9 min read

Learning Objectives

By the end of this module, you will be able to:

  • Choose between a Block Storage volume and a Network File Storage (NFS) share for a given access pattern
  • Match a Block Storage class or an NFS cluster size to the throughput and IOPS a workload needs
  • Plan around the regional, private scope of NFS and the storage-zone-versus-compute-zone asymmetry
  • Provision an NFS cluster and a share, restrict it to one client network, and mount it from several servers in the Data Center Designer (DCD)

Unit 5.1: Block and File Storage

Introduction

Module 4 gave every server its own volumes. As soon as two servers need the same files, a per-server volume is the wrong tool: copying files between servers creates drift, and a self-built file server becomes one more machine to patch and protect. The storage decision is therefore about the access pattern first: one server reading and writing its own blocks, or many clients sharing one file system.

This unit sets the boundary between the two, sizes each against its performance model, and ends by building FinCorp's shared file storage for the portal tier in the DCD.

1. One Server's Blocks or Many Clients' Files

1.1 Block Storage: The Server's Own Device

A Block Storage volume is equivalent to an Internet Small Computer System Interface (iSCSI) block device and behaves like direct-attached storage. It is managed independently of servers, up to 24 volumes can be connected to a Dedicated Core Server or a Cube, and each volume is replicated four times on distributed physical devices within the selected data center location. Unit 4.2 covered how to attach volumes and which attributes are fixed. For this unit, the point is the access model: a volume is the disk of the server it is attached to, and its file system is managed inside that server's operating system.

The class decides the performance envelope. Essential is static at 1,100 IOPS, independent of size. Balanced and Performance scale with volume size up to per-volume ceilings: 24,000 read and 18,000 write IOPS for Balanced, 45,000 read and 30,000 write IOPS for Performance, each at 4 KB block size and with a minimum server size. Match the tier to the access pattern: Essential for backups, archives and cold data; Balanced for web servers and development environments; Performance for databases and other latency-sensitive random I/O. Unit 2.4 prices the classes, and Unit 4.2 explains the 100 GB floor and the minimum server size for the SSD classes.

1.2 Network File Storage: Shared Files on a Private LAN

NFS is a fully managed, high-availability shared storage service that uses the Network File System protocol, version NFSv4.2. It provides SSD-backed, encrypted-by-default file storage over a private network, with no data transfer charges between storage and compute, and POSIX permissions at file and directory level. Data can be accessed simultaneously from multiple compute instances, such as VMs and containers, without compromising data consistency. Typical uses are shared configuration and static assets, media serving and Kubernetes ReadWriteMany persistent volumes.

A cluster contains shares, which are individual file systems that clients mount. At creation, you give the cluster an IP address from your LAN, which becomes the floating IP address of the two VMs that NFS uses for high availability. Clients mount a share on Linux with the cluster's IP address and the share's UUID: mount -t nfs <cluster-ip>:<share-uuid> <local-mount-path>.

1.3 Sizing NFS: Capacity Buys Performance

The cluster size ranges from 2 TiB to 42 TiB in 1 TiB increments, all of it usable, and it determines the bill. Size can only be increased after provisioning, never decreased. Throughput and IOPS scale with that size:

Metric Per Client Aggregate per 4 TiB
Read IOPS ~4,000 24,000
Write IOPS ~470 18,000

Peak sequential throughput is about 370 MB/s read and 319 MB/s write per TiB, so a 4 TiB cluster delivers about 1,480 MB/s read and 1,275 MB/s write. Two limits shape the design. Write IOPS are constrained by a synchronous write latency of about 20 ms, so the aggregate figure needs many concurrent operations, which means many clients rather than one mount. And NFS is not suitable for synchronous write workloads with tight timeout requirements. A database belongs on a Performance volume or a managed database (Unit 5.3), not on a share.

1.4 Scope: Regional, Private, Linux

An NFS cluster is created in one location and connected to one LAN of one VDC, and VMs on that private network connect to it. NFS is not offered in every location: the Location list shows where you can create clusters, and GB/BHX is not among them. For data that a second region must read, such as FinCorp's Berlin recovery VDC fc-prod-txl-dr, use IONOS Cloud Object Storage with replication instead (Unit 5.2).

1.5 The Zone Asymmetry

Block Storage offers Availability Zones 1, 2 and 3, plus Auto, while servers offer Zones 1 and 2, plus Auto. A storage zone is set at first provisioning and cannot be changed; to move a volume, you take a snapshot and provision a new volume from it in the new zone. Do not design a three-zone compute tier to match the storage zones: Zone 3 exists for storage only. For a redundant server pair, keep each server's volumes in that server's zone, as Unit 4.2 did.

