FAQ: what is the tx/sec speed of mainnet? when should we see Visa speeds 40Ktps?

[dckc]

from @leithaus Mar 25:

Here is the current milestones, which are very small.

• complete last finalized state (about 2 weeks from now)
• complete trie merge (about 4 weeks from now)

These will allow us to unthrottle the current network and should give us about 10K tps/shard on appropriate hardware.

Then we need the backing contract between shards (complete in about 2 months from now).

That gets us 40K tps for 4 isolated regions. That's Visa speeds.

Note that this requires hardware spec'd at the level Nash spec'd originally together with a very high speed internet between shards or at least backbone nodes in each shard.

[leithaus]

You conveniently left out all the language regarding hardware and networking requirements to make this happen.

[dckc]

apologies. I was trying not to copy stuff about other topics and left out too much. fixed.

[dckc]

Then we need the backing contract between shards (complete in about 2 months from now).

I believe that's also known as "cross shard transfers". Mike Stay did the bulk of the work a while ago in rchain/rchain#1265 ; polishing it off in a couple months seems feasible to me.

[jimscarver]

The current RChain tps really depends on what you mean by a transaction. Currently rchain is artificially drastically slowed down until certain features are complete. Currently processing of deploys in a propose are doce serially and each node proposes in a round robin fashion. Significantly are support for block merging RCHAIN-1517 allowing nodes to propose in parallel and nodes to process deploys in parallel to make better use of machine cores.

On testnet it currently does between 2.5 to 45 rev transfers per minute depending on the batching done in deploys. Since REV is not really native to rchain it does a dozen or so communications doing one rev transfer so it is technically doing up to about 7.5 primitive transactions per second. Once rchain is complete with sharding the sky is the limit. In the short term I expect to see an order of magnitude improvement.

More performance testing is needed and I am hoping to have a sandbox network to test with to see what happen testing larger number of transactions.

[dckc]

In migrating to github, note last Last finalized state Milestone, currently scheduled for July 21, as well as an Enable external validators Milestone, not (yet?) scheduled.

[dckc]

That Last finalized state Milestone is now 404. Sigh.

We continue to get regular development team updates... Community Update 127 of July 8 says things are moving, though I don't see an estimated delivery date for last finalized state, let alone block merge.

Improvements to last finalized state issued but quite a bit of work involved still. Significant progress, some of which have been released in 0.9.25 and more scheduled to be released in 0.9.25.1 after testing is complete.

[dckc]

https://github.com/rchain/rchain/milestone/2 LFS is scheduled for July 31, though significant risks are noted.

[dckc]

Back in March 2019, I played around with a market simulator to measure performance.

Each run of the simulator fit in one deploy, so it's not organized well for measuring speed-up from block merge or the like. It just measures parallelism of rholang execution within a node. I didn't observe any parallel speed-up at the time. Market size and execution time grew linearly, regardless of the degree of parallelism in the rholang code.

market-sim.zip

customers	seconds
64	10.173
128	11.545
128	14.488
128	15.393
256	19.452
512	29.147
512	34.607
768	52.504
768
1024	63.24
1024	61.254