December 21, 2018

Tari Protocol Discussion 14

Monday’s architecture discussion was a debate on the optimal block size and block time for Tari’s base layer to support its second layer ‘digital assets network’. The key questions up for debate included:

  • What is the minimum viable block interval for Tari’s base layer?
  • How might block propagation time constrain the possible range of block intervals?
  • How should the Tari protocol balance the speed requirements of the second layer ‘digital assets network’ with block orphan rates?
  • Should the base layer have a ‘transactions per second’ upper limit?

The full discussion introduced many possible block implementations including a proposal for a testnet experiment to ascertain the optimal block interval & size. After the architecture discussion, this Medium article and this Bitcoin data was presented on Freenode to help the Tari community understand the fundamentals of block times and network latency.

Join us for our next discussion on Freenode in #tari-dev.

Discussion times proposed by the Tari community:

Mondays: 8pm CAT (1pm EST)

Thursdays: 11am CAT (4am EST)

To keep up with the latest Tari developments, you can follow the project on Twitter.

Transcript of Monday’s discussion

1:04 PM <@cjs77> fluffypony: Does the dynamic blocksize work well in Monero? How often has it been tested (i.e. have there been sustained periods of full blocks?)
1:07 PM <@cjs77> Clearly the **correct** way to have this debate is to haggle over it for 2 years; get some suits to pick a side, then have a user uprising to force the status quo; and then hard fork away to TariFV (Fluffy’s vision). But we don’t have that much time, so hopefully we can come to broad consensus in a few days ;)
1:09 PM <simian_za> Optimistic, if it could be done that way surely someone would have tried it already?
1:10 PM <moneromooo> There was an attack which required such a sustained high traffic period. The increase worked. That was before a lot of the recent related changes though.
1:10 PM <moneromooo> The best way to test it is to run testnet locally with a couple daemons, and spam away.
1:10 PM <@cjs77> And can you share why Monero switched to 2min block intervals from 1min?
1:11 PM <moneromooo> And maybe mine with 45% of the hash keeping empty blocks, and 55% mining default rules blocks.
1:11 PM <moneromooo> s/maybe/probably/
1:12 PM <moneromooo> The block target thing was to decrease orphan rate, and also size to some extent.
1:12 PM <moneromooo> Maybe other reasons, pony might know/remember.
1:13 PM <mikethetike> the limiting factors on minimizing the block interval in my mind are: 1. The time to broadcast across the entire network. 2. The number of transactions bundled together for privacy
1:14 PM <mikethetike> too low a block time and there will be higher possiblity of chain splits
1:14 PM <mikethetike> but also, there will be sparse blocks
1:15 PM <mikethetike> not sure if that is as big a problem for mw as it is for ring sigs
1:17 PM <mikethetike> but I’m no monero expert, so disregard if I’m talking rubbish
1:18 PM <@cjs77> I think ideally we’d want 1min blocks (or quicker) to support the 2nd layer (digital assets network); there have been huge strides in BTC to improve network propagation times. IIRC BlueMatt saying on some podcast that FIBRE has brought this down to a few seconds
1:19 PM <simian_za> Is there a rust implementation of FIBRE?
1:19 PM <@cjs77> ha ha
1:19 PM <@cjs77> I’m sure it’s next on Matt’s list after Lightning-Rust :>
1:20 PM <Hansie> Cayle: Please explain 1 minute reasoning, say vs. 2 minutes.
1:21 PM <mikethetike> What are our options given that we are merge mined with monero?
1:21 PM <mikethetike> can it be larger than monero’s 2 mins?
1:21 PM <simian_za> It should be fairly simple to model a block size vs mean propagation time that determine the rates of oprhan blocks and forks caused by the longer latencies
