Am Montag, 17. September 2001 00:20 schrieben Sie:
> > > > Not that I would recommend swapping over USB anyway....
> > >
> > > Why is that? The USB 2.0 disks seem like bandwidth won't
> > > be an issue. Is it the iffy error handling going through SCSI?
> >
> > No, it's basically the same problem as swapping over nfs.
> > Sometimes not only GFP_ATOMIC is needed, but failure due
> > to a lack of memory is not an option either.
>
> Hmm, I know that Solaris and SVr4 swapped over NFS since
> back in the SunOS 4.0 days. What's the deal? There would
Exhaustion of the free and clean pools.
To recieve nfs packets you need to allocate memory.
It is concievable that someone has managed to dirty all pages
and someone has used GFP_ATOMIC to allocate all pages reserved for atomic
allocations.
Then you'd need to swap which you can't do because you have no freeable
memory left.
> seem to be two options: (a) swapping code has a way to cope
> with temporary failures, maybe try to swap other pages, and
You can't guarantee that there are other pages which would behave differently.
> (b) only if no pages can be swapped, panic. Panics weren't
> something I remember hearing about, so I don't think they
> can ever have been common..
There are patches to enable swapping over nfs in Linux. They work ... most of
the time.
> I know there's been a lot of discussion/work about making
> memory management better, none of which I've paid attention
No connection to this issue.
> to. Is the situation that Linux only has (b), not (a)? Or is
> the concern instead to completely eradicate (a)? Or maybe
> just make it happen a lot less? :)
AFAIK the vm subsystem assumes that a failed swap means that there's an io
error on your swap partition. The affected task is killed.
>
> > Plus of course, the issue of disconnect on a device holding a swapfile.
>
> That one's more workable. A lot of the USB 2.0 controllers
> are set up with internal connectors. It's practical to have
> internal disks on USB, using serial cables rather than nasty
> IDE ribbon cables complicating airflow ... and still get decent
> I/O performance (except for server style loads).
Well yes, but you can't assume it might not be disconnected. Disconnecting is
legal everywhen. AFAIK usb storage doesn't time out a request, but waits for
reconnection. You could have init stuck in state D.
Regards
Oliver
_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
