Sending images to clients in parallel using multicast
Imagine you have 10,000 hosts that all simultaneously need a copy of the same image, and furthermore all of the hosts are a single layer 2 fabric (or in any case, multicast is feasible). A particularly efficient way to distribute the image to all of those hosts is to just start sending the image out over a multicast group all the clients know to subscribe to. To deal with a client showing up late, the source can just keep repeating the image in block order until it can tell that all clients are done.
Blueprint information
- Status:
- Not started
- Approver:
- None
- Priority:
- Undefined
- Drafter:
- Mark Washenberger
- Direction:
- Needs approval
- Assignee:
- None
- Definition:
- New
- Series goal:
- None
- Implementation:
- Unknown
- Milestone target:
- None
- Started by
- Completed by
Related branches
Related bugs
Sprints
Whiteboard
TBH Mark, if we really have 10,000 hosts IMO we'd better think about "zero-copy" approach for the image consuming instead of using "copy" traditional idea. Or CoR (copy-on-read) approach instead of pure CoW based "zero-copy", it could prevent the issues around "zero-copy" approach like stability, and also could prevent network latency. Like to know you thoughts. --zhiyan
rosmaita wishlist 2014-04-25