Unix on a VAXCluster ??

[Dick Dramstad]

Ron,
	You write:

>Up until recently, a VAXCluster is a set of VAX's around interconnected by
>the CI bus.  The CI disk/tape controller (HSC50) which is the heart of the
>VAXCluster is very hostile to non-VMS/TOPS-20 operating systems, as it has
>OS specific code in the controller microcode.  I assume DEC is planning to
>provide Ultrix support in the HSC.

>Lately DEC is selling things called (humorously) local area VAXClusters.
>This is essentially a DEC proprietary equivelent to the traditional
>diskless workstation.  It requires no specific hardware other than an
>DECNET compatible network interface. Note that due to unexplained phenomena,
>you can't support diskless Local Area VAXCluster nodes from a real CI
>VAXCluster.

	Understanding why you can't support LAVC nodes from a real CI
VAXcluster will not only take the mystery out of this "unexplained
phenomena", it might even take the "humor" out of the situation.  And
"traditional diskless workstation"?  My, how quickly nice innovations
become traditional!  Wonder when I'm going to get a traditional
personal supercomputer on my desk?  :-)

	VMS's CI driver is just another path to disk and tape drives,
like MASSBUS or Unibus device drivers.  The DEC workstation folks
looked at VMS and said, "Hey, we'll just diddle the CI driver and make
it talk over the Ethernet instead of those expensive funny blue CI
cables and then we can have a traditional diskless workstation."  (I'm
paraphrasing, of course...)  Ethernet is slower than the CI, but for
MicroVAXs it does a reasonable job.

	Now, if you substitute the Ethernet CI driver on the boot node
(the moral equivalent of a file server), you can't get to CI-connected
devices because VMS "knows" that you only have one CI.  So, you can
have one or the other, but not both.  

	DEC will tell you (although probably not officially) that
they're working on "hierarchical" VAX clusters that do what you'd like
them to, but given the amount of work it took to make regular VAX
clusters work (Version 4.n, basically), I wouldn't be surprised by a
long wait.  (Then again, I'm sure the changes to the Unix kernel to
support NFS were no less pervasive than VMS changes to support
VAXclusters.  I've heard it said that yes, you could "drop NFS" into a
regular BSD kernel, but it "makes a big splash."  Hierarchical
clusters will probably cause a flood.)

Dick Dramstad
rad at mitre-bedford.arpa