Theme
Storage Center
A compute centre answers "how fast can we compute"; a storage centre answers "can we feed it, keep it and move it".
The Jumborca Storage Center is built on the JDC AI storage foundation and delivers petabyte-scale storage as a service: customers subscribe to capacity per TB per month, buy no hardware and hire no team, while Jumborca runs the racks, power, the links to compute nodes and day-to-day operations. SupInsight schedules and observes storage and compute together, JDC-Vault keeps a physically isolated vault for core data, and SupDRC with SupDrill keep the service within its SLA through failures and drills. The centre can sit in a customer-designated facility, a Jumborca partner data centre, or be built as the storage module of a compute centre.
Service forms, the three-layer architecture, nodes and certifications are on the service page.Open Storage Power Center
What customers face
Stated in the language of the industry, with no product assumed.
- After compute expands, storage becomes the bottleneck: GPUs wait for data, training throughput is held back by I/O, and every storage expansion restarts a procurement cycle.
- Multimodal training data is huge and mixed hot and cold: images, video, logs and checkpoints share the same tier, hot data occupies fast media and cold data has no cheap home.
- Storage hardware has long lead times and large one-off spend: from approval to rack is measured in quarters, while capacity demand changes monthly.
- Separating compute and storage leaves two operations teams: each watches its own monitoring and incidents are reconciled across both.
Solution architecture
- 01
Ingest and multi-protocol access
Training clusters, inference services and pipelines connect over file, object and block protocols; mounts stay the same after expansion.
- 02
JDC storage foundation · hot / warm / cold
JDC holds datasets, corpora and checkpoints, moving them across three media tiers by access frequency; capacity is subscribed per TB per month.
- 03
Storage and compute scheduling and observability
SupInsight puts capacity, bandwidth and compute-node state in one view; capacity alerts and anomalies surface on the board first.
- 04
Protection · isolated vault and disaster recovery
JDC-Vault keeps a physically isolated golden copy; SupDRC fails the storage service over by runbook and SupDrill verifies it in regular drills.
Core capabilities
One product per block; open the product page for detail.
- JDC AI storage foundation High-throughput storage for training and inference with hot, warm and cold tiers, multi-protocol access and capacity subscribed per TB per month; expansion does not change how applications mount it.
- SupInsight Storage and compute observability Capacity, bandwidth and compute-node state in one view; anomalies surface on the board first and are handled by the operations team.
- JDC-Vault Isolated vault Physically isolated copies of core datasets and model checkpoints, out of reach even when the production network is compromised.
- SupDRC Disaster recovery and failover The storage service itself has a recovery plan; node or site failures fail over by runbook, verified regularly with SupDrill.
Typical rollout path
Four steps, each with an output that can be checked.
- 01
Assess
Inventory datasets, checkpoints and access patterns, set the hot/warm/cold ratio and growth curve, and produce a capacity and bandwidth requirement.
- 02
Implement and migrate
Rack JDC nodes in the designated facility or partner data centre, connect them to compute nodes, migrate datasets in batches and verify checksums.
- 03
Drill and verify
Run one node failover and one isolated-copy restore with SupDrill, keep the auditable record, then admit training workloads.
- 04
Operate
The Jumborca operations team handles inspection, expansion and capacity alerts; the customer sees the same data on the SupInsight board.
Delivery options
- Capacity subscription
- Subscribed per TB per month, fixed-price contract optional; hardware, racks, power and operations are included.
- Delivery in a customer-designated facility
- The centre can sit in the customer's own facility or a designated data centre; data stays inside the boundary the customer draws.
- Built alongside a compute centre
- Delivered as the storage module of a compute centre, with the storage-to-compute links planned by Jumborca.
Bring your system inventory and recovery requirements; we start with a reference architecture, then talk pricing.