On Fri, Oct 17, 2014 at 6:37 PM, Karim Tawfik <[email protected]> wrote:
> Thanks alot for your quick reply.
>
> What am confused about and think that is contradicting, is the first point
> mentioned that memcached server as separated and didn't know anything about
> each other, which means, if there is a data replicated on 2 server, and each
> have it own memcached, there would be no kind of keeping the data consistent
> between them, Am I right?
No, I think you are still missing the concept. The servers don't
know/care about each other. It is the clients that know about the
number of servers and split the keys across them. So nothing is
replicated. One key/value goes to one server only, and the hashing
math makes all the clients pick the same one.
> If yes, is this a good practice, or for all my clusters I should have one
> and only one memcached for'em all?
First, remember that it is just a cache, so the client needs to be
prepared to get the data from a persistent store if it isn't in the
cache. Then think about the percentage of misses that the backend
storage can handle, timing wise. The more memcache servers you have,
the smaller the percentage of misses you'll have if one goes offline.
--
Les Mikesell
[email protected]
--
---
You received this message because you are subscribed to the Google Groups
"memcached" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/d/optout.
