Hi everyone,
Does anyone have experience building Elastic Stack on BareMetal? Its depends on the price of SAN, which is expensive. If anyone has done it, could you share with me how you did it and the pros & cons?
Thank
Hi @api_developer, welcome to the community!
First, it's important to note that Elasticsearch performs exceptionally well on bare metal. Thousands of clusters (possibly tens of thousands) have been built on bare metal, making it the preferred deployment option "in the early days."
Today, Elasticsearch is frequently run in containerized environments such as Docker, Kubernetes, and OpenShift. The Elastic Cloud Hosted platform (the official hosted Elastic Stack) operates efficiently and with high performance in these containerized environments.
Now, regarding your question and why perhaps no one has responded yet:
Yes, Elasticsearch can run extremely well on bare metal. However, your question is quite broad, and you will need to provide more context to get a better response.
Here are some points to consider:
-
Why do you want or need to run on bare metal? Is it because you have leftover hardware? Do you operate in a dedicated data center? Are you seeking extreme performance? Or is it simply for another reason?
-
What specific use case are you trying to address—logging/observability, search, or SIEM (Security Information and Event Management)?
-
What general scale are you considering? For example, are you looking at 10 GB/day, hundreds of GB/day, 1 TB/day, or 10 TB/day?
-
Are you planning to run the free basic version or a subscription?
You mentioned an expensive SAN (Storage Area Network); keep in mind that the cost is relative.
Many people choose to run on bare metal to maximize performance, which typically means using local SSD storage.
All these factors need to be taken into account. There are several threads on this forum discussing similar topics that might also provide valuable insights.
Tell us more or look around a bit, and come back.
1 Like
We have done this with plenty of customers, of course!
Aside from the disk question (SAN is usually less recommended, but workable) there are some additional configurations you might need to change from the defaults which assume virtualization is in place (for example the default number of threads, etc).
Like Stephen mentioned - the devil is in the details. For small operations, don’t worry about it too much.
For any significant scale, I’d recommend keeping a close eye on key performance metrics and optimize from there. See Pulse for Elasticsearch as a great free tool to get you started on getting those sizing insights.
I have 7 node cluster runing baremetal on older hardware from few years without any issue at all. I just replace HDD wuth SSD. started with cluster version 6.x and now on 8.x
all nodes are data and out of them three are master/ingest/transform as well.
1 Like