Re: ACID et al

David Cressey wrote:
> ...
>
> I disagree. But my view is heavily influenced by DEC VMS and DEC Rdb. The
> DEC VMS lock manager was integral to the OS, and very comprehensive. DEC
> Rdb was able to use a subset of the lock manager services for its own
> purposes. I don't think either engineering group was shortsighted,
> compared to most.
>
>

Certainly not as much as the people who designed the OS I'm using while I write this!

Having spent much more time in the Dec world, you'd know better than I would how things worked on the inside. I was impressed by VMS's equivalent to IBM's VSAM access method, even though I can't remember its acronym (too long ago that I used it and that was only once) because it had locking of a sort that VSAM didn't have, with some notion of transaction, built in. As I recall, 'keys' were locked but they couldn't be just any old key that a program might make up, the keys had to have values that were keys in that access method's stored records. Still it was an improvement over VSAM. I'm not so sure about AS400 aka System 38 as we weren't allowed to know about it for fear that IBM might sue us! Actually, I've read more than once that even IBM didn't want too many people to know about the S/38 since it was designed, built and sold by a 'junior' division that wasn't supposed to cut into the big-ticket sales of the senior division.

In the big IBM OS's, you could make up values that you wanted locked and synchronize things with other tasks or address spaces as long as they co-operated. But I believe the OS designers didn't consider very important the possibility of handling thousands of such locks at the same time, which is what the DB engineers would have liked.

p Received on Thu Dec 08 2005 - 02:28:43 CET