Scalability vs Throughput myths -- A thread

Scalability - validating more transactions on the same hardware

Throughput - processing more transactions across the whole system

It's easy to increase throughput, but it always comes at a cost. It's very hard to improve scalability.

The difference between "process" and "validate" is critical here too. If your mechanism gains throughput by not validating things, it's not a scalability improvement

Myth:
Rollups improve scalability

Reality:
Rollups increase throughput

A rollup node still has to validate every transaction, just like an L1 node does. The overhead to validate both the rollup and the L1 imposes higher hardware requirements for the same usage

Myth:
Higher gas limits imply scalability

Reality:
No.

Higher gas limits are equivalent to higher hardware requirements. No scalability improvement. It increases throughput by kicking some nodes off the network and weakening the consensus process https://twitter.com/peter_szilagyi/status/1273921434700730369

Myth:
Sharding improves scalability

Reality:
Sharding increases throughput

Sharding means that nodes stop validating all transactions across the system. It turns fully-validating nodes into partially-validating nodes https://twitter.com/jadler0/status/1125481385932873728

So how can we improve scalability?

- Parallelization. Reclaim unused CPU time!
- Algorithmic improvements. Validate more efficiently!
- Keep state in memory. Don't spend time on SSD reads!
- Increase the semantic content of a tx. Make off-chain protocols use less chainspace!

ETH2 pushes towards stateless ethereum + static state access (SSA). This _is_ a scalability improvement, achieved by making Eth2.0 more like the (inherently more scalable) UTXO system found on other chains. It scales by shifting complexity off-chain, and moving state to memory

Furthermore, Eth1 delenda est.