As long as you're running a 64-bit operating system, there is IMO only one compelling reason to *not* run as one big node: in the case when you want to flush your cache or deploy a new version of memcached, if you're running as multiple little nodes, you can do the flush / rollout gradually.
But unless deploying new versions or flushing your cache are frequent operations, I'm not sure I would optimize for them. Running as one big node means you'll get much better CPU efficiency for large multi-key "get" requests from clients, you'll have less memory wasted on network I/O buffers (assuming you're using TCP), you'll have better cachewide LRU behavior (though once you're over a few GB, the difference is probably negligible from the client's point of view) and you'll have fewer processes to monitor. Most of our memcached servers are 16GB boxes running one big node each. That said, if you're not running a 64-bit OS, then you're pretty much stuck with multiple nodes per host. -Steve On 6/13/07 1:28 AM, "[EMAIL PROTECTED]" <[EMAIL PROTECTED]> wrote: > Hi Memcache users, > > > whats a good nodesize for [multiple] dual quadcore 32 gig box. > do you define 1 big node? or do you all stay with the 32-bit maximum > of 2 Gb ? > > > > Would be nice if you share your thoughts with me. > > > Thanks, > > Alex
