Full node influence on scalability

Question

Since transactions are propagated in the network using full nodes—can it be said that the number of full nodes influences the maximum Tangle scalability? To paraphrase, even if transaction validation is done in parallel each time a transaction is incoming, can the nodes get overburdened and cause network congestion, if there are not enough full nodes?

To expand:

1. this answer states that a full node does have to process every transaction eventually
2. this answer states that spamming can harm the network if the workload is unevenly balanced (many nodes are dormant and the used ones are being spammed)
3. this question, at this moment lacking an answer, is suggesting that the entire network can be attacked by attacking a single full node since all valid transactions will be relayed to the network eventually.

This leads me to conclude that the maximum tps is determined by the maximum processing power of a full node in the Tangle. Am I oversimplifying this?

Answer 1

You are asking two questions that are somewhat contradictory:

1. "[...] can the nodes get overburdened and cause network congestion, if there are not enough full nodes?"

2. "[...] that the maximum tps is determined by the maximum processing power of a full node in the Tangle. Am I oversimplifying this?"

The first asks if more full nodes increase scalability, the second suggests that it doesn't help but that scalability is limited by the strength of the average full node.

In the current state of the network (and in the foreseeable future) it is true that adding more full nodes does not make the network more scalable. Every transaction has to be processed -- and this means not only recorded but validated by checking it's complete ancestry -- not only by each full node but most importantly by the coordinator (a single node as far as I know). See also this question. Adding a new node means that more work is done redundantly but it doesn't reduce load on the other nodes at all.

You could argue that full nodes around the coordinator can shield it from malicious spam by filtering out (not relaying) obviously contradictory transactions*. However, in order to do this, they will have to do more work than the coordinator (they will have to know and verify all transactions as well in order to assess whether an incoming tx is valid) and become new bottlenecks themselves.

* The notion that "edge nodes" can protect inner nodes is brought forward by the developers and IOTA proponents from time to time. Here is an example of a purported such statement by CfB: https://www.tangleblog.com/2017/06/15/the-tangle-is-safe-a-commentary/ (read from "To do a sudden 34% attack")

Answer 2

I guess you are right for the current state of mainnet.

However concepts like sharding, swarmnodes, horizontal scaling and hardware support in the form of JINN (and/or Q?????) will eventually lead to hopefully a highly scalable solution. All these features together will enable some form of collective intelligence where full-nodes are cooperating (without needing to know everything themselves).

A bit vague but these are just my thoughts.

References:

Swarmnodes/sharding: From IOTA's roadmap

Another approach planned to enable the IOTA client running in these very resource restrained environments is to shard the core logic and database amongst different devices that then collectively run it. Similarly to swarm intelligence, this enables a cluster of devices to efficiently make transactions without being a full node, but having reduced trust requirements from SPV and light clients.

JINN: From an old webpage from Jinn Labs (founded by CfB and David)

How powerful will a Jinn processor be?

Jinn should not be compared to processors that you buy for your computer today. Those are based on vertical scaling, meaning that the processors get more and more powerful. Jinn on the other hand utilizes horizontal scaling. I.E the increase in computational power comes from a network of Jinns (distributed computing).

But how powerful will one singular Jinn processor be?

This is unknown and also not important. Only performance/electricity or performance/density matters.

How is Jinn's horizontal scaling different from what is already being employed to mitigate the exhaustion of Moore's Law (I.E. just adding more processors) ?

Answer: Architecture is being developed in such a manner that it avoids limits set by Amdahl's law (http://en.wikipedia.org/wiki/Amdahl's_law). Architecture of x86-64 can't afford this due to marketing issues.

Q: Nothing formal is known about Q. The following article speculates about what Q might be. For sure it will be a game-changer.

“Q” brings distributed computing to the IOTA protocol with JINN processors (or emulators). It scales horizontally, benefiting from the global computational power of all other JINN devices (or emulators).