Session 02 Mysticeti v1
Summary
TLDRThe speaker discusses the engineering team's breakthrough in transaction finality, introducing 'Misti', the world's fastest consensus engine aiming to achieve instant finality for shared objects. They highlight the importance of reducing latency for a better user experience and outline the innovative multi-leader DAG-based consensus protocol, which significantly reduces latency and promises unlimited throughput, setting a new standard for web 3 and global decentralized networks.
Takeaways
- 🚀 The engineering team has over 70 PhDs working on cutting-edge technologies like consensus, cryptography, and storage for web 3 and the broader web.
- 🔍 They are developing a solution called 'Misti' to achieve instant transaction finality for shared objects, similar to the speed of single-writer objects.
- 🏁 Misti aims to reduce transaction finality time significantly, with current industry-leading times being 2 seconds, and the goal being under half a second.
- 📉 Latency in user experience negatively impacts satisfaction, with each second of delay causing a 19% worse experience, highlighting the importance of reducing wait times.
- 🔄 The company's blockchain has unlimited throughput, emphasizing the need to focus on reducing latency beyond just increasing throughput.
- 🛠️ Different transaction types have varying finality times; single owner objects are instant, while shared object transactions take longer and are the focus of optimization.
- 🔧 The 'fast path' allows single object transactions to bypass consensus, significantly reducing their finality time to 500 milliseconds.
- 🌐 The company uses a Directed Acyclic Graph (DAG) based system for consensus, which offers advantages over linear-based systems in terms of resilience and performance.
- 📈 The introduction of 'MRTI' (presumably a next-generation consensus mechanism) promises an 80% reduction in latency, with block times as low as 80 milliseconds.
- 🎯 MRTI's design allows for a more efficient consensus mechanism, which does not require verifying the DAG for each block progression, streamlining the process.
- 📚 The company's research and development are peer-reviewed and scientifically robust, with educational institutions set to teach MRTI as a course.
Q & A
What is the main focus of the engineering team's work as described in the script?
-The engineering team is focused on developing advanced technologies related to consensus, cryptography, and storage, aiming to revolutionize the world of web 3 and the web as a whole.
What is the current challenge the team is trying to address in transaction finality?
-The team is working on making transaction finality for shared objects as fast as single writer objects, aiming to achieve instant finality even when transactions require coordination.
What does 'Misti' stand for and what is its significance?
-Misti is the name given to the world's fastest consensus engine developed by the team. It signifies the breakthrough in achieving faster transaction finality and reducing latency in blockchain transactions.
Why is latency a critical issue in blockchain technology according to the script?
-Latency is a critical issue because it directly impacts user experience. Research has shown that every second of latency can worsen the user experience by 19%, which is why reducing latency is a priority.
What is the current transaction finality time for shared object transactions on the blockchain mentioned in the script?
-The current transaction finality time for shared object transactions is between one and a half to two seconds.
How does the script differentiate between single owner objects and shared object transactions?
-Single owner objects, such as asset transfers, happen instantly within 500 milliseconds. Shared object transactions, like marketplace transactions and collaborative games, currently take longer to finalize, between one and a half to two seconds.
What are the key components of transaction finality that the script mentions?
-The key components of transaction finality mentioned in the script are block time, transaction finality, and settlement finality, which all contribute to the overall speed and efficiency of the blockchain.
What is the significance of the 'fast path' in the context of the script?
-The 'fast path' allows single object transactions to bypass the consensus process, leading to a faster transaction finality of 500 milliseconds, showcasing the potential for high-speed transactions.
What are the two types of consensus algorithms discussed in the script?
-The two types of consensus algorithms discussed are linear-based consensus algorithms and DAG (Directed Acyclic Graph) based systems, each with their own advantages and disadvantages in terms of latency and throughput.
What is the goal of the 'MRTI' consensus protocol as described in the script?
-The goal of the 'MRTI' consensus protocol is to combine the benefits of a DAG-based system with the low latency of a linear-based system, aiming to achieve an 80% reduction in latency and an 80-millisecond block time.
