[clue-tech] "rewindable" drive?

[Nate Duehr]

The problem is that he wants the equivalent of a snapshot every time a  
single block is written.  Snapshots work great, but triggering them  
every minute is a huge loss in system processing power as the  
filesystem/dataset gets larger.

While as he says "someone will do it" and maybe even has, things like  
RAID 1 can handle disk redundancy, and much slower frequency snapshots  
can handle most reasonable needs to "roll back".

Getting down to atomic roll backs of garbage data going into an entire  
filesystem (meaning garbage data out in true GIGO form right onto a  
filesystem by a poorly designed application -- compared with a  
database and a well-designed app that clearly separates the UI from  
the data being collected and acted upon, which is a different story)  
means something is severely wrong with the input application's ability  
to confirm the accuracy of the input data from users or that there is  
deliberate tampering or severe stupidity at play, both of which are  
human issues better "fixed" by policy not by the technology.

"Dear User/Employee: Cause enough damage to our work, get fired...  
Because we KNOW it will mean we have to drop back to an hour/day/week  
old snapshot.  We're not willing to spend the kind of money/time  
necessary to protect you down to the individual transaction because we  
think adults can handle data entry without doing it in a way that will  
damage the system.  Here's a tested procedure that works.  Use it or  
you'll become personally and intimately familiar with Colorado's  
unemployment insurance process."

Policy like the above should be far  easier and cheaper than trying to  
do atomic change protection of an entire filesystem and every data  
write.

I've seen one billing system that had to do a COW on every write, but  
the copy was not meant as a real-time recovery tool.

It was just a legal record of "what happened" and was stored on a  
write-only filesystem only readable by the auditors.  No deletes were  
allowed and none of that data was ever copied back to the "sending"  
system(s).

The application that did the COW was completely proprietary and custom  
and the source had been audited by an outside 3rd party for accuracy.   
(It handled billing records in near real-time between major telco  
carriers and this was saving all parties so much money, they could all  
easily afford to pay for such a customized system and the auditing.)

--
Nate Duehr
Sent from my iPhone

On Nov 21, 2008, at 18:27, "Collins Richey" <crichey at gmail.com> wrote:

> On Thu, Nov 20, 2008 at 2:02 PM, Nate Duehr <nate at natetech.com> wrote:
>
>
>> A hardware or software "snapshot" solution would work, but there's  
>> added
>> complexity if there's a DB involved.  (You have to quiet the DB  
>> before the
>> snapshot... some DB's have commands to do this -- Oracle, etc -- some
>> don't)...
>>
>
> I can't comment about the other hardware or software solutions, but
> the Storagetek/Sun solution I described is an instantaneous
> snapshot.Every block of the chosen file/filesystem is marked as part
> of the snapshot without any intervening I/o, so even databases can be
> rolled back to their instantaneous and valid state.
>
>
> -- 
> Collins Richey
>     If you fill your heart with regrets of yesterday and the worries
>     of tomorrow, you have no today to be thankful for.
> _______________________________________________
> clue-tech mailing list
> clue-tech at cluedenver.org
> http://www.cluedenver.org/mailman/listinfo/clue-tech
>