The Globus alternative for AI and petascale.
Globus is a familiar way to orchestrate and share research data across institutions. But it is grid-era: it manages transfers over GridFTP and still depends on your DTNs being well tuned, so at AI and petabyte scale your throughput is only as good as that tuning. Zettar co-designs the whole data path to hit line rate, turnkey.
Orchestration is not the same as line rate.
Globus schedules and manages the transfer, but the speed you actually get still comes down to your endpoints — the data-transfer nodes, storage, and tuning underneath. Its grid-era design was built for moderate research flows, not for feeding GPU clusters or moving petabytes on deadline, so it often leaves much of a fast link idle. Making it fast becomes your tuning project, on top of the service.
Globus vs the Zettar zx Appliance.
| Globus | Zettar zx | |
|---|---|---|
| Architecture | Grid-era orchestration over GridFTP | Co-designed: storage, host, network & software |
| Realized bandwidth | Depends on your DTNs and tuning (ESnet measured zx at 2×–7× GridFTP) | ~90%+ — line rate, pretuned |
| What you operate | A managed service plus DTNs you tune | One turnkey appliance, ready on day one |
| Petascale at line rate | No published petascale line-rate result | 1 PB in 29 hours at 96% utilization |
| Data path & control | Orchestrated through a cloud control plane | On hardware you control, on-prem and in-country |
| Workloads | File transfer & sharing | File, object, and live streaming in one stack |
| Tuning | Yours to do and maintain | Co-designed and shipped pretuned |
Globus is good at moving and sharing research data between institutions. When the goal is line rate at AI and petabyte scale — with nothing to tune — that is what the co-designed appliance is for.
Co-designed for line rate, not just orchestration.
The Zettar zx Appliance co-engineers storage, host, network, and software with an NVMe burst buffer to run at line rate — roughly 10× typical movers, with no software ceiling — and ships pretuned on proven enterprise hardware. With SLAC and the U.S. DOE it moved 1 PB in 29 hours at 96% link utilization over a 5,000-mile path, and Taiwan’s NCHC has deployed it. Winner of the Supercomputing Asia 2019 Data Mover Challenge.
Compare Zettar with: rsync · Globus · Aspera / Signiant · Read: rsync, Globus, or an appliance?
Common questions
Is Zettar a replacement for Globus?
For high-speed movement at AI and petabyte scale, yes. Globus is good at orchestrating and sharing research data, but it is grid-era and its speed depends on your DTNs being well tuned. The zx Appliance co-designs the whole path and ships pretuned to run at line rate.
Does Globus not already move data fast?
Globus manages the transfer; the throughput still comes from your endpoints and tuning underneath, so at AI scale much of a fast link can go unused. zx is co-designed across storage, host, network, and software to run at line rate (about 90%+ of available bandwidth), proven at 1 PB in 29 hours.
Can Zettar work with our existing research setup?
Yes. It ships as a pretuned appliance on HPE ProLiant enterprise servers and runs on-premises and in-country, moving file, object, and streaming data. It can run alongside your existing tools, including Globus.
Line rate, nothing to tune.
Bring a transfer your DTNs struggle with — we’ll show you what a co-designed appliance does, measured on your data.
