Running multiple seed nodes on the same server

[wackou]

Ha, it looks like you're always a couple of steps ahead of me

I think we both agree then that a single node with combined capacity is better for the network. Also, I believe it would be beneficial for the BitShares network to be a small-world network, and an abundance of "hub" nodes does help a network to be more like a small-world network, see: http://en.wikipedia.org/wiki/Small-world_network#Properties_of_small-world_networks

However the same IP with different ports could be a router hiding different machines behind.

Yes, however if your router falls off the internet, then it doesn't really matter how many computers/processes/client you have behind it...

[emski]

There shouldn't be a benefit of holding multiple client instances except the minimum redundancy due to crashes.
However the same IP with different ports could be a router hiding different machines behind.

Similar past thread: https://bitsharestalk.org/index.php?topic=7746.0

[wackou]

I was recently wondering whether it would be more beneficial for the network to run multiple seed nodes on a same server, connecting on different ports, or a single one with combined capacity, eg: is it better to have 2 nodes with 100 connections each, or 1 node with 200 connections?

I saw on the current list of seed nodes (https://github.com/BitShares/bitshares/blob/469cd826066b05d4399aaa081eccf783a776c0e6/libraries/client/include/bts/client/client.hpp#L70-L89) that there are some hosts with multiple instances running on the same IP, so someone must have thought that this was a good idea. Intuitively I would tend to say that a single combined node is better for the network, as it makes it more tightly connected, but I might be wrong. Also, a lot of connections between multiple processes might be duplicate, while the single node ensures distinct connections.

An argument in favor of multiple instances is to provide redundancy, but you can use something like http://supervisord.org/ to mitigate for crashes and restart the client automatically, and in the case you lose network connectivity you're no better off with multiple processes than with one.

Any opinions / arguments in favor of each solution?