Investigate slow block queries and insertions

[Mirko-von-Leipzig]

See #657 (review)

Run the account generation again first for 100K accounts (to compare with prior results) and then for 1M accounts to see if the results @igamigo got were representative.

Regarding the results, if these were in fact representative, we should create an issue to investigate what exactly is driving the performance. The database is not that huge (i.e., 2.5GB) and so, it shouldn't be taking ~800ms to retrieve block inputs or 1 - 2 seconds to insert a block.

[TomasArrachea]

I ran the seed-store for 100k and 1M accounts and got somewhat better results.

• For 100k accounts took ~3.5 mins: https://hackmd.io/w0x1h5q4QbSu5uBZtwv-IA
• For 1M accounts took ~48 mins: https://hackmd.io/_5OIqh9UQ625NyU4wgeo2A

[bobbinth]

Nice! Thank you! So basically, block insertion goes from ~200ms when the database is empty to ~400ms when we have 1M accounts in it (and GetBlockInputs goes from 15 ms to 120ms). These numbers look much more reasonable - so, we probably don't need to spend more time on this now, but we should definitely investigate this more in the future.

So, maybe we close this issue and create another one specifically for optimizing block insertion and block input retrieval.

[bobbinth]

we should definitely investigate this more in the future.

I think the target we should aim for is when the DB has 100M accounts, block insertion should be around 200ms, and GetNoteInputs should be around 50ms.

btw, assuming 5-second block, this throughput corresponds to about 50 TPS, or for 3-second blocks this would be ~85 TPS.

[TomasArrachea]

Superseeded by #777