Logical Volume Support

[sharwood]

Is this on the AppLogic roadmap?

[PeterNic]

Can you elaborate on the capability you are asking about?
Regards,
-- Peter

[sharwood]

Hi Peter,

Essentially the ability for applications to see all drives on a physical node (or the entire grid) as one drive. If I understand correctly, right now, the size of the volume that an application can have is limited to the size of the largest hard drive on the node which the application is running.

I see in 2.0 that mulitiple drives are supported, will this address the above issue?

Steve

[PeterNic]

Hi Steve,

Yes, AppLogic 2.0 supports multiple hard disks per physical server, presented as a single disk.

For example, if the server has 2 disks of 750 GB each, you see this under AppLogic as a single disk of 1.5 TB (slightly less for the AppLogic boot partition itself). As a consequence, the largest volume you can have is approximately of that size (of course, mirrored on another server for redundancy).

Regards,
-- Peter

[sharwood]

That's excellent. Thanks Peter.

[acw]

Are there ways to have a logical volume from disks across servers?

[PeterNic]

Albert,

Yes and no. A single volume currently needs to fit within a server; it may span multiple drives within the server but not across servers. (Of course, it is mirrored across two or more servers.)

You can still create larger volumes inside appliances -- by using mdraid or lvm inside appliances to span multiple logical volumes within the appliance.

Does that help?

-- Peter

[acw]

mmm... not sure if that can help. Is this what you are saying :

- create x number of data vol, and call each vol1, vol2, etc...
- attach all of them to an appliance
- use mdraid or lvm to pool them together into a single mount point for nfs, and then other servers can attach using that single mount point, and have all the vols appear as 1 directory ?

But there is a limit of 12 vol for an appliance isn't it?

What I need is looking for a way to increase a usable file system without having to resort to building it into the app some kind of dir management. The only choice I see right now is to partition the data storage, and have the application track / assign locations, which I'm wondering if we can avoid.

[PeterNic]

Yes, that's what I meant.

If you want to go above 8 or so volumes, I would definitely recommend looking at clustered or distributed file systems (in appliances).

-- Peter

[acw]

something like Lustre? any info if anyone is using this or something similar on applogic?

[PeterNic]

Yes, lustre, GFS and the like; maybe even Oracle's OCFS2; see some others at http://www.linux.com/articles/30433. I know at least one company who's using OCFS2. If you don't need a posix-compliant API (fopen/fread, etc.), then you may also use things like MogileFS (used by digg) or AFS, as well as other distributed file systems (http://en.wikipedia.org/wiki/List_of_file_systems#Distributed_file_systems).

Regards,
-- Peter

[PeterNic]

more links at: http://www.yolinux.com/TUTORIALS/LinuxClustersAndFileSystems.html

[acw]

Thanks Peter. MogileFS looks very interesting, but the lack of POSIX compliance would mean quite a bit of work to implement.

Any plans from 3Tera on developing special functions / capability or appliance for this type of need?

[PeterNic]

Yes but nothing definitive at this time. You may also want to ping our partners. (If anyone's interested in setting something up, please contact me or post here.)

We do plan to release a NAS headend appliance from one of our partners; it is like the NAS appliance but instead of storing locally on grid's disks, stores over a web-based external storage service. However, it sounds like that is probably not what you're looking for.

If you or anyone else decides to build it, we can help. If anyone already has it and wants to share -- free or paid -- ditto.


Regards,
-- Peter

[acw]

Hi Peter - the web based storage - not really looking to do that as current option, but may become an option. We'll take a crack at the few options you mentioned and check them out. Will let you know. Thanks.