When trying to bootstrap a node in a VM on Google Cloud Platform I see the performance running fine for approximately 30 minutes before it completely tails off.
You can see this from the graphs below which show the CPU + IO for the past day on the instance. You can clearly see the point at which I start the tezos-node as everything spikes up. CPU goes up to 20% - about 80% of a single CPU (I guess Tezos doesn't make use of multiple CPUs) and the IOPS steadliy climbs up, peaking at about 3,500 IOPS or 16MB/S.
However after 30 minutes everything just tanks to the point where it takes several seconds for each block to validate. Not sure what's going on here.
This is a n1-highmem-4 instance (4 vCPUs, 26 GB memory), although I have seen the same behaviour on various other instance type in GCP. And can be reproduced very easily on any instance type I have tried.
This one has been running for about 17 hours now and only got to block BL6SwDDDhqHkQEpfGWVBFBZ6LX1Byg4eibHSEvQbmS3bL7Z8b89 (3-Aug)
Update
After the comment about disk size I provisioned a bigger SSD (200GB) and this appears to resolve the problem. Will do some research to see what the minimum disk size is to achieve the consistent 3,500 IOPS that Tezos seems to need.
df -i