[Gluster-devel] Choice of Translator question

[Csibra Gergo]

Friday, December 28, 2007, 1:02:51 PM, Gareth Bult wrote:

> I think we're mis-reading the conversation a little.

Oh. I read through the whole thread, and I understand what you want.
You want AFR for data reliability and striped reads for speed. And for
self heal you want to heal only the bad chunks.

> The issue is is that with 3 stripes, the size of each stripe is (n)/3.
> For a 90GB file, this is still 30GB per stripe.
> Given restarting glusterfsd breaks the "mirrors", this will still
> require at least 30GB to be copied on self-heal.
> (which isn't workable)

> The way in which this "would" be usable if if the striping process
> broke the file down into "many" files, typically down to the chunk
> size .. which would make the self-heal process reasonably
> quick. Unfortunately creating a config file with 5,000 stripes also
> seems a little unworkable.

Well, if you have traditional storage with RAID 1, you need to
remirror the whole disk (which can be 500GB too) if it breaks.
Glusterfs do it better, it mirrors only the files :)

For remirror bad chunks only, you need to store version information
for every chunk, and need to keep them in shape in every read/write. I
think it's can be have more overhead than remirror (heal) one big(ger)
file once.

I think the striped read from AFR is not a big deal, if you have some
programming experience, can do it too :)

-- 
Best regards,
 Csibra Gergo                            mailto:gergo at csibra.hu