Re: Setup for perfromance?

[Harrison Fisk]

Jacob Christensen wrote:
> Hi,
>
> Thx for you replay...
>
> A setup like this below will then be good for
> performance?
>
> Com1: hostname= 10.0.0.1
>
> Com1: hostname= 10.0.0.1
> Com2: hostname= 10.0.0.2
> Com3: hostname= 10.0.0.3
> Com4: hostname= 10.0.0.4
> Com5: hostname= 10.0.0.5
> Com6: hostname= 10.0.0.6
>
> Com5: hostname= 10.0.0.5
> Com6: hostname= 10.0.0.6
>
> Is it importent to do somekind of loadbalancing
> between the API's? If I only send requests to API
> on com1, will the load then automaticly be spread
> out or do I have to do this myself?

There are some operations that are done on the API side and some done on the DB side. The main things that are done on the DB side currently are index requests. So when you do an index lookup, that is handled by the underlying DB nodes whereas otherwise it is done in the API node. In the future much more processing will be pushed down to the DB side as to allow for more automatic load balancing.

With that said, you will still want to have some load balancing in the API. The more concurrent operations you have going on, the higher the effective performance will be. ie. if you need to do 1000 q/s, then with 5 API nodes that would only be 200 q/s each, which would be more attainable than using 1 API node doing 1000 q/s itself.

The actual method of loading balancing you use doesn't have to be complex, a very common setup is to have one API node running locally for each application server (web server, etc...) that you have. The JDBC driver can also handle this for you automatically in recent releases.

> Is it only importent to have lot of RAM in the DB
> nodes or also in the API nodes?

As I said before, some processing does occur in the API nodes. For example if you do a query that requires sorting that can't be done with the index, then the API node will need to sort and hence use sort_buffer_size. However there is no data cache, so it isn't too much memory that is required.

Harrison Fisk, Trainer and Consultant
MySQL AB, www.mysql.com