YT YouTube2Text

YouTube2Text → Transcripts → Blockchain Trilemma SOLVED: Turbocharged Scaling with RLNC | High-Performance Web3

Blockchain Trilemma SOLVED: Turbocharged Scaling with RLNC | High-Performance Web3 — Transcript

by Optimum · 5,146 words · 778 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:04All right. Uh I'm here today uh with
0:09Nicholas Nicolawo and Kisher Conir. So,
0:13Professor Nicolauo, long-term
0:15collaborator
0:16uh now member of Optimum. Uh Dr. Conoir,
0:21very long-term collaborator
0:24uh and uh now CCO
0:27at Optimum. Uh my name is Muriel Madard
0:30and I am co-founder along with Kishori
0:34of Optimum and uh I've been uh for many
0:37many years professor at MIT uh holding
0:40the NEC chair in software science and
0:42engineering for professor in the
0:44electrical engineering computer science
0:46department and leading the network
0:48coding and reliable communications
0:50group. Um so just as background optimum
0:54which is the company that all three work
0:56on um is the company that is doing the
1:01IP for web 3 the internet protocol for
1:04web 3 looking at decentralized
1:06uh data optimized hence the name optimum
1:10movement um and today we're going to
1:13chat about the trilmma something that
1:17many folks in web3 and blockchain know
1:20discuss us um [snorts]
1:23comment on uh and you know we have done
1:27some recent work on that which you know
1:29we we've posted we've had some posts but
1:31we thought it'd be fun just the three of
1:33us as collaborators friends uh
1:36colleagues to chat about it and and
1:39share our discussions with a broader
1:41audience. So we're we're hoping um that
1:44you'll enjoy our chat and maybe this
1:46will lead to your own discussions with
1:48your own colleagues online in person and
1:52just to um start some some more
1:55intelligent discussions. So with that
1:58I'll start with um
2:01you know let's say the the origins of
2:04the trillemma and particular the work uh
2:07that Nancy Lynch uh just as a little bit
2:11of background n my colleague Nancy Lynch
2:13actually held the NC chair in software
2:15science and engineering before me uh and
2:19she and I co-supervised Kishori when he
2:21was a postoc uh at at MIT and of course
2:25she has worked extensively ly with
2:27Nicholas. So, uh, you know, she's a bit
2:29of a she's an adviser to company. She's
2:31a bit of conjunct intellectual
2:33conjunctive tissue here. Um, so, you
2:37know, I I'll just open that that
2:39question of, you know, the origins and
2:40what the the the underlying theorems
2:43are. Uh, maybe I'll start with you,
2:45Nicholas, and then Kishor, if you want
2:47to chime in.
2:48>> Yeah.
2:50>> Uh, thank you, Muriel. Uh, yes. I think
2:55the origin of the Tai lians stems from
2:58initially from the cap theorem where
3:02that was proposed
3:04uh if I I remember correctly in in a
3:08conference and uh and Nancy with a
3:11colleague said Gilbert they they took
3:14the opportunity to prove that PLMA the
3:17CPA that was correct and uh that had to
3:21do with consistency partitioning and um
3:24availability of data. Uh
3:29they prove in that uh in in a paper that
3:32indeed there is a problem in
3:34guaranteeing all three of those uh
3:38properties in a single distributed
3:40system. Meaning that we cannot guarantee
3:44we can either guarantee uh consistency
3:47with partitioning but not availability
3:50uh partitioning with availability
3:52without consistency or consistency and
3:54availability without having
3:56partitioning.
3:57And uh in in simple terms
4:01all three cannot coexist. This is what
4:03the the tri says. uh and um within that
4:09um uh that pybe they managed to show
4:12that indeed there was a problem with the
4:15uh CPA and all three cannot coexist.
4:18However, two of them can coexist in a
4:21distributed system. Um following that uh
4:26we have the trilma that is applied in
4:29blockchain which it talks about three
4:31other properties uh related to the first
4:36however uh termed in a different way
4:39which uh the three properties that we
4:41are we have in the blockchain trimea is
4:44the decentralization
4:46scalability and security and upon this
4:51there were there were many discuss
4:52questions that uh we can go through and
4:56and take a a closer look I guess in the
4:58in our discussion.
