Pick rounds with the most inputs available to mix first #2278

UdjinM6 · 2018-09-09T20:50:16Z

After discussion in previous PRs I can't stop thinking that maybe we shouldn't introduce some arbitrary variance and fight the consequences of mixing with "no duplicate txids" rule and instead we could use it to our benefit somehow and the variance will appear naturally. So here is the idea: scan all coins with all rounds first to figure out how many inputs available for each of them and pick the one that has max inputs. For example, let's say that we have <R, N1>, <R+1, N2> (where R is some number of rounds; N1, N2 is some number of available inputs and N1 > N2). We mix M inputs, so after this we'll have something from <R, N1-M> to <R, N1> (it depends because previous txes could have more than 1 output) for R and <R+1, N2+1> for R+1. Note, that this way we consumed up to M inputs with unique txids for R rounds but we only added 1 input to R+1 rounds (because of "no duplicate txids" rule). What's interesting is that 1) there is already a variance for R which comes from previous txes and 2) if N2+1 is greater than N1-M then we kind of shuffled them and R+1 now has higher mixing priority (hence even more variance).

I wonder if this makes any sense or if I'm missing smth :)

See ~~6cebae0~~ eddb4d7 for the implementation, ignore all the commits from previous PRs included here, will rebase later.

UPDATE1: we can include inputs from the very next round too if there are some, this should bring more variance I think, see ~~541f1f3~~ 7c632e7f3. This will however "squash" them into the same single tx which will decrease mixing speed due to "no same txid" rule.

UPDATE2: reverted 7c632e7f3, it tends to have the same issue with lower and lower number of inputs on higher rounds which could probably make cluster analysis easier.

UPDATE3: also randomly skip some inputs and try to add up to PRIVATESEND_ENTRY_MAX_SIZE of every needed denomination to balance out "no duplicate txids" rule and make it harder to analyse.

PastaPastaPasta · 2018-09-09T21:42:27Z

Does this mean that only one round is / will be mixed at once? For ex a 2 round input won't be able to be submitted with a 3 round input in the same mix?

UdjinM6 · 2018-09-10T07:14:49Z

@PastaPastaPasta yes, this was always the case and it is the same here too, not touching this part.

Antti-Kaikkonen · 2018-09-11T14:08:21Z

I think this solution might increase mixing throughput, but it could take longer to get a positive mixing balance, especially if the user has hundreds of inputs.

What is the reason for trying to select inputs with the same number of rounds? Have you considered a solution like this?

UdjinM6 · 2018-09-11T14:19:49Z

@Antti-Kaikkonen

What is the reason for trying to select inputs with the same number of rounds?

Because that's the shortest path to get some PS balance. When we calculate the number of rounds each input was mixed, we pick the worst possible estimation to make sure that each input was mixed at least required number of times. So when you pick 2 inputs inputs, one with 2 rounds and the other one with 5 rounds to mix in the same session, they'll end up both having min(2,5)+1=3 rounds.

Antti-Kaikkonen · 2018-09-11T23:15:55Z

@UdjinM6 Oh that's right. I didn't think about that before.

…ix first

UdjinM6 · 2018-09-15T19:07:05Z

Rebased.

Note: this is an alternative to #2275

InhumanPerfection · 2018-09-20T12:52:02Z

I think "no duplicate txids" rule making it easier to analyse. Not easy to explain this only with words so I made some slides ;)

…eady

UdjinM6 · 2018-09-20T13:14:54Z

Interesting :) what about 33c12af as of a countermeasure?

InhumanPerfection · 2018-09-25T18:06:25Z

Yes, 33c12af will solve this exact issue (however it will speed down mixing by half), but my suggestion is in PrepareDenominate instead of:

    int nStepsMax = 5 + GetRandInt(PRIVATESEND_ENTRY_MAX_SIZE - 5 + 1);

do something like this:

    int nPairsMatch = 0;
    for (const auto& pair: vecPSInOutPairsIn) {
        if (pair.second.nRounds >= nMinRounds && pair.second.nRounds <= nMaxRounds) {
            nPairsMatch++;
        }
    }
    int nStepsMax = 1 + GetRandInt(std::min(nPairsMatch, PRIVATESEND_ENTRY_MAX_SIZE));

it will:

also resolve this issue
increase randomness in inputs selection when less than 6 inputs (I think it will be majority of cases with "no duplicate txid" rule)

(IMHO the whole idea of putting input selection in strict constraints (like with "no duplicate txids" rule) reduces randomness and probably facilitates analysis, idk...)

