[SAC] New OSGeo Virtual Machine Creation

[Howard Butler]

On May 12, 2010, at 4:18 AM, Seven (aka Arnulf) wrote:
> Can you explain what criteria qualify a project to "get their own" or
> what "grouped into generic Project VMs" exactly means?

My opinion is the answer to this is "after a project has demonstrated that it is consuming enough resources of a shared instance to warrant its own".  A big challenge with each project getting its own instance is the fact that users each instance most likely do not totally consume the resources, and we end up partitioning a lot of resources and having highly variable utilization.  We don't want to get into the position where we have 10 VMs doing basically nothing and 1 that is completely saturated no room to grow.  

Another challenge with lots and lots of VMs is that the creation and maintenance of each one increases our administrative overhead cost.  

> We do not mind to
> share any resources as long as they are responsive and not impacted by
> outages as this will not do for publicly visible services.

The "contract" with using TelaScience's systems has always been "be prepared to move your stuff quickly in case of a long outage."  Most of the projects affected by the recent outage were prepared and able to move a DNS pointer and set up a new home in under the time of the DNS update once it was clear the outage was going to be more than a few hours.

Moving to systems hosted by OSGeo will not remove the fact that projects are still responsible for their own setup, backup, and migration of their custom services in the case of an outage (not withstanding common ones like trac, mail, etc).  Otherwise, a project not prepared for this is at the mercy of the BOFHs.

We're guests at TelaScience, and we should remember to continue to treat our relationship as such.  Moving all of our services that are hosted at TelaScience to our own systems removes the guest aspect, but not the volunteer one.  

</preach>

Howard