[Info-vax] Marketing ideas for VSI ?

[Stephen Hoffman]

On 2018-12-14 16:38:51 +0000, Dave Froble said:

> Consider RAID 1.  Mirrored disks.  Lose one, but your data is on the 
> other.  Nothing lost.
> 
> Now, not exactly the same, but perhaps consider cluster nodes as 
> mirrored systems.  Lose one, and others continue with the processing.
> 
> You don't get that when you have to recover to a passive system.

Replication can provide what you're describing.  It's fairly common, 
particularly for environments and configurations that don't offer 
active-active configurations.  This includes what HPE refers to as 
"zero-data-loss" recovery configurations.

https://h20195.www2.hpe.com/v2/GetPDF.aspx/4AA4-4251ENW.pdf

There are also options here for two-phase commit or analogous into 
remote memory—which is much faster—and two-phase commit into remote 
storage, using either server- or controller-based replication.

As for the remote memory?  Was just reading details on a recent Intel 
Optane NVMe product with 2 GBps read and write rates, and a ~7 PBW 
ratings.  Which means stuffing journals and logs onto those can work 
quite well, if not the entirety of "smaller" (~400 GB) wads of data.  
~7PBW is not a small amount of data.  Ponder 2PC (Paxos or otherwise) 
across a 10GbE network into that NVMe, and shoveling (just) the app 
data across the network connection.  And yes, there are software 
packages available for various platforms that can help with this, too.  
 (OpenVMS has seen work toward porting some specific and related 
packages too, though I've not checked the status recently.)

https://www.the-paper-trail.org/post/2009-02-03-consensus-protocols-paxos/
https://www.the-paper-trail.org/post/2014-08-09-distributed-systems-theory-for-the-distributed-systems-engineer/ 

http://www.gtug.de/HotSpot2018/download/Presentation/B406-Cameron.pdf


-- 
Pure Personal Opinion | HoffmanLabs LLC