Elastic & Auto-Scaling

One architecture. Any volume. No shards.

Legacy PACS can't scale — so enterprises shard, running multiple separate instances of the same software, each with its own database and infrastructure. Sirona serves your entire multi-site, multi-state, multinational practice on a single elastic architecture that scales automatically with demand.

See Sirona in Action

How Sirona is Different

Legacy vendors shard because they can't scale

Most legacy PACS architectures hit a ceiling between 1 and 3 million exams per year. After that, the only option is sharding — running multiple separate instances of the same software, each with its own database, its own servers, its own maintenance burden. Large enterprises end up operating many separate shards — each a separate system to manage, patch, monitor, and keep in sync. Sirona's cloud-native architecture eliminates this entirely. Kubernetes orchestration with horizontal pod autoscaling and Karpenter node provisioning means the platform scales continuously — from 100,000 exams to 10 million — on a single architecture with a single database, a single worklist, and a single pane of glass.

Infrastructure that scales itself

Kubernetes orchestration, automatic node provisioning, and stateless microservices — designed to handle any volume without manual intervention.

Kubernetes Orchestration (EKS)

Horizontal pod autoscaling adjusts service replicas based on real-time CPU, memory, and request load. No manual scaling, no capacity planning.

Karpenter Node Provisioning

When pods need more infrastructure, Karpenter automatically provisions the right-sized EC2 nodes in seconds. When demand drops, nodes are reclaimed.

Stateless Microservices

Every backend service — viewer, reporter, worklist, DICOM ingestion — is stateless and horizontally scalable. Add replicas without coordination.

Multi-AZ Redundancy

Services run across multiple AWS Availability Zones. If an entire AZ goes down, traffic automatically routes to healthy zones with zero downtime.

Global CDN Edge Caching

CloudFront delivers images from edge nodes worldwide. Scaling reads doesn't require scaling origin servers — the CDN absorbs the load.

Elastic AI Compute

AI inference workloads auto-scale independently from core services. Deploy new models without impacting viewer or reporter performance.

Scaling that works like cloud should

Architecture

Single-tenant isolation, multi-tenant efficiency

Sirona's multi-tenant architecture isolates each customer's data while sharing the scaling benefits of a unified infrastructure. When a legacy platform needs many shards, it's because the architecture can't separate data isolation from compute scaling. Sirona does both — your data is isolated at the database and storage layer while compute resources are shared and scaled elastically. This means a 50-radiologist practice benefits from the same infrastructure investments as a 500-radiologist health system, without paying for dedicated hardware.

Data isolation at database and object storage layer — not compute layer

Shared elastic compute pool — resources flow where demand is

Single architecture from 100K to 10M+ exams per year

No sharding, no instance splitting, no multi-system management

Operations

Volume spikes that the platform absorbs automatically

Surgery days at a hospital system. Monday morning telerad backlogs. A new site onboarding 200,000 historical studies. M&A integration doubling your exam volume overnight. These are the scenarios where legacy PACS falls over — and where Sirona doesn't blink. Kubernetes horizontal pod autoscaling detects increased load and adds service replicas in seconds. Karpenter provisions new compute nodes when existing capacity is exhausted. The DICOM ingestion pipeline, the viewer rendering service, and the worklist query engine all scale independently based on their own load patterns.

Horizontal pod autoscaling responds to load in seconds

Karpenter provisions new nodes when pod density hits limits

Each microservice scales independently — ingestion, viewing, reporting

Historical study migration runs as background jobs without impacting live reads

Economics

Usage-based compute replaces hardware CapEx

On-premise PACS requires you to buy servers for your projected peak load plus headroom — then depreciate that hardware over 3-5 years while it sits idle most of the time. Sirona flips this model. Compute scales up during peak hours and scales down overnight. You never pay for idle capacity. There's no hardware to procure, no refresh cycles to plan, no CapEx budgets to defend. When your practice grows — new sites, new radiologists, new modalities — the platform grows with you instantly. No lead time, no procurement, no migration.

