[Twg] [Lustre-discuss] cmu RAIDFrame

[Shipman, Galen M.]

Rita,

OpenSFS has funded design work to support advanced layouts (beyond just simple RAID 0). This work is managed by the OpenSFS Technical Working Group.

If you want to keep up to speed on this work you can join the TWG email list:
 http://lists.opensfs.org/listinfo.cgi/twg-opensfs.org

Here is a brief summary of the work:

Project 1. Layout Enhancement Design
Technical Description and Approach
Lustre file layouts represent how file data is distributed over OSTs. Only simple RAID0 (striping)
layouts are supported today, and enhancements are required to implement future features such as Data
on MDS, Data Replication, live Data Migration, and RAID1/5/6 or erasure coding. Design goals are
listed below.
• Extend the Lustre layout to include the following formats:
o N-way stripe replication. This is a requirement for the Data Replication project. It
generalizes RAID0+1 to allow multiple replicas of the data in a Lustre file. This will
ensure that data can remain available in the face of multiple OST failures - e.g. multiple
short- or long-term OST failures.
o RAID5/6 and erasure codes. These anticipate future developments to enable more space efficient
replication techniques.
o Layout extents. This allows different layouts in different extents of a Lustre file.
• Determine additional requirements for algorithmic layouts – e.g. CRUSH.
• Compatibility and version interoperation. The design work must address requirements to support
rolling upgrades.


Thanks,

Galen



On Jan 16, 2014, at 7:15 AM, Rita wrote:

I was curious if the CMU RAIDFrame (network) has been implemented for Lustre.

http://wiki.lustre.org/index.php/Architecture_-_Server_Network_Striping


If not, are there still plans to implement this feature? It seems many distributed filesystems have some sort of redudancy built in but lustre is lacking this and I was curious if these types of features are in the spotlight.


--
--- Get your facts first, then you can distort them as you please.--
_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at lists.lustre.org<mailto:Lustre-discuss at lists.lustre.org>
http://lists.lustre.org/mailman/listinfo/lustre-discuss