In modern virtualised environments, memory often becomes the constraint before CPU does. That problem is more challenging today as memory pricing remains volatile and supply less predictable, driven by AI-focused hardware demand. Expanding DRAM is no longer a simple upgrade decision. It carries real budget, procurement and planning impact.
There is also an indirect domino effect on licensing and even sustainability. In many environments, key infrastructure software is licensed per CPU core, so when memory limits VM density, licensed CPU capacity sits underutilised. That inefficiency can inflate software licensing costs and, over time, contribute to a larger infrastructure footprint than necessary, with flow-on impacts to power and cooling.
This is where VMware Broadcom’s Advanced Memory Tiering can make a real difference. It extends effective memory capacity using a more cost-effective NVMe-backed tier, helping improve VM density and overall platform utilisation.
This can deliver benefits like:
- Higher VM consolidation per host
- Stronger CPU and licence utilisation
- Lower memory-driven scaling pressure
- Reduced DRAM upgrade dependency
- Improved overall platform efficiency
What Advanced Memory Tiering Does
Advanced Memory Tiering introduces an NVMe-backed secondary memory tier alongside DRAM. The ESXi VMkernel continuously monitors memory page activity and makes placement decisions in the background. Frequently accessed pages remain in DRAM, while less active pages are moved to the NVMe tier to free DRAM for active working sets. When access patterns change, pages are dynamically promoted back into DRAM.
This is proactive tier placement managed entirely within the hypervisor, rather than reactive guest swapping under memory pressure. There are no guest OS changes and no application modifications required. From the VM’s perspective, nothing changes. The guest operating system remains unaware of where individual pages physically reside.
By default, tiering is configured at a 1:1 DRAM to NVMe ratio. A host with 1 TB of DRAM and 1 TB of NVMe can effectively support 2 TB of total system memory. Validated configurations extend up to 1:4 in appropriate scenarios, subject to workload behaviour and hardware capability.
Performance and Security
NVMe is not DRAM, so tiering is not a replacement for proper sizing. However, benchmark testing shows that for suitable workloads, meaningful density gains can be achieved with minimal and predictable impact when active memory remains within DRAM.
Because NVMe is non-volatile, security teams often ask what happens if a drive is removed. VCF 9 supports AES-XTS encryption for tiered pages, with each ESXi host generating random kernel-level keys. Encryption is handled within the hypervisor and does not require an external key manager. If a device is removed, tiered data remains encrypted and unreadable.
Workload Suitability and Ideal Use Cases
Advanced Memory Tiering is best suited to environments where memory is the constraining resource and CPU headroom still exists. It aligns well with VDI, test/dev and general enterprise workloads with lower active memory ratios.
It is less suited to consistently memory-hot or highly latency-sensitive workloads. When properly designed, it enables higher consolidation while maintaining predictable performance.
Want to Dive Deeper?
For detailed architecture, sizing guidance, performance validation and design insights, explore the official Advanced Memory Tiering posts from the VMware Cloud Foundation team on the VMware blog.
Browse Advanced Memory Tiering content on the VMware blog
I hope you found this helpful. Feel free to comment if you have any questions.
You must be logged in to post a comment.