VMware vSphere¶

This guide describes how to prepare a virtual machine for ARTESCA on VMware vSphere 8 and 9.

Prerequisites¶

VMware and its components must be installed with:

VMware vSphere 8 or 9.
The ARTESCA OS ISO must be uploaded to a vSphere datastore. To download it from the Asia-Pacific region, use this link instead.

Make sure to review the virtual hardware requirements to size the virtual machine according to your ARTESCA deployment, and the Recommended vSphere Settings below, because you apply several of them while creating the virtual machine..

Recommended vSphere Settings¶

Apply the following settings to get predictable performance and to align the virtual machine with the ARTESCA redundancy model. Unless stated otherwise, they apply to both single-node and multi-node deployments.

Firmware¶

In VM Options > Boot Options, set the firmware to EFI (UEFI). Leave Secure Boot disabled unless your site policy requires it.

CPU Topology and Scheduling¶

Align Cores per Socket with the physical host NUMA topology so the guest is presented with a correct virtual NUMA (vNUMA) layout. Keep the vCPU count within a single physical NUMA node when possible to preserve memory locality. For larger virtual machines, map the sockets to the physical sockets.
Do not enable CPU Hot Add. It disables vNUMA and forces the guest onto a single NUMA node, which degrades memory locality for large virtual machines.

Resource Shares and Reservations¶

Set CPU and Memory shares to High (or set explicit reservations) so the ARTESCA virtual machines keep prioritized access to compute when the host is contended.
Enable Reserve all guest memory (All locked) to prevent memory ballooning and hypervisor swapping, which severely degrade storage performance.
Set disk shares to High, or use Storage I/O Control (SIOC), so the ARTESCA virtual machines keep prioritized access to datastore I/O under contention.

Storage Controllers¶

Keep the OS and service drives on the NVMe controller.
Attach the data drives to VMware Paravirtual (PVSCSI) controllers for high IOPS and low CPU overhead.
Distribute the 12 data drives across multiple PVSCSI controllers (up to 4) to parallelize I/O and avoid a single controller queue becoming a bottleneck.
Set the advanced parameter disk.EnableUUID = TRUE so the guest exposes stable disk identifiers for the multiple ARTESCA drives.

Disk Provisioning¶

Provision the service and data VMDKs as Thick Provision Eager Zeroed for production. Eager zeroing pre-zeroes the blocks at creation time, removing the first-write penalty and delivering consistent write performance for the storage workload.
Use Thin Provision only for demo or space-constrained environments, where space efficiency matters more than sustained write performance.

vSAN Storage Policy (Multi-Node)¶

If the datastore is backed by vSAN and you deploy a multi-node ARTESCA cluster, assign the ARTESCA VMDKs a storage policy with no replication (Failures to tolerate = 0, that is, RAID-0 / no mirroring). ARTESCA already provides data redundancy across its nodes, so vSAN mirroring duplicates the protection, doubling capacity usage and write amplification without adding resilience.

Warning

Failures to tolerate = 0 removes vSAN protection for an individual virtual machine. Only use it when redundancy is guaranteed by a multi-node ARTESCA cluster and the node virtual machines are placed on different hosts. For a single-node ARTESCA system, keep the default vSAN protection policy.

Placement and Availability (Multi-Node)¶

Create a Distributed Resource Scheduler (DRS) anti-affinity rule (Separate Virtual Machines) so that ARTESCA node virtual machines never run on the same physical host. A host failure then impacts at most one node.
Prefer separate datastores or failure domains per node to avoid a shared storage single point of failure.

Time Synchronization¶

Ensure the ESXi hosts and the ARTESCA virtual machines synchronize against the same reliable Network Time Protocol (NTP) source to avoid clock drift across the cluster.