Re: [Bigdata-developers] Bigdata HA architecture design document (RFC)

[Bryan T. <br...@sy...>]

I've added some language on HA for the standalone versions of bigdata:

Standalone
In addition to the clustered database, there are two standalone bigdata versions.  One version is based on the WORM store and is suitable for between one and two billion triples are write-once data sets.  The other version is based on the RW store and is suitable for larger data sets and read-write workloads.  The standalone versions have less overhead for full transactions and have no overhead for RMI.  However, they have less total capacity and significantly less potential write and query throughput.

Highly available standalone systems utilize the same write replication and master election mechanisms.  Load balancing for query is possible, but must be done at the application layer, e.g., by round robin of different SPARQL queries to the different nodes in the failover chain.  Like scale-out, the master for standalone is dynamically determined and writes must be directed to the master.

Bryan 

> -----Original Message-----
> From: Bryan Thompson 
> Sent: Monday, February 22, 2010 1:26 PM
> To: 'big...@li...'
> Subject: Bigdata HA architecture design document (RFC)
> 
> All,
> 
> Please find attached a design document for the bigdata HA 
> architecture.  I believe that this captures most of the 
> inputs from various people over the last several months.  
> 
> Comments are hereby solicited.
> 
> Thanks,
> 
> Bryan
>