Virtualisation is a powerful technology that can significantly reduce the amount of physical hardware required while also improving disaster recovery capabilities and providing a more centralised and manageable IT infrastructure. However, virtualised environments introduce their own challenges and must be designed carefully to avoid performance degradation.
Not every server can be virtualised without consequences, and performance identical to dedicated hardware should not be assumed. In the context of IDEA, the most critical component is the Microsoft SQL Server hosting the IDEA database.
Because resources such as CPU, RAM, and storage are virtualised, the performance of one VM can be heavily influenced by other VMs running on the same host. SQL Server, in particular, is sensitive to poorly configured virtual environments and contention for shared resources.
To ensure stable performance, it is essential to optimise both I/O throughput and CPU efficiency on the VM running SQL Server.
A recommended best practice is to use PVSCSI controllers for storage. Further details can be found in VMware’s knowledge base: https://knowledge.broadcom.com/external/article?legacyId=1010398
The socket-to-core ratio is another important factor when sizing a SQL Server VM. You should never assign more virtual sockets than the physical host provides and, in general, no more than four sockets should be used. For IDEA, a configuration of 1-2 sockets with 2-4 cores each is typically sufficient.
Even better results can be achieved by aligning the VM’s virtual CPU topology with the host’s physical NUMA layout. This applies not only to CPU resources but also to memory allocation. Crossing NUMA boundaries can significantly impact performance due to increased memory access latency.
Finally, be aware of a known issue related to certain host CPU types. Over the years, this has caused severe delays, often between 10 and 30 seconds, for certain SQL Server queries on affected systems. While the “host” CPU type is generally considered the most powerful due to its feature pass-through capabilities, specific CPU bugs can dramatically increase memory read latency and even cause temporary freezes.
Based on our experience, particularly with customers running Proxmox, switching the VM’s CPU type to x86-64-v2-AES has consistently resolved these delay issues immediately.
Dirk has been working at IDEA since 2010, and is one of our veterans with 4,695 tickets solved in the last 2 years alone. Dirk is on hand to give expert support! This column answers your most common questions!
Some of the columns allow you to set/check fields in IDEA (“IHM Relevant” or the “Show in Logs and Rounds”) by entering a TRUE or FALSE.