Assuming I'm reading this right, that's<p><pre><code> 40x i3.metal
70x db.r5.24xlarge
300x m5.2xlarge
</code></pre>
Or, assuming my napkin math is right, $900k/mo at list prices, not counting bandwidth costs, things like S3 and EBS.<p>No one at this scale pays list prices, but still, ouch.
Saw this, I was wondering what they mean by PSQL Cluster where they need 70(!) instances. Is this just replication or also sharding of some kind?<p>Also, everyone always talks about instance cost first, but consider this: 100-120gb Internet egress per minute comes out at another $200k-$300k per month.
Wow, those are some hefty server requirements. Those resources should be able to support a DAU in atleast millions - if not 10s of millions.<p>What was parler's DAU at their peak?
While most hosting providers won't touch them for political and/or moral reasons, it does seem like if they had the capital for it, owning their hardware and colocating it would be much cheaper long-term (assuming there is a long-term for their platform).<p>Also, it seems like an obscene amount of RAM is required for such a platform, and I'd assume for most hot content that RAM would be relevant, but given their scale I think they could get by with a lot less.
What’s all this hardware for? Storing and serving tiny tweets can’t need this?<p>Is there something non-obvious in their service that could use a huge fraction of the hardware?
Media transcoding?<p>If I were them I’d try to get back up as soon as possible with 10% that hardware even if it meant e.g no access outside the US or no media transcoding at all or whatever MVP version they can create.
That's ... not that much. ~6/7 racks of servers.<p>Any good sized centrally located DC would be able to meet those requirements, but you'd need to drop a couple mill up front for the servers.