[Simh] RE : RE : How to setup a simh VAX VMS cluster that handles more than 16 disks.

[Boucher, François]

Hi,

Just a quick suggestion about your cluster device naming ...

>DISKS SERVING

>VMS in a cluster supports disks when they are named differently.
>
>Since you can have a cluster done with identical machines, their device
>names would be the same!  This situation would create a disk name
>collision!.

>I think you mean "... identical machine configurations" rather than
>"identical machines".


Well, yes and no, as all my vaxes in the cluster are microvax 3900, they are identical (simulated) machines.  But they have different disk images attached to each of them, so a different configuration.


>To avoid this, you have to modify the ALLOCLASS parameter in MODPARAMS.DAT
>to have a different number of ALLOCLASS for each node.  For example, the
>ALLOCLASS=1 will have disks of the node having a $1 prefix added for the
>disk device.  Also, on REGIE, an ALLOCLASS=3 will add the $3 prefix etc.
>
>(define a different ALLOCLASS number for each node).

>ALLOCLASS is used to great effect when disks are on shared storage bus such
>as DSSI or CI or they are dual-ported (which are the only options for VAX;
>on Alpha and Integrity you can also have FibreChannel and SCSI shared disks
>-- but SIMH only does VAX.)  Correct use of ALLOCLASS allows a given
>physical drive to be accessible to all cluster members by more than one
>path, thereby improving fault-tolerance.

Absolutely true, but you need the hardware to provide such dual paths.  In simh. even if you define a disk as dual pathed (command is accepted in VMS 5.5-2 but gives an error on OpenVMS 7.3)  It is not really dual pathed as disk access arbitration is done in software by the OS instead of by hardware on the SCSI bus, for example.

>However if all the disks in your cluster are "local" disks (i.e. directly
>attached to their respective host nodes) which is typical for the disks in a
>SIMH setup, much confusion can be avoided by setting ALLOCLASS to 0 on all
>nodes.  With ALLOCLASS=0 the device names change to node$DUAnnn: where
>'node' is the VAX nodename (SCSNODE parameter) and 'nnn' is the unit number.

Also completely true.  The only advantage is that the device name is shorter with $1$dua1: than $extra$dua1:

>Of course, if you have a VAXcluster of SIMH instances running on a single
>host computer (e.g. a single PC running Windows or Linux), it might be
>possible to have each simulated VAX connect to the same disk image.  In
>which case you would have shared storage and need to set ALLOCLASS to some
>positive integer > 0 (the same value on each machine which shares the disk).

Effectively, I have a single desktop running 3 simh instances, because it contains a quad core CPU that permits this.  I did try to attach locally a disk image to more than one simulator instance, and unfortunately, it does not work!  I can mount the disk image in VMS on say two nodes, but if one node write any modification to it (let's say it creates a new file), the other simh instance is not aware of the data modification on the disk, as if the host OS (linux in my case) does not send a signal to the other node to make it aware that a change in the disk has occured.

The only way i found to successfully overcome this problem is to mount a given disk locally on only one simulated node, and to share it through MSCP to the other nodes in the cluster.  MSCP makes other nodes aware of modifications to the contents of a given disk.  Of course, you need a good network performance, and then you need to apply a tuning to the ethernet driving of simh.  (see the tread about [Simh] 10ms delay on ethernet communications - the solution! <http://mailman.trailing-edge.com/pipermail/simh/2008-January/001539.html> )  However, my desktop is using a SATA disk, and it might be possible that using a fibre channel external disk permits this locally attached disk sharing.  This is a thing that SRI does with their Charon Vax setup.  I just did not test this for simh, so I cannot say true or false about this.  If anyone did successfully mount locally a disk image to two or more simh vaxes, please let me know how you did it!!!  I would be most happy to know about this.

Thanks for your suggestion, Jeremy

Francois Boucher ing.



Regards,

        Jeremy Begg


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20080117/5a2d6a8d/attachment-0003.html>