Ethereum Foundation Core Devs, 8 Sep 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Ethereum Core Devs Meeting #24 [9/8/17]

Description

Agenda: https://github.com/ethereum/pm/issues/22

A

Check-Check, okay, see, can you hear me I didn't see you come in yeah here, yeah, excellent.

A

Alright, it looks like we're up. Welcome. Everybody here is a link to the agenda.

A

Resting serious face the first item, subtleties or questions. We need to work out, yo Ichi, you were you put something in about VIP 96. Is that still warranting any discussion so.

B

When I swing to the pretty place to contain the three different programs, three different hex codes and one of them should be deployed. Well, so one is: do you need to code? Id I, don't have to think about this, but there are different code in the request. I think I know which one is the right one, but the other one should be moving.

C

Let's okay hold on I'm gonna check myself, so.

C

Yuuichi, are you talking about the like? Word says: cash contract code assets you above apply and then down below again. It says something about the my code in the VM in it code, exactly exactly okay hold on. So let me just I'm gonna, compile the Serpent's and see what that gives you.

D

A

And while the talipes.

E

Looking that up is.

A

There any other implementations that need to be checked as their a solidity. One.

C

No, this is the only implementation.

F

Pawel add some comment there about being a sign comparison: okay,.

C

Yeah, oh so these young code at the top is just like something old incorrect and I'm. Just gonna ignore that or I'm gonna edit. That rather I'll do this right now.

A

A

Okay, great I think that resolves that right.

B

Sure it's a minor Victoria issue.

A

Great okay: the next item is VIP 603, Matthew d4 ought a has a comment and the agenda that kind of explains his reasoning for implementing.

G

A

Different curve, so oh of a talk, were you gonna say something yeah.

C

I was just gonna say that I'm in carousing, Lyon I am more a bit more in favor of this, and I was the last time basic we're just because of seeing that we don't have or just seeing how inefficient the other curve is. So.

A

Yeah, let's have, let's have Matt who's on the line, explain it to everybody and then we can. We can go through the pros and cons, but it does. It does seem like a needy IP, so Matt. If you want to go ahead, sure.

H

Thanks yep, okay, cool I.

E

H

I mean the the EIP itself is just kind of like a copy of the other one, but the comment goes into the detail of I mean generally. For me, the the biggest reasons are the fact that it's like widely supported widely implemented curve and people don't want me to roll their own.

H

You know implementations of things going forward, I mean you know, like all guinness relatively new name yeah, it's inefficient. If you use the alt swan right, the the real efficient one version of that curve requires like keep execution which is disabled on the vast majority of things.

H

Beyond that, like yeah, you know it's the ability to check, manipulate curve points on on nature transactions, in fact that, like tons of other coins, use native-like SEC PTO today, six-second P, 256, K 1, is native. You know like allows you to do nice. Pegs between chains on that and I mean the biggest. The biggest thing I see.

H

Is that, like you know, using that curve securely is, is less trivial than using SEC p2 66k one right, there's only one group on that curve, and you can't you can't really mess it up, as if you know if you use, if it issues like the normal core operations or is whereas on on the all TN, well um bein 256 anyway, the curve there's two groups: they have slightly different properties. If you choose the wrong one, you break, the CDH is easy.

H

If you don't realize that there's two groups, you know you there's way more mistakes there, like you know generally, the stronger support is the community support and the fact that so much sports 256 K one and it's way way faster. But you know so that's my my rationale and you know it seems silly to add a curve and not also add the curve that you know it's. At least the one supported natively I.

F

C

Don't have anything I, don't have anything to say that I haven't said before well.

A

Is this the same curve? That's in Bitcoin, yes,.

C

A

Yeah I mean this seems like a good idea. Does anyone have feedback or any opposition or questions not to.

C

A

It immediately just like.

C

Look, this is obviously not growing not going into Byzantium, but but.

C

Let's see, there is okay, one just question: I have is them how easy like for if anyone's in here has experience with the sexy libraries so like the stuff, some people that we already here's four of the ec recover, how easy or difficult is it to call specifically the the parts that you look, the Corvette again multiplying so.

