Possible use for memcache?

[mike]

On 11/27/05, David Phillips <david at acz.org> wrote:

> If I understand your approach correctly, it seems to lack consistency.
>  What happens when two servers try to access the same file due to
> memcached dying?  How do you handle the switch over?

The same that happens if two memcache client instances try to
write/access the same key - the memcache protocol will work it out, I
assume. Whichever one wins will be the "representative" for that file
location.

> It sounds like you have some requirements and non-requirements in mind
> already.  Once you formally identify all of those, it becomes much
> easier to create a matching solution.  There might be a simpler method
> given your constraints.

> You would actually want OCFS2, as OCFS is specifically designed for Oracle.

Yeah - I just simplified it by saying OCFS :) I didn't want to
necessarily tie my explanation to a specific version. But OCFS2 is
what I've been looking at as well.

> What exactly do you mean by "the same hardware"?

I mean the same physical storage array, being accessed by multiple
"initiators" - I suppose would be the iSCSI or AoE terminology for it.

Normally a GFS concept with a central locking manager seems to manage
all that. However, I believe (but I could be wrong) that simply using
a software-based middle layer that actually creates a [reverse]
"fence" (basically "allow this server to access this" - not "DENY this
server" like it seems typical I/O fencing does) just by only allowing
one physical client to access the file will remove the problem of data
corruption due to a non-GFS filesystem mounting the same physical
hardware device.

> If you're going with centralized storage, what's wrong with using a
> single NFS server to access it?  A NetApp or EMC product seems
> appropriate here.
>
> This is an interesting discussion and topic, but I still don't have a
> clear understanding of the problem you are trying to solve.

What do those devices cost? Lowest price for a Netapp will run $7-9K
if you're lucky and have a good sales rep. That's for a single head
unit. While it has redundancy inside of it, it's still not two
physical units for unit-level redundancy.

Don't get me started on EMC's either... I don't have an enterprise budget. :)

> > AoE is what I would hope to use, but may have to use iSCSI since it's
> > better supported. There's a thread I found whilst researching this
> > stuff where someone vehemently was trying to disspell supposed truths
> > about AoE and it's multi-targetting and other features.
>
> Can you provide a link?

Yeah, I was too lazy to look it up originally. Here's the thread. He
starts getting into it in this specific email:
http://lists.centos.org/pipermail/centos/2005-November/014509.html

- mike