Architecture and storage selection guide

Start with the workload—not the product name

Choose storage and virtualisation architecture by access semantics, failure boundaries, recovery objectives and growth. Product selection should be the result of that process, not its starting point.

A practical selection sequence
1 · WORKLOAD
and I/O behaviour
2 · ACCESS
file, block or object
3 · AVAILABILITY
and recovery targets
4 · SCALE
and operating model
RESULT · a supportable architecture and a shortlist to validate
Workload shapes I/OLatency, concurrency, capacity and retention are different requirements.
Semantics before speedThe application must support the chosen file, block or object interface.
HA, DR and backup differContinuity and historical recovery solve different failure events.
Growth changes topologyScale-up, two-node HA and distributed scale-out have different boundaries.
Step 1 · Workload

Identify what the application needs from storage

The same capacity figure can describe very different systems. Start with access pattern, concurrency, consistency, recovery and operational ownership.

VIRTUALISATION

VMware, Hyper-V or eEVOS

Prioritise: supported shared storage, multipathing, consistent failover and VM-level recovery. eEVOS is also a complete virtualisation alternative with integrated Backup & Disaster Recovery.

Open the virtualisation guide →
DATABASES

Transactional and latency-sensitive I/O

Prioritise: durable writes, predictable tail latency, path resilience and application-consistent recovery. Benchmark the real database profile.

Compare block-access choices →
CONTAINERS

Kubernetes persistent data

Prioritise: CSI integration, access modes, failure-domain alignment, lifecycle behaviour and protection outside the Pod.

Open the Kubernetes guide →
AI & HPC

Parallel data pipelines

Prioritise: aggregate throughput, metadata behaviour, network design and feeding multiple compute nodes. GPUDirect Storage is not implied.

Open the AI and HPC guide →
MEDIA

Editing, ingest and rendering

Prioritise: concurrent sustained streams, shared namespace, project protection and predictable collaboration across clients.

Open the media guide →
SURVEILLANCE

Continuous recording and evidence

Prioritise: sustained write load, retention, camera/VMS compatibility, failure behaviour and practical retrieval.

Open the surveillance guide →
FILE SERVICES

SMB and NFS collaboration

Prioritise: identity, permissions, locking, client mix, namespace and service failover—not only aggregate bandwidth.

Review file-access characteristics →
OBJECT SERVICES

S3 for applications or service providers

Prioritise: API compatibility, tenancy, protection, lifecycle, quotas and operating model. MSP requirements deserve a separate design.

Open the enterprise S3 guide →
Step 2 · Access model

Choose the interface the workload can use correctly

A protocol is not a performance guarantee. Hosts, networks, path design, queue behaviour, storage media and the application all influence the result.

AccessData modelGood starting pointDesign trade-offs to validateTechnical guide
SMBFileWindows-oriented file services, collaboration, media and application sharesIdentity, ACLs, locking, dialect and multichannel/client supportSMB & NFS
NFSFileLinux/Unix applications, VMware NFS and shared namespacesVersion, locking, identity mapping, mount behaviour and failoverSMB & NFS
iSCSIBlock over EthernetBroadly understood shared block storage for virtualisation and applicationsMultipathing, network isolation, host filesystem ownership and queue designiSCSI →
Fibre ChannelBlock over FC fabricEstablished enterprise SAN environments with operational FC expertiseZoning, dual fabrics, HBAs, multipathing and lifecycle costFibre Channel →
NVMe/TCPNVMe block over TCP/IPModern Ethernet block storage without an RDMA requirementHost qualification, multipathing, loss/latency, CPU and end-to-end supportNVMe-oF →
NVMe/RDMANVMe block over RDMALow-overhead, latency-sensitive fabrics where the complete path is qualifiedRoCE/InfiniBand engineering, NICs, switching, congestion and operational skillsRDMA →
S3ObjectApplications, archives, backup targets and multi-tenant services using an S3-compatible APIApplication compatibility, consistency expectations, object sizing, lifecycle and tenancyEnterprise S3 →

Do not translate file access into block access casually. With block storage, the host normally owns the filesystem and coordination. Multiple hosts require an application or clustered filesystem designed for shared block access.

Step 3 · Availability and recovery

Select the failure boundary—not an abstract “HA” label

Decide which faults must be tolerated automatically, which events may use a recovery procedure and how much history must remain independently recoverable.

STANDALONE

One storage server

Focused and economical, but server-level failure is not masked by a second storage node. Add independent recovery and an operational plan.

SYNCHRONOUS LOCAL HA

Automatic two-node failover

Protects local service continuity through synchronous Mirror or dual-controller shared storage. Latency, quorum and paths matter.

ASYNCHRONOUS DR

Separate recovery copy

Provides a recoverable point at another system or site. The data gap follows the last completed replication cycle.

DISTRIBUTED

Scale-out continuity

Places data across multiple nodes and failure domains. Quorum, protection policy, free capacity and network design set the tolerance.

HISTORICAL RECOVERY

Backup and retained versions

Protects against deletion, corruption or unwanted change. A second live copy alone is not a historical backup.

Availability

Can the service continue after a component, path or node fails?

Disaster recovery

Can a service be restored beyond the local failure boundary within its RPO and RTO?

