If the global Visanet does 3k tps, why would a crypto need a spec of 100k tps in the short or mid-term future? There isn't even a need to cover that many txs because those txs don't exist.
Microtransactions.
And 6 billion people who don't have credit cards now.
Visa has about 50% global market share. So double to 6k. Now multiply by 7 if you want to expand the market. That's already 42k/sec average rate (peak is likely much higher) and you haven't even touched microtransations.
And that's just domestic consumer payments.
Throw in various other financial markets and products (b2b, insurance, futures, etc) and 100k tps soon becomes quite limiting.
Some financial products and markets need more than just raw throughput though. I don't think we'll see fortune 500 stocks traded on crypto-platforms as that requires a HFT capable solution (< 1ms trades), which after spending some time looking at various tricks to achieve it, I struggled to get even < 1 sec reliably without implementing some forms of centralization. Ergo its likely impossible in a truly decentralized manner due to CAP and other information theories.
Of course there is also an adoption period, so by the time you're hitting 100k tps, the technology available can probably do 200-300k tps.
Bottom line though, I'm yet to see any tech around here actually achieve anything even close to 10k tps in a real world testing environment, let alone 100k tps. Right now there isn't anything that could absorb the transactional load should one of the current crop of platforms get some serious mainstream traction and take off.
I find that quite worrying, because it could happen and all of that transactional load will perform poorly and put the users off immediately.....then they'll never come back.
Probably the most vocal regarding supposed load capability other than myself is BitShares. They claimed (and still do) 100k tps, but that was in a controlled test with a super computer
In reality when it came to the crunch the best the could do in their test network was 1000 tps, and it crashed the network IIRC, so it was throttled to 100 tps. They may have fixed that specific problem now and are able to squeeze out some more, but I doubt its anywhere like the 100k tps they claim.
IOTA has the ability to handle quite a lot of load though. I was looking at similar structures to a DAG/Tangle over 2 years ago now and the performance was promising. Ultimately though we discarded that tech as it wasn't able to support some of our additional critical requirements other than outright speed. IOTA doesn't have all the requirements we do, so it can probably utilize the DAG in a less stressing manner and see possibly more performance than we did (lets leave the Iota POW debate where it is
).
Quite a few other projects also have claimed 10k+ tps speeds but have never actually shown any evidence of anything at all.....so....yeah
Pretty much everything so far in terms of "claimed speed" has been total hot air, in most cases < 1% of actual claimed speeds has been attainable with regular commodity hardware.
I myself am also guilty to some degree of that, some might remember about 6 months ago there was quite a lot of noise about what eMunie is able to do in theory with the channels and partitions. In theory we should be able to support in excess of 100k tps quite easily, but as of yet we haven't performed any test to try and hit it. Part of that comes from the fact that actually generating 100k tps is a pretty tall order unto itself, and to do it in a way that simulates real world to some degree (not just 1 huge box in a room spitting out transactions to a 2 node network) is more of a logistical challenge.
That said, I think the eMunie platform is still the fastest on the block at the moment, as we've done a number of speed tests in our beta network and posted the results with our fastest so far being ~2500 tps.
Still a long way off the theoretical 100k+ though....hopefully I wont have to eat too much of my hat like everyone else so far