Behind the Exabytes: A Field Note from Inside the Cloud

Vriti Magee | Jul 9th 2025

Where Cold Storage Runs Hot: Inside the Vault Built for Exabytes. Illustrated by ChatGPT & DALL·E

At Cloud Field Day 23, Scality presented a practical view of how its storage architecture supports some of the world’s most data-intensive organisations—many of them operating outside the public cloud.

What Lies Beneath

In 2019, I visited one of Iron Mountain’s Australian facilities while reviewing its capabilities as a third-party provider for a Big Four Australian Bank. The bank, a long-standing partner of Iron Mountain for information management services, was using its infrastructure to support invoicing and payment processes. At the time, my task was to assess operational and technical suitability. What stood out was Iron Mountain’s emphasis on continuity and control—priorities that felt foundational rather than reactive.

At Cloud Field Day 23, that same emphasis surfaced again—this time through Scality, the software platform now underpinning Iron Mountain’s rearchitected object archive service. As Scality’s team explained, Iron Mountain had previously encountered architectural issues with an earlier platform. In response, they transitioned to a fully online Scality RING deployment, aiming to ensure that data remained both accessible and resilient.

Rather than leading with trend-driven messaging, Scality used its sessions to demonstrate how the RING architecture, first introduced in 2010, supports exabyte-scale growth, internal cloud services, and lifecycle-based data management. These are not hypothetical capabilities. The platform is now used by global banks, European space agencies, and national service providers—often as the foundation for internal, sovereign cloud infrastructure.

Ask most technologists where enterprise data lives, and the answer will likely default to the cloud. What that overlooks is who—if anyone—owns the cloud. At Cloud Field Day 23, Scality made the case that its platform, the RING, now underpins internal cloud infrastructure for entities ranging from global banks and intelligence agencies to national space programmes. It is not a general-purpose storage system but a tightly honed, multidimensional architecture—one built, as its founders put it,

“by French engineers to solve a very American problem.”

From Postcards to Petabytes

Scality’s origin story is refreshingly analog. In 2008, Comcast and other service providers realised they were storing cat pictures and PowerPoint jokes using financial-grade SANs—an economically untenable architecture. They admired the hyperscalers’ use of cheap servers and infinite-scale object storage but lacked the developer capacity to build such systems themselves. Enter Scality: a storage firm built around the principle of scale across all dimensions, not just capacity. From the outset, the architecture was disaggregated—compute, storage, and metadata could each scale independently.

This idea has aged well. According to Giorgio Regni, the firm’s co-founder and CTO, Scality now manages 12 exabytes across its global customer base, supporting over 6 trillion objects and 447,000 drives. A typical large customer runs 2,000 or more internal applications, each treated as a separate S3 client. The architecture supports tens of millions of buckets—such as one telco mapping 35 million subscribers to individual S3 buckets—and petabytes-per-day ingest rates for some customers, with latency in the 2–5ms range for flash-backed workloads. No less impressively, the average customer sees just one service disruption every five years.

🛠️ Architectural View

In one deployment, Scality processes over 1 billion S3 operations per day with latencies between 2 and 5 milliseconds. The platform supports up to 60,000 operations per second per cluster, backed by a hybrid tier of flash (for metadata) and high-density disk (for object bodies). Write operations are distributed across nodes for scalability and resilience. Multi-site configurations use S3-compatible cross-region replication with indefinite retry logic to address transient network failures.

Customers of a Certain Scale

Scality’s most compelling stories come not from product demos, but customer deployments.

At one major European bank—Scality RING has become the core of a global internal cloud. Over 2,000 applications, ranging from document archives to developer platforms and internal services, now write to S3 buckets managed by Scality. The platform, which began with just 1PB in 2017, now supports 100PB across six independent installations, with one production and one disaster recovery RING in each global region.

