For the complete documentation index, see llms.txt. This page is also available as Markdown.

Selecting an Instance Type

General Information about Instance Type

Instance Type Overview

Naming

AWS naming of instance types is broken down here:

Choosing an Instance Type

Question

Focus on this

Does the software utilize multiple cores?

mem2_ssd1_v2_x16

How much memory does the software use (per core)?

mem2_ssd1_v2_x16

How much disk space is needed for the software (per core)?

mem2_ssd1_v2_x16

Always use version 2 of an instance type!

mem2_ssd1_v2_x16

Instance Classes and Cores

  • Each class (like mem1) is scaled so that each core in an instance has access to the same amount of memory/disk space:

  • Example: mem1_ssd1_v2_x2:

  • 4 Gb total memory / 2 cores =

  • 2 Gb / Core

  • Example: mem1_ssd1_v2_x8:

  • 16 Gb total memory / 8 cores =

  • 2 Gb / core

Choosing a Good Instance Type

  • Scale usage/instance type according to usage statistics and dataset size

  • If it doesn't utilize all resources

  • Use a smaller instance type

  • Runs out of memory, or is slow

  • Consider using a larger instance type

Multistep Workflows

  • Each stage of a workflow is run by a different set of workers

  • Each stage can be customized in terms of instance type

Last updated