5:01>> Yeah, thank you. Thank you Nicholas
5:02Kishor. Did you want to add something to
5:04that? Yeah. No, this is a great point
5:06actually Nicholas that mentioned that
5:09although the trma in blockchain talks
5:12about three different you know
5:15properties but the very style of
5:18reasoning or thinking originates in cap
5:20theorem because you'd see system
5:23designers who are not doing directly any
5:25theoretical work they use the cap
5:27theorem as a means to adjust the
5:30tradeoff which two to pick and you can
5:33have any truth but not all in a
5:35realistic network. So now that result uh
5:40was also you know there are some lot of
5:44gray areas that when you actually bring
5:47the model to realistic systems what does
5:51it mean to be asynchronous if a
5:53[clears throat] connection is disrupted
5:54for a long time is the connection is it
5:57asynchronous or it's fail and there's a
6:00very nice article by professor Lynch and
6:03Gilbert Seth in one of the comm ACM
6:06communication issue dedicated entirely
6:09to the capture. And now when it comes to
6:11the blockchain world, it's very
6:13interesting that they actually modeled
6:15in a very similar way or frame but
6:18except it is not proven theoretically
6:20and that's where you know a lot of the
6:23discussions or confusions you can say
6:27you know continues and it's very
6:29important to understand get deeper into
6:33>> agree agree and and maybe just for folks
6:35because um you know the the consistency
6:39part which is in the original cap
6:41theorem um I think is a little bit
6:44misunderstood
6:45um and maybe that's a little bit why as
6:49you mentioned you know the trillemma is
6:53a an informal
6:55representation of the cap theorem but in
6:58that informality
6:59maybe some uh that some of the original
7:05rigor of the cap theorem has been lost
7:07and in a way the trilamic uh is a little
7:12bit too
7:14um pessimistic, right? Uh and partially
7:18because you don't have those
7:19definitions, right? And in particular,
7:21what you don't see is consistency
7:24>> that right cons in the in the in in the
7:27in the original cap theorem there's
7:28consistency which has a very specific
7:31meaning. uh and uh in you know in the
7:36trilmma you don't have consistency and I
7:38think this may be one of the core places
7:41where there's some level of uh
7:45disconnect between the informal let's
7:47call it folk theorem of the trillemma
7:50which has issues and has in my view but
7:54I'd like to hear everybody's view sort
7:56of made people think too pessimistically
7:59about what can be done versus you know
8:01the original cap theorem which is a
8:03theorem you know with a proof is correct
8:05uh but a proof that is correct for those
8:07very particular very
8:10uh specific definitions.
8:15>> Yeah.
8:16>> Yeah.
8:17>> Yeah. So I I'd like to hear h how do you
8:20guys like to explain you know when
8:22you're t trying to tell somebody about
8:23consistency you know who who's who's not
8:25a who's not
8:27>> somebody who's been working in this
8:30particular area of distributed and
8:32decentralized system computer science
8:33how do you like to uh uh explain
8:36consistency I can go first I could say
8:38how I give it as an example and then I'd
8:41love to hear is you know what I I tell
8:42people about consistency is like you
8:45know you're storing when you store. It's
8:48like you have a closet, right? So, um I
8:52go and I hang in a dress, right? We're
8:54sharing we're sharing
8:57uh you know, you have reading and
8:59writing. So, the the writing is I stick
9:02stick stuff in the closet, right? So, I
9:03go and I I I stick in a dress, right? Uh
9:06and then Kishori comes in later and he
9:09sticks in a jacket, right? Now,
9:13Nicholas comes in, opens the closet to
9:15to read, to see what's in there.
9:17Ideally, he sees the most recent state,
9:19which is the dress and the jacket. He
9:21might see an old state. He might see
9:23just the dress, but he's not going to
9:25see just the jacket that never happened,
9:28right? That that's how I like to
9:30describe it. So, so that's my that's the
9:32way I describe it, you know, when I'm
9:33just talking to to folks informally.
9:36It's like, what is this? But I'd love to
9:38hear what you guys how you guys describe
9:40consistency.
