What if you could extend memory capacity… without buying more memory?
Memory tiering in VCF9 gives architects flexibility in how they balance performance, density, and cost. The key design choice comes down to memory tiering ratios which dictates how much NVMe-backed memory is added relative to DRAM.
Each ratio serves a different purpose, and choosing the right one depends on workload behavior, risk tolerance, and financial goals.
1:1 Ratio – The Safe Starting Point
1TB DRAM + 1TB NVMe = 2TB usable memory
This is the most conservative and predictable configuration. The majority of active memory remains in DRAM, while cold pages are transparently moved to NVMe. For most workloads, the performance impact is negligible.
This ratio is ideal for teams adopting memory tiering for the first time or environments where performance consistency matters more than maximum density.
Think of it as: doubling memory capacity without doubling DRAM cost.
1:2 Ratio — Where Density Improves
1TB DRAM + 2TB NVMe = 3TB usable memory
This is where memory tiering starts to change infrastructure economics. More cold memory pages are placed on NVMe, allowing significantly higher VM density per host while still protecting hot working sets in DRAM.
This ratio works especially well for VDI, mixed workloads and environments full of oversized VMs that rarely use their full allocation.
Think of it as: fewer hosts, better CPU utilization, and lower cost per VM.
1:4 Ratio — Cost Optimization First
1TB DRAM + 4TB NVMe = 5TB usable memory
This is the most aggressive option and should be applied selectively. It assumes workloads have large amounts of allocated memory that are rarely accessed. As long as hot memory fits within DRAM, performance remains stable but your mileage may vary accordingly.
This ratio shines in test/dev, analytics, and highly overprovisioned environments where cost efficiency outweighs peak performance sensitivity.
Think of it as: minimizing DRAM spend on a secondary cluster while maximizing usable memory.
The goal is simple:
- Keep hot memory fast
- Push cold memory to cheaper media
- Stop designing for worst-case allocations
Are you leveraging memory tiering with VCF9 yet?
I would love to hear more about your experience so far!
