A common assumption is that flash storage in a hyper-converged platform delivers much higher performance than traditional...
hard disk drives, but at a significantly higher cost. There is also an assumption that in order to mitigate that cost a mix of flash storage and HDDs must be utilized. It is here where the flaw originates. Flash storage performance advantages are true, but the cost disadvantage is no longer a given, especially when comparing flash storage to 15,000 or 10,000 rpm high-performance HDDs.
Another faulty assumption is that all flash storage is created equal. There are huge differences in performance and cost in the different types of flash storage including: DDR4 DIMM flash storage, PCIe flash storage, NVMe flash storage, SAS flash storage, SATA flash storage, write-optimized flash storage and read-optimized flash storage. This doesn’t even take into account the differences in underlying flash NAND technologies from various flash vendors.
Flash storage is a tool to manage performance in a cost-effective manner. It is generally less expensive in all its forms than memory and generally more expensive than much slower HDDs. But that second generalization is changing rapidly.
The reality is that every hyper-converged platform is different. There is no single, omnipotent type of hyper-converged infrastructure. Remember that hyper-converged infrastructure is hardware, hypervisor, software, storage and networking integrated in a single turnkey package. Every vendor’s configurations and options are limited to their tested certification matrix and what they are willing to support. Even the software-only hyper-converged infrastructure vendors support only specific hardware configurations. In general, hyper-converged infrastructure vendors require all server nodes in the hyper-converged infrastructure cluster be configured with the same hardware specifications.
So what about the flash storage? Since every vendor is different, determining how much flash is necessary in a given environment should be boiled down to a few general steps:
- Determine all of the hyper-converged application workload performance and capacity requirements. Performance is measured as the desired and maximum allowable response time and throughput. Capacity planning is based on the amount of storage required over the life of the application or hyper-converged platform.
- Determine the budget that is allocated for the hyper-converged infrastructure project. This includes the budget for acquisition, implementation, professional services (if required), maintenance, operations, upgrades, software license subscriptions, power, cooling, fixed overhead allocation, rack space, floor space and administrators.
- Find out how each hyper-converged infrastructure system will meet the performance and capacity requirements of the project for the life of the project. Some will be able to meet requirements with a combination of flash storage and HDDs. Others will be able to meet requirements without HDDs but with a combination of different types of flash storage. And still others will be able to meet requirements with a single type of flash storage. For each vendor, compare the hyper-converged infrastructure’s quality of service. QoS is the percentage of time that their hyper-converged infrastructure configuration will consistently meet the performance requirements of the application workload. Mission-critical application workloads will have the most stringent QoS requirements.
- Compare costs. Comparisons should be based on total cost of ownership (TCO) and effective capacities. TCO includes all the costs previously mentioned. The high storage high densities, low power and low cooling costs of flash storage make it more cost effective or cost-equivalent to HDDs, depending on the type of flash storage utilized. The way that each hyper-converged platform manages effective storage capacities is unique as well. Some replicate the data multiple times to provide access to all data on all nodes even if a node fails or is removed. Many utilize deduplication and compression to expand the total usable storage in the hyper-converged infrastructure cluster. Use of zero-capacity snapshots is also an important factor in calculating cost. Non-zero capacity or thinly-provisioned snapshots can exacerbate hyper-converged platform costs.
After narrowing down the choices to two, test them both. Best way to test is to utilize a software or appliance tool such as Load DynamiX to closely emulate the loads that the production environment will actually see. Best practice is to test from 1.5 to 2.5x more than the expected loads to make sure the hyper-converged infrastructure will meet unexpected needs. You will then be ready to make a selection based on how well the hyper-converged infrastructure meets the criteria in performance, capacity and TCO.
Using hyper-converged platforms and existing hardware
Hyper-converged offerings creep into SAN and NAS market
Major players in the hyper-convergence market
Making the decision to move to hyper-convergence