1.6 The FinCorp Decision

FinCorp's portal servers fc-app-01 and fc-app-02 both need the same document templates, branding assets and configuration. A volume on each server would mean a sync job and drift between the two halves of the pair. FinCorp therefore creates one 2 TiB NFS cluster in Frankfurt on fc-app, one share for the assets with a quota, and a client group limited to the fc-app subnet with root squash. Berlin gets its copy of the same assets from Object Storage, not from the Frankfurt share.

Need Choose Why
A server's boot or database disk Block Storage, Performance class Lowest latency, random I/O per volume
Files shared by servers in one VDC NFS share One copy, POSIX permissions, no traffic charge
Data read across regions Object Storage with replication An NFS cluster serves the private LAN it is attached to
Cold data on a single server Block Storage, Essential class Lowest block cost

DCD Implementation Walkthrough

You will create FinCorp's NFS cluster on the fc-app LAN of fc-prod-fra-core, add a share restricted to that LAN, and mount it on both portal servers. This realises the "one copy, many clients" decision from Section 1.6.

Prerequisites: the portal servers from Unit 4.2; a free address from the managed-service range .2 to .9 of fc-app, recorded in the address plan of Unit 3.1; an account with the Access and Manage Network File Storage privilege, or a contract owner or administrator.

Build goal: Provision shared file storage mounted by multiple clients.

Steps (in the Data Center Designer):

  1. Go to Menu > Backup & Storage > Network File Storage and select Create Cluster.
  2. Enter the Cluster Name fc-nfs-portal. In Location, select Germany / Frankfurt. Set Size to 2 TiB with the slider; it can grow later but never shrink. Keep the File System Version at NFSv4.2.
  3. In Datacenter, select fc-prod-fra-core. In Datacenter LAN, select fc-app. In Private IPv4 IP, enter the reserved address of the fc-app subnet with its prefix. Save the cluster and wait until its state is AVAILABLE.
  4. In the cluster list, open OPTIONS > Manage Shares and select Create Share.
  5. Under the share properties, enter the Directory Name portal-assets and a Quota in MiB that caps the share. Keep the default Group Id and User Id of 65534 unless your file ownership model needs others.
  6. Under Add Client Groups, select NFS Squash Mode "Map root user to anonymous user id". In IP Networks, add the fc-app subnet in CIDR notation. Leave Hosts empty: the IP Networks setting supersedes the host list.
  7. Select Save, then copy the share ID from the share's ACTIONS menu with Copy ShareId.
  8. On fc-app-01 and fc-app-02, mount the share with mount -t nfs <cluster-ip>:<share-uuid> <local-mount-path>, using the cluster address and the copied share UUID.

Common mistakes:

  • Choosing "No squash mode". It is not recommended, for security reasons.
  • Filling both IP Networks and Hosts and expecting the host list to restrict access. The network entry wins.
  • Starting with a large cluster "for later". Size only grows, and the size is the bill; grow it when the throughput or capacity is needed.
  • Putting a database or a latency-critical synchronous writer on the share. Use a Performance volume or a managed database.
  • Expecting the Berlin VDC to mount the Frankfurt share. Replicate the data through Object Storage instead.

Summary

Block Storage is one server's device, sized by class and volume; NFS is one file system for many clients on one private LAN, sized by cluster capacity. Choose by access pattern, size NFS for throughput as well as capacity, and keep cross-region sharing in Object Storage. Storage zones include a Zone 3 that compute does not have, so pin volumes to their server's zone rather than designing around it.

Key Points:

  • A Block Storage volume behaves like direct-attached storage; the class sets its IOPS and throughput ceilings.
  • NFS clusters range from 2 TiB to 42 TiB in 1 TiB steps, grow only, and gain throughput with size.
  • NFS write latency of about 20 ms rules it out for synchronous writers with tight timeouts.
  • An NFS cluster serves the private LAN it is attached to; cross-region data belongs in Object Storage.
  • Block Storage has Zones 1, 2 and 3; compute has Zones 1 and 2.

Important Terminology:

  • Share: an individual file system inside an NFS cluster, with its own quota and client groups.
  • Client group: the set of networks or hosts allowed to mount a share, with a squash mode.
  • Root squash: mapping a client's root user to an anonymous user ID, so that a compromised client cannot act as root on the share.