[
https://issues.apache.org/jira/browse/CASSANDRA-8060?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Donald Smith updated CASSANDRA-8060:
------------------------------------
Description:
We have three data centers in the US (CA in California, TX in Texas, and NJ in
NJ), two in Europe (UK and DE), and two in Asia (JP and CH1). We do all our
writing to CA. That represents a bottleneck, since the coordinator nodes in CA
are responsible for all the replication to every data center.
Far better if we had the option of setting things up so that CA replicated to
TX , which replicated to NJ. NJ is closer to UK, so NJ should be responsible
for replicating to UK, which should replicate to DE. Etc, etc.
This could be controlled by the topology file.
The replication could be organized in a tree-like structure instead of a
daisy-chain.
It would require architectural changes and would have major ramifications for
latency but might be appropriate for some scenarios.
was:
We have three data centers in the US (CA in California, TX in Texas, and NJ in
NJ), two in Europe (UK and DE), and two in Asia (JP and CH1). We do all our
writing to CA. That represents a bottleneck, since the coordinator nodes in CA
are responsible for all the replication to every data center.
Far better if we had the option of setting things up so that CA replicated to
TX , which replicated to NJ. NJ is closer to UK, so NJ should be responsible
for replicating to UK, which should replicate to DE. Etc, etc.
This could be controlled by the topology file.
It would require architectural changes and would have major ramifications for
latency but might be appropriate for some scenarios.
> Geography-aware, daisy-chaining replication
> -------------------------------------------
>
> Key: CASSANDRA-8060
> URL: https://issues.apache.org/jira/browse/CASSANDRA-8060
> Project: Cassandra
> Issue Type: Wish
> Reporter: Donald Smith
>
> We have three data centers in the US (CA in California, TX in Texas, and NJ
> in NJ), two in Europe (UK and DE), and two in Asia (JP and CH1). We do all
> our writing to CA. That represents a bottleneck, since the coordinator nodes
> in CA are responsible for all the replication to every data center.
> Far better if we had the option of setting things up so that CA replicated to
> TX , which replicated to NJ. NJ is closer to UK, so NJ should be responsible
> for replicating to UK, which should replicate to DE. Etc, etc.
> This could be controlled by the topology file.
> The replication could be organized in a tree-like structure instead of a
> daisy-chain.
> It would require architectural changes and would have major ramifications for
> latency but might be appropriate for some scenarios.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
