Le vendredi 11 mars 2005 à 17:41 +0100, Magnus Hagander a écrit :
> > > Do *not* do this with a production database.
> > >
> > > Vmware does *not* correctly handle fsync()s (or O_SYNC or any of
> > > those) thruogh to disk. If your host PC crashes, your database will
> > > almost certainly be corrupted. fsync() on the client just
> > puts it in
> > > the RAM cache on the host. Not even in the write cache on
> > the disk/raid.
...snip
> It's possible this works fine if you use direct disk access in vmware
> (giving the session a native disk to access), but I haven't tried that.
OK! I understand your worries now. I always do this because initial
reading through the different disk modes when 3.0 came out made my hair
stand on end. The speed and size of disks today means that each virtual
machine can treat a part of the disk as its own as far as I'm concerned.
The other disk modes always seemed strange to me - maybe they have uses
for others... When I am in my virtual machine I like to see the HD diode
go on each time I do a save, improves my tan =:-D
> After some looking around (and with some hints from Dave Page) for my
> own needs of virtualising linux-on-linux, I've moved to linux-vservers.
> While it doesn't virtualise everything, it's good enough for me. I
> suggest you at least look at it before going down the vmware path - it's
> also free software unlike vmware.
Looked at that. It requires heavy guru voodoo magic at host OS install
time. VMware (I already own the licence I was going to use) can be
installed on a machine that is up and running.
Thanks
Tony