Resources · Glossary
Data-movement glossary.
Plain-English definitions of the data-movement, data-transfer, and high-speed networking terms enterprise and research teams search for most.
- AI factory
- A datacenter built to manufacture intelligence — GPUs, high-speed fabric, and storage producing trained models and tokens at scale. Its output is gated by how fast training data arrives: idle accelerators still bill by the hour, so a co-designed data-movement appliance that stages petabytes at line rate keeps the factory producing. See AI & Neocloud.
- Air-gapped transfer
- Moving data within disconnected or classified networks with no external connectivity. Zettar runs on-premises with no call-home dependency, so it can operate in air-gapped enclaves. See Aerospace & Defense.
- ALCOA+ / data integrity
- The data-integrity principles regulated life-sciences pipelines must uphold — Attributable, Legible, Contemporaneous, Original, Accurate, and more. Zettar's unconditional end-to-end checksums give provable integrity in transit. See Life Sciences.
- Append streaming
- Moving live, fast-growing data as it is produced (for example from high-speed instruments) instead of waiting for a complete dataset. See zx Streaming.
- Bandwidth-delay product (BDP)
- The amount of data in flight on a link: bandwidth × round-trip time. A 100 Gbps path with a 100 ms round trip holds about 1.25 GB in flight, so a single TCP stream needs a window that large to keep the link full — one reason single-stream tools stall on long, fast links. See the guide to transferring petabytes.
- Burst buffer
- A fast NVMe staging tier between production storage and the network that absorbs I/O bursts and decouples storage from the WAN — a key building block of high-speed data-movement architecture.
- Checkpoint / model weights
- The large state files AI training writes and reads — checkpoints and weights — that must move fast between storage, clusters, and regions to keep GPUs from idling. See AI & Datacenters and how to keep GPUs fed.
- Checksum / end-to-end integrity
- Verifying that every byte arrives exactly as it was sent. Zettar applies unconditional end-to-end checksums on every transfer, so data is provably intact.
- Cloud-path penalty (cloud tax)
- The throughput lost when data moves in and out of public clouds through standard transfer paths and tools — measured at 30–50% (Fang et al.). It is separate from the provider's egress fees, which are set by cloud pricing: Zettar's co-designed path bypasses the throughput penalty; it does not change egress pricing. See Hybrid-Cloud Movement.
- Co-design (hardware-software co-design)
- Engineering the data mover, host, storage, and network together as one system instead of optimizing each in isolation. Co-design is why Zettar realizes throughput that general-purpose tools leave on the table — the network is rarely the only bottleneck.
- Congestion control
- The TCP algorithm (CUBIC, BBR, Reno) that governs how fast a sender transmits. On well-provisioned research and education networks the choice barely matters; Zettar performance is largely insensitive to it.
- Cryo-EM (cryo-electron microscopy)
- Imaging that produces multi-terabyte datasets per session, which must move to HPC and the cloud. Zettar moved a KEK cryo-EM dataset 5.9× faster than aws-cli. See Life Sciences.
- Data gravity
- The tendency of large datasets to attract applications and compute to wherever the data already lives. Fast data movement lets you put compute where it is cheapest instead of being pinned by data gravity.
- Data migration
- A one-time or phased move of data to new storage, a new system, or the cloud. With Zettar, a petabyte moves in about a day on a 100 Gbps link — days, not months. See Migration.
- Data movement
- The discipline of moving large volumes of data between storage, systems, sites, and clouds — efficiently, securely, and verifiably. At terabyte-to-petabyte scale, data movement (not compute) is often the real bottleneck. See the zx Appliance.
- Data mover
- Software or an appliance purpose-built to move data fast and reliably at scale. Zettar zx is a unified, scale-out data mover for file, object, and streaming.
- Data replication & sync
- Keeping copies of data consistent across sites or clouds for disaster recovery and collaboration, within tight RPO/RTO windows. See Replication & Sync.
- Data sovereignty
- Keeping data under a nation's or organization's own jurisdiction and control. Zettar runs on-premises, with no forced cloud, so sovereign-AI and regulated data stay in-country. See Sovereign AI.
- Data transfer
- Sending data from one location to another over a network. High-speed data transfer at scale needs the parallelism, tuning, and integrity checking that general-purpose tools (rsync, scp, FTP) were never built for.
- Data Transfer Node (DTN)
- A server dedicated to high-speed data transfer, common in research and HPC environments. Zettar turns DTNs into a turnkey, pretuned appliance. See Research & Academia.
- DICOM
- Digital Imaging and Communications in Medicine — the standard format and network protocol for medical images and their metadata. In a PACS or VNA migration, DICOM records are reconciled while the underlying image files are copied. See PACS and VNA migration at scale.
- DPU / IPU
- A data-processing (or infrastructure-processing) unit such as NVIDIA BlueField that offloads networking and storage from the host CPU. Zettar zx can run on DPUs for denser, more power-efficient data movement.
- File transfer vs. object transfer
- File transfer moves files and directories; object transfer moves objects in systems like Amazon S3. Zettar moves both — and between them, with no format wrappers. See zx File and zx Object.
