Probably decreasing tokens number can help to mange big cluster?


Best regards, Vladimir Yudovin, 

Winguzone - Cloud Cassandra Hosting






---- On Mon, 21 Aug 2017 19:38:37 -0400 Eduard Tudenhoefner 
<[email protected]> wrote ----




We've been doing successful testing with multi-DC setups and 500 nodes per DC. 
However, I agree with Jon here. Certain things are easier/faster with e.g. 
5x100 node clusters than 1x500 node cluster.



Cheers



On Mon, Aug 21, 2017 at 10:16 AM, Jon Haddad <[email protected]> 
wrote:

As far as I know, those 75K nodes are not in a single cluster.  If memory 
serves correctly (and this article seems to indicate that it does 
http://www.techrepublic.com/article/apples-secret-nosql-sauce-includes-a-hefty-dose-of-cassandra/),
 you’ll see clusters of 1,000 nodes.  



Things start to get a little hairy once you go above a couple hundred nodes.  I 
would rather run 5 100 node clusters than a single 500 node cluster.  In 
theory, once you’ve built out the tooling to manage 2 clusters you should be 
able to apply it to manage 20 (reality always gets in the way though…)



Jon



On Aug 21, 2017, at 9:15 AM, techpyaasa . <[email protected]> wrote:



Thanks lot for reply :)



On Aug 21, 2017 6:44 PM, "Vladimir Yudovin" <[email protected]> wrote:



Actually there are clusters of thousandths nodes: Some of the largest 
production deployments include Apple's, with over 75,000 nodes storing over 10 
PB of data



Best regards, Vladimir Yudovin, 

Winguzone - Cloud Cassandra Hosting






---- On Mon, 21 Aug 2017 08:35:37 -0400 techpyaasa . 
<[email protected]> wrote ----




Hi 



Is there any limit on having number of nodes in c* cluster.

Right now we have c*-2.1.17 cluster with 3 DCs each DC with 3 groups & each 
group has 21 nodes.



We wanted to increase the cluster capacity by adding 6 nodes per group as many 
of nodes disk usage crossed 65%.



So just wanted to clarify is there any limit/drawback having huge cluster/too 
many nodes in a c* cluster



Thanks in advance

TechPyaasa









































Reply via email to