9:41>> Yeah. Go ahead, Nicholas.
9:43>> This is a very nice example and uh I
9:45really like it to bring it to you know
9:48in the real life uh problems and issues.
9:51Uh and indeed consistency this is this
9:54is what consistency is. Sometimes when I
9:56try to explain consistency I I said I'm
9:59I'm taking in a terms of of copies of
10:02data. You have number of copies of data
10:05and what you want to do actually is to
10:07keep those copies the same, right? So if
10:11you make a change on a copy, you have to
10:14make sure that all the copies get this
10:16change. Okay? And this is this is a
10:19consistency and and as you said Muriel
10:22the basically the problem stems the caps
10:26theory stems from the assumptions that
10:28we make in the system that there is a
10:32synchrony so we don't move in
10:35synchronous steps inside the system and
10:38also uh the other thing is failures that
10:42uh we we do not control and we don't
10:45know whether they happen if we do have
10:48synchron
10:49uh and that's why consistency it gets
10:53tougher to guarantee when we have these
10:56assumptions in the system whereas if we
10:59had synchrony and we know that
11:00everything is ideal and nice and
11:02everybody moves in steps then if I I
11:06change a copy then I could change all
11:09the copies at the same step so everybody
11:12will move in the same
11:14>> step what do you think
11:16>> yeah I I think uh it's it's a very
11:18interesting concept. We know it
11:20intuitively. We feel it depending on
11:23different situations. At the same time,
11:26it's so hard to formally describe and
11:30argue about it. But then it's very far
11:32away from the it feels it's very far
11:34away from the practice. For example, if
11:36I'm writing a program, I update a
11:40variable then I read it. I expect
11:42whatever just I update it. But now if I
11:46go to multi-threaded program that
11:48variable might be touched by different
11:49threads that are scheduled by the
11:52operating system. Then you go to
11:54distributed systems it's not just
11:56threads in one machine. Now
11:58>> it's a bunch of machine connecting over
12:01to the same location and they may fail
12:03and in blockchain they may be vyantine.
12:06So now what is the underlying then
12:08behavior that we are expecting uh when
12:11it then it's the you know the the
12:14researchers they define it as an object
12:17abstract object and it has some behavior
12:21and it is essentially the behavior and
12:24it does not directly talk about how you
12:26implement it. It's a behavior you
12:28exhibit and that's very very very uh
12:32interesting. Cap theorem was about a
12:35data store data access
12:38and blockchain you'd care about a
12:41different kind of consistency maybe it
12:43respects the overall definition of the
12:45blockchain how the EVM or in the case of
12:49Ethereum should work is the behavior
12:51consistent
12:52>> uh that that needs to be you know
12:54carefully discussed and thought about
12:56>> absolutely absolutely and I think you
12:59know from uh from closets to
13:01multi-threaded programs to you know
13:04copies everywhere that you have to
13:06update quickly. Um I mean this this
13:09illustrates you know the the the
13:11inherited difficulty and I would say not
13:13just distributed but really
13:14decentralized right because to some
13:16extent you know if it's distributed then
13:20you know you can still have a leader
13:22that tells people what to do even if
13:24it's a synchronous you basically you
13:26know the leader can say I'm not going to
13:28do anything until I hear from everybody
13:30etc. Right? when it's truly
13:31decentralized like in blockchain like
13:34you know
13:35>> everybody's just doing their thing good
13:37luck with this
13:38>> you know um
13:39>> uh so so I wanted to go back to you know
13:42I wanted us to spend a little bit of
13:44time on on consistency because I feel
13:47that that's maybe one of the concepts
13:48that's most tricky you know uh and
13:52somehow
13:54diverges but I'd like to hear what you
13:56think it diverges most in its
14:00definition from the informal triilma
14:03definition, right? Because what we're
14:04talking about here is effectively what
14:07is the mismatch between the cap theorem
14:10which is theorem theorem theorem and the
14:13trilama which is more of a directional
14:18type of um
14:22folk theorem let's call it that way
14:24right um so
14:27you know what do you think are the the
14:30main points of divergence between the
14:32cap theorem and the trilmma.
14:36>> Nicholas, do you want
14:39>> Yeah, I can I can try to to go first.
