question about multiple server setup

Ivan Krstić krstic at
Wed Aug 30 18:17:26 UTC 2006

Evert|Rooftop wrote:
> or.. can I add in all my servers in one big pool and make my webservers
> all share the same pool.

This is what memcached was designed for, yes.

> 1. lets say I have 80 apache forks, if I would use the big pool this
> could mean I could end up with 80*6 = 460 open sockets.. Is this feasable?

Memcached uses libevent, which in turn uses the most scalable async
connection backend on your platform. On Linux 2.6, this is epoll, which
will happily scale to tens of thousands of connections.

> 2. Are all servers accessed to see if there's a cache available, or is 1
> server picked for every request

One server is picked for each key, based on a trivial hash.

> 3. If for every request 1 server is picked, do I need to replicate data
> across servers?


> 4. If only 1 server is picked, and this server happens to be one of the
> older boxes (with less memory) does this mean 50% of my requests have a
> smaller cache to work with and will therefore be slower?


Based on your questions, it sounds like you don't at all understand what
memcached is supposed to do. You might wish to read the available
documents and Google around a bit.

Ivan Krstić <krstic at> | GPG: 0x147C722D

More information about the memcached mailing list