-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
The idea is to compare the execution time of SC2 with different image-pulling configurations:
- Guest pull (regular) image-pull: add benchmark comparing image pull mechanisms #18
- Lazy guest pull (w/ nydus) image-pull: add benchmark comparing image pull mechanisms #18
- Host mount (w/ nydus) image-pull: add benchmark comparing image pull mechanisms #18
- Hybrid
Do for two different applications:
-
helloworld-py
image-pull: add benchmark comparing image pull mechanisms #18 -
tflite
inference (or something from a benchmark) image-pull: add benchmark comparing image pull mechanisms #18
For each mode and application, we will have:
- Unencrypted images image-pull: add benchmark comparing image pull mechanisms #18
- Encrypted images
Also, we need to remember to include in our measurements:
- A bar for function's runtime image-pull: add benchmark comparing image pull mechanisms #18
Known issues:
- TFlite does not work with host-mount (?)
- Hello World seems to trigger an error with host mount (seems to be in the transition between snapshotters)
- Host mount performance does not seem reasonable
Metadata
Metadata
Assignees
Labels
No labels