14:42Um, I think okay, one of the main parts
14:47that is missing from uh from uh the
14:52uh the blockchain trial trial as we as
14:55we know it right now is uh as you said
14:58there is the consistency. Consistency is
15:01something that is is not referred uh
15:04explicitly into the the blockchain
15:08triuma. Uh although uh I believe that
15:13notion is hidden without the properties
15:15that they are looking at. Uh
15:19first of all if you if you think about
15:20blockchain blockchain it needs
15:22consistency in its core. Why? because we
15:26need a single copy of a blockchain to be
15:30replicated among distributed
15:32decentralized peers, right? And
15:35everybody needs to see a consistent view
15:38otherwise we're going to have a problem,
15:41right? So in order to generate this
15:44consistent view what they are uh uh what
15:47we are using in in blockchain define
15:51exactly these three suffers from these
15:54three properties that we have not suffer
15:56but it's dependent I would say uh
15:59because in order to establish
16:00consistency we exactly use consensus
16:04where everybody agree on a single copy
16:07or on a single
16:08>> or enough people agree [laughter]
16:10>> exactly enough agree and when enough
16:13people agree on on that on that copy
16:16then we assume that okay this blockchain
16:19is what we're are going to be using from
16:22uh a point in execution and and above.
16:25So uh the three uh the three properties
16:30that they have in the blockchain trilmma
16:33essentially focus on the solvability of
16:36consensus because we have
16:38decentralization
16:40how much we can decentralize.
16:43So as we have more people participating
16:46in this uh uh uh in the consensus in the
16:51agreement protocol uh we do have uh
16:54security. How can we protect consensus
16:58from malicious attacks? And we also have
17:00scalability which means how can we solve
17:04consensus as fast as possible once we
17:08have decentralization in the system
17:11and um and essentially everything comes
17:14down and collapses to having consistency
17:18in blockchain. [clears throat]
17:19>> So
17:21>> so that's why I think they are very
17:23related. Yeah. What do you think? In
17:25>> in fact I feel that uh when we talk
17:29about the trilma we talk about security.
17:32I mean security mean might mean a lot of
17:34things. One somebody could be u you know
17:37lost their money because somehow
17:39somebody got access u to their keys or
17:42etc. But it does not mean it's a
17:44security violation of the underlying
17:47chain. To me that security needs to be
17:50defined in a more formal way maybe as a
17:52sequential specification. How should
17:55this system behave when you clients
17:57interact with it? As long as that
18:00behavior is respected and not violated
18:04then to me it's a security
18:07essentially it might be called safety in
18:09other literatures. Safety is preserved
18:12property.
18:13>> I think safe that's a really good point.
18:15Sorry go ahead.
18:16>> Yeah. Yeah. So that is now one other
18:19aspect uh trilmma brings I find it
18:21fascinating is that in most of the
18:24formal way of analyzing a distributed
18:26systems we often think that local
18:29computations are actually negligible
18:32compared to network communication. Here
18:36we are executing a smart contract at
18:38every you know in different nodes. So it
18:41is doing computation actually and that
18:44also brings another aspect to
18:46scalability. So it is indeed you cannot
18:49just execute the code of a smart
18:51contract in negligible time. So those
18:54are several dimensions we need to think
18:55about.
18:57You bring up an excellent point,
18:58Kishure, and I'm I'm going to just uh
19:00pick on that because actually remember
19:02when we started working together with
19:04Nancy uh on bringing you know coding to
19:09uh con consensus systems. You know, one
19:12of the big things we started looking at
19:13together was the cost of communication,
19:16the cost of networking, which was
19:18generally not taken into account in the
19:21literature up until that point. Remember
19:23that? That was definitely one of the the
19:25big uh the big motivating. Yeah.
19:28>> When we started working together what 12
19:31years ago that's that's not that's not
19:33what I'm even trying to think.
19:34>> For example, stories cost. Everything is
19:35about stories cost uh different cost.
19:38>> Exactly. Exactly. And I think that
19:41that's also something that the trilmma
19:44suffers from which is that absence of
19:50um
19:52taking into account the cost of the
19:55propagation, right?
