[Interest] Processes - was Click and hold on close button
roland at logikalsolutions.com
Sat Aug 31 18:37:12 CEST 2019
On 8/31/19 5:00 AM, Konrad Rosenbaum wrote:
> On Friday, 30 August 2019 12:34:20 CEST Roland Hughes wrote:
>> Actually it's more the wanna-be OS world.
> ..or griping for real OS'es! Who needs a hierarchy in a file system! Why have
> complex storage systems if you can have records in your files!
This statements makes absolutely no sense. Perhaps cutback on the
alcohol before responding? <Grin>
When DEC introduced RDB it was state of the art. On a VMS cluster, which
could span much of the globe, you could create an RDB (Relational
DataBase, I know, creative naming) database and scatter its storage all
over the cluster. To any user on the cluster it appeared as a single
local database. I worked at one client site which had multiple locations
in USA, Germany and PR which had a database spanning all of those
locations. You could do selects across all of the tables.
Each logical drive had its own hierarchy. A logical drive could be one
or more physical drives and could consist of bound volumes. Most shops I
was at also defined a logical along the lines of ALL_DRIVES which mapped
every drive in the cluster. We just didn't call it root and give it a
single character name.
> For another: most modern systems (including Linux, Windows, MacOS/X) have a
> fully developed concept of processes, threads, tasks and kernel threads that
> is not that different from the concepts of systems running on what used to be
> called "big iron".
"big iron" was only ever IBM, AMDAHL and a few other mainframe vendors.
Once the DEC-10 and DEC-20 went away DEC only made midrange computers.
AS/400, RS/6000 and a few other models were the IBM entries into this
>> When you get into OpenVMS,
>> Z/OS, AS/400, TANDEM, etc. you get real processes and real threads.
> You do realize you are talking Bullshit now - right?
> I haven't worked with those other three, but OpenVMS does have processes very
> similar to Unix - the main difference being that child processes cannot
> survive their parent and there are a couple of IPC mechanisms that just don't
> make sense outside VMS. Whether that is better or worse is debatable.
> This is my personal opinion, but I find the process related concepts on e.g.
> Linux (processes, tasks, cgroups, namespaces) much more mature than what you
> get on a modern OpenVMS. This is not VMS' fault - it simply doesn't have the
> same number of developers.
It isn't excrement of any kind. Your comparison is much like claiming a
Ford Ranger pickup is very much like an MWRAP.
If one can scope the statement to: Both burn fossil fuel, can be driven
and can haul a sack of groceries, then and only then is it correct.
These things which you gloss over are what defines the weight. It isn't
just that all child process are killed off when the parent dies. It is
that all resources are cleaned up by process management. All
non-committed journaled file writes, pending database transactions,
memory, etc. are cleaned up across the entire cluster without having to
hope backdoor tricks like network timeouts do such a thing for you. No
matter what node in the cluster your process is running on, every node
is notified the process went away and process level cleanup occurs.
Each process also has a process level logical name table which can be
inherited by child processes. The parent process can also create a JOB
level logical name table which is inherently shared by all child
processes. Children can both read and set values in the JOB level table.
Many developers use this along with ASTs to implement IPC. There are
better ways to do IPC on the platform, but this poor man's hack works too.
While each node may have its own UAF (User Authorization File) for
booting and/or standalone operation, when it joins the cluster it uses
the cluster wide UAF. Any existing processes on said node have their
systems resource limits altered to match the settings in the new UAF as
well as rights identifiers for Access Control Lists (ACLs).
The list goes on and on.
> I guess that you also realize that you are implicitly accusing the chief
> designer of OpenVMS of making a conceptually much worse OS with Windows NT -
I was actually under contract to DEC for a project when Cutler was
developing WNT for the Alpha. WNT got its name because each was one
letter up from VMS. That version really was a GUI VMS. It fully
supported Files-11 with file versioning, journaling, etc. It had logical
name tables and the same proper weight processes VMS had with all of the
trimmings. I got to see it on the early Alpha machines in the lab. It
was stunning. It could even cluster.
The tiny minds at Microsoft couldn't wrap their brains around it. They
stripped it back to Nothing-but-DOS, keeping only a tiny portion of the
code Cutler had developed. All of the really important features were
removed. The processes were stripped back to the lightweight things
which exist today.
Roland Hughes, President
More information about the Interest