oles@ovh.net
06-18-2009, 10:04 PM
Hello,
We are working on a new type of cluster called
"Leclerc" (such as Leclerc tanks) cf
http://en.wikipedia.org/wiki/Leclerc
What it is?
Imagine a hard disk connected to SAS / SCSI on two different servers. These two servers have access to the same files as in NFS, but directly through a File System. This is a special system that "locks" between the two servers to manage.
http://en.wikipedia.org/wiki/Global_File_System
Now imagine a rack disk 1PB or 4PO data
connected in the same way on 2 HG servers. The 2 servers
working on the same data in parallel. The 2 HG servers
change exactly the same volume of data. If one changes the
data, one can see immediately. You stick in a load balancing system
and you have an indestructable Leclerc cluster.
If one disk breaks, one of the servers breaks, one of the racks
Data breaks everything continues to function with any sort of
redundancy.
Why so much redundancy?
From several tens of TB of data, the question of
very, very high availability arises. This means that
everything must be redundant. The system should accept any sort of
down, while continuing to operate (in gradient but
without any unavailability). Therefore system working
on the same volumes of data in real time and without any
"single point failure".
We hope you submit something that meets this specification
load shortly. Certainly on Linux, can be
Windows, but not on Solaris (SUN bought cluster
probably in order to integrate into the ZFS experience
cluster, it awaits the result with impatience).
Hopefully, we can then offer you the same thing with
64 servers in parallel ...
Amicalement
Octave
We are working on a new type of cluster called
"Leclerc" (such as Leclerc tanks) cf
http://en.wikipedia.org/wiki/Leclerc
What it is?
Imagine a hard disk connected to SAS / SCSI on two different servers. These two servers have access to the same files as in NFS, but directly through a File System. This is a special system that "locks" between the two servers to manage.
http://en.wikipedia.org/wiki/Global_File_System
Now imagine a rack disk 1PB or 4PO data
connected in the same way on 2 HG servers. The 2 servers
working on the same data in parallel. The 2 HG servers
change exactly the same volume of data. If one changes the
data, one can see immediately. You stick in a load balancing system
and you have an indestructable Leclerc cluster.
If one disk breaks, one of the servers breaks, one of the racks
Data breaks everything continues to function with any sort of
redundancy.
Why so much redundancy?
From several tens of TB of data, the question of
very, very high availability arises. This means that
everything must be redundant. The system should accept any sort of
down, while continuing to operate (in gradient but
without any unavailability). Therefore system working
on the same volumes of data in real time and without any
"single point failure".
We hope you submit something that meets this specification
load shortly. Certainly on Linux, can be
Windows, but not on Solaris (SUN bought cluster
probably in order to integrate into the ZFS experience
cluster, it awaits the result with impatience).
Hopefully, we can then offer you the same thing with
64 servers in parallel ...
Amicalement
Octave