[Gluster-users] Delay for geo-replication?
Anand Babu Periasamy
ab at gluster.com
Fri May 27 21:15:22 UTC 2011
inotify like mechanisms do not scale for very large volumes and
particularly distributed file systems (across multiple nodes). Another
limitation of inotify is, events have to be captured and journaled in
a log centrally. This can become inconsistent easily. We came up with
a better approach to let the file system remember these events
persistantly without meta-data server model. Apps can query old events
when ever they want across the nodes through the extended attribute
API. GeoReplication daemon uses this mechanism to replicate changes
continuously. Even if you kill the daemon for some time and restore
it, it will continue from where it left off. Events produced this
offline period wont be lost.
-AB
On Thu, May 26, 2011 at 11:42 PM, Kaushik BV <kaushikbv at gluster.com> wrote:
>
> Hi Neil,
> Please find the answers inlne.
> On Thu, May 26, 2011 at 8:27 PM, Neil Aggarwal <neil at jammconsulting.com>
> wrote:
>>
>> Hello:
>>
>> Let me ask this question a different way:
>>
>> Can someone please give me an overview of
>> the geo-replication architecture.
>>
>> Is there some process that is called every x
>> minutes?
>
> The geo-replication crawls the volume continuosly for changes (does an
> intelligent crawl, i.e crawls down the fs hierarchy) only when there are
> changes beneath, and records those changes in the slave.
> Some changes are done by geo-replication module by itself (for
> directory/symlink creation )
> Some changes are done by invoking external utility to sync them (rsync).
>
>>
>> According to the docs, the geo replication is
>> asynchronous. I assume that means it is not real-time.
>>
>> How much of a delay should I expect for files to be
>> replicated across a network?
>
> The delay by which is you can expect the files to be synced at the other end
> depends many factors such as the Network bandwidth, the time rsync takes to
> sync them (if it's a new file the whole file sent across the network, if it
> is just an edit, then only portions of it would be sent across network).
> Regards,
> Kaushik BV
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>
>
--
Anand Babu Periasamy
Blog [http://www.unlocksmith.org]
Imagination is more important than knowledge --Albert Einstein
More information about the Gluster-users
mailing list