On (de)centralization and (defining) latency:
First, there are two types of centralization: physical vs. logical. Examples: X(.com) is logically centralized, yet logically decentralized. A PoS network can be logically decentralized yet physically centralized (like Hyperliquid whose validators are predominantly in or close to Tokyo, Japan).
What's the relationship between centralization and latency? The short answer is that *logical* centralization can always provide (potentially) lower latency of many different types, at the expense of censorship resistance.
How exactly do we define latency more accurately?
This is my framework: to specify latency we need three parameters: (1) start of measurement, (2) end of measurement, and (3) guarantee provided.
For example, the start of measurement could be when a user sends out a transaction to a node or it could be when a sequencer has received that transaction.
Ultimately, the guarantees for system participants and users are the most important. Here are some latency that are of interest:
1. Inclusion confirmation latency: the start is when a user sends out the transaction, end is when a user receives a confirmation that the transaction will be included with some guarantee.
2. End-to-end execution latency: the start is when a user sends out the transaction, the end is when the user receives the full transaction effect (e.g. execution price of a defi market swap).
3. End-to-end execution & finality latency: like 2 above, but the end is measured when transaction is also finalized by consensus.
Again, all of these latency measure could be lower for logically centralized but physically decentralized systems, at loss of censorship resistance (and other properties that come with decentralization).
Latency is misunderstood in blockchain.
System response is meaningful - but that’s “intrinsic latency” meaning the delay internal to the system. E.g. from the moment you get a tx to the leader how long until it’s confirmed by the network or sequencer?
“Extrinsic latency” is about how long it takes data to get to the leader — on average — which is extremely importation. A single sequence will always have bad extrinsic latency because it only exists at one point on earth
A perfectly evenly districted POS network (population weighted) is the best you can do on a single block producer for extrinsic latency
Multiple concurrent leaders can shorten this even more.
912
0
Innehållet på den här sidan tillhandahålls av tredje part. Om inte annat anges är OKX inte författare till den eller de artiklar som citeras och hämtar inte någon upphovsrätt till materialet. Innehållet tillhandahålls endast i informationssyfte och representerar inte OKX:s åsikter. Det är inte avsett att vara ett godkännande av något slag och bör inte betraktas som investeringsrådgivning eller en uppmaning att köpa eller sälja digitala tillgångar. I den mån generativ AI används för att tillhandahålla sammanfattningar eller annan information kan sådant AI-genererat innehåll vara felaktigt eller inkonsekvent. Läs den länkade artikeln för mer detaljer och information. OKX ansvarar inte för innehåll som finns på tredje parts webbplatser. Innehav av digitala tillgångar, inklusive stabila kryptovalutor och NFT:er, innebär en hög grad av risk och kan fluktuera kraftigt. Du bör noga överväga om handel med eller innehav av digitala tillgångar är lämpligt för dig mot bakgrund av din ekonomiska situation.