[nycbug-talk] Approach for the NFS cluster

[Brian Cully]

On 19-May-2009, at 20:18, Jesse Callaway wrote:
> I think scalable, available storage is something everyone is thinking
> about. If anyone feels like presenting their experience to the group,
> please submit on the website's 'about' page.... maybe a few short
> presentations in one meeting like a X vs Y vs Z. I can see heated
> debate already. Who knows if it will get scheduled, but certainly good
> to post ideas.

	Databases have already solved many of these problems to the best they  
can be solved. If you need robust network-available storage, use a  
database and all that entails. If you want their qualities without  
giving up current POSIX APIs, you'll be waiting a long, long time.

	In other words, the difference between one-to-many and many-to-many  
is insurmountable: failure is not an option, it's a requirement.

	To re-re-rephrase, your requirements determine your solutions. Most  
people don't need ACID for most of the things they do, so they  
likewise don't need a proper database and all its requirements. By the  
same hand, most people are fine with simple storage solutions. What  
you require depends on your application and its requirements. Perhaps  
what the world needs is a configurable HA system, rather than one  
which tries to do one thing that works for everyone. To my knowledge,  
there is none such animal.

-bjc