What are Ethereum performance dynamics?

Question

Date Dec 16th 2017.

Based on the block stats of the few last days, block gas limit being around 8 million, block time little short of 15 sec and the block used gas / gas limit ratio approaching 95 %, the current theoretical transaction cap appears to be around one million (1.0 M) transactions per day.

Apparently the block gas limit was raised somewhere around Dec 4th as the block time and size jerked upwards and block count downwards. The most noteworthy change was the ratio of uncles, which jumped from avg. 12.2% to avg. 26.1% (Byzantium era since Oct 16th). If I've understood correctly, uncles are a nuisance to those whose transactions ended in them, and detrimental to the efficiency as they have to be reprocessed?

We're probably crossing that one million txn/day milestone as I write this, or at least during the very few next days.

Questions:

To make space for more transactions, the block gas limit should be raised again?

This will either increase the block time, or, if the block time needs to stay at 15 secs, the difficulty must be lowered?

Both longer block time and lower difficulty lead to more uncles?

More uncles deteriorate the efficiency and hence the performance of the blockchain?

Longer block time / lower difficulty also exposes the chain to attacks?

How sensitive is the system for additional increases to block gas limit, and is there any other very short term measures to accommodate the exponentially increasing transaction load?

Answer 1

To make space for more transactions, the block gas limit should be raised again?

Yes. "A higher gas limit means that more more actions could be performed per-block." (link)

This will either increase the block time, or, if the block time needs to stay at 15 secs, the difficulty must be lowered?

"The difficulty dynamically adjusts so that on average one block is produced by the entire network every 12 seconds (ie., 12 s block time)." (link)

Both longer block time and lower difficulty lead to more uncles?

Yes.

More uncles deteriorate the efficiency and hence the performance of the blockchain?

That is a good question. Certainly it will lead to a loss of efficiency.

Longer block time / lower difficulty also exposes the chain to attacks?

The chance of malicious branch to be accepted in this scenario might look higher with a lower difficulty, but because this is a PoW algorithm and the number of miners hasn't decreased, the answer to your question is NO. It is not exposing the chain to attacks.

How sensitive is the system for additional increases to block gas limit, and is there any other very short term measures to accommodate the exponentially increasing transaction load?

In my opinion, it will mostly affect the size of blocks in the chain and how it grows. I'm sure that number of transactions in particular block affect mining process less than a difficulty does, which is automatically adjustible.
The measure to accommodate the growth of transaction load is one of the largest concerns of ethereum team. For now, ethereum can still comfortably handle 1 million to 2 million daily transactions, but thereafter it may start struggling. Demand, however, is not at those levels yet. Once it rises there, the capacity upgrades might be ready, which may further be complemented with Raiden, Plasma and other second layer protocols.
One more thing to add, switching to PoS might shorten the release time between blocks. I recommend you to read this question.

Commentary: I think I will return to this answer to add some more information or links and to make particular answers more specific. But if you find that this answer is missing something important or it misleads - please let me know in the comments.

Answer 2

[A partial answer... I haven't covered all the points.]

Preamble: As the size of the block increases, it will take longer to propagate through the network. This means that there is more chance it will become orphaned. This is what led to the recent increase in uncle rate.

Preamble#2: Uncles, and the use of the GHOST protocol, exists because of Ethereum's short block times, and the associated increased chance of multiple miners finding the PoW for the same block. (Where "short" is a relative term when considered against other chains - e.g. Bitcoin's block time.)

To make space for more transactions, the block gas limit should be raised again?

This would almost certainly cause a further slow-down in transactions being propagated and processed, so would likely not be a preferred solution. It would unfairly impact those miners on lower powered systems unable to deal with the larger blocks promptly, hence a proportion of the mining community would be unlikely to vote for it. It might also lead to a greater number of empty or partially full blocks, as miners start to worry about their large blocks propagating slowly, becoming stale, and leading to them missing out on the block reward.

Both longer block time and lower difficulty lead to more uncles?

No, to the first part. Imposing a longer block time would result in fewer uncles. Uncles exist as a result of Ethereum's short block time, as per Preamble#2. A lower difficulty, without an associated decrease in hashing power, would lead to shorter block times, which would lead to an increase in uncles.

More uncles deteriorate the efficiency and hence the performance of the blockchain?

It depends how you define "efficiency". Uncles exist to provide additional security to the chain, by indirectly incorporating the hashing expenditure of a losing miner's PoW exertion. You could argue that an increase in uncles helps to offset the greater waste that would occur without them.

(Related: What is an uncle/ommer block?)