openstack team mailing list archive

[John Dickinson <me@xxxxxx>]

That's absolutely what is planned. The idea for global replicas involves three basic pieces:

1) A separate replication network
2) Exposing a new tier of uniqueness (the "region") in the ring management tools
3) Implementing proxy affinity so that a proxy server chooses closer replicas rather than more distant ones

The first two are simple and already have much of the infrastructure in place. The third is a little more tricky.

We will probably also need the ability to increase or decrease the cluster's replica count to make this useful to more deployers.

--John




On Nov 1, 2012, at 9:49 AM, Alejandro Comisario <alejandro.comisario@xxxxxxxxxxxxxxxx> wrote:

> John, what i think would be terrific ( i hope is not implemented, if not im gonna feel a dunce ) if, for latency matters, suppose you have 4 zones, 2 on each datacenter, and on each datacenter, you have 2 proxies for example.
> 
> De idea would be that there were some kind of mechanism to tell the ring, what nodes are under what proxies, to give some kind of prefferece regarding latency, and for example not having a proxy on DC1 to cross the country to get an object from a datanode in DC2.
> 
> Have you thought anything like it?
> If not, im wondering how, for example rackspace handles this kind of issues ( ignoring all the CDN thing )
> 
> Cheers.
> 
> ---------------
> Alejandrito
> 
> On Thu, Nov 1, 2012 at 12:55 PM, John Dickinson <me@xxxxxx> wrote:
> This is already supported in Swift with the concept of availability zones. Swift will place each replica in different availability zones, if possible. If you only have one zone, Swift will place the replicas on different machines. If you only have one machine, Swift will place the replicas on different drives.
> 
> There are active discussions right now about how Swift can support a tier above these availability zones: "regions". A region would be defined by a hogher latency link and can provide additional data durability, and, depending on your deployment details, better availability. http://swiftstack.com/blog/2012/09/16/globally-distributed-openstack-swift-cluster/ has more info on the ideas we're talking about.
> 
> --John
> 
> 
> 
> 
> On Nov 1, 2012, at 8:45 AM, Leandro Reox <leandro.reox@xxxxxxxxx> wrote:
> 
> > Hi guys,
> >
> > Any plans to implement something like hadoop rack-awareness where we can define "rack" spaces to guarantee that a copy of an object is stored for example on another datacenter, on another coast. Or should this be managed by container sync to the other datacenter
> >
> > I think that this can be a nice-to-have feature, i dont know if its on the dev roadmap
> >
> > Best
> > Lean
> > _______________________________________________
> > Mailing list: https://launchpad.net/~openstack
> > Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> > Unsubscribe : https://launchpad.net/~openstack
> > More help   : https://help.launchpad.net/ListHelp
> 
> 
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp
> 
>

Attachment: smime.p7s
Description: S/MIME cryptographic signature