Backup

Can known historical data be recovered independently and verified?

Step 4 · Implementation model

Compare euroNAS examples after the requirements are clear

The platforms below solve different operational problems. None is universally “best”; the right shortlist depends on the architecture already defined.

Decision pointeuroNAS PremiumHA ClustereEKASeEVOS
Primary roleStorage OS for one physical server or Virtual Storage ApplianceAutomatic two-node storage serviceDistributed Ceph storage platformVirtualisation platform with integrated data protection
TopologyStandalone, scale-upTwo nodes: synchronous Mirror or dual-controller shared storageMulti-node scale-out without a fixed two-node boundaryCompact, two-node, shared-storage or scale-out virtualisation designs
Client servicesFile, block and S3 servicesHighly available file, block and S3 servicesCeph block, file and S3 servicesVirtual machines; storage can be internal, external shared or Ceph-based
AvailabilityDepends on the server design; asynchronous replication can provide a separate recovery copyAutomatic local service failover within the supported two-node architectureDistributed service and data protection across nodes and failure domainsVM availability, live migration and cluster operation according to the chosen topology
Growth modelScale up inside one serverScale up inside the two-node designAdd qualified nodes and capacity; rebalance across the distributed clusterAdd compute nodes at any time; they use the configured internal, external or Ceph storage
Recovery scopeSnapshots and asynchronous replication for storage-level recovery workflowsAutomatic failover plus snapshots and separate asynchronous replication optionsDistributed protection plus application-appropriate backup or external recovery processesIntegrated Backup & Disaster Recovery, including Instant Backup & Recovery, network backup and cloud backup
S3 operationsS3-compatible service and integrated S3 File Manager; bucket-count quotasS3-compatible service and integrated S3 File Manager; bucket-count quotasCeph S3, S3 File Manager, bucket-count and capacity quotas; billing functions for service-provider useNot the general-purpose S3 storage platform in this comparison
Important distinctionNo automatic two-node storage failoverFixed local two-node architecture; not a distributed scale-out clusterRequires appropriate node count, network, capacity reserve and distributed-system planningA virtualisation platform rather than a storage product. It does not provide storage services itself, but can use internal storage, external shared storage and Ceph-based storage.
Best first fitFocused standalone storage with a defined scale envelopeCompact automatic local storage HALarger scale-out storage, Ceph services and MSP-oriented S3Organisations wanting virtualisation, VM operations and recovery in one platform
Product detailPremium →HA Cluster →eEKAS →eEVOS →

The table is a first-selection aid, not a compatibility statement or sizing result. Confirm the required protocols, host versions, hardware, network design and application behaviour for the proposed configuration.

Shortlisting logic

A product follows the operating model

Use these questions to narrow the architecture, then validate the workload and protocol in detail.

  • Do you need a complete virtualisation platform or storage for another hypervisor?
  • Must storage continue automatically after a complete server failure?
  • Is the growth boundary one server, two nodes or a distributed cluster?
  • Are backup and recovery required for VMs, storage data or both?
  • Which team will operate the network, hosts, storage and recovery procedures?
Need an integrated virtualisation platform?YES
Evaluate eEVOS, including VM operations, Backup & Disaster Recovery and Instant Backup & Recovery.
Need storage services that scale across multiple nodes?YES
Evaluate eEKAS and design Ceph protection, failure domains, network and capacity reserve.
Need automatic local storage failover in a two-node design?YES
Evaluate HA Cluster and choose the supported synchronous Mirror or dual-controller shared-storage model.
Need a focused standalone storage server or VSA?YES
Evaluate euroNAS Premium and add the appropriate independent recovery plan.
Reference patterns

Four architectures cover very different boundaries

Standalone storage

Application
hosts
One storage
server or VSA

Simple ownership and a defined scale-up envelope. Add a separate recovery destination where required.

Premium example →

Two-node storage HA

Storage
node A
Storage
node B

Automatic failover with synchronous local protection or a shared-storage controller model.

HA Cluster example →

Distributed storage

Client
endpoints
Ceph nodes
and services

Block, file and S3 services distributed across qualified nodes and explicit failure domains.

eEKAS architecture →

Virtualisation platform

eEVOS
compute
Internal, shared
or Ceph storage

VM lifecycle, availability and integrated recovery with storage chosen for the required topology.

eEVOS architecture →
Before a final decision

Validate the complete service, not only the storage appliance

A credible design combines measured workload data with current compatibility and a tested operating procedure.

Application support

Supported protocol, versions, filesystem ownership, consistency and recovery requirements.

Measured workload

Capacity growth, block and object size, read/write mix, concurrency, latency and burst behaviour.

Failure boundaries

Device, node, switch, rack, power and site events that the service must tolerate.

Recovery objectives

RPO, RTO, retention, restore granularity, immutability and responsibility during an incident.

Network and paths

Bandwidth, redundancy, multipathing, zoning/VLANs, RDMA configuration and operational visibility.

Growth and operations

Expansion steps, maintenance windows, rebuild reserve, monitoring, skills and lifecycle ownership.

Turn requirements into a shortlist

We can review the workload, access method, failure domains, recovery objectives and growth plan for a business storage or virtualisation project.

Discuss your architecture
Scroll to Top