Enable Parallel Pull on smaller instance sizes #2452
Description
What would you like to be added:
Parallel Pull is currently only enabled for bigger instance types, however, for our use case we would like to be able to enable this on smaller instance sizes.
Why is this needed:
We have some nodes that are dedicated to a particular workload, so a new node is created and only pods for this specific workload can be scheduled onto these nodes. The container image for this workload is big, 6GB compressed / 16GB uncompressed! The node we're using has 4 vCPUs and 16GB of memory, but I believe the requirements for parallel pull are >= 8 vCPUs and >= 8GB of memory if I have read this code correctly.
Because this node is essentially doing nothing until this workload is running, we're happy to have the CPU and memory used to pull the image as quickly as possible as it's not doing much at this time anyway.
Therefore, allowing parallel pull on smaller instance types would be great for us. I'd be happy for this to only be enabled where an extra setting is given to show that we're doing this purposely and understand the risks.