Tomcat – Mutliple VMs for Tomcat cluster vs Multiple Tomcat instances on one physical box

Snapshots: Preferred Approach

1. Create a base VM on your machine. For your purposes, you will be perfectly fine using a compact/thin-provisioned disk image file, so don't use a fixed-size one.

2. Shut down your instance of the base VM. (Suspend/hibernate is not sufficient.)

3. Take a snapshot.

4. Distribute it to the developers after you have taken the snapshot.

5. When the time comes, fire up the base VM and update it. Any changes that you make will be written to new "delta" files, which will only be as large as the accumulated changes made since you took the snapshot. Because you have taken a snapshot, that enormous disk image file will remain 100% unchanged.

6. Shut down your instance of the base VM. (Suspend/hibernate is not sufficient.)

7. If you have made major changes (e.g. installing a large service pack), consider taking an additional snapshot now so that you will not have to re-transfer the current snapshot when you make your next update.

8. Overnight, (a) either reboot each developer machine or programmatically kill the hypervisor process on each developer machine (no, the developers will not remember to close it themelves) and then (b) copy over the changed files using ROBOCOPY /MIR or the like.

Of course, in order for this to work, you will need to transfer all changed files associated with the VM, which may include configuration files containing metadata about the snapshots in addition to the snapshot/delta files themselves. You also will need to ensure that, on each developer machine, the VM is stored in exactly the same location as it is on your own machine.

Rsync: Alternate Approach

An alternative approach would be to use fixed-size virtual disks, no snapshots, and an rsync client/server like DeltaCopy to copy only the changed portions. This is not the preferred approach because:

1. It would require installing the rsync software on each developer machine.

2. The actual copy process would take longer, despite no more data being transferred, because rsync would need to analyze the entire virtual disk file file to determine which portions need to be transferred.

Conclusion: You could use rsync for this, but it seems like overkill if your hypervisor supports snapshots.

"Would we take a major performance hit by running nginx on the VMs as opposed to running it natively on the hardware?"

It depends on the underlying hardware and how you're allocating it. Are you going to run several other virtual machines on the same hosts? How many virtual CPU's are you going to allocate versus logical CPU's available?

There is a CPU toll when it comes to virtualisation, but it's extremely little. It can be mitigated by "pinning" the virtual CPU's to logical CPU's with careful planning.