UdjinM6 · 2018-09-26T15:26:47Z

Pushed your suggestion @InhumanPerfection (f8d4a51) but then realized that it's not going to work for sessions with different denoms. We don't do this anymore(/yet?) but strictly saying the code was incorrect for this specific case. How about ab86bd0 instead? And we can also play with GetRandInt(2) (i.e. replace it with GetRandInt(10) == 0 or smth of that nature) to adjust mixing speed vs randomization ratio.

InhumanPerfection · 2018-09-26T15:59:04Z

Indeed, my example was incorrect for sessions with different denoms (and every time I read this legacy part of code, I wonder why it's still here?). And ab86bd0 looks good.
I like the idea about GetRandInt(10) == 0. And more, I like idea to make 10 adjustable and put it in parameters to application 😝 (I would like to have ability to adjust speed-randomness depending on my current needs)

InhumanPerfection · 2018-09-26T16:00:18Z

src/privatesend-client.cpp

            CAmount nValueDenom = vecStandardDenoms[nBit];
            if (pair.second.nValue == nValueDenom) {
                CScript scriptDenom;
                if (fDryRun) {
                    scriptDenom = CScript();
                } else {
                    // randomly skip some inputs when we have at least one of the same denom already
-                    if (vecSteps[nBit] > 1 && GetRandInt(2)) break;
+                    if (vecSteps[nBit] > 1 && GetRandInt(2)) {


vecSteps[nBit] >= 1 ?

…peed i.e. `GetRandInt(5) == 0`

UdjinM6 · 2018-09-26T16:40:39Z

..and every time I read this legacy part of code, I wonder why it's still here?

Yeah, that an old legacy thing. Smth to refactor in another PR I guess ;)

And more, I like idea to make 10 adjustable and put it in parameters to application 😝 (I would like to have ability to adjust speed-randomness depending on my current needs)

Good idea, added TODO

vecSteps[nBit] >= 1 ?

Ooops 🙈 This indeed doesn't match the comment above. Fixed :)

codablock

utACK (only looked for general errors...PS remains a black box for me)

nmarley

I feel same as @codablock , but didn't see any glaring issues. A couple typos identified inline.

nmarley · 2018-10-09T16:41:36Z

src/privatesend-client.cpp

-    int nRoundStart = GetStartRound(fMixLowest, fScanFromTheMiddle);
-    int nRoundEdge = GetStartRound(fMixLowest, false);
+    std::vector< std::pair<int, size_t> > vecInputsByRounds;
+    // Note: liqudity providers are fine whith whatever numner of inputs they've got


nit: s/whith/with/

s/numner/number/

omg, 2 typos in one sentence 🙈 fixed :)

Actually 3 typos 😆 "liquidity"

This is the old one actually https://github.com/dashpay/dash/pull/2278/files#diff-8bdec64a6a2adb83b9696e5ca6d5522bL1196 but yeah..

UdjinM6 · 2018-10-09T20:18:29Z

Fixed typos, pls re-review.

nmarley

utACK

* Alternative solution: pick rounds with the most inputs available to mix first * randomly skip some inputs when we have at least one of this denom already * More randomization in PrepareDenominate * fix `vecSteps[nBit] >= 1 ?` and adjust speed/privacy ratio for more speed i.e. `GetRandInt(5) == 0` * fix typos * no comments

UdjinM6 force-pushed the mixmaxrounds branch from 6cebae0 to 541f1f3 Compare September 10, 2018 14:33

UdjinM6 mentioned this pull request Sep 10, 2018

Add more variance to coin selection in PS mixing #2261

Merged

UdjinM6 force-pushed the mixmaxrounds branch from 541f1f3 to 7c632e7 Compare September 11, 2018 10:31

UdjinM6 force-pushed the mixmaxrounds branch 2 times, most recently from 12c441a to 3cfb25e Compare September 11, 2018 18:54

UdjinM6 force-pushed the mixmaxrounds branch from c3f52b6 to 7ab14f2 Compare September 15, 2018 18:56

Alternative solution: pick rounds with the most inputs available to m…

617581e

…ix first

UdjinM6 force-pushed the mixmaxrounds branch from 7ab14f2 to 617581e Compare September 15, 2018 19:02

