Hacker News Reader

What's the benefit of going from 100Gb/s to 800Gb/s through kernel/hardware optimizations as opposed to adding more machines to meet the same throughput in this case? I'd be curious at what point returns on the engineering effort is diminishing in this problem.

View on HN

I think it's quite obvious that instead of 8 machines you then only need 1. This results in reduced costs for machinery, storage (as each machine would have its own storage) and probably power consumption too. Also, same room of servers can throughput 8 times more content.

Edit: Whoops, apparently this tab has been open for four hours and of course someone already had responded to you, lol.

IIRC a lot of these boxes are deployed at actual ISP's so they're closer to customers. I'd imagine the rackspace is therefore limited and the more you can push from a single machine, the better.

Reply on news.ycombinator.com