[Info-vax] [OT] Data/system integrity, was: Re: HBVS, was: Re: WASD 11.0 announced.

[Kerry Main]

> -----Original Message-----
> From: Info-vax [mailto:info-vax-bounces at rbnsn.com] On Behalf Of
> lawrencedo99--- via Info-vax
> Sent: 30-Jun-16 7:03 PM
> To: info-vax at rbnsn.com
> Cc: lawrencedo99 at gmail.com
> Subject: Re: [Info-vax] [OT] Data/system integrity, was: Re: HBVS, was:
> Re: WASD 11.0 announced.
> 
> On Friday, July 1, 2016 at 1:10:05 AM UTC+12, Kerry Main wrote:
> 
> > ... the CAP "theorem" is outdated ...
> 
> On the contrary. It’s never been disproved. All that’s happened is that
> the hard requirements for one or the other have been loosened
> somewhat as a compromise.
> 
> Was there some kind of suggestion further up this thread that somehow
> VMS clustering was a way around CAP? It wasn’t. Remember the
> concept of cluster quorums? Consider the following sequence:
> 
> * You have a process running on node N of your cluster.
> * The cluster becomes partitioned.
> * Node N is in a partition that is unable to achieve quorum.
> 
> What happens to the process? To guarantee consistency, it has to stop
> processing until that node can rejoin the quorum. Which means you no
> longer have availability for that process. Or you continue processing and
> risk inconsistency.
> 

Assume A-A cluster with volume shadowing:
As soon as the VMS cluster node N gets partitioned, it loses quorum &
crashes. Any write in progress completes on all shadow volumes or is 
rolled back from all volumes.

Client auto-reconnects to any remaining node in the cluster & processing
continues. Data is consistent on all nodes, service is available on the
remaining nodes.

> So you see, in the presence of partitioning, you can either keep
> consistency or availability, but not both.
> 
> CAP theorem QED.
> 

CAP theorem RIP. 

Also, please read link I provided for another (albeit much more detailed)
viewpoint which agrees with the latter and not the former.

:-)


Regards,

Kerry Main
Kerry dot main at starkgaming dot com