[MarkLogic Dev General] SSD and fast data directory
mike at blakeley.com
Tue Oct 1 20:26:41 PDT 2013
I think it's a good bet. An SSD is usually a good 10x faster than a traditional HD. So the average journal write should still return more quickly than it would if you dedicated a single disk to forest journals and stands. One potential risk is that an SSD might display pathological response times when the device is nearly full, or due to a firmware bug.
If the entire database fits on SSD with merge space to spare, then I'd set the data directory there to begin with. As far as I know the server uses the fast data directory for journals, plus as many stands as will fit. So if the fast data directory never fills up, the main data directory should stay empty.
On 1 Oct 2013, at 16:20 , Will Thompson <wthompson at jonesmcclure.com> wrote:
> The example in ML's docs suggest an 8-core server with 4 forests, all
> sharing one SSD fast data directory. Is it fair to assume that even under
> heavy load the single-threaded speed of an SSD will always outperform the
> 4 forests of magnetic disks? It seems like there might be scenarios where
> having multiple IO threads would be beneficial, even if they're not SSD
> speeds. I realize this may depend on exactly *how* fast of an SSD and HDD,
> but are there any rules of thumb (or exceptions to look out for)?
> Also, assuming you could fit your entire database on an SSD, what
> differences are there between putting the forest directly on the SSD vs
> putting it on HDDs with the fast data directory configured on the SSD?
> General mailing list
> General at developer.marklogic.com
More information about the General