[Info-vax] FREESPADRIFT

[Paul Sture]

On 2016-06-17, David Froble <davef at tsoft-inc.com> wrote:
> Paul Sture wrote:
>> On 2016-06-16, lawrencedo99 at gmail.com <lawrencedo99 at gmail.com> wrote:
>>> On Friday, June 17, 2016 at 6:24:46 AM UTC+12, Michael Moroney wrote:
>>>
>>>> Don't forget that Spiralog was newer than ODS-5/ODS-2.  And Spiralog was
>>>> to be a different file system for VMS (not a mod to ODS-5) but was 
>>>> cancelled.
>>> Linux was able to add journalling to an existing filesystem (turning
>>> ext2 into ext3). VMS never did the same?
>> 
>> A mentioned by another poster, work was started on a new filesystem.
>> 
>> We're back to where you spend the available resources - retrofitting
>> a new feature to an existing system or implementing a replacement?
>> 
>> The existing VMS filesystem doesn't cache writes as much as other file
>> systems, and on-disk pointers are built in such a way that a power loss
>> doesn't result in the damage you see on other file systems.
>> 
>> I'd put it that the current VMS filesystem didn't need journalling as
>> urgently as other filesystems.
>> 
>> And yes, the relative certainty that when you write something to disk
>> it the current VMS world, it does actually hit the disk, has been a
>> limiting factor in performance.
>> 
>
> But it sure helped when that $10M in accounts receivable wasn't lost ....

To be honest I cannot remember a single incident where either a crash
or power failure caused wholesale loss of data on a VMS disk.

Yes, that backup/image restore or indexed file reorganisation which
didn't complete due to a crash or power failure would be toast, but it
would be pretty obvious what had happened.

Real life example from the early 2000s when a less-uninterruptible 
power supply took out the power to a whole building:

    The VMS systems came back minus a few disks which didn't spin up.
    Volume shadowing and the replacement of dead disks brought the
    systems back to normal working order.  The applications on these
    systems used a mix of Oracle Rdb and Oracle Classic, set up by
    competent DBAs.
    
    In contrast some of the Unix systems (there were no Linux
    deployments at that customer at that time) required full filesystem
    restores from tape before they could be brought back into
    production.  This took several days for the larger systems.

Footnote: some concrete figures for money lost by extended downtime
came out of that incident and management was not afraid to invest
in beefing up their disaster recovery capabilities.

-- 
There are two hard things in computer science, and they are cache invalidation,
naming, and off-by-one errors.