Distributed systems are composed of a set of computers (nodes) agreeing to run a protocol or suite of protocols to accomplish a common goal. This goal could be sharing a file as defined by the BitTorrent protocol or folding a protein using Folding@Home.
The most effective protocols gain resources as nodes join the network. A file hosted by BitTorrent, for example, can be downloaded much faster on average if many peers are concurrently downloading it. The speed increases because the peers provide resources while also consuming them. This characteristic is what one typically means when stating a distributed system scales.
The challenge with the design of all current cryptocurrencies is that they actually are not designed to be scalable. Blockchains, for example, are usually an append-only linked list of blocks. The security and availability of a blockchain protocol relies upon many nodes possessing a full copy of the blockchain data. Thus, a single byte of data must be replicated among N nodes. Additional nodes do not provide additional resources.
This result is the same for transaction processing and the gossiping of messages throughout the system. Adding more nodes to the consensus system does not provide additional transaction processing power. It just means more resources have to be spent to do the same job. More network relaying meaning more nodes have to pass the same messages to keep the whole network in synchronization with the most current block.
Given this topology, cryptocurrencies cannot scale to a global network on par with legacy financial systems. In contrast, legacy infrastructure is scalable and has orders of magnitude for more processing and storage power. Adding a specific point, Bitcoin is a very small network relative to its payment peers, yet struggles to manage its current load.
Our scalability goals for Cardano are greatly aided by our consensus algorithm. Ouroboros permits a decentralized way to elect a quorum of consensus nodes, which in turn can run more traditional protocols developed over the last 20 years to accommodate the needs of large infrastructure providers such as Google and Facebook10.
For example, the election of a quorum for an epoch means we have a trusted set of nodes to maintain the ledger for a specific time period. It is trivial to elect multiple quorums concurrently and partition transactions to different quorums.
Similar techniques could be applied for network propagation and also sharding the blockchain itself into unique partitions. In our current roadmap, scaling methods will be applied to Ouroboros starting in 2018 and continue to be a focus in 2019 and 2020.