Meta is moving to the Cloud 🌩

I am not sure at all I am following any of your conclusion here at all.

3 Likes

Maybe, I miss something or just don’t understand something!? Based on the Rafael’s answer just though that having an owned dedicated cloud setup, intends of using leased infrastructure is more cost-effective in a long run.

Regardless your “hosting” costs are going to be a rounding error compared to the cost of hiring staff to watch over the software etc.

If you are a hobbyist and do not care about SLAs and just want cheap Discourse now. Go with Digital Ocean, you will be happy it will be cheap.

If you need 24x7 monitoring and SLAs and need to know this thing will always be running no matter what. AWS can work and bare metal can work. Buying hardware and colocating a rack will be cheaper and faster than most of your AWS options out there, building the infrastructure required to run this with this kind of SLA will take you years.

6 Likes

Thank you Sam :slight_smile: In terms of SLA, well you can get a good one with many reliable Cloud and/or IT infrastructure providers. ASW are definitely not the best and definitely not the cheapest of all enterprise-grade, reliable IaaS solutions available on the market. I’m not saying this to argue on anything, it is just a conversation. May I suggest that having a any High Availability or even business continuity and Fault-Tolerance (for the processing operations) Cloud infrastructure based on VMware would be comparable in terms of cost (probably cheaper) than the monthly service fee paid to any major cloud. The colocation cost will be 40% percent of the monthly bill and the cost of the hardware depreciate over 4 to 5 years added to the bill, meant that the overall Infrastructure expenses would be … 30% - 40% in a four of five year period. Please note that I’m not just discussing your setup or suggestion what you guys should do :slight_smile: not at all. This is particular post is about to say that owned infrastructure, not matter whether it is a bare-metal or not decreases the TCO, compared to the leased one when it comets 3 - 5 year period or time. Through the years I have been working as IT infrastructure consultant for a number of companies who cannot afford interruption of services, including gaming and gambling providers. None of them is on the major clouds. All of them have owned dedicated clouds, due to Sea reasons and reliability concerns.

Thank you for your time Sam!

My point here which should not be missed is that SLA is not just about hardware being reliable. In fact, achieve a great SLA you have to design your architecture in such a way that a single machine can fail and the party can keep going.

To have that kind of design and to maintain it and to keep upgrades going you need people. And people cost a lot more than servers.

8 Likes

I agree about the SLA. Have you tested the VMware’s Automated Fault-Tolerance. It is a service that creates a mirrored, failover virtual machine (VM) for each and every server installation. It is fully synchronized with the main VM. In case of failure of the OS or the physical host, the mirror VM takes on, without any service interuption. This is a 100% infrastructure uptime solution. It is so bulletproof that Amazon themselves signed with VMware to use their technology on AWS, as they lacked their open technology that offers operational continuity.

I agree that management effort (human resources) worths more than the technology. However, with the VMware Fault-Tolerance, one does not need commit more than 1-2 hours of infrastructure management per month. Therefore, one can get is as a Managed Dedicated Cloud service contract. Any other system administration would be on an applications level, something you guys do anyway!

1 Like

FT only solves one problem: host SW/HW failure

If the VM itself fails due to a software problem, it’s still down. You still want to have multiple application servers to tolerate failure and do a seamless application upgrade (so FT has no real benefit for the app servers).

You also need redis/PG failover and FT would help there for the primary.

8 Likes