[CRIU] CRIU VS other approach

[Andrei Vagin]

On Sun, Aug 14, 2016 at 07:20:52PM +0100, Thouraya TH wrote:
> Hi all,
> 
> Please, i have a question about the architecture and the use of CRIU.
> 
> As I know, the size of a container’s snapshot depends on the size of the RAM.
> The CRIU images contain only what's in memory, not what's on disk.

If we are talking about migration, you need to look at p.haul. It
migrates a countainer with memory and files and does this iteratively.

https://criu.org/Live_migration

> 
>  
> 
> Many prefer for example to snapshot only the layered file system of the
> container and discard the memory, under the assumption that the checkpoint is
> smaller this way and it is also faster to restart the container and then the
> state saved in the file system.

In many cases a downtime is smaller in case of iterative migration in
comparison with restarting a container.

When you restart a container, you drop all connections to it. If you
migrate a conainer, it is migrated with all connections, so end users
don't notice this migration.

> 
> This is a common approach for container migration.

This is a commont approach for offline migration. CRIU allows to make
online "zero-downtime" migration.

> 
>  
> 
> Please, what’s the motivation of choosing CRIU ?

* reduce a system downtime
* don't drop online users
* migrate a current runtime state

> 
> 
> Thanks a lot for all explanations.
> 
> Kind regards.
> 

> _______________________________________________
> CRIU mailing list
> CRIU at openvz.org
> https://lists.openvz.org/mailman/listinfo/criu