H

I've implemented this in guess, like you know, on my own, it's it's very easy yeah, it's like a five lines, but like five lines of code for off code, really, okay,.

A

And you're thinking that be the same across different bindings for different languages, yeah.

H

Because this so there's two different ways to add this to go, there's you can use the you know the C bindings, which again those are you know, portable across languages there's also the BTC HD, which is yeah, which is also like a clear go implementation. If you know people want to use that for priority as well I mean it's has the same exposes the same clinics. It's easy recovery kind of requires like at that level of acts to the bindings anyway,.

A

Okay, any other feedback.

A

What I'm thinking is, if there's no opposition, this will be something that will come up in a future. A future quartet meeting, potentially the next view and we'll make sure that that EIP is up to standards, have one of the editors review it and then reappropriation.

A

H

A

Think I might have heard someone speak up. Was there anyone else?

A

No okay, great!

A

The next thing we have hope I, think yo Ichi dropped off for a second Google's, been having some problems this morning, I think they updated YouTube and dropped everybody who was logged on to it at like 858, at least all my screens refreshed. So maybe that's what's happening updates to testing, so this would be Martin, Jared, Casey and yo Ichi. If he comes back Martin. What is the summary for testing unless someone else wants to take that.

F

A

Back as well, yeah.

F

It's also felt that their.

F

A

You each we were just going over testing and we were saying if anyone wants to give kind of an overall update, I'm, not sure who's been most active in it, because I, don't think Dimitri's here right now to give a solid up or to give like a full summary.

B

Okay, so now on high C beta numbers for bathroom, we have less than ten minutes. That's nice for CPP and we've seen somehow 700 errors, but I cannot reproduce them in Opera, so it might be just some pretty close to mastering CPP, but not much the tests of something that that could be reasoning, I mean III, haven't been able to reproduce this and then I, which I started to receive many.

B

So many people pointed out mistakes in tests in this week. So that's a good sign, I! Think it's getting to ship in shame against everybody else. So we envision B.

A

Okay, great so the next thing would be, and basically agenda, items, BC and well. What's coming up as C part, two are about gas prices and client implementations.

A

So I think unless someone has a better idea, we'll just go with the client updates first for implementation of VIPs and then go through the gas price gas pricing for the opcodes. So we'll start with death. It looks like they're done with all their a IP is Peter. Do you have an update, yeah.

D

So we are the same place as two weeks ago, so essentially we are mostly done as far as I know. All new tests that have been added are still passing, then the only eight failing tests that we currently have anything investigative, but there are some known issues with some tests, so I'm I'm not sure whether they're these AIDS failures are because the tests are wrong or something else is wrong, but a friendly we're mostly done so we're just trying to do a cat sure.

A

Parody I think that would be our kitty.

I

I'm also working on fixing test issues, I think there is only a couple of tests state tested currently not pass him and also need to implement watching tests. I think we'll be ready in two or three days: okay,.

A

E

A

Year Martin mentioned something about the parity test or Martin. You were just speaking up, go ahead, yeah.

F

So I'm wondering if it's possible, if it's have you it's a disabled mal, so that it's not possible to configure parity to run on Byzantine rules.

F

Because it's failing all but on the Byzantium tests and hi.

I

Yeah, it would be to use a branch okay. Well, thank you branch name, perfect,.

F

A

Okay, great next, we have CPP aetherium who's in here. I know that Christian couldn't make it.

A

It looks like or yuuichi actually yeah yo, Ichi I think you might be up to date. On that.

B

A

Oh Andre sorry, the Andre go ahead now.

J

Well, it's actually mostly taken care of president Jim updates. Now, in so I know, there was several minor fixes during these two weeks, so.

J

A

I'm, looking at the meta implementation, progress, PR and the CPP aetherium repo- and it looks like 86- has a few 100 has a few and 211 has a few from the Google Doc that's kept up, but yeah. That doesn't seem too. That doesn't seem too bad looks like Jared and yo eg covered a lot of them.

