Great question and it is very hard to say.
There are a dozen technical bottlenecks that only devs know about.
At the current max blocksize for Hive, we increase by 225 GB per year,
But Hive blocks are never full so we have no idea if we can increase that limit.
We have to stress-test the network in the wild to know for sure.
That only comes with real adoption.
You are viewing a single comment's thread from:
It is actually more than that. Max block size voted for by witnesses is currently just 64 kB (which is also smallest allowed max size). With 10.5M blocks per year we get over 641 GB of potential new raw block log. Now, we have a compression since HF26, which cuts it in half on average, plus like you said, blocks are rarely full - it is apparent considering 7 years of block log are not even 380 GB after compression.
When it comes to consensus/witness node, during work for HF26 I've run flood test with just custom jsons (it was to expose any places in code that could be optimized that don't depend on the operations themselves). Once some optimizations were in place I wasn't able to saturate 1 MB blocks, even while sending presigned transactions. True, the test was running six witness nodes and the flooding threads on a single computer. Later we've run flood on mirrornet (with mix of actual transactions translated from mainnet, just squashed to smaller number of blocks) and 2 MB blocks (maximal allowed max block) were just fine. But flooding hived for limited time (we were running out of source transactions) is one thing. There are other essential services that must work, f.e. Hivemind, and that is a whole different story. I don't know about flood tests being run on "full stack". Hivemind in particular has a lot of work to be done still, unfortunately some of it involves changes in APIs which is always a pain to synchronize with those that are using it.
One more unknown is how the RC will behave in environment of bigger blocks and more sustained traffic. Observing flooded mirrornet was not enough due to short time frame. Personally I think witnesses should increase max block size to 96 kB (or even 128 kB), so we can observe how whole system behaves in real conditions. It won't affect us in any way during normal traffic (blocks are 22 kB on average) but should cover occasional spikes as well as prolonged high traffic during some events in Splinterlands.