Questions on operating a dedicated membase cluster
I'm trying to do some capacity planning for our soon to be introduced cluster of membase. I'm having a hard time finding any anecdotal information on operational best practice. I have found a few writings that have highlighted a few things that I would appreciate some feedback on:
a) I've read in a number of places that people have gone with one thread per CPU core. Why is this? If membase is multithreaded, processes requests quickly, on a fast system, why wouldn't it be faster with a few more threads per CPU core? I understand diminishing returns, but is there a sweet spot?
b) Most cases I've read have a standard gigabit ethernet setup. Our application makes a ton of requests to our current local PO cache. I worry about saturating the network between our app servers and the membase cluster now that we're looking to move to a dedicated cluster. Does this worry have merit? What have others done?
c) The default max number of connections is 1024 and I've read instances of running out of connections. Is there any harm in setting this value to say something like 10240? What might be the negative impact of this?
I have most of the memory settings figured out, which is more specific to our use. Is there anything else I should worry about that deviates from the default setting?
Thanks for your time.