A

Cool, let's see a theory MJ and welcomed McHale. This is his first meeting he's from the etherium jteem yeah.

K

Thank you guys. We have implemented to all the APIs and we are passing blockchains state and transition tests successfully and now I'm working to bring you up to date.

K

We M tests and transaction tests over should should be ready in two days. I guess we just need to check that we are passing them so I'm dating then bringing them to new format enough in our code. So we can support new format at this those tests, so we can run and check how how it works. So that's it.

A

So I forgot is aetherium J compatible with hive or.

K

Has been compatible with hive some time ago, when I we learn done with the test checks, I'm going to do work and hive to make it fully compatible. We make a pair MJ compatible hive great.

A

Definitely talk to the hive guys about that reach out. If you need me to get you in a getter channel or anything like that,.

K

A

Alright, thanks for the update atheria may s I, think that would be KC.

L

Yep, as of this morning, we merged the elliptic pairing Cree compiles and almost all the state tests are passing now, thanks to the great I've heard sup.

L

Recently, joint contributors still have a couple of loads about a dozen failures to to debug and blockchain tests even or I have progress on those so expect to be passing on the blockchain test. Pretty soon too.

A

Okay, great, thank you. Let's see, yellow paper any relevant updates on that yo Ichi! Yes,.

B

This morning, line notice that we pointed out one ape is missing on the that he was correct: difficulty, palma today and the property water change. This was I had forgotten about it. So I added opinion about today. Okay,.

A

Awesome, thank you. Paya app metallic, not really not really any progress since last time. Okay and I don't see any other clients on the call. I need to reach out to the Haskell team, I think they're, the only ones that are missing that are actively kept up with me as I. Believe Ruby is deprecated, okay, so the next one is determining the gas prices and opcodes. That was tasked last time with Arkadiy and Martin, and I saw that Martin keeps up with a repository called benchmark and the etherium I'm gonna post the link to it.

A

But Martin you can go ahead with your overall update any issues or just the summary.

F

Yeah so in summary, we had a chat about this earlier in the week me and Katie and Vitalik, and so based on the on the benchmarks, there's an analysis with some suggestions and we've made some concrete suggestions based on that and basically let me just find that the analysis here.

F

So the current suggesting that we would suggest is that, for addition, we would keep the current cost, which is a 500 reasoning, being that, even if we increase the file some, it doesn't really matter because there's a 700 call cost which we can count in on that. So the total cost would fill be 1200.

F

So we can David mod X, they analysis suggested, which is multiply it by around 1.5 or maybe two, but then it turned out that both parity and CPP use quite an optimized version of mortix I.

F

Think the Tallaght we are just multiplying it by five. Instead, that write the tonic.

C

um For mod X, yes, and my reasoning is that by multiplying it by 5i, but, like other people, feel free to check this. But I believe that the exponentiation of 4096 bit numbers by three year, would that still only cost. Something like five thousand gas, which is still like, not very high.

F

F

Continuing on scalar multiplication, we analysis warmed up somewhere in the range of 30 to 50, K and yeah I think you said, 40 K for it to be usable and in the hope that it would vary, could be further optimized and clients and.

F

F

Case your italic, you guys, having notes about bid, said about pairing.

C

I remember that I think we had decided that the one that the current schedule that was in PI ethereum, so 100,000 based on TV thousand four points is like, is so fine and we should just not change it.

F

Yeah, because that would leave us about a factor of two away from easily.

G

F

For for is Ramez recover, but still were they unacceptable, like a gas per second ranges.

F

For the for the hardware work that we do in the past, you know.

A

Okay, so looking at this, what is left to be benchmarked, or are we comfortable starting to make some decisions about the gas price for Bicentennial.

F

So I would hope that we can today just decide. Yes, we'll use that scheme and.

C

That's my hope.

F

To if, during the coming weeks we find- or someone finds that there's another service with vulnerability with specific input vectors, then we might have to go back on that in some kind of emergency, roll back or change, but I.