- Full-motion video (FMV)
- Continuous video from ISR sensors, aircraft, and drones — high-volume streams that must move from collection to exploitation fast and intact. See Aerospace & Defense.
- Goodput / link utilization
- The share of a link's capacity actually used to deliver data. Zettar sustains high utilization — 96% in the SLAC/ESnet record — where general-purpose tools stall at a fraction of the link.
- GridFTP
- An extension of FTP for high-throughput, parallel, and striped transfers, long used across research grids and historically by Globus. See Zettar vs. Globus.
- Ground station / downlink
- Where satellite data is received from orbit. Downlinked imagery must move from globally distributed ground stations to cloud, archive, and processing. See Earth Observation.
- Hybrid-cloud / multi-cloud data movement
- Moving data between on-premises, hybrid, and multiple public clouds. Zettar bypasses the 30–50% cloud-stack penalty that slows standard cloud transfer paths. See Hybrid-Cloud Movement.
- Latency-insensitive
- Performance that holds up over long distances. Zettar has moved data over 5,000- and 12,375-mile paths with no throughput collapse — distance does not wreck the transfer.
- Line rate
- The full usable rate of a network link. Zettar runs at line rate — sustaining roughly 90%+ of whatever bandwidth your end-to-end path can deliver, even on a shared link — so the network, not the software, is the limit.
- Managed file transfer (MFT)
- Enterprise software for secure, governed file exchanges — workflows, scheduling, partner onboarding, and audit trails. It is built for governance rather than for moving petabytes of bulk data at line rate. See managed file transfer vs. high-speed data movement.
- Object storage (S3)
- Storage that manages data as objects through an API, such as Amazon S3. High-speed object data movement within and across clouds is a core zx capability. See zx Object.
- Offline data transfer (sneakernet)
- Moving data by shipping drives or transfer devices instead of sending it over a network. It can make sense with no usable link, but door-to-door time includes copying data on and off, transit, and verification. See ship drives or use the network?
- PACS / VNA
- Picture Archiving and Communication System / Vendor-Neutral Archive — medical-imaging stores holding petabytes of radiology and pathology studies that must move between sites, to the cloud, and into AI. See Healthcare and PACS and VNA migration at scale.
- Parallel file system
- High-performance storage (Lustre, BeeGFS, GPFS) that stripes data across many servers for throughput. BeeGFS was the storage used in Zettar's 1 PB record-setting transfer.
- Parallel streams
- Splitting a transfer across many TCP connections, files, or nodes at once, so throughput is not capped by a single stream. Parallelism only pays off when storage and hosts keep up too. See the guide to transferring petabytes.
- Petascale / petabyte (PB)
- A petabyte is 1,000 terabytes. Petascale data movement means reliably moving data at the petabyte scale — Zettar moved 1 PB in 29 hours with SLAC and ESnet.
- RDMA / zero-copy
- Techniques that move data with minimal CPU involvement (RDMA, kernel bypass) to reach high throughput. zx uses the host efficiently so the CPU is not the bottleneck.
- RPO / RTO
- Recovery Point Objective and Recovery Time Objective — how much data you can afford to lose and how fast you must recover. Wire-speed replication helps hit tight RPO/RTO windows at petascale. See Replication & Sync.
- S3 multipart transfer
- Splitting large objects into parts moved in parallel for S3-compatible object storage, so one big object never caps throughput — a core zx Object capability. See zx Object.
- Scale-out (symmetric, peer-to-peer)
- An architecture where every node both sends and receives and clusters self-organize, so aggregate throughput grows as you add nodes — with no software-imposed bandwidth ceiling. The network fabric, not the data mover, sets the limit.
- Science DMZ
- A network design pattern, popularized by ESnet, that puts dedicated data transfer nodes on a high-performance path at the network edge with targeted security controls, so large research flows are not throttled by general-purpose enterprise network equipment. See Data Transfer Node.
- Seismic data
- Subsurface survey data, often petascale, that moves from remote fields and vessels to processing centers and HPC. See Energy.
- Sequencer / NGS data
- Next-generation sequencers output terabytes to petabytes of genomic data per run that must move to analysis fast. See Life Sciences.
- Storage tiering
- Moving data between performance and cost tiers (for example file to object) transparently and in native formats, with no lock-in. See Transparent Tiering.
- Tape-out
- The final chip-design data handed from designers to a foundry — large files that move between design centers and fabs worldwide, where IP must stay encrypted and intact. See Semiconductor & EDA.
- Throughput
- The actual rate of data delivered. zx runs at line rate — filling the bandwidth your end-to-end path can sustain — and scales out across servers with no software ceiling.
- WAN optimization
- Appliances or software that speed up branch-office traffic with deduplication, caching, compression, and protocol acceleration. Effective for repetitive traffic; much less so for new, already-compressed, or encrypted bulk data. See why the network isn't the bottleneck.
- Wire speed
- Moving data at the full rated capacity of the underlying network link, with minimal overhead. Wire-speed transfer means the network, not the software, sets the limit.
Get started
Move your data at wire speed.
See the appliance behind the numbers — or schedule a demo.
