Has anyone migrated NTA to use Google Cloud SQL?

Question

I am interested in migrating our Solarwinds system to Google Cloud Platform.  We currently have NPM and NTA.  Has anyone succesfully migrated their system to GCP? If so, are you running instances of Windows Server 2016 for the databases? Is it possible to use Cloud SQL? I know it is technically not supported yet but I want to know how some of you migrated to GCP and what problems you ran into. Thanks for any information.

mferrucci · Answer

Thank you for sharing that info! We are going to try to migrate the database to Cloud SQL. Fingers crossed! I will update this thread with anything I found to be potential issues once we try. I am fairly confident it will work but with it being unsupported, I cannot be certain.

mesverrum · Answer

I run my environment in GCP, it's been there for about 2 years.  About 9000 nodes right now.  At the time we made the move Cloud SQL wasn't able to do MS SQL, it was only doing the more linux native styles so we stood up a instance of Server 2016 and installed SQL.  Also hosts a couple other databases related to our team but Orion is the biggest load by far.  I wasn't here while they did the actual migration itself but I am in that server every day and it is functionally exactly the same as if I had it in a local DC except I don't have to argue with any storage admins about insisting they can't give me high performance disks.  We run all our important SQL drives on what adds up to about 2 TB of SSDs and performance is always solid on that end.  Read and Write latencies are consistently below 2ms except during our nightly maintenance, but thats not exactly a surprise since that's the point where we hammer the server hardest.

Between SolarWinds and other tools we manage, 37 compute instances and a few load balancers and such, our entire GCP project costs a bit under $10k per month, but I know there is a lot of room for optimizing that.  Initially the company just wanted to lift and shift and ignored costs, now they are looking where we can trim and I expect that when I'm done I'll have my team down in the $5-6k monthly range.

As an example, the team who spec'ed out my SQL instance really over did it on RAM, 16 cpu 624 GB. They just maxed out the sliders for the type of instance.  I did some testing over the last few weeks to confirm and we actually would get almost exactly the same performance as we have now with the server down around 150gb, so I'm making that change this week and that change alone reduces the bill by over $1500 a month.  Also looking at using stacking poller licenses to cut us down from the current set of 9 compute instances running APE's to 3 with stacks since the marginal cost of adding more resources to a single instance is more cost effective than running many smaller separate windows servers.  That change will reduce the bill by another $2500 or so each month.