19:58uh versus just and and as you said, you
20:00know, the idea that the compute
20:04>> is taken into account. Maybe the storage
20:06is taken into account but that you know
20:08if you have computer you have storage
20:10and you don't have a way of moving the
20:12data back and forth between the storage
20:14and the compute so that the compute can
20:16happen because the compute has to happen
20:17on data otherwise what are you computing
20:19on that that missing part is uh is is
20:24you know is still reflected um in the
20:27current trial. That's my view. That's my
20:30opinion. But you know, opening it up for
20:32for you both to to comment on that.
20:35>> Yeah. What do you think about it?
20:39>> Uh
20:40yes, I mean um indeed talking about the
20:44storage and computation uh right now is
20:48uh is the big part of of what we are
20:51doing in uh uh in blockchain, right?
20:54uh whatever we do, we do uh we store
20:57information and then we try to um uh
21:01compute as you said over smart contracts
21:04in order to be able to reach uh a state
21:07that everybody's going to be there and
21:10uh uh computation is a big part of it.
21:13So we cannot distinguish um comput
21:17computation from communication because
21:20one great part of of the whole ecosystem
21:23is uh is going to the uh computation.
21:27uh however I I believe that
21:30communication even though um is
21:35computation is a big part of blockchain
21:37communication remains as one of the
21:40central um uh overheads that we have in
21:44in the system. So uh we need to find
21:49ways to see how communication can be
21:54reduced or can be optimized in in a
21:56sense so as uh uh we can uh have uh uh
22:03either solve or see if there is a
22:05solution to the problems that we face in
22:07the trial lima or to show that okay even
22:11if we do have this optimization of
22:13commitation we still face some
22:15bottlenecks that uh may lead to the the
22:18triing.
22:19>> Yeah. So actually it's very interesting
22:21that uh when you talk about diff
22:25decentralization
22:26and each node is involved in straight
22:30transition of the you know replicated
22:32state machine they may diverse then you
22:35have actually violated security or
22:38safety.
22:38>> Yes. The only way to keep them in check
22:41is to have guardrails of communications
22:43because how would they know that they
22:45are not moving apart from each other in
22:48terms of the state?
22:50>> Yes. Yes. Actually, even if you only had
22:53two nodes, one which was basically your
22:55CPU and wrote is your memory and you
22:58know it's very clear who does what and
23:00there's only a single copy of the data
23:02in the memory and somehow you know you
23:05had a very narrow bus or columns between
23:09them. You know, your system's going to
23:10be terrible. [laughter]
23:12>> It's just like
23:14>> and it's not because you know there's
23:16some clever issue here about
23:18decentralization or anything. It's just
23:20like you can't get there, right? Um, so,
23:24so I think that that's that that's cool.
23:27Um,
23:28>> you know, we together have have actually
23:30talked about, you know, the end of the
23:33triilma. You know, certainly we've had
23:35some sort of fun uh fun content uh on uh
23:40uh on optimum about the you know, tril
23:43triinati and so on. Um and and maybe we
23:46can go a little bit more now that we
23:49understand
23:51that there is
23:54a theoretical underpinning to the
23:56trillemma
23:57but it's not the trilmma
24:00and in particular you know one should
24:02not over
24:06interpret the trilmma right because in a
24:09way it's been it's too loose it's not
24:12rigorous uh and and therefore it's very
24:16easy to
24:19misinterpret
24:21rather than just let's say a helpful
24:25uh vague reinterpretation of the cap
24:27theorem. One one can think that it's
24:29actually something that's telling you uh
24:32impossibility results that are that are
24:34not there. um want to to get you feeling
24:38get the work that we do, the work that
24:40we do together, you know, at Optimum,
24:42how it, you know,
24:46makes it so that
24:49some of the interpretations of the
24:51trillemma were not correct, you know. Um
24:54and maybe, you know, one of the things I
24:56can do is is is start with an example,
24:58right?