UdjinM6 changed the title ~~[WIP] Alternative solution: pick rounds with the most inputs available to mix first~~ Alternative solution: pick rounds with the most inputs available to mix first Sep 15, 2018

randomly skip some inputs when we have at least one of this denom alr…

33c12af

…eady

UdjinM6 force-pushed the mixmaxrounds branch from 0d8f845 to 33c12af Compare September 26, 2018 15:11

More randomization in PrepareDenominate

ab86bd0

InhumanPerfection reviewed Sep 26, 2018

View reviewed changes

fix vecSteps[nBit] >= 1 ? and adjust speed/privacy ratio for more s…

4f45bf6

…peed i.e. `GetRandInt(5) == 0`

UdjinM6 mentioned this pull request Oct 9, 2018

Try to mix more than one input by default #2275

Closed

UdjinM6 changed the title ~~Alternative solution: pick rounds with the most inputs available to mix first~~ Pick rounds with the most inputs available to mix first Oct 9, 2018

UdjinM6 added this to the 12.4 milestone Oct 9, 2018

codablock previously approved these changes Oct 9, 2018

View reviewed changes

nmarley previously approved these changes Oct 9, 2018

View reviewed changes

fix typos

d0cf51a

UdjinM6 dismissed stale reviews from nmarley and codablock via d0cf51a October 9, 2018 20:17

nmarley previously approved these changes Oct 9, 2018

View reviewed changes

no comments

9f7f141

UdjinM6 dismissed nmarley’s stale review via 9f7f141 October 10, 2018 14:55

UdjinM6 merged commit f7b0b57 into dashpay:develop Oct 15, 2018

UdjinM6 deleted the mixmaxrounds branch November 26, 2020 13:27

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Pick rounds with the most inputs available to mix first #2278

Pick rounds with the most inputs available to mix first #2278

UdjinM6 commented Sep 9, 2018 •

edited

Loading

PastaPastaPasta commented Sep 9, 2018 •

edited

Loading

UdjinM6 commented Sep 10, 2018

Antti-Kaikkonen commented Sep 11, 2018

UdjinM6 commented Sep 11, 2018

Antti-Kaikkonen commented Sep 11, 2018

UdjinM6 commented Sep 15, 2018

InhumanPerfection commented Sep 20, 2018

UdjinM6 commented Sep 20, 2018 •

edited

Loading

InhumanPerfection commented Sep 25, 2018

UdjinM6 commented Sep 26, 2018

InhumanPerfection commented Sep 26, 2018

InhumanPerfection Sep 26, 2018

UdjinM6 commented Sep 26, 2018

codablock left a comment

nmarley left a comment

nmarley Oct 9, 2018

nmarley Oct 9, 2018

UdjinM6 Oct 9, 2018

InhumanPerfection Oct 10, 2018

UdjinM6 Oct 10, 2018

UdjinM6 commented Oct 9, 2018

nmarley left a comment

Pick rounds with the most inputs available to mix first #2278

Pick rounds with the most inputs available to mix first #2278

Conversation

UdjinM6 commented Sep 9, 2018 • edited Loading

PastaPastaPasta commented Sep 9, 2018 • edited Loading

UdjinM6 commented Sep 10, 2018

Antti-Kaikkonen commented Sep 11, 2018

UdjinM6 commented Sep 11, 2018

Antti-Kaikkonen commented Sep 11, 2018

UdjinM6 commented Sep 15, 2018

InhumanPerfection commented Sep 20, 2018

UdjinM6 commented Sep 20, 2018 • edited Loading

InhumanPerfection commented Sep 25, 2018

UdjinM6 commented Sep 26, 2018

InhumanPerfection commented Sep 26, 2018

InhumanPerfection Sep 26, 2018

Choose a reason for hiding this comment

UdjinM6 commented Sep 26, 2018

codablock left a comment

Choose a reason for hiding this comment

nmarley left a comment

Choose a reason for hiding this comment

nmarley Oct 9, 2018

Choose a reason for hiding this comment

nmarley Oct 9, 2018

Choose a reason for hiding this comment

UdjinM6 Oct 9, 2018

Choose a reason for hiding this comment

InhumanPerfection Oct 10, 2018

Choose a reason for hiding this comment

UdjinM6 Oct 10, 2018

Choose a reason for hiding this comment

UdjinM6 commented Oct 9, 2018

nmarley left a comment

Choose a reason for hiding this comment

UdjinM6 commented Sep 9, 2018 •

edited

Loading

PastaPastaPasta commented Sep 9, 2018 •

edited

Loading

UdjinM6 commented Sep 20, 2018 •

edited

Loading