Hi Mike.
On Thu, Jun 11, 2009 at 10:49:12AM +0930, Mike O'Connor ([email protected])
wrote:
> In your announcement to day you mentioned that no one wants to move from
> NFS, we in my case I really want something to replace it as it has just
> to many limitations and is just slow.
>
> I need to be able to build high reliability storage servers and the
> tools out there right now just do not do the job.
>
> Options
> 1. I need two separate storage servers (12 disks) which are kept in sync
> 2. I need a number of servers running vservers (www.linux-vservers.org)
> to access the storage on network.
> 3. I need to be able to add extra disks with our affecting uptime.
>
> NFS:
> 1. Very hard with open source software
> 2. When we do this the just one of the vserver guest can cause the whole
> system to slow down.
> 3. There are no local file system which would allow this easily and
> maintain local raid copies. (btrfs is the future here)
>
> POHMELFS:
> 1. Once stable would perform this effortlessly.
> 2. From everything I have read and from my own understanding of the code
> I believe this would not happen.
> 3. Again this is an issue but with btrfs this would not become an issue.
>
> So overall if you going to move POHMELFS away from being a very
> effective network filesystem protocol, I do not know what I can use instead.
It is still possible to work with the elliptics storage backend like
with the network filesystem, but its data representation will be
different compared to common directory hierarchy.
There will be no common names and optionally objects may live in the
database, but this allows to move away from non-scaling metadata servers
and need for the manual sync of the exported directories in case of
network failures and effectively limited scaling - we can not increase
the storage size when using NFS model.
--
Evgeniy Polyakov
_______________________________________________
Pohmelfs mailing list
[email protected]
http://www.ioremap.net/cgi-bin/mailman/listinfo/pohmelfs
