLVE -1 -2 0 => -1 0
 1  0   RESOLVE -1 0    =>  0

In essence, we resolved 2 clauses to prove that x1 is TRUE, then resolved another two to prove that x1 is FALSE, and then we resolved “x1” with “-x1” to arrive at EMPTY=TRUE, i.e. 0=1, which is obviously nonsense.

The RUP notation

In the so-called RUP notation (“Reverse Unit Propagation”), the above is simply written as:

a  1 0
a -1 0
a 0

Basically, we simply write down each derived clause one after the other as “added” (=”a”). It’s called Reverse Unit Propagation, because it’s possible to prove each of these clauses to be correct, by starting from the top, and performing simple propagation [where propagation is substitution of the values into the equations, and checking if any variable is forced to a value] . Let me explain. Say we want to prove “1 0” is correct. Then, we set all the literals’ inverted values, here, “x1=FALSE”, and propagate on the original clauses plus any clause that’s above “1 0” (and hence has been proven to be correct). If we set x1=FALSE, then the original formula’s clause “1 2 0” will propagate x2=TRUE and then the original formula’s “1 -2 0” conflicts immediately (all its literals are now FALSE). Hence, propagation lead to a conflict! This means that indeed, “1 0” must hold.

What’s important to realize here is that each of the lines of the proof must be checkable by simple propagation. Let’s say we have this set of constraints:

8 9  2  3 0
8 9 -2  3 0
8 9  4 -3 0
8 9 -4 -3 0

Given these constraints, it is fairly easy to check that “8 9 0” holds. However, we can’t just write into the proof that “a 8 9 0”. Because for that, the propagation check as explained above should succeed: when I set both x8=FALSE and x9=FALSE, I should get the a conflict. But I don’t! Instead when I substitute x8=FALSE and x9=FALSE in to the equations, I get this reduced formula:

 2  3 0
-2  3 0
 4 -3 0
-4 -3 0

Which does not force any variable to any value (i.e. they don’t propagate anything) and none of them are the empty clause, so I’m stuck. The proof verifiers literally will say they are “stuck”. So, how I can prove “x8 OR x9”? Let’s do it like this:

a 8 9 -3 0
a 8 9  3 0
a 8 9 0

Let’s see how this works. First, let’s try to validate “8 9 -3” by setting v8=FALSE, v9=FALSE, v3=TRUE. Then the original constraint “8 9 4 -3 0” will propagate v4=TRUE, and the original constraint “8 9 -4 -3 0” will be immediately conflicting (all its literals are now FALSE). So simple propagation fails, hence the clause “8 9 -3” must be correct. While this is not the goal (that’s “8 9 0”), we are getting there. We now prove “8 9 3” in a similar fashion, and then we prove “8 9 0” in the same fashion. Easy!

Faster Proof Verification: DRUP

While RUP is indeed a good format, we don’t always need all the constraints to be in the active set in order to prove UNSAT. Let’s quickly re-visit this CNF, and try to prove “8 9 0”:

8 9  2  3 0
8 9 -2  3 0
8 9  4 -3 0
8 9 -4 -3 0

In order to reduce memory load, let’s delete all the clauses we don’t need after we used them! These steps are going to be noted with “d” (=”delete”) instead of “a” (“add”):

a 8 9 -3 0
d 8 9  4 -3 0
d 8 9 -4 -3 0
a 8 9  3 0
d 8 9 -2 3 0
d 8 9  2 3 0
a 8 9  0
d 8 9 -3 0
d 8 9  3 0

NICE! So we first derived “8 9 -3 0” and then removed both of the clauses we used to create it, they are not needed anymore! Begone “8 9 4 -3 0” and “8 9 -4 -3 0” ! Same thing with “8 9 3 0”. Then, once we derived “8 9 0”, we even deleted the intermediary constraints “8 9 -3 0” and “8 9 3 0” we previously derived. Begone! Not needed anymore, all we need is “8 9 0 “.

