The fastest, petascale-proven, turnkey data mover — 10× typical commercial and freeware tools, wire speed the day it's racked. We don't sell AI. We move the data that powers it.
Zettar moves data at the full speed your network and storage allow — using nearly all the bandwidth you have, sustained from first byte to last, not a peak.
Moving a petabyte for the U.S. DOE, it was capped at 80 Gbps to spare the shared network and still finished in 29 hours — on a full 100 Gbps link, that's a petabyte a day. Your infrastructure sets the ceiling; Zettar reaches it.
Proven at the largest scale, not benchmarked in a lab.
With SLAC, ESnet, and the U.S. DOE, Zettar moved an actual petabyte over a 5,000-mile 100 Gbps loop in 29 hours at 96% bandwidth utilization — encrypted and checksummed. Winner, Supercomputing Asia 2019 Data Mover Challenge. Read the record → See the paper → DMC release →
No integration project. No tuning team. Record-setting throughput the day it's racked.
Pretuned and system-engineered — every component selected and matched to zx. The proven core appliance is the HPE ProLiant DL380 Gen 11.
Lower spend, faster teams, freed-up engineers.
Open, standard enterprise hardware with no proprietary lock-in — and none of the cost of a months-long DIY transfer project.
Once a fast link is in place, throughput is set by storage I/O, the host, and the software — not the wire. Zettar co-designs all three, so a fast link actually delivers.
A fast link only pays off when everything behind it keeps up.
The data mover, tuned Linux host, and hardware are engineered together — no DIY tuning. CUBIC default congestion control is sufficient; packet loss stays negligible on R&E backbones.
A single stream can't fill a high-speed link — not even close.
zx moves data with many concurrent streams across cores and nodes at once, so the link runs full instead of half-empty — and throughput scales out as you add nodes.
Slow storage can't throttle a fast transfer.
A high-speed NVMe burst buffer stages data between production storage and the WAN, keeping the link fed end to end.
Move data across the building or across the planet at near line rate.
Validated across 10/50/100 ms paths, with real transfers spanning 5,000 and 12,375 miles.
Speed only counts when every byte arrives intact, the pipeline keeps running, and it slots into the systems you already operate.
Trust what landed — no silent corruption, no clear-text exposure.
Unconditional end-to-end checksums plus TLS encryption verify and protect every transfer.
Pipelines that survive failures and busy windows.
Built-in high availability, load balancing, and scheduling keep data flowing around the clock.
Make movement a service your platform calls, not a manual step.
RESTful APIs and an SDK let you script and embed zx directly into your workflows.
Skip the months-long DIY tuning project.
One system-engineered appliance replaces a custom build across firewalls and heterogeneous hardware.
| Capability | Zettar zx | rsync / scp | Globus | Aspera / Signiant | DIY build |
|---|---|---|---|---|---|
| Wire speed at petascale | ✓ | — | Partial | Partial | Partial |
| Linear scale-out across nodes | ✓ | — | Partial | — | Partial |
| File and object in one tool | ✓ | — | Partial | — | — |
| Live (append) streaming | ✓ | — | — | — | — |
| Turnkey, pretuned appliance | ✓ | — | — | — | — |
| End-to-end checksums + TLS | ✓ | Partial | ✓ | ✓ | Partial |
| No proprietary lock-in | ✓ | ✓ | Partial | — | ✓ |
| Open-hardware economics, no lock-in | ✓ | Partial | Partial | — | — |
Those tools were not built for petascale. zx is roughly 10x faster, unifies file, object, and streaming in one stack, and is petascale-proven and turnkey — one validated appliance instead of a months-long DIY tuning project or a costly box that's just software pre-loaded on hardware nobody co-designed.
Once a fast link is in place, the bottleneck is no longer the network — it is storage I/O, host architecture, and software design. zx is co-designed across the whole path, using NVMe burst buffers for staging and host-level optimization to reach line rate where general-purpose tools stall at a fraction of it.
It is proven in production, not a lab demo: 1 PB moved in 29 hours at 96% link utilization with SLAC (U.S. DOE) and ESnet, and winner of the Supercomputing Asia 2019 Data Mover Challenge. For more, see Fang et al., arXiv:2512.15028 (CC BY 4.0).
No. The appliance is platform-neutral: HPE is the proven flagship today, AIC and MiTAC builds are engineered, and zx can be qualified on the enterprise platform you standardize on — open, standard hardware with no proprietary lock-in.
Lower total cost of ownership: no months-long tuning project to staff, open hardware instead of a closed, proprietary stack, and far less ongoing engineering time — your team runs it without becoming data-transfer specialists. And it delivers results a DIY stack simply can't match.
Fastest, petascale-proven, turnkey, and co-designed — built on open, standard hardware that scales out across nodes. Let us show you wire-speed data movement on your workload.