Scale up during peak, scale down overnight — pay for actual usage

No server procurement, no hardware depreciation, no refresh cycles

Growth is instant — add sites or volume with zero infrastructure lead time

Eliminates CapEx for PACS infrastructure entirely

AI at Scale

Elastic compute makes AI-native radiology possible

Running QualityAssist, Anatomic Navigator, Auto-Fill Comparison, AI Prior Summary, and third-party AI algorithms on every study requires compute that scales with volume and model complexity. On-premise GPU clusters are fixed-capacity — you can run three algorithms today, but what about ten tomorrow? Sirona's elastic architecture provisions AI compute on demand. When you enable a new AI model, the platform allocates the resources it needs. When volume surges, AI processing scales alongside clinical workflows. This is the infrastructure foundation that makes AI nativity real — not a feature checkbox, but a compute architecture that can actually support it.

AI inference scales independently from viewer and reporter services

New model deployment doesn't require hardware procurement

GPU compute provisioned on demand — no fixed-capacity constraints

Foundation for running 10+ AI algorithms per study at any volume

Scale without limits

1

architecture for any volume — no sharding required

Zero

shards, at any exam volume

Multi-AZ

redundancy with automatic failover through every scaling event

0

servers to buy, maintain, or refresh

Growth without infrastructure constraints

Everlight RadiologyRead the partnership announcement

“This partnership marks the beginning of the cloud-native era in radiology software globally. Sirona has delivered an architecture that is fundamentally different from anything else in the market today — Sirona is unified, and cloud-native from the ground up. For teleradiology, that architectural distinction is existential, not incremental.”

Andy Donaldson

CTO, Everlight Radiology

Epsilon Health

“In 12 months on Sirona we went from zero to more than a million exams a year. That growth was only possible because we deployed our own AI on top of the Sirona platform — our radiologists consistently read X-rays 20–40% faster on workflows we could not have built on any other platform. You cannot build a modern practice on a legacy PACS — and Sirona is the only modern enterprise-grade platform in radiology today.”

Rustin Rassoli

Founder & CEO, Epsilon Health

See all customer stories

FAQs

What does 'sharding' mean and why is it a problem?

Sharding means running multiple separate instances of PACS software because a single instance can't handle the volume. At enterprise scale, legacy PACS deployments can end up running many separate shards. Each shard requires its own database, servers, maintenance, and monitoring — multiplying complexity and cost. Sirona eliminates sharding entirely with a single elastic architecture.

How many exams can Sirona handle?

There is no hard ceiling. Sirona's Kubernetes-based architecture scales horizontally — adding compute resources automatically as volume grows. Legacy PACS typically hits limits between 1 and 3 million exams per year. Sirona's architecture is designed for continuous scaling well beyond those thresholds on a single instance.

How fast does auto-scaling respond to demand spikes?

Horizontal pod autoscaling detects increased load and adds service replicas in seconds. When existing node capacity is exhausted, Karpenter provisions new EC2 compute nodes — also in seconds. The entire response from load detection to additional capacity is measured in seconds to minutes, not hours.

Do I need to plan for capacity in advance?

No. Sirona's auto-scaling handles capacity automatically. You don't need to forecast exam volumes, request additional servers, or plan upgrade windows. The platform scales with your actual demand in real time.

How does elastic compute support AI workloads?

AI inference workloads — QualityAssist, Anatomic Navigator, third-party algorithms — auto-scale independently from core clinical services. When you deploy a new AI model or increase volume, the platform provisions additional compute automatically. No GPU procurement, no fixed-capacity constraints.

What happens during an M&A integration that doubles volume?

The platform scales automatically. Historical study migration runs as background jobs without impacting live reading performance. New exam volume from the acquired practice flows through the same elastic architecture. There is no hardware procurement step, no capacity planning meeting, and no lead time.

How does cloud pricing compare to on-premise hardware costs?

Cloud compute scales with actual usage — up during peak hours, down overnight. On-premise hardware is provisioned for peak plus headroom, then depreciates while sitting idle most of the time. Sirona eliminates server procurement, refresh cycles, power, cooling, and dedicated IT staff for hardware management.