17 Jul 2018 |
komodobridge | [discord] <Mylo(KomodoPioneers)> Yes, can we run a small scale for 12 hours doing 2k tx/s with stats working the whole time?
What is the cost difference between 2k, 4k, 8k tx/s for the purpose of this test? | 08:07:54 |
komodobridge | [discord] <blackjok3r> 64 chains seems to cap out at about 7000tx/s of single payment, but sustaining that for long periods is difficult... if you increase it out to 3-4 payment TX's it should be possible to run indefinitely. | 08:09:01 |
komodobridge | [discord] <blackjok3r> Or I can go back to an earlier iteration, that uses 2 TxBlasters to blast for unlimited time. | 08:09:28 |
komodobridge | [discord] <blackjok3r> It was just not possible to use that version at very large scale of chains, dude to limits of hardware available. | 08:09:54 |
komodobridge | [discord] <Mylo(KomodoPioneers)> Is there a bottleneck with the single payment/ | 08:10:52 |
komodobridge | [discord] <blackjok3r> yes on the transaction generating side... it requires 2 marketmakers/komodods to broadcast enough transactions to fill the mempool. | 08:11:21 |
komodobridge | [discord] <blackjok3r> When you are using one wallet to simulate over 1000 wallets, its a little hard. | 08:11:38 |
komodobridge | [discord] <blackjok3r> If we had 1000 wallets to send TX from.... then it would not matter | 08:11:49 |
komodobridge | [discord] <blackjok3r> Once again, limited by hardware. | 08:12:15 |
komodobridge | [discord] <blackjok3r> Short of actually using it for a national currency, its all but impossible to get that many users. | 08:12:45 |
komodobridge | [discord] <Mylo(KomodoPioneers)> so 64 chains doing single payment ~= 7k tx/s
doing 5 payments allows for marketmaker to keep up with blasting to fill mempool quick enough by sending transctions containing more bytes but sustainable for longer. Load testing endurance rather than throughput.
?
Users will be machines and dApps. | 08:17:51 |
komodobridge | [discord] <blackjok3r> Yes, I have not actually tested a threshold for payments per TX that allows sustained full blocks, but 100 needs nothing, 10 chains will be blasting at any one time over 64 chains. but 1 payment is not really possible to keep mempool saturated. For this I used a mining pause at the start to fill all the mempools up. | 08:19:38 |
komodobridge | [discord] <blackjok3r> I can definatly do that, although like I said there is an older version, that can sustain single payments, it just needs an extra CPU per chain. | 08:20:06 |
komodobridge | [discord] <Mylo(KomodoPioneers)> Nah let's not use older version imo. Endurance is another metric that will need to be known once throughput is grokked. | 08:20:53 |
komodobridge | [discord] <blackjok3r> Not really, because the transactions being generated in this way do not apply to any real world use... they are a simulation. | 08:21:33 |
komodobridge | [discord] <blackjok3r> What is being tested is the amount of confirmed TX that is possible. | 08:21:46 |
komodobridge | [discord] <blackjok3r> Testing the TxBlaster API over endurance is pointless, and I think this is why James said a 15 minute window of all chains with full blocks was all that was required.. I did have a single chain locally runnign for over 5H with full blocks though, so it can be run for a long time | 08:23:25 |
komodobridge | [discord] <Mylo(KomodoPioneers)> OK - very specific goal. Got it. | 08:25:49 |
komodobridge | [discord] <blackjok3r> However if you just want data to say a blockchain can run at full block capacity, you could just point to BTC at the end of last year... every block was full for months straight. | 08:26:48 |
komodobridge | [discord] <blackjok3r> I dont see a need to prove what has already been proven. | 08:27:07 |
komodobridge | [discord] <Mylo(KomodoPioneers)> OK fair point. | 08:29:39 |
komodobridge | [discord] <Mylo(KomodoPioneers)> What has been the largest scale test tx/s to date? | 08:30:05 |
komodobridge | [discord] <blackjok3r> The 1024 chain test. | 08:30:18 |
komodobridge | [discord] <blackjok3r> but they were 100 pyamnt TX's so the 20,000 number is far lower than theortically possible. | 08:30:42 |
komodobridge | [discord] <blackjok3r> The 256 chain test, to test kubernetes should eclipse that number easily... theoretical maximum is ~37k tx/s. | 08:32:27 |
komodobridge | [discord] <blackjok3r> If we need higher than that we can use 512 chains to double it. | 08:33:12 |
komodobridge | [discord] <blackjok3r> Either size will give us the data we need. | 08:33:31 |
komodobridge | [discord] <Mylo(KomodoPioneers)> It's pretty much processing bytes of tx data across a decentralized chain cluster. Correct?
100 chains ~= 200MB/min
1000 chains = ~=2000MB/min
10000 chains = ~=20000MB/min | 08:50:52 |
komodobridge | [discord] <Mylo(KomodoPioneers)>
https://cdn.discordapp.com/attachments/449949868904022036/468701212317843466/Screen_Shot_2018-07-17_at_6.50.21_pm.png | 08:51:17 |
komodobridge | [discord] <Mylo(KomodoPioneers)> I want to build a template/HOWTO so others can take our work and implement their own test. | 08:51:48 |