If you’re an IT leader who’s currently planning a database infrastructure refresh, you’re likely trying to balance competing demands. On the one hand, growing datasets and the rise of memory-intensive virtualized workloads are demanding more hardware capacity and performance than ever before. Fully populating a server with dynamic random-access memory (DRAM) can help you deliver the highest levels of performance. 

On the other hand, memory-related hardware costs are continuing to surge. Buying less DRAM keeps costs down, but it also can limit how many virtual machines (VMs) and database workloads each server can host. You also may want or need upgraded memory, but you don’t want to pay premium prices for it. 

VMware Cloud Foundation (VCF) 9.1 memory tiering on Intel Xeon 6 processor-powered servers offers a middle path. Memory tiering, controlled at the hypervisor level, combines DRAM and a dedicated NVMe SSD into a single logical memory space. VMware ESX keeps frequently accessed memory pages in DRAM and moves less-active pages to the NVMe tier.  

If you’re weighing memory configurations for a server refresh, the practical question is whether memory tiering can deliver the database performance you need while spending less on DRAM. We put the technology to the test to find out. 

How we tested VCF 9.1 memory tiering performance and value 

We tested three memory configurations on the same Intel Xeon 6745P processor-powered Dell PowerEdge R770 server:  

  • 1TB of DDR5 DRAM 

  • 512GB of DDR5 DRAM 

  • A VCF 9.1 memory-tiered configuration: 512GB of DDR5 DRAM plus 512GB of capacity from a 3.84TB NVMe drive  

On each configuration, we ran the HammerDB TPROC-C online transaction processing (OLTP) workload across multiple database VMs, measured throughput in new orders per minute (NOPM), and compared results. We calculate value with estimated memory-related hardware pricing from Dell. 

Key findings from our testing 

The 1TB DRAM configuration delivered the highest transactional database performance in our testing. However, VCF 9.1 memory tiering on Intel Xeon 6 processor-powered server with 512GB of DRAM and an NVMe SSD delivered 85 percent of the 1TB DRAM configuration’s performance at a lower cost, and it supported twice as many VMs as the 512GB DRAM server. Here’s how the memory-tiered configuration compared to both the 1TB DRAM and 512GB DRAM configurations across several key metrics: 

Memory-tiered configuration vs. 1TB DRAM configuration 

  • 35 percent lower memory-related hardware costs 

  • 85 percent of the throughput* 

  • 6 percent more NOPM per total server acquisition dollar 

  • Same number of database VMs 

Memory-tiered configuration vs. 512GB DRAM configuration 

  • Twice the number of database VMs 

  • 50 percent higher total NOPM 

  • 85 percent CPU utilization vs. 51 percent, putting more of the server’s processing power to work 

*Results may vary depending on configuration. 

What this means for IT teams 

Teams that need strong database performance but want to control memory spending have a practical option in VCF 9.1 memory tiering on Intel Xeon 6 processor-powered servers. Compared to the 512GB DRAM configuration we tested, the 512GB DRAM plus 512GB NVMe memory-tiered configuration we tested supports more VMs and more throughput for only the added cost of the NVMe drive. It also delivers better value in NOPM per dollar than the 1TB DRAM configuration we tested. 

For more in-depth information about our VCF 9.1 memory tiering performance and value tests, check out the detailed report below.