[Linux-cluster] Vmware clustering and GFS

[Robert Ruge]

I have a rather obscure problem that I am hoping someone has seen
before.

I have a functioning Redhat 4 cluster with GFS and I have been running
Vmware Workstation 5 instances out of the GFS filesystem with no real
problems. However, now I am attempting to setup a two node Microsoft
cluster server using Vmware and shared disk files all stored on GFS. I
have created the shared disk files and edited the Vmware configuration
file as per the instructions for setting up a cluster. However, when I
start the Vmware cluster first node it locks up before the Windows
login screen, cannot be killed and requires a hard reboot of the
Redhat server. But if I move the local part of the Vmware machine
(configuration and C: drive file) to an ext3 filesystem but leave the
shared disks on the GFS filesystem it all appears to work ok.

There are no errors in the log file but the server load climbs
dramatically after the lockup.
Any suggestions?

Here is the relevant bit from my vmx file.

#!/usr/bin/vmware
config.version = "8"
virtualHW.version = "4"
memsize = "512"

# Shared bus settings
diskLib.dataCacheMaxSize = "0"
diskLib.dataCacheMaxReadAheadSize = "0"
diskLib.dataCacheMinReadAheadSize = "0"
diskLib.dataCachePageSize = "4096"
diskLib.maxUnsynchedWrites = "0"
disk.locking = "FALSE"

# INTERNAL SCSI HBA
scsi0.present = "TRUE"
scsi0.virtualDev = "lsilogic"

# INTERNAL disk 1
scsi0:0.present = "TRUE"
scsi0:0.fileName = "adc-g-1.vmdk"
scsi0:0.redo = ""

# Shared disk HBA
scsi1.present = "TRUE"
scsi1.virtualDev = "lsilogic"
scsi1.sharedBus = "virtual"

# Shared Quorum disk
scsi1:0.present = "TRUE"
scsi1:0.fileName = "/export/VMware/adc-g-cluster/Quorum.vmdk"
scsi1:0.redo = ""
scsi1:0.mode = "independent-persistent"
scsi1:0.deviceType = "disk"
scsi1:0.virtualDev = "lsilogic"

# Shared Data disk 1
scsi1:1.present = "TRUE"
scsi1:1.fileName = "/export/VMware/adc-g-cluster/SharedData1.vmdk"
scsi1:1.redo = ""
scsi1:1.mode = "independent-persistent"
scsi1:1.deviceType = "disk"
scsi1:1.virtualDev = "lsilogic"

Robert Ruge   School of Information Technology, Deakin University