What are the expected improvements with the implementation of MRTT in the mainnet by July 2024?
-The implementation of MRTT is expected to reduce consensus latency from 1.9 seconds to 390 milliseconds and settlement end-to-end latency from 2.2 seconds to 690 milliseconds, making it the fastest chain for latency in the world.
Outlines
🚀 Introduction to the Engineering Team's Innovations
The speaker introduces the engineering team's efforts to revolutionize transaction finality in shared objects, aiming to match the speed of single-writer transactions. The team's expertise in consensus, cryptography, and storage is highlighted, with a focus on Misti, a new consensus engine designed for instant finality. The current state of blockchain transaction times is discussed, with a goal to reduce latency and improve user experience, as every second of delay negatively impacts satisfaction.
🔄 Understanding Transaction Lifecycle and Consensus Mechanisms
This paragraph delves into the lifecycle of a transaction on the blockchain, from creation to finalization, and the complexities involved in shared object transactions. The concept of 'fast path' for single object transactions is introduced, which bypasses consensus for speed. The engineering team's challenge to improve performance for consensus objects is presented, along with an explanation of linear-based and DAG-based consensus algorithms, and the advantages of the latter in terms of avoiding bottlenecks and providing higher throughput.
🌐 Introducing Misti: The Future of Consensus Protocols
The speaker reveals Misti as a groundbreaking consensus protocol that significantly reduces latency, with an 80% improvement over traditional methods. Misti operates on a DAG-based system, offering a lower block time of 80 milliseconds and faster transaction finality. The benefits of Misti include its multi-leader approach, which ensures robustness against poor performance or downtime, and its potential to finalize transactions without full consensus in some cases. The paragraph concludes with a preview of the dramatic reduction in latency that Misti will bring to the blockchain.
🏆 The Impact of Misti on User Experience and Future Developments
The final paragraph emphasizes the transformative impact of Misti on user experience, promising an unmatched level of satisfaction in a decentralized network. The speaker outlines the expected latency times for single writer and shared object transactions, which are set to be the fastest in the industry. The potential for Misti to be taught in universities and its peer-reviewed status underline the protocol's significance. The paragraph concludes with a look forward to further integration and improvements in the second half of 2024, solidifying the protocol's position as a leader in user experience and throughput.
Mindmap
Keywords
💡Engineering Team
💡PhDs
💡Consensus
💡Cryptography
💡Web 3.0
💡Transaction Finality
💡Misti
💡Latency
💡DAG (Directed Acyclic Graph)
💡Throughput
💡Settlement Finality
Highlights
Our engineering team includes over 70 PhDs working on consensus, cryptography, and storage innovations.
Misti is the world's fastest consensus engine, aimed at achieving instant transaction finality.
Current blockchain performance allows single writer transactions to finalize in half a second.
Shared object transactions currently take about 1.5 to 2 seconds to finalize on average.
The focus is on reducing latency and improving user experience by minimizing transaction finality time.
Execution of transactions can happen within 10 to 20 milliseconds, but consensus currently takes longer.
Misti aims to reduce latency from 2 seconds to milliseconds for faster transaction finality.
Misti is expected to have block times of 80 milliseconds, making it the fastest blockchain in history.
Misti uses a DAG-based consensus protocol to improve performance and reduce latency.
DAG-based systems offer more robust performance without being affected by single leader failures.
Misti's consensus mechanism allows for finality in 0.5 round trips, significantly reducing latency.
The introduction of Misti is expected to cut latency by 80%, with end-to-end transaction finality in 690 milliseconds.
Misti will be integrated into mainnet by July 2024, promising the best user experience for decentralized networks.
Misti's improvements will lead to unmatched user experience and performance in DeFi protocols.
Misti's innovations are recognized and taught in universities, highlighting its academic and practical significance.
