[Gluster-devel] Pluggable interface for erasure coding?

[Pranith Kumar Karampuri]

8th March works for me. 9th I am travelling. Xavi is well versed with this
part of the code, so even if I am not available, you guys should pick the
time which is convenient for Xavi.

The following may be of help for this discussion.
https://github.com/gluster/glusterfs/blob/master/doc/developer-guide/ec-implementation.md

If my memory serves right, Xavi's implementation outperformed liberasure,
that is the reason we didn't integrate with liberasure sooner.


On Fri, Mar 3, 2017 at 7:01 AM, Per Simonsen <per.simonsen at memoscale.com>
wrote:

> Hi,
>
> I forgot to add the time zone: the suggested time is 10 am (GMT+1).
>
> Best,
> Per
>
> On Thu, Mar 2, 2017 at 5:19 PM, Per Simonsen <per.simonsen at memoscale.com>
> wrote:
>
>> Hi,
>>
>> I suggest that we set up an online meeting next week to discuss the
>> erasure coding features as well as possible implementations of a plugin
>> architecture. We also have some experience integrating with the liberasure
>> library mentioned by Prashant which we can share.
>>
>> Does 10 am on Wednesday(8th of March) or Thursday(9th of March) next week
>> work for you guys?
>>
>> Best,
>> Per Simonsen
>> CEO
>> MemoScale
>>
>>
>> On Thu, Mar 2, 2017 at 12:00 AM, Xavier Hernandez <xhernandez at datalab.es>
>> wrote:
>>
>>> Hi Niels,
>>>
>>> On 02/03/17 07:58, Niels de Vos wrote:
>>>
>>>> Hi guys,
>>>>
>>>> I think this is a topic/question that has come up before, but I can not
>>>> find any references or feature requests related to it. Because there are
>>>> different libraries for Erasure Coding, it would be interesting to be
>>>> able to select alternatives to the bundled implementation that Gluster
>>>> has.
>>>>
>>>
>>> I agree.
>>>
>>> Are there any plans to make the current Erasure Coding
>>>> implementation more pluggable?
>>>>
>>>
>>> Yes. I've had this in my todo list for a long time. Once I even tried to
>>> implement the necessary infrastructure but didn't finish and now the code
>>> has changed too much to reuse it.
>>>
>>> Would this be a possible feature request,
>>>> or would it require a major rewrite of the current interface?
>>>>
>>>
>>> At the time I tried it, it required major changes. Now that the code has
>>> been considerably restructured to incorporate the dynamic code generation
>>> feature, maybe it doesn't require so many changes, though I'm not sure.
>>>
>>>
>>>> Here at FAST [0] I have briefly spoken to Per Simonsen from MemoScale
>>>> [1]. This company offers a (proprietary) library for Erasure Coding,
>>>> optimized for different architectures, and  with some unique(?) features
>>>> for recovering a failed fragment/disk. If Gluster allows alternative
>>>> implementations for the encoding, it would help organisations and
>>>> researchers to get results of their work in a distributed filesystem.
>>>> And with that, spread the word about how easy to adapt and extend
>>>> Gluster is :-)
>>>>
>>>
>>> That could be interesting. Is there any place where I can find
>>> additional information about the features of this library ?
>>>
>>> Xavi
>>>
>>>
>>>
>>>> Thanks,
>>>> Niels
>>>>
>>>>
>>>> 0. https://www.usenix.org/conference/fast17
>>>> 1. https://memoscale.com/
>>>>
>>>>
>>>
>>
>


-- 
Pranith
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.gluster.org/pipermail/gluster-devel/attachments/20170303/4fb5b1fb/attachment.html>