Extended resolution for proofs

Okay, I’m gonna cut this part short, mostly because I am terrible at this, but we have talk about the elephant in the room: RAT, or, more beautifully, Resolution Asymmetric Tautologies. So, what is it? Essentially, it allows you to declare new variables in your proof and use them later. This is absolutely amazing, because with this, you can provide so-called extended resolution proofs (warning: 1968 research paper), which allows us to express certain proofs more compactly, in fact, exponentially more compactly!

Here’s how to do it. Let’s say you want to use the definition “x1 OR v2 = x99” in your proof, because it would make your proof smaller. This can happen, for example, because your constraints often have “x1 OR x2” in them, and now you can replace all of those parts with simply “x99”. Well then, you write this into your proof file:

a -99 2 1 0
a  99 -1 0
a  99 -2 0

Notice that this is just a simple OR gate definition. The only thing RAT requires is that (1) the first variable is the one that’s being defined and (2) the introduced variable must not be part of our input formula — so our input formula here should have at most 98 variables. That’s it. Now you can use x99 anywhere you like, as long as it makes sense. For example, you could take a clause “1 2 3 4 0” and convert it to “99 3 4 0”:

a  99 3 4 0
d  1 2 3 4 0

Notice that “99 3 4 0” is simply a RUP proof at this point: setting 99=FALSE, 3=FALSE, 4=FALSE will immediately propagate x1=FALSE, x2=FALSE (due to the clauses “99 -1 0” and “99 -2 0”), and x1,x2,x3, and x4 all being FALSE will immediately trigger a conflict for clause “1 2 3 4 0”, so we have the RUP property and we are fine. Then we can delete “1 2 3 4 0”, we don’t need it no more. Notice that I can’t write the reverse:

d  1 2 3 4 0
a  99 3 4 0

If I wrote that, “a 99 3 4 0” will not work, since it relies on “1 2 3 4 0” but we already deleted that, oops!

Besides the above trivial example with the OR gate, extended resolution can simulate, in polynomial time, some algorithms that are extremely hard (read: exponentially hard) for normal resolution to express. This means that using RAT, we can express even complicated mathematical concepts in this extremely simple system, for example in this case, pseudo-boolean reasoning.

What the solver does vs. what the proof checker does

So let’s say you implement DRAT proof trace into your SAT solver and your proofs are always verified. All good! Not so fast. One day you decide to check the proof that DRAT recovered, and to your absolute horror, you realize that DRAT thinks your proof to be substantially different than what the SAT solver thought it was! How can that be?

Well, the DRAT checker recovered a proof. It’s a valid proof, and it can be recovered given the set of clauses that haven’t been deleted using the “d” operator, as above. But notice that we never told the proof checker which set of clauses we resolved on to get to a clause! It just kind of figured it out by itself. Now, we can look at this and say, this is great because (1) I don’t have to do hand-holding of the proof checker, and really interestingly, (2) the proof checker could actually recover different proofs from the same proof trace(!). Ambiguity can be useful. Well, that’s great, but what if I really-really want to know what the SAT solver actually did?

Enter FRAT. This proof format is super-close to DRAT, with the following changes. Firstly, FRAT numbers each clause with a unique clause ID. This is needed because the same clause can be in the solver’s memory more than once, and we need to be able to distinguish them from each other — they may have been derived in completely different ways! Secondly, FRAT allows an optional hint at how the clause was derived. Let me give a hands-on example, it should be quite clear. As a reminder, here is the input formula we had before:

 1  2 0
 1 -2 0
-1 -2 0
-1  2 0

And here is the DRAT proof:

a  1 0
a -1 0
a 0

Now, for the FRAT proof:

o 10  1  2 0         -- input clause, let's set its ID to 10
o 11  1 -2 0         -- input clause, let's set its ID to 11
o 12 -1  2 0         
o 13 -1 -2 0         -- last input clause
a 14  1  0 l 10 11 0 -- first resolvent
d 10  1  2 0 
d 11  1 -2 0 
a 15 -1  0 l 12 13 0 -- second resolvent
d 12 -1  2 0 
d 13 -1 -2 0 
a 16 0 l 14 15 0     -- empty clause! It's UNSAT!
f 14  1  0           -- finishup from down here
f 15 -1  0
f 16 0

