This brings up a feature that I really miss from the now missing Gemstone app server. It had the ability to cluster VMs on a single machine (as well as across a network), and allow an admin to partition an app across those VMs if needed. It would also cycle VMs on a schedule, start up VMs if a particular load threshold was achieved and shut down unneeded VMs. Although I am not an admin type, I took one of their admin classes and I saw the performance differences by allowing the server to add VMs as the client load increased.

I do not remember too many of the details as it was a few years ago, but each VM had a configuration specifiying heap size, shutdown schedule, etc. Maybe with modern VMs this isn't necessary anymore, but on big hardware, like the 16 CPU Solaris box we were using, it seemed a shame that if a single VM shutdown for whatever reason that that $250k server was now dead as an app server.

my .02 :-)
Robert

Craig Wohlfeil wrote:

I think this should be implemented within the VM as well. Imagine the overhead of having another copy of the server running for every application. I also think that you shouldn't have to have the server installed twice to run a separate VM. With a large server (with clustering support) it can more efficient to run 2/3 small VMs that have the same configuration than one large one.

Aaron Mulder wrote:

However, I'm really suggesting we implement this within the VM,
whereas the above works by running multiple VMs. If you want to run
multiple VMs, you can always just install the product twice. I prefer to
also have the option to run apps sandboxed in the same VM. Granted you
wouldn't be able to run different apps with different server components
(EJB/Web container, etc.) this way, but I think that's OK.


Aaron






Reply via email to