Hi there,
with what kind of latency (between nodes) can one operate Aerospike without XDR in a stable manner?
Hi there,
with what kind of latency (between nodes) can one operate Aerospike without XDR in a stable manner?
Aerospike supports a geographically distributed multi-site cluster where inter-node latencies are larger. There is no upper limit as such on the latency, but certain parameters need to be tuned for such deployments to work well typically by our solution architects.Can you provide more details of your use case?
We have a setup distributed across three data centers. The highest avg. network latency between servers across datacenters is 25ms (there can be spikes where it is higher ofc.).
Would it require tuning at this latency level?
You will need Enterprise license for a cluster size above 8, and a 3-site cluster may exceed 8 nodes. XDR synchronized clusters are also supported only in the Enterprise version. If you are running with an Enterprise version, I would suggest you reach out to Support to have consultation with a knowledgeable solution architect. If not and if you are ok to be contacted by email, I can provide your email to someone who can discuss the options. (Corrected the Community Edition cluster size limit to 8.)
So 25ms won’t work without XDR and enterprise?
XDR provides a way to synchronize data across multiple clusters, and it is supported only in the Enterprise version. The inter-node latency would be much less of an issue for XDR, as opposed to within a single cluster where the core functions depend on the regular inter-node heartbeats. CE may work; please give it a try - probably will require adjustment to heartbeat config.
It should just work… if the latency is acceptable from your application (read and write would be impacted differently and would depend on where the client application is running from). The key thing for the cluster to be stable is the heartbeat configuration (the heartbeat.timeout and heartbeat.interval) and even the default should be fine if the network is robust and simply have a consistent 25ms latency between nodes.
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.