24:59um and then and then to hear hear
25:03everybody's opinions or your own
25:05examples or maybe comments on mine. Um
25:09let's think of scaling, right? Um, if
25:12you don't code and you have a network
25:17where you know even before you're trying
25:19to do computation or anything, you're
25:21just trying to get data across and you
25:24have some fraction either because of
25:29failures, Byzantine attacks or whatever,
25:31some fraction of shards that don't get
25:33through. And I just have a the simplest
25:35network, just a daisy chain, link one,
25:37link two, link three, etc. Right? And if
25:40I code
25:42only end to end so you know the kind of
25:47uh legacy codes that are currently used
25:50by other companies in blockchain say a
25:53readman code or um something like uh I
25:59don't know like a you know traditional
26:01rateless code um then what will happen
26:06is actually my throughput will not
26:09scale. It will go down exponentially
26:12with the number of notes. Right? So to
26:13me that's a very clear idea that okay oh
26:17well I can't scale right um but while at
26:22the same time remaining you know
26:24reliable let's call it reliable and safe
26:28you know uh reliable meaning that I have
26:30a reliable throughput uh and um you know
26:35keeping everything else that I need. So
26:37but actually if I do RLNC and I recode
26:41then I can do that then I can actually
26:45get a perfect throughput that's to say
26:47the maximum available throughput and
26:49that throughput is now scalable it
26:51doesn't depend on the number of hops
26:53right whether I have two hops or a
26:55million hops my throughput stays the
26:57same basically I was talking about the
26:58fact that you know what what the example
27:01I just mentioned is one of those you
27:03know whatever anzat no lost
27:06You know, it's an example of what what
27:08people believe, you know, which is kind
27:09of like what the trma is, right? It's a
27:11it's a something that people believe,
27:13which is has has some root in reality,
27:16of course, uh but is maybe not a correct
27:20um
27:22uh understanding of the theoretical
27:23limits as more, you know,
27:27an interpretation of something which is
27:29theoretically correct but which is not a
27:31correct interpretation. And then also
27:33bringing in empirical experiences that
27:36people have had and which say oh well I
27:38I'm not able to do that right so you
27:40know the scalability that I just
27:42mentioned here you know if you were to
27:44say well you know this would be an
27:46example of why I can't scale but yet if
27:48you do the math correctly and you recode
27:51you actually can scale so you know
27:53that's my favorite if you will sort of
27:57uh counter example and again you can't
28:01completely give a counter counter
28:02example to something that wasn't
28:03completely defined, you know,
28:05rigorously, but you, you know, a counter
28:07example to that, you know, general
28:10sentiment, let's put it that way. So,
28:11that's my favorite counter example. I'd
28:13[clears throat] love to hear yours,
28:14Kishori, and yours, Nicholas.
28:15>> Yeah. So, uh, no, this is very uh
28:18interesting. So, I'll come uh I arrive
28:21at almost the same conclusion, but in a
28:23slightly different angle. So what I
28:26think is that the trilmma as it is
28:28stated even though it's not proven in
28:30any uh you know under certain specific
28:33assumptions or at least I do not see any
28:36well- definfined um it's it's like a
28:39folklore now different people will look
28:42at it from different points of view some
28:44people will think you know it's too
28:46difficult to create a system others will
28:48think hey I have actually beaten that
28:50amount and that's because there is no
28:53one clearcut definition based on a
28:55formal way of expressing the U
28:58statement. Now
29:00>> even with that if you look at it
29:03decentralization
29:05and you know um security and scalability
29:10if communication were extremely good
29:13imagine and almost like a you know near
29:18you know as best as it could get then
29:20you can actually have a much better time
29:23with it and you can circumvent this uh
29:26pessimistic aspect. Why? Because if
29:29communication is low and reliable, you
29:32can actually make the time slot
29:34synchronous steps which is easier to
29:36deal with much smaller. Essentially, you
29:39are making time go faster.
29:41>> Yes.
29:41>> Because and that is the key I think
29:45because the other two are not you know
29:48you cannot compromise securely. It's a
29:50definition of what you mean by safety.
29:54uh number of participants you cannot
29:55eliminate otherwise you'll wind up being
29:57a centralized system. The only access
30:00that is remaining to be tackled actively
30:03is this communication and tied to
30:06scalability. That's how I look at it. I
30:08would love to hear.
30:10>> Yeah. Yeah. That's a really good point.
