Skip to content
This repository has been archived by the owner on Dec 20, 2024. It is now read-only.

Pulling a image with dragonfly is actually slower #1128

Closed
JiaDong007 opened this issue Dec 7, 2019 · 3 comments
Closed

Pulling a image with dragonfly is actually slower #1128

JiaDong007 opened this issue Dec 7, 2019 · 3 comments
Labels
kind/question all questions or confusion about this project

Comments

@JiaDong007
Copy link

Question

I am use docker registry,
When using Dragonfly to pull two nodes at the same time:
real 12m39.620s
user 0m2.326s
sys 0m0.892s

When not using Dragonfly to pull two nodes at the same time:
real 6m57.331s
user 0m0.475s
sys 0m0.432s

@allencloud allencloud added the kind/question all questions or confusion about this project label Dec 7, 2019
@allencloud
Copy link
Contributor

The statistics you collected is reasonable. Your cluster scale is very small. When your cluster only has 2 nodes, the time is 2 times native pulling times.

Because the supernode will first download the image from registry. Then node will download image from supernode. Then real 12m39.620s is reasonable.

If you have lots of nodes in your cluster, the parallel downloading will consumes all the bandwidth of the registry, then use p2p could save your bandwidth and improve your efficiency. @JiaDong007

@allencloud
Copy link
Contributor

I am closing this issue. I am not sure if my answer satisfies you. If you have further questions, please feel free to re-open this issue. Thanks. @JiaDong007

@JiaDong007
Copy link
Author

Now i have 100 dfclient nodes,but i docker pull the same image in parallel ,them spend more than 20 minutes on average,some cost 1 hour.What is the reason? @allencloud

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
kind/question all questions or confusion about this project
Projects
None yet
Development

No branches or pull requests

2 participants