Quantum has sped up part-object retrieval from its ActiveScale tape backend with a Ranged Restore feature.
It says customers can now restore only the specific byte ranges they need from large objects, rather than rehydrating entire files, significantly reducing retrieval times. More ActiveScale software updates also deliver substantial performance gains when reading small objects from the cold or tape tier, enabling more than five times faster restores. With these updates, Quantum says, tape-based archives can now operate as responsive, query-ready data lakes designed for AI, analytics, and high-performance computing workloads.
CEO Hugues Meyrath said: “AI changes the value of archive data. Organizations need to store large volumes of data affordably, but they can’t afford to wait hours or days to retrieve the data that feeds training pipelines and analytic engines. With these new ActiveScale enhancements, we’re redefining what cold storage means, making massive, long-term archives readily accessible, intelligent, and primed for AI at exabyte scale.”
It’s certainly an advance in object retrieval speed from tape, though it doesn’t bring tape retrieval up to disk speed, as a look at ActiveScale’s architecture shows.
ActiveScale is a two-tier, S3-compliant, object storage system, with an SSD+disk front end for warm or active-class storage, accessed with S3 Standard class APIs, and a tape library back end for cold data storage. This is accessed using S3 Glacier-class API calls.
Metadata is stored in NVMe SSDs for fast access. Objects are erasure-coded, with up to 19 nines (99.99999999999999999%) durability, and ActiveScale systems can be grouped in a RAID-like way called RAIL, or Redundant Array of Independent Libraries.
Incoming S3 Standard class objects are written to disk or SSD, and split into erasure-encoded fragments or shards. Incoming smaller objects are grouped together and the group is then sharded, with each shard possibly containing several objects. ActiveScale metadata tracks this.
Writing objects to tape involves grouping them into so-called larger blobs, hundreds of GB in size. This enables large streaming writes, minimizes the work the tape robot needs to do loading and unloading tapes, and also reduces tape wear.
If policies specify cold storage, or an S3 Glacier API is used to put an object into the ActiveScale system, then, instead of writing individual objects to tape, a collection of them is streamed to tape to improve tape writing efficiency.
Restoring objects from disk or SSD requires shard identification and reading followed by object reconstruction, unless it’s a small object and contained in a single shard.
Restoring objects from tape means locating and reading the blob from tape and writing it to disk or SSD, and then reconstructing the object from its shards. This may well initially involve mounting a tape from its library shelf and then streaming through the tape to find the BLOB. Object restoration from the tape library will typically take several minutes.
Quantum says traditional Amazon S3 Glacier operations require full-object retrieval, often hundreds of gigabytes at a time. The ActiveScale Ranged Restore feature “allows selective restore of just the required data segments. This capability significantly reduces wait times, compute cycles, and egress impact, making archives more usable and operationally efficient.”
The tape storing the required blob holding the Ranged Restore bytes could be sitting on a library shelf, and need fetching and mounting in a drive by a robot. It will take 30 to 90 seconds or so for this, and then potentially minutes for the tape streaming to the blob’s location. It’s finding the object location on tape that takes the longer time.
Quantum says it’s the only provider currently offering this custom S3 Glacier Ranged Restore extension for tape-based cold storage.
The small object restore performance boost is enabled by intelligent batching and ordering of restore requests, significantly accelerating high-volume retrieval workflows such as AI training. Quantum says: “This enhancement is particularly impactful for AI model training, data validation jobs, compliance lookups, and automated pipeline-driven restores.”
Quantum chief product officer Geoff Barrall said: “Cold data is no longer offline data. By eliminating the legacy limits of Glacier-class archives, ActiveScale turns tape into an active asset – fast, API-accessible, and ready for AI and analytics at scale.”
These new enhancements are available now. More information here.
Bootnotes
- We understand that finding an object or file at the end of a full LTO-9 tape involves winding the tape to the correct position, which takes approximately 8 to 10 minutes for a full tape.
- Spectra Logic has a partial object restore feature with its BlackPearl disk-based Deep Storage Gateway to its Spectra tape libraries. This allows a user to request a piece of an object rather than the whole object.