F

Yeah, hopefully enough, that's hope enough can happen so I, don't think we need to do any more benchmarking, the four to make a decision we'll make it a session and hope we don't have to go back on. Okay,.

A

That sounds good. What's the best? Sorry, if there's noises that is my cat playing with the mouse that squeaks. So what is the best way to organize making these decisions? Should we use the benchmarking repo they're, not just small enough to just talk about it over get her I.

F

Think we can just I mean everyone who has had an interest in these things has been invited to see themselves read the analysis. I talked about it on all product, so I'm I prefer just else does anyone in this call object to the proposed figures and unless someone speaks up, I think we should just put in the numbers in the eeks and then no find out for those that piece. Each has now been updated asymptomatic and then we just throw it out in the tests see city were strong science passes or not.

A

Okay, that sounds good to me. Anybody have any comments or objections. No.

D

I kind of agree at this point: you should just because whatever number we're currently at and think they are reasonable and then just roll with it, because it's way to date already.

A

Sounds great: okay: the next item on the agenda review the time estimate for testing and release.

A

As far as that goes, some of the thinking behind that was that if we felt that testing was close enough to being finished or finished, we could set a test net block number, maybe even today, and just shoot for that, so that we can stay on target for something after which, as we get closer to the test net or at the start of the test net, we can set a block number form a net depending on how well the test net goes.

A

We would put that in there with the caveat that, if testing, if, if, while on test net, we find a vulnerability or something we need to fix, we can delay what does everyone think of that.

C

Let's find a new.

A

So today is September, 8th and a week will be. The 15th two weeks will be the 22nd I. Think block times are gonna hit above 30 seconds mid month. I think.

C

They're gonna hit 30 seconds at the end of the month. Rather, but I can check again yeah yeah, it's community, it's I'm, pretty sure it's gonna be at the end of the month.

C

But I will oblige with another.

C

Four to five one: nine, three: six, let me just get the timestamp and they can run like the piece of code. I've run way too many times, one more time great and while we're getting that how long a.

A

So the caller says that.

C

We are gonna hit a girl from go up to a 30 seconds on the 22nd and it's we are going to go up to a 39 seconds around October 27th.

A

Okay, so is, as far as testing goes: Oh awesome, welcome, Demetri and actually Demetri able to hear me yeah awesome. We were just talking about if testing what, if the testers from the testing team and the client teams felt like they were comfortable, declaring a test net block number today that we could do that, and then we could declare a main net block number sometime after the test net starts or right before it starts, and then, if test net goes well, we'll keep the main net block number.

A

Otherwise we can delay it if we find some issues in your opinion, Demetri. Where are we in terms of testing and are we comfortable enough to launch the test net and the next week or two.

G

Yeah I think we comfortable to manage the test net, but still we have don't have any news from Paraty, client or other client, the just that only get is working on a highest consensus issues that they are having right now. I.

A

Think I talked recently I talked to Arkady and Arkady. Is it still the case that you believe in the next week or less at this point, I guess, 7 to 10 days that parity will be passing most of the hive tests should.

I

Be personal tests some of these ok.

A

That sounds good, so, if that's the case, then I think that doing something like having the test at 10 days from now. So something like the 18th might be a good idea.

A

What's everyone's opinion and on doing something like that.

F

hmm That's good sooner, the better yeah.

A

I agree and I mean the reason for that is it's a Monday I feel like Mondays are better to launch things because you don't have to work the weekend if you launch it on a Thursday or a Friday, and things go wrong or anything happens.

A

So it looks like let's see if we do that on the 18th, and then we were on the test net. For two weeks we could launch on October 2nd. How long should the tests that be run for cuz I've heard figures between two and four weeks? I know there's some period of time where you know it doesn't really make a difference. How long you run it after a while, we just kind of know, it's gonna be good or bad. What's what's the opinion based on previous experience with this.

C

So this is how long we run the test that, before the switch over to the main Network correct.

A

That's the question: okay,.

C

um So my answer is probably still gonna be around 3 to 3 to 4 weeks. Okay,.

A

Peter, did you have an opinion? I saw you talk for just a second, no.

D

I just want to say that, basically, if things go wrong, I can imagine that they will go wrong. Pretty fast I guess I would have said 2 to 3 weeks doesn't really matter, but yeah I don't make sense to run it for much because things break, it will break ok.

A

Well, the good news is, we don't have to decide on a main net number today, but it sounds like the main that number may be on the second or the 9th of October, so something like that would make everyone feel comfortable. It sounds like also doing it on the 18th would give us 10 days to finish up testing so that that's getting give us a little more time than we need, probably from what it sounds like we can personally.

C

I'd perform the ninth over the second, but no okay.

A

Yeah I, don't see why we should you know playing it. Safe sounds better yeah.

C

And so like in yeah, so the block number will probably ends up being like either four point: three five million or four or four point somewhere between nine and four point: four million. If it ends out budge written dragging on later and that's from Maine. That.

A

Right, yeah forbid note, okay and then.

D

Just was like no last means not besides the final block number before the test, nut forks, because I think for for the homestead relief. We picked a number which was only a preliminary number and then we kind of delayed it and that resulted in actually, apparently, they had release with the old number and then they did a new release with the new number and some of the clients worked over because they haven't updated and they thought that they already have the correct block number.

D

So, if possible, let's try to see the main test network first and then finalize the main. That.

A

Okay, so as far as the test that number, what's a good number for that, that would fall roughly around the 18th yeah. The eighteenth.

L

One note about the chestnut/black number is because the difficulty is very low, especially relative to the main net.

L

If more hashing power comes online and then starts mining on the test net, then it can accelerate the the data or time when the when the block number is reached, so should probably allow for one to two days target at one to two days later than the intended date. That way, if, if the hash rate goes up by 100 X, then it at least it doesn't the you don't reach the block number earlier than intended.

A

Mmm-Hmm, okay, so what would you recommend.

A

G

A

Doing a that would be doing a block number that is a bit ahead of the 18th I guess or am I am I thinking of this backwards.

L

Yeah, that's right: I haven't um done the numbers myself, but I can work on it. Real quick here, yeah.

A

No problem, we still have some other agenda items, so we can kind of come back to this at the end. Whoever wants to try to come up with a nice clean number for that.

F

A

F

Like to suggest I mean since the test that can be easily influenced, wouldn't it be better to if we say today that 40 Testament on the 18th and set the block number, we don't decide that your block time today decided later there's less variance. Yeah.

A

I'm: okay, with that, what would be the minimum amount of time before before a test net is released, that a client would need to be released with these with these numbers.

D

They're already pushing the limit.

A

Yeah I guess what I'm saying is like. If we decide the number of three days before, then everyone will have to update their clients, which is okay, but I, don't know if that would cause problems. If a lot of people stay on the old client and keep keep running Rob stone.

L

Some clients will definitely be on the old Robson, there's already a lot of clients on on various Forks of reps and anyway. The important thing is: is the hash power that pointed at that the hash powers updated that they so that the miners update more than all clients.

A

Okay, so something like deciding the block number on the 13th of the 15th. Would that be a combination of enough lead time, but avoiding a lot of variance in the block number beforehand.

I

A

Okay, next Wednesday sounds good because it's midweek, so if there's any complications, we can deal with it. Let's say the 13th. Anybody opposed or want to push it to the 15th I. Don't know enough about. You know how fast these things can change to really make a determination, I'm, just kind of throwing out numbers, I, think I do know they change quickly.

A

By the way, this price effect to this as far.

C

As my prison yeah, so yeah right, the price going down increases the likelihood that mining power goes down, which makes block x be a longer, but well it has the long term. So mix walked on some longer, and so it does have the one kind of happy side to it, which is that it's what it's slightly delays the next the next two having's yeah.

A

Won't mention it on this channel. Just everyone check, get her real, quick, but basically there's a rumor that would potentially decrease the price. So that's kind.

C

Of why I'm thinking.

A

C

Just a rumor looks like it's in from everything I see inside the Chinese channels, very likely true, yeah.

