On Tue, Oct 28, 2014 at 7:41 AM, yuelongguang <fasts...@163.com> wrote:

> clewis:
> 1.
> i  do not understand the use case of  multiple regions and multiple zones
> and multiple rgw.
> in actual use(upload, download) , users  have dealings with rgw directly,
> in this process , what do regions and zones do?
>

The regions and zones are modeled on Amazon's S3.  It's up to you to decide
if you need to have multiple clusters spread across the globe, or if you're
fine with all of your data in a single location.  Every zone (even the
replication zones) have their own URL.

Regions and zones gives you the ability to create multiple S3 sites in
different locations.
Replication gives you the ability to make a copy of each S3 site.

How you choose to combine those two things depends on your needs.



> 2.
> it is radosgw-agent to sync data and metadata.   if  data synchronization
> is finished  right after metadata is synced , or conversely?
> if not,  how is it when user access a file throught another region/zone?
>

radosgw-agent syncs metadata and data.  metadata is done first, because
metadata include the users and buckets, and we need buckets to exist before
we can copy the objects in those buckets.

There is a bit of delay between metadata and data sync.  Replication is
asynchronous, and it generally takes a least a minute before the secondary
cluster has the data.  It may be longer, depending on how long it takes
things to copy.

Because each zone has it's own URL, the easiest way for a user to access
the secondary zone is to change the URL's hostname to use the secondary
zone's URL.  The bucket, object, and user credentials are the same.


 3. during using multiple regions/zones,rgw , do they have any relations
> with cdn(Content Delivery Network)? or how to use multiple
> regions/zones/rgws with cdn?
>


If you want to get more complicated with your load balancer and geo-DNS,
you can.  It's certainly possible to direct writes to the primary zone, and
geo-balance reads across all the slave zones.

It's complicated, and I have no intention of attempting it any time soon.
I simply don't have the scale to make doing this myself practical.  I'll
keep paying Akamai.
_______________________________________________
ceph-users mailing list
ceph-users@lists.ceph.com
http://lists.ceph.com/listinfo.cgi/ceph-users-ceph.com

Reply via email to