The system is not only large, but also structured to support consistent storage policies across thousands of applications. Application teams configure replication, immutability, or versioning through internal tooling that abstracts the complexity of S3 APIs. All operations—from node replacement to hardware migration—are performed online. The result: over 300 billion objects, 200,000 buckets, and a billion operations per day, all served with 75% lower cost per terabyte than the bank’s prior infrastructure.

🛠️ Architectural View

The deployment spans six physical regions, each with independent RING instances and asynchronous replication between production and DR platforms. All lifecycle and security controls are exposed via S3-compatible policies, wrapped in internal automation tools. Bucket configuration defaults to immutable, versioned storage with CRR enabled—ensuring resilience without backup dependencies.

A second story came from CNES, France’s national space agency and part of the European Space Agency. With a mandate to preserve decades of environmental satellite data—and to ingest 10–12PB per year—the agency turned to Scality to build a storage system that was cost-effective, energy-aware, and transparent to end users. The result is a three-site stretched RING, fronted by real-time compute pipelines and backed by tape libraries, where cold data is automatically offloaded. Retrievals mimic AWS Glacier APIs, but without the egress charges or foreign dependencies.

🛠️ Architectural View

CNES uses Scality to manage a three-site stretched RING, combining real-time compute with robotic tape libraries for cost-effective cold storage. To manage long-term archive data, Scality enables automated tiering to tape once objects become cold, while keeping all metadata active in the RING. Applications retrieve archived data through a Glacier-compatible API that returns a pending status, with notifications delivered once the object becomes available. This design allows users to interact with both hot and cold data in a single namespace, enabling long-term preservation of more than 20 years of satellite imagery without compromising accessibility or operational efficiency.

Scaling Without Drama

If Scality’s engineering story has a thesis, it is modularity over monolith. The RING’s architecture, Regni explained, is not only microservices-inspired but formally disaggregated. Storage nodes run IO daemons. Metadata lives on flash, but can be relocated. Connectors are stateless and ephemeral. Customers scale them not as a cluster but as a set of loosely coupled planes, adding more of whichever tier becomes a bottleneck.

This independence matters. A full-flash “hot tier” can now run separately from a tape-backed “long-term tier,” both within the same namespace. A stateless S3 endpoint can reside near an AI pipeline, while storage nodes live in the core data centre. Even multi-site deployments remain manageable thanks to Scality’s peer-to-peer CHORD-based protocol: each node only needs to maintain contact with three to five others.

No less crucial is Scality’s aversion to traditional backup logic. As Paul Speciale, the CMO, noted: no one backs up an exabyte. Instead, Scality supports asynchronous replication, immutable buckets, and object versioning as standard practice. The RING does not merely store files; it preserves state, manages lifecycle, and provides observability.

🛠️ Architectural View

Scality supports native erasure coding, geo-distributed availability zones, and hot–warm–cold tier transitions via policy. Customers can scale metadata layers independently to handle edge cases such as 35 million buckets (as seen in a large mobile operator deployment). Internal connectors support RESTful access paths in addition to S3—used by latency-sensitive AI pipelines that depend on consistent 2–5ms performance.

A Platform by Another Name

It is tempting to label Scality as simply an object storage company. It is not. What emerged at Cloud Field Day 23 was a platform—not in the sense of an all-consuming suite, but as a backbone on which regulated enterprises are building internal clouds. It competes less with MinIO or Ceph than with the instinct to build one’s own cloud-native stack from scratch.

At a time when vendors tout every product as “AI-ready,” Scality’s restraint is notable. It has customers ingesting a petabyte per day, supporting real-time telemetry pipelines and AI workloads, and preserving 20 years of satellite imagery—but the pitch remains infrastructure-first. As Regni said, with a touch of dry understatement:

“Refreshing we didn’t talk about AI.”

It’s easy to overlook systems that simply work. But the institutions that underpin much of the global economy—banks, governments, space agencies—require exactly that. Scality positions RING as a platform for enterprises building internal clouds—one they turn to when object storage becomes mission-critical.

🔍 Links for Further Reference

Watch the full Cloud Field Day 23 sessions: