[OpenAFS] AFS suggestions

lamont@scriptkiddie.org lamont@scriptkiddie.org
Wed, 27 Jul 2005 14:19:18 -0700 (PDT)


On Wed, 27 Jul 2005, Chris Huebsch wrote:
> On Wed, 27 Jul 2005, ERIC K. CHEU wrote:
>> thanks... I'm actually not a novice AFS administrator.  But my bosses have 
>> just asked me to ask what other people to provide for more redandant 
>> storage and for charge back model.
>> We are already going to have a raid 5 system implemented to hold our AFS 
>> volumes.  And my boss was trying to think of ways to make AFS volumes more 
>> redundant while at the same time being available for change (something 
>> that is hard to do when you only can make read-only replicas).  Now
>> if there were such a thing as read-write replicas, then that'd solve a lot
>> of our problems...
>
> But there are not (yet). There are some strategies using ro-clones,
> which can be transformed to rw-volumes if worse things happen.

I'm intrigued by the "(yet)"?

I also have applications for r/w replicas.  It would be entirely 
acceptable for me to mirror writes transparently across two fileservers 
and be able to treat that just like a normal r/w volume, with notification 
of exceptions due to one of the fileservers failing.  I'm going to wind up 
writing code like that myself otherwise and playing shuffle magic with 
volumes -- a transparent r/w mirror would save a lot of development 
time...

> And a further hint from me: Do never trust Raid 5. The disks are like
> twins. They have the same histroy, they lived the same live, they die
> the same day.

And RAID controllers can puke and ruin your whole day no matter how you 
architect your array...