Why was Spiralog retired?

[Robert_Boyd]

A friend of mine was telling me there was an essential flaw in the algorithm for Spiralog that eventually came to light and doomed it.

He thought there was a technical article that explained the problem, but I haven't been able to find any reference to it.

Anybody know where I should be looking?

Robert

[Hein van den Heuvel · Accepted solution]

Mostly it was a day late and a dollar (sic)short. Besides the technical challenge, there was the challenge of a remote engineering group (Dollar (the town), Scotland
Timing/Politics (internal to VMS, and the future of VMS as seen at the time)/Engineering resources and such all played a role.

Supposedly one trivial/predictable technical flaw did it in:
It was a log based file system right?
So before you actually did something to the data, you would add an entry to the log to describe what was going to happen (to be able to recover from failures).
So if you delete a file, you first need to add data to the log, then the delete can take place.

Now what if you try to delete a file because you are out of space: disk full?
You can not do the delete, because you need space, but to make space you need to do the delete. Ouch.

Supposedly that was the nail in the coffin.
I'm sure you and I can come up with solutions to this problem (reserve some space for deletes comes to mind :-), but none was deemed robust enough to solev this problem for all folks under all circumstances.

fwiw,
Hein.

[Uwe Zessin]

That's what I heard, too:

- the file system turned into read-only mode once it became full and you had to recover with a full backup/restore cycle :-(

- DEC was too much in love with Windows NT

I think this group was also working on another product (Snapshot Services for OpenVMS) that had even made it into the price list, but was killed anyway.

[Stanley F Quayle]

Actually, I sometimes have a need for reading a Spiralog file system. Anyone have any code for it?

[Uwe Zessin]

How an old version of VMS? ;-)

[Stanley F Quayle]

> How an old version of VMS? ;-)

Ending with which version?

[Tom O'Toole]

I remember the problem with filling up a spiralog disk. I found that one immediately during my testing. And yes, one would think reserving some internal space should have allowed this problem to be resolved.

I thought the file system was great, but never used it in production for the above reason. It's too bad this wasn't continued. It would be appropriate for apps needing write performance, like database journals, or disk to disk backup.

[Rob Young_4]

Yo Tom... how goes it?

>It's too bad this wasn't continued. It would be appropriate for apps needing write performance, like database journals, or disk to disk backup.

I disagree. When you look at the percentage
of VMS disk (in total megabytes world-wide) that has writeback cache turned
on, it is probably quite a high percentage.
Shoot, I know storage that has 64 GBytes of
cache (not MByte, GByte). Writes that take
about a millisecond to complete. I'm looking
forward to the day when VCC_WRITEBEHIND and
VCC_WRITE_DELAY are default settings and
writes are low micro-second. I had
thought by now that MRAM would be maturing
but it is taking time. Some day when
MRAM (or similar) is quite common, that is
probably when we see OSes hitting cache
with their writes and still be crash consistent.

Spiralog looked great 10 years ago (for just what you mentioned - write in place). But the hardware moved the goalposts.

Rob

[Wim Van den Wyngaert]

I'm getting old. Read this about MRAM.

http://en.wikipedia.org/wiki/MRAM

Wim

[Uwe Zessin]

Now, that reminds me about one of my previous employers who played with
http://en.wikipedia.org/wiki/Bubble_memory