Manage your installation capacity

In ECE, every host is a runner. Depending on the size of your platform, runners can have one or more roles: Coordinator, director, proxy, and allocator. While planning the capacity of your ECE installation, you have to properly size the capacity for all roles. However, the allocator role deserves particular attention, as it hosts the Elasticsearch, Kibana, APM, Enterprise Search nodes, and the relevant services.

This section focuses on the allocator role, and explains how to plan its capacity in terms of memory, CPU, processors setting, and storage.

You should plan your deployment size based on the amount of data you ingest. Memory is the main scaling unit for a deployment. Other units, like CPU and disks, are proportional to the memory size. The memory available for an allocator is called capacity.

During installation, the allocator capacity defaults to 85% of the host physical memory, as the rest is reserved for ECE system services.

To adjust the allocator capacity, prior to ECE 3.5.0; it is necessary to reinstall ECE on the host with a new value assigned to the --capacity parameter. From ECE 3.5.0 onwards, just use the ECE API :

curl -X PUT \
  http(s)://<ece_admin_url:port>/api/v1/platform/infrastructure/allocators/<allocator_id>/settings \
  -H “Authorization: ApiKey $ECE_API_KEY” \
  -H 'Content-Type: application/json' \
  -d '{"capacity":<Capacity_Value_in_MB>}'

For more information on how to use API keys for authentication, check the section Access the API from the Command Line.

Here are some examples to make Elastic deployments and ECE system services run smoothly on your host:

ECE uses CPU quotas to assign shares of the allocator host to the instances that are running on it. To calculate the CPU quota, use the following formula:

CPU quota = DeploymentRAM / HostCapacity

Consider a 32GB deployment hosted on a 128GB allocator.

If you use the default system service reservation, the CPU quota is 29%:

If you use 12GB Allocator system service reservation, the CPU quota is 28%:

Those percentages represent the upper limit of the % of the total CPU resources available in a given 100ms period.

In addition to the CPU quotas, the processors setting also plays a relevant role.

The allocated processors setting originates from Elasticsearch and is responsible for calculating your thread pools. While the CPU quota defines the percentage of the total CPU resources of an allocator that are assigned to an instance, the allocated processors define how the thread pools are calculated in Elasticsearch, and therefore how many concurrent search and indexing requests an instance can process. In other words, the CPU ratio defines how fast a single task can be completed, while the processors setting defines how many different tasks can be completed at the same time.

Starting from Elasticsearch version 7.9.2, running on ECE 2.7.0 or newer, we rely on Elasticsearch and the -XX:ActiveProcessorCount JVM setting to automatically detect the allocated processors.

In earlier versions of ECE and Elasticsearch, the Elasticsearch processors setting was used to configure the allocated processors according to the following formula:

Math.min(16,Math.max(2,(16*instanceCapacity*1.0/1024/64).toInt))

The following table gives an overview of the allocated processors that are used to calculate the Elasticsearch thread pools based on the preceding formula:

This table also provides a rough indication of what the auto-detected value could be on newer versions of ECE and Elasticsearch.

ECE has specific hardware prerequisites for storage. Disk space is consumed by system logs, container overhead, and deployment data.

The main factor for selecting a disk quota is the deployment data, that is, data from your Elasticsearch, Kibana, and APM nodes. The biggest portion of data is consumed by the Elasticsearch nodes.

To calculate the disk quota, use the following formula:

Diskquota = ICmultiplier * Deployment RAM

ICmultiplier is the disk multiplier of the instance configuration that you defined in your ECE environment.

The default multiplier for data.default is 32, which is used for hot nodes. The default multiplier for data.highstorage is 64, which is used for warm and cold nodes. The FS multiplier for data.frozen is 80, which is used for frozen nodes.

You can change the value of the disk multiplier at different levels: