Switching backup tools starts with the first backup of every account. It’s the heaviest night, and the admin wants to know what it costs the server and the storage. On the same accounts, with the same rules, how long do Salvo and JetBackup take for the first backup, and what does each leave on the storage?
Why it matters to whoever runs the server
- 45% less time (5 min 41 s versus 10 min 18 s): the first night ends sooner.
- 166× fewer files on the storage, with 13% less space and half the write latency. Fewer files means less work for the storage tonight and every night after (Lab 15).
- Where Salvo loses: more CPU, load and memory on the source server (1.3 GB at peak, versus 50 MB). It runs at the lowest priority so the sites go first, but the admin needs to know.
| JetBackup | Salvo | |
|---|---|---|
| Total time | 10 min 18 s | 5 min 41 s |
| Space used on the storage | 47.6 GiB | 41.2 GiB |
| Files and folders created on the storage | 701,685 | 4,217 |
| Average write speed | 82.6 MiB/s | 123.4 MiB/s |
| Average write latency on the storage | 90 ms | 42 ms |
| Operations per second on the storage (average) | 489 | 359 |
| Storage CPU (average) | 16% | 12.7% |
| Sent over the network | 50.2 GiB | 41.4 GiB |
| Peak load on the source server | 14.2 | 23.0 |
| CPU used on the source server | 69% | 87% |
| Peak tool memory | 50 MB | 1.3 GB |
Method
Source server: production cPanel, 8 cores, sites active, during business hours. 10 accounts, about 55 GB and 660 thousand files; the largest has about 29 GB and 270 thousand files.
Storage: the same backup server for both (ZFS, lz4 compression, one virtual disk). Each tool wrote to its own empty volume, created for the test, and each one’s space and disk operations were measured separately, every 5 seconds.
Same rules: the same 10 accounts, 3 accounts at a time and an SSH destination on both. One at a time, with a 1-minute pause between them to measure background noise. JetBackup 5.4.1.4 with an incremental job and SSH destination; Salvo with zstd compression at the fastest level, encryption, lowest CPU and disk priority and a 3 GB memory cap.
Why the difference: JetBackup recreates each account’s whole tree on the storage, one file for every site file; Salvo packs everything into a few large packs. Every file on the storage is future work: checking, hardlinking, deleting at retention. Salvo also compresses before sending; the storage’s ZFS gains nothing on top of that (1.00× ratio), while on JetBackup ZFS compressed 1.18× and the backup was still larger.
Where Salvo loses
CPU and load on the source server: compression is the costliest step (Lab 07). With 3 large accounts at once, load reached 23 on an 8-core server, versus 14 on JetBackup. Salvo runs at the lowest priority, so the sites go first, but the load is real. And memory: 1.3 GB at peak, versus 50 MB for JetBackup.