[Info-vax] Re; Spiralog, RMS Journaling (was Re: FREESPADRIFT)

[Stephen Hoffman]

On 2016-06-18 18:31:49 +0000, Kerry Main said:

> One only has to power off a system in the middle of a benchmark on that 
> system to see what the effects could be in the real world.

I've worked more than a few OpenVMS systems that have to be manually 
rebooted, and with non-trivial recovery processing.  There's no APIs 
for coordinating with backups, no integrated means of checkpointing 
application data, and a whole pile of other limits.

> With new disks like SSD and really big memory technologies emerging, 
> one needs to ask whether the difference in perf is really worth  the= 
> potential for data integrity issues?  Especially when one considers the 
> considerable pain  one must go through to recover from some significant 
> event like a system crash or hardware error or power failure on a write 
> back designed system.

Yeah, and I've had that same issue with long-running jobs on OpenVMS, 
where the primitive tools force that into the database and particularly 
onto the end developer,   There's nothing even remotely close to Time 
Machine in OpenVMS, either.   Then there's that RMS is incapable of 
getting consistent live backups.

> Perhaps we will see more strategies like mirroring a RAM disk with a 
> SSD disk so that reads come from the RAM disk, but writes also go to 
> SSD?

Mirror the whole server.   Then there's the support for configuring and 
deploying servers, for replicating servers and checkpointing and 
process fail-overs in a cluster, for containers and app-stacking, and 
any number of other features that might leverage the strengths of 
OpenVMS.  Which all tend toward the "here's a box of parts kiddo, go 
have some fun" answer.

VSI needs to acquire and maintain stable and preferably growing profits 
and — with apps and vendors still leaving the platform — the folks ay 
VSI also need to grow the installed base, too.

That means taking a very careful look at when write-through is 
appropriate and when it's not, when server replication — server 
shadowing, in OpenVMS-speak — might be the way to go, hauling the 
cluster configuration and management forward from the 1980s, 
implementing a file system that's targeted for flash memory and — for 
those folks using older servers and older hardware — SSD storage, and a 
whole pile of other projects far too large to list.  Which will be no 
small effort to design and fund and schedule and deploy and..., of 
course.



-- 
Pure Personal Opinion | HoffmanLabs LLC