A

So, basically, due to that information that I won't say because it's just rumor at this point that would kind of make things weird, anyways man. That is interesting.

A

Basically, let's decide it next Wednesday then I guess: unless does anyone have any opposition for deciding the test net block next Wednesday.

L

Just for an estimate, the testament Rapson is currently at one point, six, three million. So if we do ten days, then we'd be about one point: six, nine million unless I'm mistaken and so one extra day would be one point: seven million, which is a nice round number.

I

A

That sounds good.

A

One point: seven one point: seven million would be a nice round number. Okay, I. Have that noted.

A

Just Shh, oh my cats freaking out as far as the main net test block. So just basically what I'm wondering is yeah if the price drops dramatically. What does that affect the block? Time is the thing I'm wondering because that's what a lot of people in the community would be wondering: um how does what affect the walk time? The price dropping? How that affects the I guess? Oh.

C

Yeah so in like basically because because it makes the difficulty defective, the difficulty bumps like much stronger, but that's only if mining, if hash power really drops in our previous experience, even large hashmap one price drops have not in the short swords and mining power, drops. Okay,.

A

Great, so it sounds like the risk: isn't that high for the economic factors here, yeah cool alright, so we have that block number done. What's the next item on the agenda, any other comments on the block, numbers and stuff. In summary, something around 1.7 million is an estimate for the block number on Rob's tonight.

A

It next Wednesday to see how the vary or to see how difficulty and other things change and robson between now and then, and we're tentatively trying to hit September 18th for the tests net block number, at which point we would start deciding on a main net block number. That would hopefully be two to three or sorry three weeks after rob stone which would put us at October. Ninth, that is an unofficial number for all the coin media that keeps saying September.

A

It's not September, so the October 9th, potentially, unless there's something goes wrong with the test net. So item number two on the agenda is the snappy compression for dev p2p. That's the Ayatollah.

F

A

Go ahead, yeah.

F

So on, when we decided that since number, could we please just make it quick yeah, some folks suggest them to the oil core dev channel and get her those who posts can raise their hands. Oh yeah,.

A

F

A

That's a good idea: we're not gonna have a call next Wednesday I should I should have made that more clear. It would just be a getter chat, great.

D

A

It would just be a get er chat because we already have somewhat of a number and if things change, we can just adjust that number for the block. Number I should say cool. Any other comments.

L

Probably just to get her chat up, they at the same time as the regular scheduled calls here.

A

Yep you're talking about next Wednesday right, yeah,.

L

A

So same time as the regularly scheduled call so 1400 UTC, we would have a discussion on the Gator channel about the block number I'll put that in my notes, as well so 1400 and Wednesday okay.

A

So the next item is a IP 706. This is a snappy compression for dead, p2p Peter came up with this. It reduced the sink bandwidth by 60 to 80 percent and would affect the network layer of the protocol. I've reached out to swarm developers. I think Daniel who works on swarm, has already commented a little bit on it, and we've received more comments since then in the EIP. Let me post it here so Peter. If you can just give a quick summary and then we can discuss.

D

Sure so the crux of the IP is that previously a few I think when Experian launched, we had some plans to extend that beauty. Beer. Basically, that's the base, that's working, the RFA theorem so that it employs compression because currently just encrypts, whatever upper layer that work protocols throw at it and just transmits it. That way, and we had this idea, but nobody ever pursued it because it was kind of bashed together with some other changes and and the last few days, I was running some benchmarks to see what what furfle.

D

Basically, how compression would impact performance- and it's quite staggering so in using the etherium to do a fast sink on main, that it produces download traffic from 34 gigabytes to 13. So basically, that's about the 60% saved and on rinkeby down it reduces from 2.6 device to about less than 500 megabytes. So on rinkeby, it's more spectacular because it's quite full of spam, oh, but but the idea is that the compression really can do an enormous save on on the East protocol.

D

So so I created I try to I, try to figure out how how much code or how much effort it would take to actually implement that be compression or I mean compression in general for for a deaf-mute appear and eternal that it's really simple, so it all it takes is whenever so before. A message actually gets to be encrypted and put on the wire.

