ElasticSearch Deployment recommendation

Elastic Stack Elasticsearch
1

Hi,

What would be the cons of having elasticsearch () running on a server that has another server application running that servers up lot of data and potentially 500 to 1000 users. Can we increase system specs to have them run smoothly or would it be recommended to have elasticsearch on a separate server? and why?

Thanks,
Imad.

2

It is just easier to get performance information if you know there is only Elasticsearch on the box. It isn't the end of the world to have other things running but I'd avoid it.

You want to avoid anything from the internet getting access to Elasticsearch. No publicly open ports, etc.

3

My ElasticSearch would only be interfacing with couchdb on same system so it would be behind firewall.

Can you please elaborate on any potential problems that we can run into if we are are deploying ElasticSearch on same server machine with other server application? (Would just need more supportive argument for the customer).

4

As I said - it is easier to get performance information if it is on its own machine. You also can be sure of what resources it has if it is on its own machine. You can also be sure that something won't go wrong and it won't starve the other system of resources. If you are willing to accept those risks it ought to be fine.

Just remember that Elasticsearch makes heavy use of memory mapped files so when you pick the ES_HEAP_SIZE make sure that you leave enough free memory for the kernel to do proper disk caching for the files.

5

Thanks @nik9000 for the reply.

Also just want to check what would be the best RAM size you would recommend for a server that has elasticsearch and couchdb on it. Customer would have 100 clients and each would have about 5000 users/roles and location data. Would 16Gb be a recommended size of RAM for this much of data?

Quoting from ElasticSearch website "A machine with 64 GB of RAM is the ideal sweet spot, but 32 GB and 16 GB machines are also common. Less than 8 GB tends to be counterproductive (you end up needing many, many small machines), and greater than 64 GB has problems that we will discuss in Heap: Sizing and Swapping."