Important: This documentation is about an older version. It's relevant only to the release noted, many of the features and functions have been updated or replaced. Please view the current version.
GEM hardware requirements
This page outlines the current hardware requirements for running Grafana Enterprise Metrics (GEM). Grafana Labs reserves the right to mark a support issue as ‘unresolvable’ if these requirements are not followed. See the Grafana Labs Enterprise Support SLA for more details.
CPU and memory
GEM should be deployed on machines with a 1:4 ratio of CPU to memory, so for every CPU core there should be 4 gigabytes of memory. For most clusters, Grafana Labs recommends deploying GEM onto machines with 16 CPU cores and 64 gigabytes of memory. All the nodes in the cluster should be of the same type. This is a good mix of CPU to memory for the type of workloads that GEM usually performs.
Disk
Various components of GEM (ingester, alertmanager) require fast, persistent, local disk resources to be available to the host machine. For example, in the case of the ingester component, all incoming data is sent to a write-ahead log (WAL) to help withstand unexpected node termination. Additionally, a large amount of data is flushed to disk, by default, every 2 hours, in preparation for uploading it to long-term object storage. The following are supported configurations for several cloud providers as well as guidance for custom hardware:
Amazon Web Services (AWS)
GEM is tested to run with io1 Provisioned IOPS SSD EBS volumes to ensure adequate performance to run the system correctly. The io1 storage must be provisioned at 50 IOPS per gigabyte, with a minimum of 150Gi allocated to ensure performant I/O.
Google Cloud Platform (GCP)
GEM is tested to run with pd-ssd SSD persistent disks to ensure adequate performance to run the system correctly.
Microsoft Azure
GEM is tested to run with Premium SSD SSD persistent disks to ensure adequate performance to run the system correctly.
Custom cluster hardware
GEM requires fast disks to run. Build your cluster with fast, locally attached SSD-based disks.
Network
All components of GEM require fast network access. Nodes on which the software runs should be connected by 10 gigabit/second or faster network connection speed.
Object storage
Various GEM components require object storage for config storage as well as long-term data storage.
Amazon Web Services (AWS)
GEM is tested to run with AWS’s S3 object storage service using the Standard storage class.
Google Cloud Platform (GCP)
GEM is tested to run with GCP’s GCS object storage service using the STANDARD storage class in both regional and dual regional storage locations.
Microsoft Azure
GEM is tested to run Azure’s Blob Storage object storage service using the Standard storage class with replication type LRS.
Unmanaged object storage
GEM generally works with object storage installations which support the popular AWS S3 API. However, vendors have various performance characteristics for each solution and installation, so performance testing with your individual solution will be necessary to determine if the performance profile will work for your use case. Slower storage backed by hard disks might be acceptable for less intensive workloads, but more intensive workloads will likely require more performant object storage solutions backed by SSDs.
Was this page helpful?
Related resources from Grafana Labs
