IT has been on vacation for 20 years, time to get back to work.

By Richard Tengdin, Enterprise Solutions Architect, Nth Generation

*This image was generated by Adobe Firefly.

Since the dawn of server virtualization IT operations has managed their servers as a pool, letting application-level management fall back to the application owner. Memory was cheap, storage was cheaper, and software was licensed by CPU. IT wasn’t allowed (or able) to charge back resource usage to departments or projects, they just ran the pool as overhead cost.

Boy, those were the Good Old Days.

Enter the AI boom and VMware changing hands (again and again). Now memory is dear, NVMe-based storage is dearer, and vSphere virtualization is (expensively) licensed per core. The costs of IT infrastructure are rising at a rate that is hard to believe, so Right-Sizing new servers is not just a cost-saving measure but one of corporate survival. The question is what are the issues  and what tools are available today to help with the process?

Virtual machines are sized by number of virtual CPUs, amount of virtual RAM, and assigned storage. Our habit has been to let the end users throw a lot of resources at the workload and allow features like DRS and storage Thin Provisioning to take care of the mismatch between workload needs and virtualization resources. There is no process to go back and assess VM performance to determine what is actually required (1 vCPU vs. 8, etc.). That policy needs to change.

The first management focus should be vCPU core counts and how they affect workload scheduling on a server. In basic terms a VM can’t execute until enough physical cores simultaneously have a free time slice so the VM can be scheduled. A VM with an excessive number of vCPUs has to wait a (relatively) long time to get a time slot plus having lots of over-sized VMs reduces the consolidation possible on the cluster. One of the reasons VDI can run with a 6:1 virtual to physical consolidation ratio is because virtual desktops aren’t normally built with large core counts. A DNS or DHCP VM can work fine with 1 vCPU, configuring them with 8 vCPUs ‘just in case’ costs consolidation now and licensing costs in the future for new hardware. A more extreme case is a database VM with 32 vCPUs that only requires 8. Rightsizing allows you to reduce hosts today and resize future hosts for tomorrow, saving virtualization licensing at the same time. The Cloud providers naturally price their VMs by vCPU and vRAM now. Time for IT to join the party.

Virtual RAM needs to be managed the same way. We’ve just thrown vRAM at workload and sized servers for 50% or lower RAM occupancy. With proper management operating at 80% occupancy or higher is achievable, and memory can be oversubscribed for those workloads with lower performance requirements. How fast does a File Server VM need to be? HPE’s Greenlake platform uses RAM consumption to bill virtual environments because VMs frequently have many more vCPUs than required. RAM overprovisioning has performance impacts but do all your VMs need full performance? Can low demand VMs use paged memory and still perform? It is worth testing now that DDR5 RAM is $500/GB and rising.

What about storage? It was managed by adding resources, it was just understood that storage management software was more expensive than just adding more disc space and backup capacity. Everyone saw the predictions about petabytes of data added per year and just shrugged their shoulders. 

What tools do we have today? 

Virtual Machine management begins with monitoring, and tools like Cloud Physics, vRealize Operations Manager, Login VSI for Citrix, and SolarWinds Virtualization Manager can show you what is going on with your environment including actual utilization information. Static analysis using RVtools shows you what you have but can’t suggest changes to correct oversizing. Shadow IT is equally valuable here. WHAT!!!!  Yes, IT can leverage cloud providers to test workloads on larger and smaller VM configurations to see if reducing that 32 core 512GB elephant to 8 cores and 128GB of RAM actually impacts performance.

On the storage front tools like Varonis and Komprise can identify stale data, address access issues, and even help migrate your oldest data to archival storage or purge it altogether. Active storage management gives IT a doubled win by reducing both the expenses for primary storage and the ongoing expenses for the backup servers, storage, and backup archive space required to protect it.

The Nth Generation Team has the tools and experience to help analyze your current environment and recommend right-sizing now and when it is time to design your next virtualization environment. We’re looking forward to working with you.

Learn more: HPE Morpheus VM Essentials | VMware Alternative | Nth Generation

Request a complimentary analysis, assessment or virtualization cost savings calculation: https://www.nth.com/hpe-request