D

We can do a compression round on it and, similarly, when, when it's a decrypted from the wire, we can do a decompression and and basically I managed to the whole code for this in GUI streaming in the go code is about thirteen eight lines, so I think it's it's worthwhile addition and I kind of brought up the cipa to discuss it. Ethical, wise, the EIP, consists of bumping.

D

The version number of deaf-mute appear to from four to five, and basically, whenever handshake arrives, client, if the handshake contain special number, five and clients and sorry clients should do snappy encoding from that point onward, they're having some discussions with corner cases and yeah IP can't find it on listed there. So one of the first questions is whether people feel confident about rolling out such a change or not. I've heard one concern about this change. From from the fight game team, that's Leah from Piper Miriam.

D

He was saying that Python itself doesn't have a cure, Python implantation of snappy, so this would entail I thought having to actually wrap the C++ library. They are a bit reluctant as I know. All other clients I mean all other languages. Go rust JavaScript. You have your implementations for snapping. So this the only concern was that one open question is whether we want to have compression mandatory or whether we want to as a maybe a compatibility flag and enticed signal whether they want to confess Allah. So with the basic micro. Two questions.

D

One is whether their general interests do include something like this and two. Whether we want to. If there is interest whether we want to make mandatory or optional mm-hmm.

C

Is there some like a wide wavy or specification of how has snappy works.

D

So I linked a Wikipedia article. I am personally not familiar with how stuff you wasn't gonna I'm, not sure what any some information I think he may have looked into it.

D

Sorry I'm asked a question, so the question is whether you have any knowledge about how complex snappy is internally at guessing miles. On thirty people, I looked at.

E

Briefly, before we're.

D

Not in great detail.

E

And then relatively straightforward, if I recall correctly and of course, there's the most compression algorithms, if you really wanted, you could um commute to dummy one that license which doesn't comprise anything I.

D

Think so in, in a sense, it's important to note that the deaf people, so the the IP proposal, is backward compatible in that a client may choose to run version four and not do any encryption. So it is perfectly fine for some clients to upgrade now and some clients to upgrade in two weeks or in four weeks or whenever they feel like upgrading the clients can do this completely another form. Another I.

A

Think it's a good idea pending more discussion.

A

So this would not require a hard fork because it's part of the networking layer, so the clients would just have to implement this correct. Yes,.

E

If not I mean I'm, how much more discussion do we need I guess I.

A

Think, or do we just we just answered Piper's question? Didn't we I guess I was unsure. If Piper's question was effectively responded to also and I guess, the other thing is like Andre and a few other people had a few sub questions like to enforce the 16 megabyte limit on compressed size and a few different things. Oh, never mind, I see: okay, Carol, okay, Peter! You commented a couple hours ago about updating the spec for replacing lazy with 16 Meg, okay, yeah.

D

So so, currently, this bag states that the decompressed size is limited to 16 months, and this essentially has the same limits enforced at the current that beautiful. So this is nice because you don't need any extra effort around lazy decompression and we can basically, we can use the same who's as we currently have. So if the message the plain text messages they decompressed messages larger than 16 max, then it's just gone out.

D

A

Arkadiy, do you want to elaborate on a comment you just made about having this be a sub protocol level.

I

Yes, I've just said two comments, suggesting that it should be negotiated for each the protocol, because some protocol might not did all that.

A

And why would they want to disable? That is that for different sub protocols that would be implemented within the client. Well,.

I

Yeah because some of them are already well at least I- have the one parity that already uses compression on a different level, and we don't want to compress everything twice.

D

Well, so so one of the problems it is that it all of a sudden.

D

This means that I explicitly mentioned this scenario in the VIP in the discussion I mean in the fact that the problem with doing this on a sub protocol level is that all of a sudden, if I roll, my every single sub protocol officer, needs to be aware of a nice to hear about compression, and so that's that's already getting murky and regarding the speed as Nick was pointing out that that's not be essentially, if Hannah, almost as fast as mem copy, it doesn't actually compress so the go-go implantation of snappy you actually make it compress to JPEG images, which are probably not really compressible, can compress of this a with about 15 gigabytes per second on a single core.

