Cedric Fung

Mixin me 25566

Latency Matters More Than TPS

Jul 16, 2020

In a service of processing data, regardless of decentralized or centralized, both latency and throughput matter. Latency and throughput are related but not the same, and to make something usable, the low enough latency is a must, then comes the throughput.

In most centralized services, due to the high performance of CPU, RAM, SSD and the Gigabit Ethernet, latency is almost not noticeable by human being. So most technology reviews will only outline the throughput factor of services, you may find benchmarks showing the read or write OPS of a database, or millions of messages per second of a messaging queue, and they don’t even bother benchmark the latency of an operation, because at most time it never exceeds some milliseconds.

However when comes to blockchain, the fundamentals change substantially, to keep the data consistency across thousands of distributed nodes, everything is slowed down. The Nakamoto Consensus mechanism used by Bitcoin has secured the network for more than a decade and proves its reliability with the tradeoff between latency and throughput. For Bitcoin, there is a theoretically hard cap about 30 transactions per second due to the 1M block size limit, and the average 10 minutes block time means its latency level, and due to the possibility of blockchain reorganization, many people treats a transaction confirmed only after more than 3 blocks passed, that would be at least 30 minutes.

Considering the efficiency of our modern life and the Internet, needless to say 30 minutes, even 1 minute or 30 seconds are unbearable for daily usage. The sad thing in blockchain world is that almost everyone is talking about TPS without mentioning the latency aspect. Guess what’s one of the unbeatable winners in TPS?


Yes, it’s the train. Assume an 1TB SSD weighs 100 grams, and a train cabin is capable of 5 tons of stuffs, then we can mount 50,000 disks in it, and a train with 20 these cabins can transport 1 million disks, i.e. about 1 exabytes of data. That’s huge! Allegedly, “all words ever spoken by human beings” could be stored in approximately 5 exabytes of data. In 2013, Randall Munroe compiled published assertions about Google’s data centers, and estimated that the company has about 10 exabytes stored on disk. Even the fastest ethernet can’t handle this level of data transferring.

Train technology has dramatically improved since it is first invented, there is more than 100 times improvement in the speed, much faster, ergo apparently lower latency and higher throughput. But we still can’t use the train as some daily data transferring mechanism, we all prefer the low latency 5G wireless network. The train is suitable whenever we want to settle some huge data center.

Bitcoin network is the train, stable and affordable, literally fast enough. Most time people will use BTC on some 5G level distributed ledger, and make some settlement on the Bitcoin train once in a while. We don’t need another train, even faster, still a train, and our goods are very limited, like one ordinary train can already transport all words ever spoken by human beings, one Bitcoin train is also capable of the settlement of cryptocurrencies. The real obstacle prevents cryptocurrencies from being accepted by people is the lack of a low latency distributed ledger which is able to finalize a transaction within a second.

That’s what we are working on since the day one of Mixin Network.


  1. Wikipedia: Exabyte

About the Author

Core developer of Mixin Network. Passionate about security and privacy. Strive to formulate elegant code, simple design and friendly machine.

25566 @ Mixin Messenger