Transcripts
[Music]
[Music]
TR
[Music]
[Music]
so I'm going to try my best to do as
much Justice to this talk as possible um
our engineering team are truly some of
the best in the industry um we have in a
company over 70 phds working on
consensus
cryptography um all sorts of really
interesting um bleed bleed N Stuff all
the way to storage that we think is
going to change the world for web 3 and
just the web as a whole um one of the
things we've been working on is how do
we
make um transaction finality for shared
objects as fast as single writer objects
right now I was explaining before that
to transact onui to do you know sending
money to someone sending a token to
someone sending an nft asset to somebody
all that happens instantly right now in
half a second what we want is when you
have a transaction that's touching a de
or any uh anything that requires
coordination how do we get that to
instant finality as well as as fast as
we have for um single writer objects
Misti is what we've called it Misti is
the world's fastest consensus engine um
people talk about throughput all the
time throughput is interesting we've
already solved that problem by having a
blockchain that has unlimited throughput
but what matters how do we go to instant
finality latency is the hardest problem
to solve Beyond
throughput in this example we're just
showing you what's happening Inu right
now from when you approve a transaction
to when it's final it's taking about 2
seconds that's actually really good it's
faster than all the other blockchains
today it's actually really good but I
think that's actually from a user
perspective can we go better than that
right it'll be amazing if that click is
effectively instant as I click the
button the finality is instant as a user
would be much more pleased there's some
research that's been undergone that
every second that you have in latency
for user experience makes it a a 19%
worse experience for the user so users
are going to be more and more upset the
worse it gets that's why many people are
un un um are not happy about blockchain
performance today but we're talking
about two seconds being the industry
leading today we're not happy we want to
go way below that we think half a second
should ultimately be the goal
now there are different types of
transaction types onu as I mentioned
there is um the single owner objects
that's just an asset transfer um imagine
a game making moves in a video game you
do that singly and that's that's
happening instantly um NF transfers all
these things happen onui within 500
milliseconds so when you click a button
it's finalized within half a second but
what about the other transaction types
we're talking about shared object trans
action types you're talking about
Marketplace transactions you're talking
about um collaborative games if you have
multiplayer games I have two or three
people playing a game or more these
things right now are taking between one
and a half to two seconds to finalize
onu um even you know public accessible
counters if you and I need to execute a
transaction that requires an increment
increment of a counter at any given time
right now that requires coordination it
goes through consensus so a bit of
discussion about sweet speed the most
important thing to think about is there
are multiple components of this right
there is block time transaction finality
and settlement finality these are there
multitude of things that's involved how
fast can you propagate blocks how fast
can you get transaction finality and how
fast can you get settlement finality
most people trying to give you
optimistic numbers but those are great
because when the world goes well but
most times the world doesn't go so well
so what are the what are the real
numbers for finality from a user
perspective when I've clicked the button
and I know it can never be reversed so
if you think about what matters it's the
protocol efficiency which is the number
of round trips across across the
internet to the performance of a node
consensus is the most expensive part
execution is very low cost execution 10
milliseconds is possible you can get
maybe 20 30 but realistically onu
execution is happening within 10 20
milliseconds very very fast in parallel
consensus as we say right now is taken
above a second and a half to to do which
is how do we get more efficient
consensus mechanisms
onu this is howu works today this is a
life cycle of a transaction onu you
would create a transaction once you
create a transaction it would get
effectively you collect a set of
certificates once you have a set of
certificates it would go for checks and
verification once it's been verified it
would go to ordering once it's been
ordered it will go effectively to
execution once it's been executed it's
final so multiple steps required which
is what's taking up to 2 seconds today
so there are multiple complicated
processes that have to happen for a
transaction to be final if you see
underneath there's something called fast
path in a fast path which is a single
object um transactions I talked about
before you bypass consensus it's not
needed so from when you verify a
transaction you can go straight to
