Does latency to the pool matter any more?

[roy7]

I'm wondering if the effort/cost to distribute additional nodes for a pool is really worth it, with stratum support in all of the major mining programs. It allows for local work generation, so the latency to/from the pool itself shouldn't impact your hashing speed much, if at all?

And if that's the case, does having globally diverse nodes for your pool make any difference? I'm working on a pool and was intending to have a few locations in other parts of the world to reduce latency for people in those areas, but given the way stratum works I'm not sure if there is any benefit at all to that now.

Thanks.

[eleuthria]

Latency still matters since stales are caused by the latency between a miner and the pool (and the pool software to some degree).  Additionally, a pool that is more globally distributed is able to push their blocks across the network faster, since having interlinked pools in multiple countries/continents guarantees the pool's block will get sent to different parts of the world faster than relying on the peer-to-peer relay, reducing the frequency of orphaned blocks.

[roy7]

Thanks. I can see getting the block update into to the client as early as possible is important to update them to the new block template to work off of.

I assume then it also makes sense to manually add as peers each of my servers, to each of my daemons, so they will share a new block internally as fast as possible. Does it make sense to manually peer to places like blockexplorer.com or blockchain.info as well, assuming they allow it?

[kinlo]

Someone with more math background should double-check my theory here, but this is how I see it:

Imagine you have a bad connection to the server, let's say ping times are 500 ms.  That means it takes 250 ms before your computer receives the command to start working on a new block. (ping times are round trip times, ie the time for a message to get to the server and back)

So when you are working on a block, you are working on it for 10 minutes on average (bitcoin rule!),  but 250ms of those 10 minutes you are working on the wrong block - you didn't yet get the message to switch.

So that means, 250 milliseconds out of 10 minutes you're doing useless work.  That equals out to 0.042 % of the time you are mining, that you are creating stales. And that's with relatively high latency.

So latency does matter, but given these numbers, even very bad latency isn't that bad...

In my opinion, the network latency doesn't really matter, the latency in the miner client is much more relevant.