Okkkay. So… the original clauses are part of the proof trace now, they start with “o”. And each clause has an ID, after o/a/d/f. This can be any number, but they must be non-clashing (obviously). Furthermore, we are allowed to have an “l” after the closing “0” when adding a clause — these are the clause IDs we had to resolve to arrive at the new clause. So, e.g. IDs 10 and 11, i.e. “1 2 0” and “1 -2 0” must resolve to “1 0” (clause ID 14). And finally, we must “finalize” all our clauses, i.e. we must account for all clauses remaining in memory, with the “f” command.

Firstly, finalization means that in case your SAT solver forgot to delete a clause in the proof trace, but deleted it from memory, finalization will fail, because the solver will not finalize that clause, and so the checker will complain. In essence, finalization forces the hand of the SAT solver writer to make sure that the “view of the world” from the perspective of the SAT solver and that of the proof verifier match when it comes to clauses in memory. This is not the case for DRAT — the proof verifier there could have millions of clauses more in memory than the SAT solver :S

Secondly, the “l” notation means that we can now be precise about what clauses were actually resolved by the SAT solver to arrive at a new clause. However, this is only a “can” — adding the “l” is not necessary, and of course can be stripped. This means that if we want to, we can have ambiguity, or if we want to, we can have precision. The best of both worlds!

Trusting trust

Proofs are cool, and are very important so we can be sure our results are correct. But… are we sure they are correct? Well, not so much. You see, the proof verifiers themselves are not verified. Say, frat-rs will say “VERIFIED” on a FRAT proof, but… frat-rs itself is over 5000 lines of Rust. And before you think I’m really stretching things here, let me just remind you that I personally found a segmentation fault in drat-trim, the official DRAT proof verifier, used at the SAT competition. (I actually found the bug accidentally while fuzzing my own SAT solver.)

So, what can we do? Well, we need another verifier. I’m not kidding. Basically, frat-rs can take a proof that the SAT solver provides, and translate it to a fully annotated, clean proof with all the deletions at their earliest possible points (see this hack). We can then write a slow, but verified proof checker, that can do the trivial checking that the resolutions are indeed correct, reaching the empty clause (i.e. 0=1). The pipeline would then be:

Airplane software code + compliance requirements ->SAT_query
solution, proof  = SAT_solver(SAT_query)
solution, clean_proof = frat-rs(SAT_query, solution, proof)
solution = verified_proof_checker(SAT_query, solution, clean_proof)

What this allows us to do is that we don’t need to trust the SAT_solver or frat-rs. We can treat them as black boxes. All we need to trust is that the verified_proof_checker is correct: given the SAT_query, the solution can be trusted. In this sense, clean_proof is simply a hint for the verified_proof_checker to do its job. Notice that I completely skipped how SAT_query can be trusted — that’s an entirely different can of worms :D

In case you want to use such a verified proof checker, there is one in lean4, available here, and one for acl2, available here. The non-verified tool to “elaborate” simpler (e.g. DRAT) proofs into annotated (LRAT) proofs, you can use frat-rs.

Conclusions

So, we talked a lot about proof traces, proof formats, and how to trust the solution. In some cases, this is extremely important — some software and hardware is verified in this way that controls trains and airplanes, and industrial systems (think: nuclear reactors, rockets… Ariane 5 crash anyone?). We certainly don’t want any bugs in there.

