[Linux-cluster] iSCSI GFS

[Gordan Bobic]

isplist at logicore.net wrote:

>> If you need the ultimate flexibility, you could always have all the
>> storage exported via iSCSI to an aggregator box, have that run a
>> software RAID to link them all up, and then re-export that via iSCSI.
 >
> I'm looking on goggle, I'm not finding exactly what this is yet. Would this 
> also be known as storage aggregation or since we're talking about iSCSI, iSCSI 
> storage virtualization?

It's not virtualization. It is equivalent to mounting an NFS share, and 
then exporting it again from the machine that mounted it.

> I am guessing that what you're talking about would mean something like the 
> servers looking at one point for their data. That one point would be called 
> the agregator. 

Exactly. You have a machine that pretends to be a SAN when it in fact 
has no space on it. Instead, it connects to all the individual storage 
nodes, mounts their volumes, merges them into one big volume, and then 
presents that one big volume via iSCSI.

> The aggregator would basically either be the central connection point to all 
> of the data (which I doubt since it means it would bog down at some point) or 
> it acts as the routing point, sending the server to the proper storage device. 

It's a central connection point AND a router, only it isn't just 
straight routing, because the data is RAID striped for redundancy.

> I am guessing that the storage device ends up being almost anything reachable 
> on the network, iSCSI, FC, what ever servers can reach. It might be storage on 
> the FC network, it might be a box with an IP, it might be storage behind 
> another server.
> 
> Is this the idea?

Pretty much.

>> It rather depends on what your requirements and constraints are. You can
>> do all sorts of clever things with a multi-level-tree organisational
>> design of your storage solution (something like having a separate
>> eggregation box as I described above), and it'll scale pretty much as
>> far as you want it. Just add an additional disk chassis into the aggregator.
> 
> My concern is also trying to keep things fairly easy to understand so that I 
> can explain it to the person that will be hired to help me eventually :). 

Good help is hard to find. Most people seem to have a fundamentally 
lacking understanding of more advanced things like SANs and clustering. 
I know because I interviewed quite a few recently. :-(

But if they have a good understanding of clustering, RAID and SANs all 
this should be pretty obvious to them.

> More than anything, right now at least, I just want fast reliable LAMP 
> services and badly need an easy storage growth path that does not tie me into 
> strict requirements. 

You might find that the SAN-of-SANs idea above, with the aggregator, is 
the most future proofed growth path as you could scale it to pretty much 
any size the underlying file systems will allow.

Gordan