> Some of the downsides of using disk as disk are 
>
> 1) Concurrency to disk: if you have multiple media servers, how to ensure
> they're cooperating peacefully when reading/writing to the same disk area.
> You can do that in several ways:
>
>       * Divide up in separate LUNs. Each media server owns it and creates
> a file system on it. Very difficult to change the amount allocated to each.
> Maybe a volume manager and advanced file system can address some of that.
>
>       * Cluster file system. Good solution. Introduces complexity and
> potentially cost. 
>
>       * "Gentleman's agreement" sharing of LUN. Access to files are
> centrally brokered by the master server. I believe this is how CommVault
> Galaxy/GridStor operates. The servers asks broker politely before writing to
> a certain disk area. Not a true cluster file system.
>
>       * 

The way that CommVault does it with Gridstor is that each media server 
shares an index cache (the equivalent of NBU's catalog/databases) via 
NFS or CIFS. I call it a "poor man's clustered filesystem". It affords 
one the luxury of pooling media servers, such that if one goes down, any 
other can take over in it's place. Since the index cache is shared, the 
job just picks up where it left off. It also allow one to load balance 
among media servers. The same goes for disk storage targets. If disk 
storage unit "A" becomes unavailable for any reason, disk STU "B" can be 
configured to take over in it's place. Massive job and component resiliency!

-- 
===================================

   Steven L. Sesar
   Lead Operating Systems Programmer/Analyst
   UNIX Application Services R101
   The MITRE Corporation
   202 Burlington Road - MS K101
   Bedford, MA 01730
   tel: (781) 271-7702
   fax: (781) 271-2600
   mobile: (617) 519-8933
   email: ssesar at mitre.org

===================================