EC2 Instance Type Changes

Below are my recommended instance type changes based on reviewing the CPU and Memory Utilization of each of the last 2 weeks.

• t2.small ($0.23/hr) -> t3a.micro ($0.0094 USD per Hour)
• t2.micro (0.0116) -> t3a.micro (0.0094)
• t1.micro (0.02) - > t3a.micro (0.0094)
• apps1 m3.large (0.133) -> m5a.large (0.086) (Note: No incl. storage on new instance)
• demo m3.large (0.133) -> t3a.medium (0.0376) (Note: No incl. storage on new instance)
• proxy1 m1.large (0.175) -> m3.medium (0.067)

In some cases the new instance types don't come with any included HDD storage, whereas the current instance type did. I believe that in at least some cases we use the included storage for system logs, but Jeremy can confirm and also make the necessary changes to not longer require the secondary / included storage.

In addition, note that on the demo instance we no are longer required to run the pdf servlet demo.

Assigned to Jack to inventory what systems are running on each instance.

Ken, what is meant by the first three bullet points?

The mentioned t1.micro, m1.large and m3.large Instance Types are all previous generation Instance Types, meaning they are PV and not HVM. After chatting with AWS Support I've learned that PV based Instance Types cannot be upgraded to HVM based Instance Types. All current Instance Types are HVM based. This doesn't mean we can't create PV based Instance Types anymore as you can still choose one of the previous generation Instance Types, but I'm not sure if that helps us.

I think we should consider switching to HVM based Instance. Not covered by this JIRA, but this should not concern our Voyent Alert! Clusters as they are already running on t2 HVM based Instance Types. So sticking to apps1, demo1 and proxy1 for now on our "old" production account. Any attached custom Volumes should be a breeze to "port" over as we just need to re-attach them to a new replacement Instance. The problem lies with the standard provided 8 GB root Volumes. These Volumes cannot be reattached to a new HVM based Instance as the Linux kernel installed on it will not run. This means that whatever we have installed on the root Volume of the mentioned Instances we need to reinstall on the respective replacement Instances.

I feel that demo1 and especially proxy1 are probably fairly easy to recreate, as proxy1 just runs Apache HTTP Server I believe, and demo1 just has Java installed on the root Volume (all the Tomcat servers are installed on the added custom Volume). The apps1 Instance is likely harder to recreate. Maybe Jeremy can make an assessment of what needs to be done to recreate the apps1 Instance if we decide to do so.

Evidently, there will be some downtime when switching over from one to the other.

Ken, what is meant by the first three bullet points?

I'm suggesting that we swap out all our matching existing instances for the new types listed, if feasible.

For both the ken.fyten@icesoft.com and engineering.support@icesoft.com accounts I did a fair amount of clean up already (not necessarily related to this case, but still...):

• Terminated Instances that are no longer required (most of them were marked "To Be Decommissioned") for both accounts:
  • Some stopped Instances on ken.fyten@icesoft.com are still there. Most of these can likely be terminated as well, but I didn't want to pull the trigger just yet. These Instances are:
    • apps1 (Broken)
    • apps1 (TEMP)
    • blog1.icesoft.private.domain
    • mail1 - To Be Decommissioned
    • proxy1 (2012-11-23)
    • proxy1.voyent.public.domain (To Be Decommissioned? 52.34.1.165)
  • Quite some stopped Instances on engineering.support@icesoft.com are still there as well. Mostly stopped Instances for QA. However, these can use another look.
• Deleted all Volumes that were not "in-use" anymore
• Deleted low-hanging fruit Security Groups that were clearly not needed anymore
• Deleted VPC and associated resources like Subnets and such
• Demo and Production environments still run on 2 Instances
• Latest, Latest1, Scaling1, Staging1 and Staging2 are now running on a single Instance

I'm currently looking into Snapshots that can be removed...