[gpfsug-discuss] How about GPFS Native Raid?

[Chris Hunter]

I'll 2nd Zach on this. The storage funding model vs the storage purchase 
model are a challenge.

I should also mention often research grant funding can't be used to buy 
a storage "service" without additional penalties. So S3 or private 
storage cloud are not financially attractive.

We used to have a "pay it forward" model where an investigator would buy 
~10 drive batches, which sat on a shelf until we accumulated sufficient 
drives to fill a new enclosure. Interim, we would allocate storage from 
existing infrastructure to fulfill the order.

A JBOD solution that allows incremental drive expansion is desirable.

chris hunter
yale hpc group

> From: Zachary Giles <zgiles at gmail.com>
> To: gpfsug main discussion list <gpfsug-discuss at gpfsug.org>
> Subject: Re: [gpfsug-discuss] Disabling individual Storage Pools by
> 	themselves? How about GPFS Native Raid?
>
> OK, back on topic:
> Honestly, I'm really glad you said that. I have that exact problem
> also -- a researcher will be funded for xTB of space, and we are told
> by the grants office that if something is purchased on a grant it
> belongs to them and it should have a sticker put on it that says
> "property of the govt' etc etc.
> We decided to (as an institution) put the money forward to purchase a
> large system ahead of time, and as grants come in, recover the cost
> back into the system by paying off our internal "negative balance". In
> this way we can get the benefit of a large storage system like
> performance and purchasing price, but provision storage into quotas as
> needed. We can even put stickers on a handful of drives in the GSS
> tray if that makes them feel happy.
> Could they request us to hand over their drives and take them out of
> our system? Maybe. if the Grants Office made us do it, sure, I'd drain
> some pools off and go hand them over.. but that will never happen
> because it's more valuable to them in our cluster than sitting on
> their table, and I'm not going to deliver the drives full of their
> data. That's their responsibility.
>
> Is it working? Yeah, but, I'm not a grants admin nor an accountant, so
> I'll let them figure that out, and they seem to be OK with this model.
> And yes, it's not going to work for all institutions unless you can
> put the money forward upfront, or do a group purchase at the end of a
> year.
>
> So I 100% agree, GNR doesn't really fit the model of purchasing a few
> drives at a time, and the grants things is still a problem.