What I didn’t talk about, is that proof traces do something else, as well: they allow us to understand how SAT solvers work. You see, while the proof is written in the forward fashion — the same direction that SAT solvers work, — the proof actually only makes sense in the reverse direction. We always start at the empty clause, at the end of the proof, and walk backwards to see how the empty clause got proven! So, if you think about it, proofs can be used to look at SAT solvers in the reverse direction. And that’s super-exciting. For example, how much work did the SAT solver really have to do to reach the empty clause? How much complete nonsense did it do that was utterly useless to prove unsatsifiability? How many times did it forget proof fragmets that it later had to re-learn? Or, what is the minimum memory footprint that its produced proof could fit in? And if we take advantage of DRAT’s and FRAT’s ambiguity: is there a proof with fewer steps it could have produced given the total same set of resolvents? In my opinion, there is a whole wealth of data in proof traces that we should do research on to learn more about SAT solvers.

(PS: Tracecheck is one of the earliest proof trace systems. I didn’t want to mention it because it’s not as relevant anymore as the above, and it would have made the story a tad more confusing)

CMSGen, a Fast Uniform-Like Sampler

Research, SAT, Tools,

Uniform sampling is a problem where you are given a solution space and you have to present solutions uniformly, at random. In some cases, this is quite simple, say, for the lotto. Just pick 5 random numbers from a box and we are done! For the lotto the solution space is very easy to generate. However, when there are constraints on the solution space, it starts to get tricky.

Let’s say that I have a function I want to test, but the input to the function has some real-world constraints like e.g. the 1st parameter must be larger than the second, the 2nd parameter must be divisible by the 3rd etc. If I want to test that this function operates correctly, one way to do it is to generate 100 uniformly random inputs that don’t violate any of the constraints, run the function, and see if all is OK. For this, I need a fast way of generating uniform samples given the constraints on the solution space.

Sampler speed vs. accuracy

There have been many samplers proposed in the literature. I personally have worked on one called UniGen, a guaranteed approximate probabilistic sampler, meaning that it’ll give approximately uniform samples most of the time, and we have a proof to back this up. It’s a great sampler and will work very fast on many instances. However, for really complex solution spaces, it can have trouble. Say, you want to generate interesting test inputs for your deep learning algorithm. Deep neural networks tend to be extremely complex when translated to binary constraints, so UniGen will likely not be fast enough. It would give very good quality samples (i.e. properly uniform samples), but if it’s too slow, we may want to exchange quality of samples for speed of generation.

There are two well-known samplers that are supposed to generate uniform samples on complex solution spaces, QuickSampler (code), and STS (code), but give no guarantees, let’s call these “uniform-like” samplers. Unfortunately, the paper by Chakraborty et al and its resulting code Barbarik showed that these uniform-like samplers are highly non-uniform. Barbarik is a pretty neat idea that basically constructs solution spaces with known solution distributions and then asks the sampler to generate uniform samples. Knowing the solution space, Barbarik can then verify how non-uniform the sampler is. Imagine having a box with 1000 balls, half of them blue and the other half green. Now if I ask the sampler to give me 50 balls at random, and all of them are green, I’d be a bit surprised to say the least. It’d be like the 5-lottery having the same numbers 3 times in a row. Possible, but… not very likely. If I do this experiment 100 times, and I always get 50 green balls, it’s fair to conclude that the sampler is not uniform.

Our new uniform-like sampler, CMSGen

Given an effective tester, Barbarik, we (Priyanka Golia, Sourav Chakraborty, Kuldeep S. Meel, and myself) thought perhaps we can follow the nowadays very successful test-driven development (TDD) methodology. All we have to do is to make our sampler pass the test of Barbarik, while being at least as fast as STS/QuickSampler, and we’ll be good to go. In fact, given Barbarik, it only took about a week of playing around with CryptoMiniSat’s options to beat both STS and QuickSampler in both accuracy and speed. This speaks volumes to how important it is to have a robust, reliable, and fast testing framework that can give immediate feedback about the quality of samples generated.

