Multiple Cluster vs Multiple DC

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Multiple Cluster vs Multiple DC

arodrime
Hi,

tl;dr : How to choose between having a Cluster with 2 DC and having 2 separate clusters as follow:

"cluster_name: mycluster
KS1 {DC1: 3, DC2: 0} and KS2 {DC1: 0, DC2: 3}"

OR

"cluster_name: cluster1
KS1 {DC: 3}
+
cluster_name: cluster2
KS2 {DC: 3}"

We can use distinct columns and row to keep distinct entries, depending on sharding, locallity and query I know how to model this.

At a higher level we have CF, Keyspace. Basically, to chose if I want to choose between a CF or a Keyspace, I ask myself if this table has a logical interest in beeing separated from existing keyspaces. This mainly depend on RF I want on each DC or if the table has a very specific workload that would need a specific attention and tuning.

But when to decide having multiple cluster instead of multiple DC with each DC its Keyspace ?

Thanks for any enlightening you can give me around this.

C*heers,

Alain


Reply | Threaded
Open this post in threaded view
|

Re: Multiple Cluster vs Multiple DC

Anuj
Multiple DC is usually useful in case u need Geo Redundancy or have distributed workload. Do u have these clusters at same physical location? If 3 nodes are at diff locations dont u need geo red for disaster ? Whats ur RF for cf? 

Thanks
Anuj

Sent from Yahoo Mail on Android


From:"Alain RODRIGUEZ" <[hidden email]>
Date:Tue, 5 May, 2015 at 2:50 pm
Subject:Multiple Cluster vs Multiple DC

Hi,

tl;dr : How to choose between having a Cluster with 2 DC and having 2 separate clusters as follow:

"cluster_name: mycluster
KS1 {DC1: 3, DC2: 0} and KS2 {DC1: 0, DC2: 3}"

OR

"cluster_name: cluster1
KS1 {DC: 3}
+
cluster_name: cluster2
KS2 {DC: 3}"

We can use distinct columns and row to keep distinct entries, depending on sharding, locallity and query I know how to model this.

At a higher level we have CF, Keyspace. Basically, to chose if I want to choose between a CF or a Keyspace, I ask myself if this table has a logical interest in beeing separated from existing keyspaces. This mainly depend on RF I want on each DC or if the table has a very specific workload that would need a specific attention and tuning.

But when to decide having multiple cluster instead of multiple DC with each DC its Keyspace ?

Thanks for any enlightening you can give me around this.

C*heers,

Alain