/biz/ - Business & Finance

, DAGs have a large difference between upper and lower limit performance bounds. The reason for this is why you'll hear that DAGs have known scaling problems. Here's an easy example: say you have a simple coin transfer tx between two accounts, your VM can do 10 of these tx/s per second, and the consensus overhead w.r.t. time is baked into the 10tx/s number. In a DAG, the performance of the network varies based on how it's used. The upper bound can be near infinite IFF different users are involved in every tx (eg: acct_a -> b, c->d, e->f, etc) -- they can all happen in parallel globally because they are causally unrelated. However, the lower bound of a DAG is determined by the case where every tx is causally related (eg a->b, a->b, etc) and in this case is 10tx/s. This has a pretty good model for thinking about it (https://en.wikipedia.org/wiki/Amdahl%27s_law -- the sequential portion of a program bounds the max speed improvement). Thus, the problem is DAGs is that their scalability is dictated by how they are used -- they aren't "CryptoKitties Resistant" in that a cryptokitties-like app can and will congest the entire network. Even if you don't get a kitty, someone that you are interacting with probably has which means that you get caught in the cascading causaility web that killer-apps trigger.
Chainweb is different in that the parallelism it uses for scalability is upfront and static. How people use it doesn't impact overall performance, the exec envs are sandboxed. This means the upper and lower performance bounds are quite tight... but that the roadmap to higher scalability is clear. If the max speed of a single chain is 10tx/s, then the min and max throughput are both 10tx/s*num_chains

>>55760596
>is already
We did it reddit