executing it and that's happening in 500
mill seconds can we get performance
almost as fast if not faster than what
we have today for single writer objects
as we have for consensus objects that
has being the task given to our
Monumental engineering team today so now
you know the anatomy of a three
transaction let's talk about how it
works with shared objects first
transaction is submitted it's processed
I get a certificate once I have a
certificate it goes straight to the
nodes for validation once it's been
validated it goes to consensus for
ordering once it's been ordered it goes
to execution once it's executed we
basically there's four R round trips
required for that and ultimately it's
finalized so you have something about
you have up to about um four and a half
rtts involved in that whole process butu
is a very complicated system what can we
do to simplify that and what can we do
to get even more performance out of the
system
today there if I want to draw your
attention of two types of consensus
algorithms there are linear based
consensus algorithms where you submit
you create a block once you've create a
block you for it you send it you you
refer to the previous transactions in a
block you you basically have another
leader propose a um um the block you
refer to that and you keep going over
and over and over again the bottleneck
is by the you can get bottleneck by a
leader that's got poor band with you
basically lead if a leader fails you
have low throughput as a result which is
a problem with most consensus based
systems today and low latency it's about
two to three round trips very very low
very very um low latency but very
susceptible to
failure dag systems which is what we
have today not bull shark onui today
onui is a dag based system much more
complicated but you can see that
ultimately what we're doing is every
single round we have multiple gossips
happening where blocks have been
proposed in parallel over and over and
over again ultimately the benefit is you
have it you have more latency but what
you have is you're not susceptible to
bad um performance of a single leader
and over time you also have um you know
it's it's a more lightweight algorithm
from a perspective of implementation
from our perspective but at the same
time you have stronger guarantees as
well so we think a dag based system is
the way to go it's we use as a dag BAS
system today but the the the downside of
a dag BAS system we have is actually an
increase in
latency so again traditional leader
driven um it's linear communication low
latency two to three round trips and
it's of course leader the con is a
leader is a Bott neck in a in a in a
protocol our dag based consensus is
effectively way faster like it's you can
get more throughput out of it as a
result but again as I touched on we
think you know we want to get better in
terms of latency and finality so what is
mrti Miss setti is is our ambition to
effectively make a dag based consensus
protocol have as low as uh um um as low
latency as you get with a linear base
system and we're able to achieve that
our consensus 1o today is final at 1.5
round trips effectively for a single
round it takes about one to five round
trips sorry with Mr setti consenso it's
half a roundtrip to create a single
block in fact the block time of mrti is
80
milliseconds it is the lowest block
block time of any blockchain in history
80 millisecond block times as a result
of
0.5% round trips to create a single
round in missii it is a game Cher it is
a industry leading mechanism and the
reason we can do that is we do not have
to verify the dag as part of progressing
the the block
so the difference is as I mentioned
consensus 1.0 require the verification
of a dag so every time you receive a dag
it has to go through a certification
with consensus 2.0 it it requires an
unverified dag so after about um two to
three rounds you've got full
verification that a transaction is final
ultimately it's it the committer done in
pipeline in parallel and multileaders
you have multiple leaders if one leader
is out it doesn't matter the chain
doesn't know notice it keeps going on it
it can take care of um of downtime um it
can take care of bad leaders it take
care of people who have very low
performance it's not affected by the
worst leader like most Protocols are but
ultimately what it gives us is what
we're seeing an 80% reduction in latency
today this is what we have in testet you
could see this is naral and bull Shar
that we're running right now that is
showing close to 2 seconds and as soon
as Miss got turned on it cuts down to 20
397 milliseconds
by the time this hits production this
will be the fastest chain for latency in
the world s we already believe it's
there but we think Mr setti just sets a
whole new bar for what users should
experience in a Global decentralized
Network another benefit as well is p50
shared object latency what does that
mean 50% of users which is going to be
the average of users that engage withu
will see finality end to endend finality
from when you click a button to when
it's Final in 690 milliseconds that is
insane before the website is loaded your
transaction is final that means the
amount of use Cas that you can engage
with is broader we just gave you a card
this card that you scanned today when
you put it on your phone and you clicked
on approve it took about 2 seconds to
finalize that is to do with now war
bullshark in a few months once it's live
in mainnet when you do that same scan
when you click on approve it will be
instantly Final in your account so that
is what it means about building the best
user experience for users over and over
again is thinking about all the way from
when a user submits a transaction to
when the chain gives you feedback that
it's final this is endtoend latency
optimistic latency is a lot lower way
lower block times of 80 milliseconds
through entend finality of 690
milliseconds But ultimately the user
will have the best experience that they
can ever have of any chain not having a
not not going to see failed transactions
not going to see latency Spike up in
times when a network is busy not going
to see a maximum throughput ultimately
the SC ch um they change skilles
infinitely will be releasing we released
a paper already called pilotfish and
pilot fish was able to show that when
you added eight times more machines to S
you got eight times a throughput so
there is no maximum throughput
separately the latency is constant no
matter how many transactions you throw
atui and the amount of um U machines you
throw at the network the latency always
stays same and the throughput goes
higher and higher this is how internet
skill infrastructure should be
built and one thing we're going to be
working on is effectively cutting down
the latency again for single writer
objects right now I explain that there
are 500 milliseconds by the time we
merge the code base between um the
ingestion phase and also the
verification phase we think we'll be
able to get latency for single writer
objects end to end down to 300 M Ms so
you have 300 milliseconds for single
writer objects Final End to End finality
and 690 milliseconds end to end finality
for shared objects in fact we've seen as
low as 500 milliseconds for shared
objects in testut as well so we think
we'll have low subsequent finality for
all transactions inui over over over and
over again um another thing of pilot
fish is we think we can actually get a
scenario where some transactions that
need consensus can be finalized without
consensus that's going into too much
detail but there are people who have way
more phds on me on this topic that can
talk about that in detail but we think
that's going to be another variation of
what we can do
here so to summarize Miss setti 80%
latency reduction consensus has gone
down from 1.9 seconds to 390
milliseconds settlement end end to end
from 2.2 seconds to 690 millisecs um it
will be in main net by July 2024 in fact
it's running in test net now the
numberers is already published it's
running the same codebase the same
validator setup across the globe and
we're seeing remarkable results and we
we want to turn on in main net in July
and we're going to be integrating um um
congestion um phase with mrti in uh
basically H2 2024 so there'll be a lot
of more upgrades coming to S as a whole
But ultimately we think D5 Protocols are
going to see an amazing Improvement in
user experience it's already Indy
leading but I think it's going to take
it to a whole new level but overall we
think more most importantly the user
experience across wi would be just
unmatched unrivaled to what any other
ecosystem as made possible this is all
possible by having world renowned
researchers in a team everything we have
is peer-reviewed and has some of the
most amazing scientific reviews um in
the world that looked at it it's all
verified all it's it's not just code
that goes out with with uh all sorts of
caveat but these are things that you
should read up on yourself we actually
have some University they're going to be
teaching Mr setti as a course in the in
universities as a result because for the
first time they can actually understand
a proper division team for tolerant
protocol with a dag based system in a
way that it's easy to actually
internalize so we believe we're making a
lot of great changes for the world but
more importantly we think we're building
the best developer platform for web 3 as
a
whole um if you want more information on
mrti do scan this QR code I'm happy to
answer any questions you may have but
ultimately this is a summary of the
elements of missii it's a multi-leader
dag based consensus protocol it has
industry-leading latency um lowest
latency that's ever been found in any
protocol it orders are maned faster than
what exists today and it's supported by
it's going to be supported by pish
effectively unlimited throughput base
system thank you very much thank you
[Applause]
Ver Más Videos Relacionados
Enabling Real-Time Applications on the EVM with Somnia by Paul Thomas, Somnia | Modular & L2 Day
Session 01 Catching Sui Alpha v2
OSDI '21 - Bringing Decentralized Search to Decentralized Services
THIS WILL DRAMATICALLY REDUCE YOUR LATENCY!
FPGAs and low latency trading - Williston Hayes - Optiver - FPL2020
In 5 minutes Unlock the API powers REST gRPC GraphQL Websocket SOAP #api #restapi #graphql #soap
5.0 / 5 (0 votes)