[Info-vax] New filesystem mentioned

[Michael Moroney]

Dave Froble <davef at tsoft-inc.com> writes:

>On 5/14/2019 9:56 AM, Bob Koehler wrote:
>> In article <qbco24$r22$2 at dont-email.me>, Dave Froble <davef at tsoft-inc.com> writes:
>>>
>>> Not that I'm an expert on VMS clusters, but it has been my impression
>>> that the OS did all the work.  But what do I know?
>>
>>     The file system is very much part of the OS.  If it's going to work
>>     in VMSclusters, it must support VMSclusters.
>>

>Why would any filesystem not support VMS clusters?

One word answer: Caching.

Any reasonably advanced file system will have a disk cache, that is memory 
containing the contents of recently/frequently accessed data.  On a system with
many processes interested in the same records, the file system can easily share its
cached content to all interested processes.  Writing is a little more complex,
the file system will have to lock out other requests for the length of time it
takes for an atomic write to be performed and the cache updated.

Now enter VMS clusters. Now there is a situation where some nodes may have the data
in cache and others not.  They had better have the same content in cache and on
disk!  Additionally, the cached data one node has is NOT visible from another node.
Only the on-disk data is.  And now if node A wishes to write to a file, it must
inform all nodes (via the DLM) that it wishes to do so, and all nodes will have to
dump (invalidate) any affected cached data before it can grant permission to do the
write.  Once the write is done, it must be certain the data made it all the way to
the disk before allowing the other node(s) to read the data back in (remember cache
was invalidated). What file systems have this level of synchronization, other than
ODS-x, even if you assume some sort of 'black box' synchronization (like DLM)
without worrying about its details?

I've been heavily involved with disk sh*d*w*ng in the past.  One of the reasons why
it is so freaking complex is that it has to do similar messy coordination when
individual nodes can come and go as well as individual member drives can come and
go, all asychronously.  That is sh*d*w*ng cannot say to a node "wait, you can't
crash until I do this and that!".

>A file system is the structure on disk, and the code to access and 
>support it.  If any node has access to the device, then it has access. 
>What more is needed?

oh if only it were that simple.