Our new uniform-like sampler, based on CryptoMiniSat, is called CMSGen (research paper here), and effectively takes CyrptoMiniSat and applies the following set of changes, through pre-set command line options:

  • Pick polarities at random. Normally, SAT solvers use polarity caching scheme, but of course we want uniform samples over all the search space, so we need to pick polarities at random.
  • Branch on variables at random. Normally, SAT solver branch on variables that will most likely lead to a conflict to maximize search efficiency (the VSIDS heuristic). However, we want to explore the solution space as evenly as possible, and so we want to approach the solution space from as many angles as possible. If you think about the search space as an N-dimensional binary cube, then we are trying to approach this cube as any ways as possible.
  • Turn off all pre- and inprocessing. Pre and inprocessing in SAT solvers are used to minimize the instance, transforming it into something easier to solve, e.g. through Bounded Variable Elimination. We later reconstruct a viable solution based on the solution to the transformed instance. However, the transformed instance may (and often will!) have a very different solution space. We cannot have that, so we must turn this off. To be fair, some pre- and inprocessing could be left intact, e.g. subsumption and self-subsuming elimination, perhaps a future paper :)
  • Restart at static intervals. Restarts are nowadays often dynamic in modern SAT solvers, or even if not dynamic, then follow a non-regular pattern. However, that could disturb how we find solution. Imagine, let’s say that solutions with variable A set to TRUE are very easy to find, but solutions with FALSE are very hard to find. What will happen? Well, in restarts where A was randomly set to TRUE, we’ll always quickly find a solution and output it. But for restarts when A was randomly set to FALSE, the system would struggle to find a solution, and after some conflicts, it will simply restart into a status where hopefully A is set to TRUE, and it can find a solution again. It is quite clear to see that this will lead to serious issues with sampling quality. Hence, we set an adjustable but static restart interval of 100 conflicts, with higher values typically leading to more uniform samples.

Performance of CMSGen

Performance of the system is on the ridiculous scale in comparison with other samplers:

When it comes to 2-wise coverage, i.e. the quality of samples, the data speaks for itself (note, UniGen is missing here because it was too slow):

Note that between STS and QuickSampler, STS is both the more uniform sampler, but also the slower one. CMSGen overcomes this limitation: it’s both faster than QuickSampler, and more uniform than STS.

And of course, the Barbarik tester gives “Accept” on CMSGen much more often than on STS or QuickSampler:

Conclusion

If you need non-guaranteed uniform but fast sampling, I’d go and try out CMSGen. It’s really a completely different beast. It’s not a guaranteed uniform sampler, but it’s incredibly effective. In fact, it’s so effective and works so well, it took me a full year to figure out how best to generate problems for it where it wouldn’t be uniform. But that’s another paper, and another blog post! In the meantime, the sampler is here, go check it out!

CryptoMiniSat 5.8.0 Released

Development, Research, SAT, , ,

After many months of work, CryptoMiniSat 5.8.0 has been released. In this post I’ll go through the most important changes, and how they helped the solver to be faster and win a few awards, among them 1st place at the SAT incremental track, 3rd place SAT Main track, and 2nd&3d place in the SMT BitVector tracks together with the STP and MinkeyRink solvers.

Gauss-Jordan Elimination

First and foremost, Gauss-Jordan elimination at all levels of the search is now enabled by default. This is thanks to the work detailed in the CAV 2020 paper (video here). The gist of the paper is that we take advantage of the bit-packed matrix and some clever bit field filters to quickly check whether an XOR constraint is propagating, conflicting, or neither. This, and a variety of other improvements lead to about 3-10x speedup for the Gauss-Jordan elimination procedure.

With this speedup, the overhead is quite small, and we enable G-J elimination at all times now. However, there are still limits on the size of the matrix, the number of matrices, and we disable it if it doesn’t seem to improve performance.

As a bit of reflection: our original paper with Nohl and Castelluccia on CryptoMiniSat, featuring Gauss-Jordan elimination at all levels of the search tree was published at SAT 2009. It took about 11 years of work, and in particular the work of Han and Jiang to get to this point, but we finally arrived. The difference is day and night.

