Virtual Hardware Requirements

This page lists the required and recommended virtual hardware for an ARTESCA Software Appliance virtual machine node. These requirements are the same for every supported hypervisor and for a physical Software Appliance deployment. Use them when preparing a virtual machine on any of the hypervisors described in this page.

General

  • Firmware: UEFI.

  • Guest OS type: Linux, Rocky Linux (64-bit), or Linux Kernel 4.x or later (64-bit).

Compute

  • CPU can be either an AMD or an Intel CPU.

  • CPU can be single or dual socket.

  • RAM (DDR) must be:

    • 128 GB or more for a production environment, depending on the configuration.

    • A minimum of 32 GB for a demo environment.

The CPU and RAM requirements depend on the number of data drives hosted per node:

Server Density

Physical CPU Cores

Memory

Up to 4LFF

12

128 GB

Up to 16LFF

24

128 GB

Up to 28LFF

32

192 GB

Up to 45LFF

40

256 GB

Up to 60LFF

56

384 GB

Up to 100LFF

64

384 GB

In a virtual deployment, provision one vCPU per physical core listed above. For predictable performance, align the virtual machine with the physical NUMA topology of the host:

  • Keep the vCPU count and the assigned memory within a single physical NUMA node whenever possible. A virtual machine that spans several NUMA nodes forces the hypervisor CPU scheduler to serve part of its memory accesses remotely, which increases memory latency and is a common bottleneck for the ARTESCA storage workload.

  • Expose a virtual NUMA (vNUMA) topology that matches the host by setting the cores-per-socket accordingly, so the guest CPU scheduler can keep processes close to the memory they use.

  • Do not oversubscribe physical cores. A high vCPU-to-physical-core ratio causes CPU ready time and scheduler contention that degrade latency.

  • Reduce the number of vCPUs, even below the minimum requirements, to improve performance.

Size the compute resources according to the server density guidelines in General Hardware Requirements for a Production System. The larger the number of data drives, the more vCPU and RAM are required.

Storage

Each ARTESCA virtual machine node is composed of several types of virtual drives.

Operating System Drive

  • One virtual disk of at least 480 GB.

  • Attach it to a dedicated controller.

Service Drive

  • One virtual disk backed by resilient and fast storage (an NVMe controller is recommended when the hypervisor supports it).

  • Avoid using thin provisioning for performance reasons.

The minimum size of the service drives depends on:

  • Maximum number of data drives on the node

  • Capacity of the data drives

  • Number of nodes

  • Partner Application

For more information on the minimum size of service drives, refer to the Required Service Disk Capacity section.

Fast storage is required for ARTESCA Data Services, the ARTESCA Storage Services index, and the ARTESCA Kubernetes Platform database.

Note

For virtual machines you do not need to set up a RAID 1 array, but make sure that the virtual drives are backed up by resilient and fast storage units.

ARTESCA Data Drives

  • Standard ARTESCA configurations require a minimum of 12 data drives.

  • ARTESCA also comes with a one-node-only configuration with 4 or 8 data drives.

  • All data drives must be identical (disk technology and capacity).

  • Provision each data drive with at least 1 TB.

  • Attach the data drives to a SCSI controller (use a paravirtual controller when the hypervisor provides one).

  • Avoid using thin provisioning for performance reasons.

Network

Provision virtual network adapters:

  • One virtual network adapter for both the control plane and the workload plane.

  • Two virtual network adapters:

    • One virtual network adapter for the control plane.

    • One virtual network adapter for the workload plane.

Use the paravirtualized adapter provided by your hypervisor (for example VMXNET3 on VMware vSphere, VirtIO on Proxmox VE) for best performance.

The link speed advertized by a virtual network adapter is nominal and is not a meaningful performance requirement on its own. A hypervisor switches traffic that stays inside a single host in memory, for example between virtual machines on the same virtual switch, so no physical interface limits it.

The bandwidth requirement applies to the physical uplinks of the hypervisor host and the underlying network fabric. These paths carry all traffic that leaves the host: physical S3 clients, the data load balancer, and ARTESCA node virtual machines on other hosts.

Size the physical network so that these paths meet the targets below. When all ARTESCA nodes and all clients are virtual machines on the same host, the physical interfaces do not constrain intra-host traffic.

Note

For the Data Network, Scality recommends using 10 Gb/s interfaces. 1 Gb/s interfaces are accepted, but Scality does not guarantee the performance of the system with these interfaces.

Important

For a 3-node cluster, Scality recommends the Data Network and the Management Network to be segregated and exposed through distinct network interfaces. Refer to Network Considerations for more information.

Guest Agent

You do not need to install a guest agent manually. During the ARTESCA OS installation, the installer detects the hypervisor and automatically installs the matching guest agent (open-vm-tools on VMware vSphere, or qemu-guest-agent on KVM/QEMU-based hypervisors such as Proxmox VE) and enables its service.

Hypervisor-specific preparation

Once you have sized the virtual machine according to the requirements above, follow the step-by-step preparation guide for your hypervisor: