Jesus arteche wrote:
> hi everyone,
>
> I'm looking for info about autoscale a cluster. I mean...with amazon
> you can generate automatically virtual machine as far as you need...if
> you configure that when the machine get 90% busy a new one will be
> created. The thing is that i'd like to do something like that for my
> database...whem my datbase get 90% busy...automatically another
> machine is created to share the charge.
>
>
> I dont know if it is possible..can anyone give some piece of advice???
this raises far more questions than answers. a few that spring to mind...
where will the data that these database nodes are operating on be
stored? will parts of it be on each node, or will each node have a
complete replica of the data?
if the data is partitioned, how will a node efficiently do a query that
requires joining data on different nodes? how will the clients know
which node to query?
if the data is replicated, what happens when two nodes try and do a
transaction that effects the same data?
if the data is stored on a shared file system (NAS, etc), how will the
nodes keep their in-memory caches coherent and keep from stepping on the
same blocks?
databases strive to be ACID, which stands for Atomicity, Consistency,
Isolation, and Durability
distributed multi-host databases like you describe are very very hard to
implement without violating one or another of those constraints.
