Elasticsearch is memory- and storage-sensitive because it relies on both the Java heap and the operating-system filesystem cache. Very small VPS plans can run development clusters, but serious search workloads need more RAM and predictable storage.
The correct server size depends on index size, shard count, query volume, indexing rate, retention, and whether the node also hosts application services.
Key takeaways
- 8 GB is a practical starting point for many small production Elasticsearch nodes.
- Heap should not consume all system memory because filesystem cache is important.
- NVMe or consistent SSD storage helps indexing and search latency.
- Shard count can waste resources even when data volume is small.
- Critical clusters need snapshots and a tested restore process.
Why Elasticsearch needs RAM
The Java heap stores cluster and query structures while the operating system caches index files outside the heap. Allocating nearly all RAM to Java can reduce filesystem cache and hurt performance.
Eight gigabytes gives a small node room for both heap and cache. Larger datasets and query loads may require 16 GB or more.
Storage and indexing
Indexing creates sustained writes, merges, and reads. Fast consistent storage can materially improve responsiveness.
Capacity planning should include replica data, index growth, merge overhead, and snapshot requirements rather than current primary index size alone.
CPU and shard design
Search, aggregation, indexing, and merges consume CPU. Dedicated compute becomes useful under sustained load, but inefficient shard design can waste resources regardless of hardware.
Avoid creating many tiny shards simply because the cluster can accept them.
Snapshots and recovery
Snapshots should be stored outside the node and tested before relying on them. Replicas improve availability but do not replace backup.
For business-critical search, cluster design and recovery planning may justify multiple nodes rather than one large VPS.
Frequently asked questions
Is 4 GB RAM enough for Elasticsearch?
It can work for development or very small indexes, but 8 GB is a more practical production starting point.
Does Elasticsearch need NVMe?
Fast storage is useful because search and indexing can generate significant I/O.
How much heap should Elasticsearch use?
Heap should leave substantial memory for the operating-system filesystem cache. Exact sizing depends on workload and node memory.
Can Elasticsearch run on one VPS?
Yes for small non-critical workloads. High availability requires a more resilient multi-node design.