Not all application resource demands fit nicely inside instance types. Once you determine your application needs, you can set your cpu/memory resource needs pe

Not all application resource demands fit nicely inside instance types. Once you determine your application needs, you can set your cpu/memory resource needs per container/service. Then you can run these containers inside your fleet more smartly. This really comes into play for memory and cpu ratio. Look at the price difference between: one m4.2xlarge = 32gb mem, 8 cores = $350 a month eight t2 medium = 32gb memory, 8 cores = $300 a month You get twice as much compute for less money with the ability to mix and match service needs instead of being constrained to individual instances.