[Gluster-users] Volume Tiering

[Mohammed Rafi K C]

comments are inline.

On 05/02/2016 05:06 PM, Lindsay Mathieson wrote:
> On 2/05/2016 2:29 PM, Mohammed Rafi K C wrote:
>> Hi Lindsay,
>>
>> Volume level data tiering was made fully supported in latest 3.7
>> releases, though an active effort is still going on to increase the
>> small file performance. You can find a starting point through the
>> blog post by Dan [1].
>>
>> Let me know if you have any additional questions.
>>
>> [1]  : http://blog.gluster.org/2016/03/automated-tiering-in-gluster/
>
>
>
> Thanks Mohammed, an informative read. When things have settled down a
> bit at work I'l setup a test volume to attach tiers to. Ironically,
> now we've migrated the production env to gluster, my ability to test
> is lessened  - can't tweak it arbitrarily now :)
>

I understand that, and I know the effort you put to achieve it :). I
really appreciate it .

In fact , tiering is for an eventual performance boost, so no hurry, we
can start eventually ;).  just kidding. :)

>
> Some questions:
>
> - Does the hot tier replication level have to match the cold tier? i.e
> brick for brick?
>
Nop. To understand it easily, You can consider hot tier and cold tier as
a separate volume which clubbed together as a single volume. Which means
you can create a different configuration in both tiers. In fact there
are some volume configuration we are not supporting on hot tier, like
disperse. There is no point of configuring a high performance ssd for
ec, right ?

You can use supported volume configuration in any magnitude.


> - I presume the files are promoted across all bricks. i.e you can't
> have different files promoted per brick.
>

I didn't get your question correctly. But I will try to answer
generically . File movement happens from one tier to another tier, as I
mentioned earlier, you can consider this as moving one file from one
volume to another volume.

As an example, let us assume, you are moving a file from hot tier
(replica configuration) to a cold tier which is a distributed disperse
configuration, then the file will be hashed to any of the distribute
set, then the file will be created with ec meta data (data and parity).

I'm  not sure, I made it more complicated ;).

> - With sharded volumes its the individual shards that are promoted?
>

Since shardes are stored as a separate files in backend, the movement
will be based on shardes.

Please feel free to ask if you have any other question of if you need
more clarification on given answers.

Regards
Rafi KC

>
>
> thanks,
>
>
> -- 
> Lindsay Mathieson

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.gluster.org/pipermail/gluster-users/attachments/20160502/c270ad4f/attachment.html>