maria-discuss team mailing list archive

[Daniel Black <daniel.black@xxxxxxxxxxx>]

On 15/10/16 03:33, Jon Foster wrote:
> I have a DB scenario that is very write intensive. Essentially its a
> large scale hit counter of sorts. Currently we're running on a single
> 12core server with 6 SSDs in a RAID6 array.

Not using RAID6 would be a good start.

> But we're looking for a way
> to scale out write volume by adding more servers, hopefully as
> conveniently as I might add Apache servers to a website. We see two
> servers laboring so we add a third to the mix and so on.
> 
> I'm still looking at the various technologies available to me and was
> wondering if someone out there had some suggestions on this front.
> Although Galera Cluster says it scales for write loads too

Where? This is dubious for the reasons below (which are the same as why
raid6 is poor).

> it also says that all servers write the same data

(well - *have* the same data rather than have it written)

> and make sure its committed by the time the transaction is completed.
> So it seems like write performance
> would never scale because all servers are doing the same write.
> 
> Maybe I'm missing something.

I don't think so. You seem to have that worked out.

> Anyhow some pointers would be appreciated.

MyRocks will be interesting to keep an eye on.

Answering Justin's question and details to what component in
hardware/MariaDB is the limiting factor is probably critical before you
go much further. Perhaps even getting professional help.