search

IHTSDO / snowstorm

Scalable SNOMED CT Terminology Server using Elasticsearch
Other
204 stars 80 forks source link

Recommended EC2 instance type #21

Closed ghost closed 5 years ago

ghost commented 5 years ago

Hello Team,

What is the recommended EC2 instance type to run the SNOWSTORM server?

Thank you, Vybhav

kaicode commented 5 years ago

The answer to this question depends on your scenario.

For general use the recommended EC2 instance type is m5.xlarge which has 16g of memory. The memory allocation should be something like 4g to Snowstorm and 6g to Elasticsearch with a few g free for disk level caching which Elasticsearch needs to work well.

As always you should perform load testing for your expected usage and use monitor the JVMs in production. Functions that take the most resources are RF2 import and some of the more complex ECL statements.

Snowstorm has been designed to scale horizontally allowing additional pairs of Elasticsearch nodes and Snowstorm instances to be added to a cluster to deal with demand but this has not yet been thoroughly tested. We know that some parts of the authoring functionality will not yet work in this configuration, that will be addressed in the first half of this year.

ghost commented 5 years ago

Thank you @kaicode

ghost commented 5 years ago

Hello @kaicode ,

One followup question, for just prototypin and internal testing, what's the minimal instance confiuration you'd recommend? Very little traffic.

Regards, Vybhav

ghost commented 5 years ago

Hello @kaicode ,

One followup question, for just prototypin and internal testing, what's the minimal instance confiuration you'd recommend? Very little traffic.

Regards, Vybhav

kaicode commented 5 years ago

It may run on an 8g ram instance. I would try giving Snowstorm 2560m and Elasticsearch 3g. Use the Elasticsearch configuration file config/jvm.options with the memory options -Xms3g and -Xmx3g. If it's a development machine you can use the spot instance type to get a better price.

ghost commented 5 years ago

Great. Thank you @kaicode . I'll give that a shot