Ethereum Miner Test — Results

By: Eyal Markovich, COO

This past month we ran a test on the Ethereum mainnet, working with large mining pools like SparkPool, BTC.com, F2Pool and others, including projects and companies working on top of ETH and running their own nodes. This is our first public testing on a mainnet and the results thus far have been extremely positive.

We had several goals in mind when designing this test; confirm that the bloXroute BDN brings value to the Ethereum network and allows it to scale, test the stability of the network when deployed in the wild, and test new features not included in the previous BCH Testing.

Our next step will include adding additional miners which will allow us to see a significant drop in uncle rates by propagating data through the BDN. The test has been very successful thus far and we are excited to share with you our initial results (expanded below) as we continue to work on scaling Ethereum and allowing it to increase its gas limit.

TL;DR:

• We saw propagation speeds increase by over 100% for some nodes using the bloXroute BDN, cutting the time until blocks are received by more than half!
• Overall we saw an average block propagation time of 190ms with the BDN versus 250ms without the BDN, which is a 25% improvement.
• In some cases, the time for a miner to receive a block was cut by 23% on average.
• As expected, non-mining nodes saw an even greater improvement compared to the better connected miner nodes.
• Block propagation to/from China is a challenge which the BDN helps alleviate.
• Ethereum blocks are currently very small, limiting the improvement bloXroute can showcase. We anticipate presenting an even greater improvement when using a higher gas limit.

These results are extremely promising and show that we can scale Ethereum.

Test Methodology

The testing was divided into two stages — with and without the BDN. Participants ran an ETH logger that reported the blocks mined by miners and the propagation of these blocks to the nodes participating in the test. Major miners ran the logger and connected to the BDN (SparkPool, BTC.com and F2Pool). Several non-bloXroute full node non-miners also reported on block propagation. Additionally, bloXroute used six benchmark nodes in three data centers (US, Europe and Asia- Pacific). Three of these benchmark nodes were not connected to the BDN in both stages of the test. The two stages ran for the same duration and during the same time range (one week later).

The Results

We measured propagation speeds of blocks mined in the US and in China to various regions around the globe (i.e. Asia Pacific, Europe, China and the US). We were able to collect the actual real propagation (excluding validation time) data on roughly 2,000 blocks mined in China and 150 mined in the US.

Overall, we saw an average block propagation of 250ms without the BDN vs 190 with the BDN (25% improvement).

The above graph shows the performance difference for the same node by using the BDN. Note the vast improvement across the board, where some nodes observed up to a 50% improvement.

For example, the utmost left point represents the average duration of blocks from China to a full non-miner node located in Oregon. Without the BDN, the average propagation time to this node was 0.42 seconds, while with the BDN it was reduced to 0.35 seconds, a 17% improvement. Even well provisioned miners saw an improvement: for example, blocks mined by a US node reached a different mining node in China in just 0.28 seconds, showing a 23% improvement. Below is the full tables of results.

Propagation to/from China

Similar to what we have seen in the BCH testing, sending and receiving blocks from China continues to be a major challenge. The graph below shows the average block propagation time from/to China. As the graph indicates, without the BDN the performance is not consistent and changes throughout the day. With the BDN the performance is both faster and more consistent, which is of huge value for the miners.

Uncle Rate Discussion

For our initial tests bloXroute connected to a limited number of nodes controlled by each pool to mitigate any operational risk, so only ~5% of newly-mined blocks were directly propagated through bloXroute. Given that 95% of the blocks were not directly propagated through bloXroute, we did not expect to see a decrease in the uncle rate. However, examining the uncle rate two days before the BDN test and two days after the test revealed a very small reduction in the rate. This reduction was not statistically significant but it is an indication to the potential improvements with the BDN.

In our next step we intend to increase the number of connected nodes, including both new and existing pools, and to register a stronger effect on the cumulative uncle rate.

Conclusion

Thank you to our partners, including SparkPool, F2Pool, BTC.com, Syscoin, and ArcBlock for participating in the test, and thank you to our talented Development Team who has worked tirelessly across time zones.

For next steps, we plan to release Version 1, perform a larger test with more hash power connected to the BDN, and potentially test increasing the gas limit on the ETH testnet with the BDN.

The best way to stay up to date regarding our V1 launch and other company information is by following us on Twitter @bloxroutelabs.

Results Appendix