Locally Repairable Codes vs Pyramid

Locally Repairable Codes vs Pyramid

[Loic Dachary]

[-- Attachment #1: Type: text/plain, Size: 973 bytes --]

Hi Ceph,

TL;DR: s/pyramid/LRC/ in the implementation of http://pad.ceph.com/p/cdsgiant-pyramid-erasure-code for clarity

Although the term "Pyramid code" has been consistently used in reference to the technique used to reduce the network requirements when repairing from the loss of a single OSD (put simply ;-), I realized today when discussing with Dan Lambright and Xavier Hernandez from glusterfs that I was unclear about what it really means.

As far as I can tell, LRC as described in http://anrg.usc.edu/~maheswaran/Xorbas.pdf suggests an implementation that is close to what has been described during the last CDS. In contrast the pyramid code paper http://research.microsoft.com/pubs/70415/tr-2007-25.pdf suggests a more sophisticated approach which I do not fully understand. 

Instead of hardcoding the word "pyramid" in the pathnames of the implementation it seems better to use "LRC" instead.

Cheers

-- 
Loïc Dachary, Artisan Logiciel Libre


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]

Re: Locally Repairable Codes vs Pyramid

[Loic Dachary]

[-- Attachment #1: Type: text/plain, Size: 1115 bytes --]

For the record and since nobody objected, I made the suggested change.

On 02/06/2014 18:00, Loic Dachary wrote:
> Hi Ceph,
> 
> TL;DR: s/pyramid/LRC/ in the implementation of http://pad.ceph.com/p/cdsgiant-pyramid-erasure-code for clarity
> 
> Although the term "Pyramid code" has been consistently used in reference to the technique used to reduce the network requirements when repairing from the loss of a single OSD (put simply ;-), I realized today when discussing with Dan Lambright and Xavier Hernandez from glusterfs that I was unclear about what it really means.
> 
> As far as I can tell, LRC as described in http://anrg.usc.edu/~maheswaran/Xorbas.pdf suggests an implementation that is close to what has been described during the last CDS. In contrast the pyramid code paper http://research.microsoft.com/pubs/70415/tr-2007-25.pdf suggests a more sophisticated approach which I do not fully understand. 
> 
> Instead of hardcoding the word "pyramid" in the pathnames of the implementation it seems better to use "LRC" instead.
> 
> Cheers
> 

-- 
Loïc Dachary, Artisan Logiciel Libre


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 263 bytes --]