WAN replication (EC2)

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



I too would be very interested in what I'd call "rack / lan aware"
replication strategies (to steal a term from the Cassandra folks).

In our case, reliability is of more importance than sheer performance,
so I'm currently splitting replication pairs across EC2 availability
zones (AZ), so that we can gracefully handle individual zone failures.
 Amazon's AZ are rather well connected (these are not WAN speeds, like
replicating across their Regions would be), but inter-zone traffic
still adds considerable latency (almost 2.0ms) compared to intra-zone
comms  (0.5ms or less).   I'm not sure if full-on async replication is
the best idea in this case, but having "nearby" node awareness (i.e.
in the same rack or zone) would be helpful.

Amazon also offers multiple Regions, spread across the world,
connected at WAN speeds/latency, which we're also very interested in
having Gluster support, although I'd guess this solution might be
technically different than the one needed for above (i.e. it would
almost certainly need to be async replication).

Having multi-datacenter support is a must-have killer feature for
gluster to move beyond HPC applications and into the world of "cloud"
storage.

-A



On Wed, Oct 20, 2010 at 9:02 AM, Craig Box <craig.box at gmail.com> wrote:
> Hi all,
>
> Some months ago<http://gluster.org/pipermail/gluster-users/2010-June/004920.html>I
> was asking about running GlusterFS on EC2, across availability zones.
>
> I read in May:
>
> Continuous WAN replication is scheduled for 3.2 release (Q4 of this year).
>
>
> Is this still the case?
>
> Our use case is that we would like to be able to replicate GlusterFS in EC2,
> across availability zones, and preferably without the resulting performance
> degradation when each node has to contact all the others to check the files.
> ?Basically, our specific use case - web uploads - is that read speed is most
> important, and getting an old file is not important.
>
> Is this likely to be addressed by the work coming in 3.2? ?Jeff Darcy, have
> you anything new and cool to report?
>
> Cheers
> Craig
>
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>
>


[Index of Archives]     [Gluster Development]     [Linux Filesytems Development]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux