Hi,
What is the recommended IOPS for fast catchup and/or relay nodes?
I understand high performance SSD is required (Fast catchup for mainnet).
Thanks!
It depends a lot on the use case.
If you are running a non-participating non-relay node, then I think you may still get away with a not-too-old SATA SSD (I don’t have the IOPS out of my head).
The worse that may happen is that you are slightly lagging and you can then upgrade.
Concretely, currently, you can run an archival node + indexer for both MainNet and TestNet on GCP VM using a balanced persistent disk (that uses more than 3TB).
See
Desempenho do armazenamento em blocos | Documentação do Compute Engine | Google Cloud for GCP performance.
If you are running a participation node with high stake or a relay, then performance relay matters and you should definitely not do the above.
//Concretely, currently, you can run an archival node + indexer for both MainNet and TestNet on GCP VM using a balanced persistent disk (that uses more than 3TB).
IOPS with the above VM config is about 18,000 IOPS (= 6 x 3000). This is good info! In my case, indexer is not needed. I’m using AWS EC2 GP3 with the maxed-out IOPS setting, 16,000. I guess it should be enough. 
FYI, I observed fast catchup failure with IOPS setting of 3,144 - this always happened after I rebooted the VM. It looks like EC2 ubuntu VM experiences high IOPS (~2,000) for a while after reboot even without algod running. Then, with fast catchup, my IOPS was peaking at 3,400 exceeding the capacity…
Thanks Fabrice!