Hello Mario,

in the end your workload defines which option(s) can be considered. They are 
different trade offs between read/write performance and the price That depend 
on your workload. E.g.
 - distribution of reads/writes
 - size of IO Requests (4k IO-Operations or 4MB..)
 - „locality“ of the IO-Operations (is there a small set of Data that is 
heavily used and other data more or less unused or. all stored data is used 
more or less equally).
 - required Bandwidth and latency
...

Usually SSD for the OSD-Journal in a replicated pool with size=3 is a setup 
that works reasonably well for most applications. But for an optimized setup 
you’ll have to analyze your requirements and then fit the setup to your needs 
(or hire someone who helps you).

greetings

Johannes

> Am 22.07.2015 um 02:58 schrieb Mario Codeniera <mario.codeni...@gmail.com>:
> 
> Hi Johannes,
> 
> Thanks for your reply.
> 
> I am naive for this, no idea how to make a configurations or where I can 
> starts? based on the 4 options mentioned.
> Hope you can expound it further if possible.
> 
> Best regards,
> Mario
> 
> 
> 
> 
> 
> On Tue, Jul 21, 2015 at 2:44 PM, Johannes Formann <mlm...@formann.de> wrote:
> Hi,
> 
> > Can someone give an insights, if it possible to mixed SSD with HDD? on the 
> > OSD.
> 
> you’ll have more or less four options:
> 
> - SSDs for the journals of the OSD-process (SSD must be able to perform good 
> on synchronous writes)
> - an SSD only pool for „high performance“ data
> - Using SSDs for the primary copy (fast reads), can be combined with the first
> - Using a cache pool with an SSD-only pool in front of the main disk-pool
> 
> > How can we speed up the uploading for file for example, as per experience 
> > it took around 18mins to load 20Gb images (via glance), in 1Gb network. Or 
> > it is just normal?
> 
> That’s about 20MB/s, for (I guess) sequential writes on a disk only cluster 
> that’s ok. But you can improve that with SSDs, but you have to choose the 
> best option for your setup, depending on the expected workload.
> 
> greetings
> 
> Johannes
> _______________________________________________
> ceph-users mailing list
> ceph-users@lists.ceph.com
> http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com
> 

_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to