So an IT engineer insisted that he has servers that comes with dual processors and CPU socket, RAMs can be hot-plugged/swapped just like HDD without rebooting server or even crashing OS.
I’m not aware of such technology nor I could find in google.
I didn’t ask him about server name, brand, CPU model because well he is older than me.
What I found so far is 1 node with multi blade servers that can be set in HA cluster. and that still a software based redundancy .
They do indeed exists at a hefty premium that is. Your “average joe” data center isn’t going to stock them but if your running something like Paypal, hospital and other “mission critical” that you need to keep the internet on the move. They can be purchased/leased at absolute premium prices. So yea they won’t be in the bargain rack.
I believe E7 processors are the ones that support this (the oldest and slowest “cycling” of hardware due to these uses). With the right chassis, boards, etc.
I had some intel itanium database servers in the late 90s that supported hot swap everything. Procs, memory the works. So they do exist. These were HP servers. Itaniums were a funny processor though and I think it was part of the instruction set.
On friendlys point the providers like PayPal just have full redundant servers, it’s cheaper to pack racks full of identical servers and remove a rack once a certain % fail rather than have someone replace parts. There is an open source open compute framework that the big providers use
I can see how that use case would be cheaper indeed. Why pay $xx,xxx+ for a single server when you can get two servers for as low as $5k for two?
The only reason why I would justifies paying the premiums would if that’s the safest way to keep the apps running for very sensitive operations. I am sure there applications that cannot stand even ms of downtime so clustering wouldn’t work for them.
In situations like that the applications I’ve used use like a reverse virtualization. I think everrun was one we used for our fire list. It ran a single virtual machine across two physical servers with ms replication. It was kinda cool. You could add a 3rd server as a quorum too…
We ran something similar when we ran “red cord” systems. We would be fined if the system was unavailable for more than 30 seconds in a rolling month.
A system that won’t tolerate a ms of downtime wouldn’t benefit from being run on a single server as you’d still need someone to swap the parts and like most things if one element fails it puts pressure on the other components.
The reason we had them was because servers were still really expensive and so was licensing so a single server with bits you could swap out hot.
Regarding legacy systems they probably wouldn’t run on the hardware and you’d be better virtualizing or reverse virtualizating as above
-swap RAM