The snapshot is large. It's ~8TB. The circumstances that produces this large snapshot are unique (1st time replication of a 20TB vm was interrupted part way through and had to be resumed. The way the backup software handled it apparently created a snapshot when it replicated the remain data and then chose to merge that data when the replication was complete). That aside, I did not realize that this operation was random I/O, which would explain the slow throughput. I don't understand why it would be random I/O or even why the deleting/merging process requires all data to be read and re-written, but that's neither here nor there. The VM is powered off, I'm not sure how much difference this makes. Also, this is the only VM producing I/O on this host, so I can confirm that all these disk requests do indeed come from this snapshot removal.
↧