Reference
Why did my backup read the full disk but upload very little?
A full read does not mean a full upload: blocks the archive already holds are skipped at the source.
The Short Answer
When no changed-block information is available for a backup, EBA-ST reads the full disk (skipping unallocated regions where the platform exposes them). But before uploading, it checks which blocks the EBA repository already holds and skips sending those - only lightweight references are recorded for them.
So a full read commonly produces a small upload: only data the archive has never seen travels to SHA.
Reducing The Reads Too
On VMware, changed-block tracking means later backups read only changed blocks. On supported CloudStack KVM backends, snagent extent maps do the same. Where neither applies, the read stays full but the upload stays deduplicated.
Related Docs