Target Phases

This one is really cool, and it’s in CaDiCaL (direct code link here) by Armin Biere, description here (on page 8). If you look at the SAT Race of 2019, you will see that CaDiCaL solved a lot more satisfiable problems than any other solver. If you dig deep enough, you’ll see it’s because of target phases.

Basically, target phases are a variation of phase saving, but instead of saving the phase all the time when backtracking, it only saves it when backtracking from a depth that’s longer than anything seen before. Furthermore, it is doing more than just this: sometimes, it picks only TRUE, and sometimes it picks only FALSE phase. To spice it up, you can keep “local deepest” and “global deepest” if you like, and even pick inverted phases.

It’s pretty self-explanatory if you read this code (basically, just switching between normal, target, inverted, fixed FALSE, fixed TRUE phases) and it helps tremendously. If you look at the graphs of the SAT 2020 competition results (side no. 19 here) you will see a bunch of solvers being way ahead of the competition. That’s target phases right there.

CCAnr Local Search Solver

CryptoMiniSat gained a new local search solver, CCAnr (paper here) and it’s now the default. This is a local search solver by Shaowei Cai who very kindly let me add his solver to CryptoMiniSat and allowed me to add him as an author to the version of CryptoMiniSat that participated in the SAT competition. It’s a local search solver, so it can only solve satisfiable instances, and does so by always working on a full solution candidate that it tries to “massage” into a full solution.

Within CryptoMiniSat, CCAnr takes the starting candidate solution from the phases inside the CDCL solver, and tries to extend it to fit all the clauses. If it finds a satisfying assignment, this is emitted as a result. If it doesn’t, the best candidate solution (the one that satisfies the most clauses) is saved into the CDCL phase and is later used in the CDCL solver. Furthermore, some statistics during the local search phase are saved and then injected into the variable branching heuristics of the CDCL solver, see code here.

Hybrid Variable Branching

Variable branching in CryptoMiniSat has always been a mix of VSIDS (Variable State Independent Decaying Sum, paper here) and Maple (multi-arm bandit based, paper here) heuristics. However, both Maple and VSIDS have a bunch of internal parameters that work best for one, or for another type of SAT problem.

To go around the issue of trying to find a single optimal value for all, CryptoMiniSat now uses a combination of different configurations that is parsed from the command line, such as: “maple1 + maple2 + vsids2 + maple1 + maple2 + vsids1” that allows different configurations for both Maple and VSIDS (v1 and v2 for both) to be configured and used, right from the command line. This configuration system allows for a wider variety of problems to be efficiently solved.

Final Remarks

CryptoMiniSat is now used in many systems. It is the default SAT solver in:

I think the above, especially given their track record of achieving high performance in their respective fields, show that CryptoMiniSat is indeed a well-performing and reliable workhorse. This is thanks to many people, including, but not limited to, Kuldeep Meel, Kian Ming A. Chai, Trevor Hansen, Arijit Shaw, Dan Liew, Andrew V. Jones, Daniel Fremont, Martin Hořeňovský, and others who have all contributed pull requests and valuable feedback. Thanks!

As always, let me know if you have any feedback regarding the solver. You can create a GitHub issue here, and pull request here. I am always interested in new use-cases and I am happy to help integrate it into new systems.

Weighted to Unweighted Counting and Sampling

Uncategorized

This blog post is to promote the tool here (paper here) that converts weighted counting and sampling to their unweighted counterparts. It turns out that weighted counting and sampling is a thing. It’s basically a conjunctive normal form formula, a CNF, with weights attached to each variable. One must count the number of solutions to the CNF, where the weight of each solution is not necessarily 1, but weighted according to the polarity of each variable in the solution. One such formula could look like:

p cnf 2 1
c ind 1 2 0
1 2 0
w 1 0.9

This basically says that we have 2 variables, a single constraint (“v1 or v2”) where a solution with v1=True is worth 0.9 points, but a solution with v1=False is only worth 0.1 points. Or, if we are sampling, we want to bias our samples 9-to-1 for solutions that have v1=True.

