[Info-vax] Databases versus RMS

[abrsvc]

On Friday, April 20, 2012 11:44:36 AM UTC-4, Ken Fairfield wrote:
> On Thursday, April 19, 2012 11:18:01 PM UTC-7, Jan-Erik Soderholm wrote:
> [...]
> 
> > The DECs SSD in SSB's (StorageWorks disks) was RAM based as far as I know.
> > That is, did not retain data at power loss. Was often used as one part
> > of a schadow-set so that all reads was doen from the RAM disks and all
> > writes went to both (and thus stored permanently).
> 
> The first Intel fab I worked at was using DEC SSD's for the "hot"
> database files.  This was VAXcluster (7800's IIRC).  I'm pretty
> sure the SSD's were in SSB's.  I *know* that these SSD's had
> rotating disks in the same package (SSB) for non-volatile storage.
> I don't recall for sure, but I *think* they had (small?) batteries 
> as well so the memory could be written to the disks in the event of
> a power failure.
> 
> What I found interesting is that, by the time I got there, these
> SSD's were suffering a fairly high failure rate.  And it was 
> universally the rotating disks that were failing. ;-p  OTOH, by
> the time they were failing, you couldn't buy such things anymore
> (from DEC/Compaq).
> 
> We were able to get as good performance for the database over FC/HSG
> with Host-Based RAID striping (and a few other configuration tricks)
> on the Alphas that replaced the VAXes, so we didn't need SSD's anymore.
> 
> [...]
> 
>     -Ken

The disks were there for backing up the memory in the event of a power failure.  There was at least one attempt to have the drives in "standby" or not spinning, but the spinup time was too long for reliable backups.  Thus, the mode was to keep the drives spinning and ready for data transfer.  This allowed for a smaller battery overall.  The power was only sufficient to save the memory contents to disk and then power down "cleanly".