Цитата:
1) One disk failure - *25% CPU host load*
2) No ssd/disk *failure admin notification at all*
3) 700 VM on 8 nodes (Nutanix easily handles much more on very similar hardware) - disk access latency is *1 second*
4) NVMe cache - when its full (but plenty of SSD space available) - the performance decrease is dramatical, IOPS are *just 15000 per node*.