Let’s go through this CNF. It has has 3 solutions, with its associated points:

solution    points
 1  2         0.9
 1 -2         0.9
-1  2         0.1

Total points: 1.9

The issue is that we have great tools for doing unweighted counting and sampling but few tools for doing weighted counting and sampling.

Converting Weighted CNFs to Unweighted CNFs

The solution is to use a converter to make an unweighted formula from a weighted formula, and then either divide the number of solutions of the unweighted formula (for counting) or simply cut off variables from the samples (for sampling). This work was done by Chakraborty et al. and presented at IJCAI’15. The basic idea is that we can approximate the weight with a binary representation and encode that in CNF. For example, 0.26 can be approximated with 1/4. If this weight is attached to v1, and there are 10 variables in the original CNF, we generate the extra set of clauses:

12 -1 0
11 -1 0
12 -11 1 0

This CNF has exactly 1 solution when v1=True and 3 solutions when v1=False. So the number of positive solutions is 1/4 of all solutions, i.e. 0.25, close to the requested 0.26.

The above set of clauses is generated from chain formulas that can be easily generated, and are only polynomal in size of the binary precision required. For k-bit precision (above: 2 bit precision), we need only k extra variables and k+1 clauses. See the paper for details, but it’s sufficient to say that it’s easy to generate and, looking through a few examples, quite easy to understand.

Example: Weighted Counting

Let’s do the full run, using our favorite counter, approxmc:

$ cat my.cnf
p cnf 2 1
c ind 1 2 0
1 2 0
w 1 0.9

$ ./weighted_to_unweighted.py my.cnf my-unweighted.cnf
Orig vars: 2       Added vars: 7
The resulting count you have to divide by: 2**7

$ ./approxmc my-unweighted.cnf
[…]
[appmc] FINISHED AppMC T: 0.01 s
[appmc] Number of solutions is: 62*2**2

$ python
>>> 62*2**2/2**7
1.9375

So, we got pretty close to the correct weighted count of 1.9. The reason for the discrepancy is because we approximated 0.9 with a 7-bit value, namely 115/2**7, i.e. 0.8984375, and we used an approximate counter. With an exact counter such as sharpSAT, we would have got:

$ ./sharpSAT my-unweighted.cnf
[...]
# solutions
243
# END

$ python
>>> 243/2**7
1.8984375

Which is pretty close to the correct weighted count and in fact corresponds to the expected error of 0.9-0.8984375=0.0015625:

$ python
>>> from decimal import Decimal
>>> Decimal("1.8984375")+\
    Decimal("0.9")-Decimal("115")/Decimal(2**7)
Decimal('1.9000000')

Example: Weighted Sampling

Weighted-to-unweighted sampling mimics counting. First, we transform our formula:

$ ./weighted_to_unweighted.py my.cnf my-unweighted.cnf
Orig vars: 2       Added vars: 7
The resulting count you have to divide by: 2**7

Then we sample it, e.g. using our favorite sampler, UniGen (research paper):

$ ./unigen my-unweighted.cnf --sampleout samples --samples 1000 > /dev/null

The samples now contain extra variables that were used for the translation, so we simply cut them off to the original sampling set size, here, 2:

 $ cut -d " " -f 1-2 samples | sort | uniq -c
 55  -1  2
 478  1 -2
 490  1  2

And we are done! The sampling worked as expected: we get proportionally more samples where v1 is True than when it’s False.

Conclusions

Weighted counting is non-trivial but can be approximated using chain formulas, blowing up the original equation in only a minor way. This allows us to use well-optimized unweighted systems to perform the counting and sampling, and then trivially transform the output of these tools to match our original weighted problem.