D

So those numbers I am wondering whether it matters if it does a round or not.

I

So I'm not suggesting completely taking it to these a protocol implementation, so it's still handled ways in their PDP. But that's the protocol will just declare one bit in one compression or not. That's it.

D

Well, yeah, but I mean who needs a heavy update, because all the capability is all the handshake message needs to be so currently the handshake message can be extended. The extra fields but I'm not sure whether the capabilities field itself can be further expanded. I, don't think that Provost force, that which means that do such a thing, it would require a major update. The deputy, your.

I

Updates new version anyway,.

D

Well sure, but if it requires a ton of update from everybody, that is much less likely to be implemented by anybody supporting.

I

Compression also requires an update from anybody right, yeah.

D

But adding 20 lines of code, your code is much simpler than adding 200 I.

E

I

E

Question is: is it with complexity.

I

Because snappy.

G

E

G

Fast and it's gonna be a big game.

E

In most cases, is it worth complicating things for I could schedule an HTTP regularly. You know gzip compression stuff, it's incompressible and it just works, because it's why 800.

I

Hz because Heather's the most don't know if.

E

That's what I'm saying I don't practice a lot of service simply serve everything compressed and gzip is, although more extreme to that snappy.

I

Sure my point is about giving choice, but.

E

Do we need the choice like? Is it actually important? Well,.

E

Given that snappy is almost as fast as just copying it and your processor and and memory bandwidth or what is it, makes you faster than your network I, don't think yeah.

A

I think I think Arkadiy. The thing is like: if there's nothing, if there's nothing currently remotely faster than that, then implementing it is a sub protocol would only complicate things rather than having everyone go with the same thing that, for the most part, shouldn't be superseded for a long time. If there's something already implemented, that's faster than snappy that everyone can come on board with that. That sounds like it would be better. Is there something that's implemented on a sub protocol level that could be implemented everywhere and a that would be faster than snappy? That's.

I

Not my point I'm not suggesting alternative to snapping we're already using snappy and show that they don't want to compress the data twice with snapping I.

E

Mean I understand that that's a bit of a code smell, but given that it's not actually going to have a meaningful impact on performance, doesn't matter.

I

I, don't think it will be more than another anyways good.

A

Ok, so it sounds like what you're saying is it's doable Arkadiy. It does add a little bit more complexity to your to your client, though,.

I

E

Its complexity to everyone's crime.

A

Not just Harrison's crime, alright know what I should say is complexity. What would they be having to change their sub protocols to not compress things twice? That's how I understood it. I.

I

D

Think it's on the table to change the parity protocol yeah.

I

It's not the problem of the network gives the same problem.

I

Well, yeah, anyway, we can discuss anything github issue.

D

A

I'm, okay, discussing it further and the github issue and getting a few more opinions and then the next chord that meeting, let's bring it up again, I think by then there will also be some more people who utilize that p2p, who might might have some other opinions on it. So but that's not a problem also. We can get kind of better descriptions, because I think there's just a little bit of confusion with what the arguments are about so well, we'll talk about it more on the IP. That's a good idea, our Katti Thanks!

A

So, let's see is there any other topics or agenda items.

A

Ok, great think of anything yeah I'm. Good too, it looks like the estimate is around the 18th. We will be doing a test net fork, potentially that is block number 1.7 million, but that is obviously can change at 1400 UTC.

A

This Wednesday we're gonna, have a getter chatroom discussion on what block number the test net will be, and then, after the test net after the test net happens, then we will decide on a main net block number that will hopefully be like this is super early, but we'll decide on a block net number between the 9th and the 16th I guess just some something in early October. Basically, the the 9th has been thrown around as a date, but that's nowhere near official, so the main that block number will be after we start the test net.

A

It sounds like, although we might have some recommendations beforehand, thanks everyone for coming. Unless there's anything else, everybody have a great day.

C

F