30:12Yeah. Yeah. Thank you, Kisher. Nicholas,
30:14I want to hear your thoughts on it.
30:17>> Yeah. Uh I mean uh this is this is
30:21exactly the the way that you described
30:23it Muriel. Indeed uh if you see from
30:27that that point of view that scalability
30:30somehow can diminish then um uh
30:33basically the trma uh it cannot be
30:37compared with the cat theorem anymore.
30:39Why? Because there we we saw we saw that
30:43we can guarantee the two properties but
30:45we cannot guarantee the third one.
30:47However, if we can guarantee scalability
30:49and we are able to um be able to deliver
30:53messages in a fast pace, it means that
30:56the centralization scalability they
30:58become one and uh there is no problem of
31:02when we decentralize to guarantee
31:04scalability and at the same time we
31:05guarantee security. So that's how uh the
31:09trila can uh uh can go away and uh and
31:14indeed the lack of formal definition as
31:17mentioned by uh this is one thing that
31:20um it's very important why because when
31:24we see when we go back to the cap theory
31:28there is a clear um relation between
31:32each property and a system um a a system
31:37behavior. For example, consistency is
31:39linked with safety. Availability is
31:42linked with um uh with a synchrony and
31:46partitioning is linked with failure.
31:49>> Right?
31:49>> So we can directly see a connection
31:52between each property and um uh a system
31:57behavior. So
32:00it must be the case that uh if we want
32:04to have like a formal trilmma there must
32:06be a connection between those properties
32:08that we say with the properties of the
32:10system and uh and for now there's a lack
32:14of definition
32:17>> which um doesn't help us to indeed say
32:21that okay this is the one and this is
32:25why or why Not it is not bad.
32:30And um as mentioned there are so many
32:33solutions that we see in um uh in the
32:37community where they claim that they
32:39solve the trilma by different um uh by
32:43different mechanism either by reducing
32:45decentralization
32:47or by having some other optimizations
32:50that may lead to uh to scalability. Uh I
32:55I believe that um right now what we have
32:58with RMC is the closest one that's going
33:03to show that scalability and
33:04decentralization can co coexist in a
33:07single system without any issues. In
33:10fact, as you mentioned, um, if I can
33:12chime in, is that that is why in
33:15optimum, you know, as you probably now
33:18realize, but I just want to bring it
33:20out, um, for the, you know, listeners,
33:23audience that whenever we want to solve
33:26a problem, we first define what is the
33:29specification behavior of that system as
33:32we interact with it
33:33>> without thinking who implemented it,
33:36what programming language, what
33:37technology. We don't even look at it
33:39first. This is the behavior we want.
33:42Once we pin down the behavior, we think
33:44about implementation
33:46and then we try to show uh and we show
33:49actually that what are the behaviors we
33:52can derive from such an protocol
33:55algorithm that claims to implement those
33:59properties and then we go to the
34:01systems. So this way you can actually
34:03bring guarantee to what your system can
34:07provide or do or cannot provide
34:10accurately and it's very good for
34:12application builders because they know
34:14exactly how your system behaves.
34:18>> Yes. Yes. And it's that sort of
34:20uncompromising engineering you know
34:23rigorous approach
34:25uh that then actually makes things you
34:28know safe and scalable.
34:30Uh because you don't go well I tried it
34:32out and it worked with five nodes and
34:34I'll be fine with a 100 notes. No it
34:35will be [laughter]
34:37like actually
34:39>> probably not.
34:40>> That's right. It's not a problem's okay.
34:42[laughter]
34:44[gasps]
34:44>> Uh I have had fun. Thank you so much. Uh
34:48this this was
34:50this was a great discussion.
34:52>> I I hope that others enjoy it. uh we may
34:55put links uh to a couple of the
34:59u uh materials that that we addressed
35:02you know the as part of this
35:04conversation. Uh so Nicholas Kishori
35:07thank you both so much. Thank you.
35:09>> Thank you.

About this transcript

This page contains the full transcript of Blockchain Trilemma SOLVED: Turbocharged Scaling with RLNC | High-Performance Web3 by Optimum, generated from the public captions YouTube serves with the video. The transcript has 5,146 words across 778 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.