Long data centre aisle between rows of server racks

Practice 02

Every dataset findable, protected and ready for AI.

Data management and tiered storage for organisations whose data has outgrown a single system: research institutes, AI teams, media libraries and public bodies. Built on Spectra Logic and Arcitecta Mediaflux.

Request a data assessment See the architecture

AI and GPU pipelinesResearch dataActive archiveAir-gapped copies

01The problem

More data than any one system should hold.

Primary storage is fast and expensive, so finished data piles up wherever there is room. It becomes hard to find, costly to keep and exposed to a single failure or attack.

Growth outpaces budget

Keeping everything on primary storage costs more every quarter.

Data nobody can find

Files without metadata are kept but never reused.

One copy is no copy

Ransomware and site failures need an off-line, verified copy.

AI needs it back, fast

Training and analysis need whole datasets staged on demand.

02Reference architecture

From source to GPU, by policy.

Mediaflux indexes every file and applies policies; Spectra Logic object storage and tape keep data at the lowest sensible cost; datasets are found by their metadata and staged back to performance storage when a job needs them.

01 Sources

Data sources

Instruments, sequencers, cameras, simulations, sensors and business applications.

NFS / SMB / S3

02 Manage

Metadata and data management

Every file indexed with its metadata; policies decide copies, tier and retention.

Arcitecta Mediaflux

03 Store

Storage tiers

FlashperformanceDisknearlineObjectBlackPearlTapeSpectra library

04 Retrieve

Retrieval and staging

Find datasets by metadata; stage them back to the performance tier automatically.

Policy-driven recall

05 Compute

AI and GPU workloads

Training, inference pipelines, analytics and HPC on complete datasets.

GPU clusters / HPC

Simplified. Designs are sized to data volume, growth, access patterns and retention rules.

03Storage tiers

Each kind of data on the right medium.

Moving data between tiers is automatic and reversible. Users and applications see one namespace; the policy decides where each file lives.

Relative comparison for planning. Access times and costs depend on the specific design.
Tier	Medium	Typical data	Access	Cost per TB
Performance	NVMe flash	Active training sets, scratch, current projects	Immediate	Highest
Nearline	Disk	Recent projects, frequently reused data	Immediate	High
Object	Spectra BlackPearl, S3	Completed datasets shared across teams and sites	Seconds	Moderate
Archive	Spectra tape libraries, LTO	Archives, compliance copies, the off-line copy	Minutes	Lowest

04Use cases

Where it pays back first.

The architecture is the same; the policies, metadata and sizing change with the work.

AI and machine learning data

Keep every training set and model version; stage the ones in use onto GPU storage.

Scientific research and HPC

Instrument and simulation output catalogued by project, kept for re-analysis.

Life sciences and genomics

Sequencer data tiered after processing, with provenance kept in metadata.

Media libraries

Decades of content kept on tape and searchable. Broadcast workflows

Government and public records

Retention rules enforced by policy, with audit trails.

Video surveillance and evidence

Long retention at low cost, with chain-of-custody metadata.

Ransomware recovery

An air-gapped tape copy that malware cannot reach or encrypt.

Multi-site and cloud

Data placed across sites and clouds by policy, under one namespace.

05Technology

Spectra Logic and Arcitecta Mediaflux.

Spectra Logic tape libraries and BlackPearl object storage hold data for decades at low cost. Mediaflux gives that data a searchable catalogue and moves it between tiers by policy.

Libraries from a single rack to exabyte scale
Air-gapped copies for ransomware resilience
Metadata-driven discovery and recall

Spectra Logic at Cinthamani

06Engagement

Start with an assessment.

We measure what you hold and how it is used before we propose anything. The assessment report is yours whether or not you go ahead.

Assess
Volumes, growth, file types and access patterns.
Design
Tiers, policies, metadata and sizing.
Pilot
One project or department proves the workflow.
Migrate
Existing data moved in, verified file by file.
Deploy
Production rollout and training.
Operate
Monitoring, media management and support.

07Questions

Enterprise data, answered.

What is an active archive?

An archive that stays online and searchable. Data sits on low-cost object storage or tape, but users and applications can still find it by metadata and recall it without a manual restore request.

Is tape still relevant for AI data?

Yes, for the copies you keep but do not use every day. Training sets, raw data and model versions grow quickly; tape keeps them at the lowest cost per terabyte, and policies stage the ones you need back to GPU storage.

What does Arcitecta Mediaflux do?

Mediaflux is a data management platform. It catalogues files with their metadata, applies policies for placement, copies and retention, and presents data across storage tiers as one namespace.

How does this protect against ransomware?

One copy is written to tape and kept off-line. Ransomware cannot reach or encrypt a cartridge that is not connected, so a clean copy is always available to restore from.

08Data infrastructure enquiries

Request a data assessment.

Tell us roughly how much data you hold and what it is for. A storage architect replies within one working day.

Sales: sales@cinthamani.com
India: +91 44 2815 3842

01Broadcast & Media

02Enterprise Data Infrastructure

03Apple Enterprise

04Audio Visual

05Cybersecurity & Cloud

06Cable TV Playout