[PetiteCloud] site changes

[Michael Thoreson]

On 11/02/2014 2:14 PM, Aryeh Friedman wrote:
>
>
>
> On Tue, Feb 11, 2014 at 3:09 PM, Michael Thoreson 
> <m.thoreson at c4labs.ca <mailto:m.thoreson at c4labs.ca>> wrote:
>
>     Have you considered GlusterFS? It has native support for qemu
>     qcow(2) images and GlusterFS has been recently made to work with
>     OpenStack http://www.gluster.org/category/openstack/.
>
>     It will give automatic scalability, fail-over and fault tolerance.
>
>     I had glusterfs working on Linux hosts as storage backing for
>     VMWare Workstation, VirtualBox and in a virtual lap I was able to
>     use it as base for high avail ISCSI and used carp for IP fail-over.
>
>
> Worth a look but a quick note on qcow2 we do not use it nor plan to 
> (it has a *CRITICAL* security flaw [like all sparse formats])... see 
> our forth coming white paper "Sparse Disk Formats Considered Harmful" 
> (not yet committed to paper) and 
> http://lists.freebsd.org/pipermail/freebsd-virtualization/2014-February/002210.html
> -- 
> Aryeh M. Friedman, Lead Developer, http://www.PetiteCloud.org
There is no reason to use qcow2, raw works just as well or any format 
supported by qemu. As I mentioned it also works with Virtualbox and 
VMWare Workstation so VDI and VMDK are possible choices. With successful 
testing of ISCI it is safe to assume any file used as a block device 
would work.

I also forgot to metion that GlusterFS recently added a file level 
stripe functionality. Basically this allows for VM or ISCSI disk images 
to be larger than any single data node. For example you could make a 2 
PB ISCI disk but only have 50 TB on each node, you just need to have 
enough nodes to strip across.

If you add replication with the stripe then you obviously get redundancy 
with it and significantly increased read performance.

So you can effectively replicate RAID 0,1,10,5+6 at the network level 
with GlusterFS.

Michael Thoreson,