#Question about Network Volumes

11 messages · Page 1 of 1 (latest)

solar kite
#

Hi!

Just a quick 2 questions about network volumes working in tandem with Serverless endpoints:

  • Does it reduce the cold start time or availability of the serverless gpus? I want to store AI models inside a network volume and access it with a serverless gpu endpoint, I would like to know the issues I may run into as cold starts and availability are pretty important to me.
  • What's the file path to a network volume if I want to set up my serverless container to use stuff in it?

Thanks all!

elfin sierra
#
  1. Actually it has been found that using a network volume actually increases the cold startup time of serverless endpoints. It also decreases response time with Flashboot. In most every case you are better off storing models, etc., directly in container storage. Network volume is the cheapest storage on RunPod, but IMHO it is not worth using it.

  2. If you attach a network volume to your image it will be attached to /runpod-volume on your serverless endpoint when it runs.

chrome ravine
#

It depends, if your container is bulky enough because of the model, or files, etc, it may be beneficial to move them model files in network volumes

stoic acorn
#

what about if it's 24gb total

#

@chrome ravine

chrome ravine
#

I guess more than 15gb is bulky alr

#

Try to keep it below 10 ish

stoic acorn
solar kite
#

On more question, my container storage setting is at 20gb, but my container is 25gb.

I’ve had no problems but this technically doesn’t make sense lol does anyone want to explain?

chrome ravine
#

its excluded from your container size, so files that will be added in run time