[Qemu-devel] Re: [BUG] Regression of exec migration

[Charles Duffy]

Pierre Riteau wrote:
> I'm afraid this solves the problem only because you disabled non 
> blocking operations on the file descriptor, effectively reverting 
> 907500095851230a480b14bc852c4e49d32cb16d.
> See comment inline.

Indeed -- fixing this oversight, I'm seeing 2m30s to write an 84mb ramsave.

This doesn't seem to square with Anthony's suggestion that the f* 
functions are to blame; we're still doing an fdopen() call and using f* 
on incoming migrations, but it's not incoming migrations where we're 
having performance issues.