ignite-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Andrey Gura <ag...@apache.org>
Subject Re: Partitioning on a non-uniform cluster
Date Mon, 14 Nov 2016 12:44:16 GMT

In order to implement resources aware affinity function you can use node
attributes (available via ClusterNode interface). But node attributes
should be initialized before Ignite node started (except of default node
attributes that can be found in IgniteNodeAttributes class).

Also you can start one Ignite instance on 16GB and two instances on 32GB.
In this case you should configure RendezvousAffinityFunction with
excludeNeighbors == true flag in order to increase cluster reliability.

On Mon, Nov 14, 2016 at 3:09 PM, Krzysztof <yazuna@gmail.com> wrote:

> Hello,
> Judging by the documentation and some discussions on this list, can you
> confirm that Ignite cache does not take into account different memory
> settings, i.e. if we have various nodes with 16GB and 32GB allocated for
> cache, there would be no two times more partitions assigned to larger
> nodes?
> In order to not to underutilize larger nodes or overfill smaller nodes we
> would have to develop our own affinity strategy via AffinityFunction in
> order to make it cache-size aware?
> RendezvousAffinityFunction seems to be completely resource-blind?
> Could you please clarify what would be the best way to achieve balanced
> distribution cluster memory-wise?
> Thanks
> Krzysztof
> --
> View this message in context: http://apache-ignite-users.
> 70518.x6.nabble.com/Partitioning-on-a-non-uniform-cluster-tp8940.html
> Sent from the Apache Ignite Users mailing list archive at Nabble.com.

View raw message