Active Archiving
Scope
Active archiving tests the design pattern where tape is online and searchable rather than shelf-bound: disk cache plus LTFS tape library plus a catalog, delivered against recall SLAs.
Core rules
- Three tiers, three jobs: backup is a point-in-time copy for recovery; deep archive is offline and slow by design; an active archive keeps data addressable and serves retrieval within minutes.
- Architecture: a disk cache holds the hot subset and metadata, an LTFS library holds the bulk, and a catalog indexes every file across every tape so locating data never requires a mount. HSM-style stub files or an object/S3 front end hide tape from users.
- Recall time is arithmetic: load and mount (tens of seconds) plus data ÷ drive throughput, divided across streams if files are striped over several drives.
- Scale comes from adding drives and slots, not from a bigger array. Because the format is open, media outlive the software vendor.
- Duplicate copies on separate media and sites, verify checksums on write, re-read a sample on a schedule, and refresh media every one to two generations to stay inside the read window.
Worked example
A 100 GB file is requested. Mount takes about 40 s; 100 GB at 400 MB/s is 250 s, so full recall is roughly five minutes end to end, with time-to-first-byte near one minute if streaming starts immediately. For a 15-minute SLA on 1 TB: 1,000,000 MB ÷ 400 MB/s = 2,500 s single-stream, so stripe across four or more drives rather than waiting for faster tape.
Traps
- Calling last night's backup an active archive - no catalog, no SLA, no retrieval test.
- Quoting recall times without the mount step.
- Designing every retrieval around a single drive.