accumulo-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Josh Elser <josh.el...@gmail.com>
Subject Re: Data Replication
Date Sun, 16 Oct 2016 18:00:23 GMT
Exactly right, Vaibhav.

vaibhav thapliyal wrote:
> I think neither of these would contribute much to load balancing. HDFS
> replication is mostly a safeguard against Single Points of failure in a
> Hadoop cluster. However, Data center replication would ensure the
> availability of an Accumulo instance.
>
> On 16 October 2016 at 21:02, Yamini Joshi <yamini.1691@gmail.com
> <mailto:yamini.1691@gmail.com>> wrote:
>
>     In other words, what helps in load balancing? HDFS replication or
>     Data center replication?
>
>     Best regards,
>     Yamini Joshi
>
>     On Sat, Oct 15, 2016 at 10:44 PM, Yamini Joshi
>     <yamini.1691@gmail.com <mailto:yamini.1691@gmail.com>> wrote:
>
>         So HDFS is for durability while replication is for availability?
>         I'm assuming that the client is unaware of the replicated
>         instance and queries the DB with no knowledge of which
>         instance/table will return the result.
>
>         Best regards,
>         Yamini Joshi
>
>         On Thu, Oct 13, 2016 at 11:46 AM, Josh Elser
>         <josh.elser@gmail.com <mailto:josh.elser@gmail.com>> wrote:
>
>             I'm not familiar with MongoDB. Perhaps someone else can
>             confirm this for you.
>
>             Yamini Joshi wrote:
>
>                 So, can I say that if I have a table split across nodes
>                 (i.e. num
>                 tablets > 1) and HDFS replication in my system, it is
>                 sort of equivalent
>                 to a sharded and replicated mongo architecture?
>
>                 Best regards,
>                 Yamini Joshi
>
>                 On Thu, Oct 13, 2016 at 11:06 AM, Josh Elser
>                 <josh.elser@gmail.com <mailto:josh.elser@gmail.com>
>                 <mailto:josh.elser@gmail.com
>                 <mailto:josh.elser@gmail.com>>> wrote:
>
>                      The Accumulo (Data Center) Replication feature is
>                 for having
>                      multiple active Accumulo clusters all containing
>                 the same data.
>
>                      HDFS provides replication as a means for durability
>                 of the data it
>                      is storing. The files that Accumulo creates on one
>                 HDFS instance are
>                      replicated by HDFS. This does not help if your
>                 entire cluster become
>                      unavailable. That is what the data center
>                 replication Accumulo
>                      feature solves.
>
>                      While both can be called "replication", they serve
>                 very different
>                      purposes.
>
>
>                      Yamini Joshi wrote:
>
>                          Hello
>
>                          I was going through some Accumulo docs and
>                 found out about
>                          replication.
>                          To enable replication,one needs to make some
>                 config settings as
>                          described in
>                 https://github.com/apache/accumulo/blob/master/docs/src/main/asciidoc/chapters/replication.txt
>                 <https://github.com/apache/accumulo/blob/master/docs/src/main/asciidoc/chapters/replication.txt>
>                 <https://github.com/apache/accumulo/blob/master/docs/src/main/asciidoc/chapters/replication.txt
>                 <https://github.com/apache/accumulo/blob/master/docs/src/main/asciidoc/chapters/replication.txt>>.
>                          I cannot seem to grasp the difference between
>                 this replication
>                          conf and
>                          the replication on HDFS level. What exactly is
>                 the use case for
>                          replication? Are the replicated instances
>                 visible to the clients?
>
>                          Best regards,
>                          Yamini Joshi
>
>
>
>
>

Mime
View raw message