1:21 PM <Hansie> Can be smaller, equal or larger
1:21 PM <@cjs77> woah.. http://bitcoinfibre.org/stats_ng.html shows 95% of nodes seeing a block in 25ms
1:22 PM <simian_za> Are we linked to moneros block time? We are just using their proof of work
1:22 PM <simian_za> We set our difficulty
1:22 PM <Hansie> Block time can be smaller, equal or larger
1:22 PM <Hansie> It is totally independent to being merged mined
1:24 PM <simian_za> What is the typical traffic on the fibre network to sync mempools?
1:24 PM <@cjs77> hansie: Users of the DAN will be used to things being fast. When the base layer gets involved (e.g. node registration), things slow down considerably. I’m thinking that the difference between a few confirmations at 1min blocks vs 2min blocks will feel like eternity
1:24 PM <@cjs77> node = Validator Node
1:27 PM <@cjs77> Actually I must be reading that graph wrong. It takes ±100ms for light to travel halfway around the world
1:28 PM <simian_za> Typical sustained pings to the US are around 200–400ms
1:28 PM <Hansie> Cayle: Ok so specific use cases of 2nd layer transactions touching the base layer will be on the value system to determine block times. Along with other practical aspects and possible attack vectors.
1:29 PM <mikethetike> trying to catch us out hey @cjs77
1:31 PM <mikethetike> “the time-to-receive a block can be seen as the time taken in excess of the speed of light through fiber”
1:33 PM <@cjs77> ah.. thanks. so 200–350ms ..that’s ridiculously quick. BTC could have 10s blocks with that kind of efficiency :)
1:36 PM <@cjs77> Hansie: yes, I think as fast as we can without orphan rate getting out of hand.
1:37 PM <@cjs77> I’d argue that other things being equal, 1min 500kB blocks > 2min 1MB blocks
1:37 PM <@cjs77> wdyt?
1:38 PM <Hansie> What about an upper limit to the tps?
1:38 PM <simian_za> typical transaction size will be around 1.5kb?
1:38 PM <Hansie> Then block size is dynamic
1:39 PM <@cjs77> “dynamic”, in that it’s not strictly limited. Grin’s weight system is pretty good
1:39 PM <@cjs77> It caps blocks to 1.3–1.5MB depending on what the txs look like
1:40 PM <@cjs77> Monero is dynamic in that the blocks can grow with demand.
1:41 PM <@cjs77> simian_za, how does Beam size their blocks?
1:41 PM <neonknight> At 1.5kb for a typical transaction, it will result in only 5.55 transactions per second.
1:41 PM <simian_za> 500kb 1-minute blocks is still about 7.2 gb a day. What is acceptable on that front?
1:41 PM <Hansie> I think an upper limit to tps, say 100 tps, would be a good upper limit to calculate maximum block size.
1:42 PM <@cjs77> I think you’ll come up with a pretty big number, Hansie
1:42 PM <mikethetike> we have a second layer that increases tps
1:42 PM <Hansie> But it is demand driven, not fixed
1:43 PM <simian_za> sorry am I off by one order of magnitude? its actually 720 meg a day…
1:43 PM <mikethetike> simian_za: only archival nodes will have that size
1:43 PM <Hansie> Just allows higher base layer throughput when needed, without scaling on the base layer
1:46 PM <@cjs77> You can target X tps, but the reality is that we’re limited by network speeds and the efficiency of our implementation. The only real way to check, imho, is to pick a block interval, then start with huge blocks on testnet and then scale back until the orphan rate drops to an acceptable level (±0%)
1:46 PM <Hansie> That would give us a sweet spot yes
1:50 PM <@cjs77> tryna think.. would difficulty & hash rate impact orphan rate? Only if diff is really low, or hashrate is pushing blocks well outside the target interval right?
1:53 PM <Hansie> I think orphan rate could be minimized with efficient communication/propagation, more than anything else. Dynamic difficulty will equalize with hash rate
2:03 PM <@cjs77> This conversation has made me realise something. BTC needs a huge block INTERVAL war, not a block size one :)
2:04 PM <Hansie> :-)