Andrea Danti - Fotolia
Solid-state drives have quickly become a staple for hyper-converged systems. Even so, workloads running on a hyper-converged infrastructure can vary widely from one another.
You should consider the type of workload the hyper-converged system will be running, and then match the workload to the type of storage that will be most beneficial based on the way the workload uses the hardware. That's why it is important to know the particular solid-state storage characteristics and how they affect hyper-converged infrastructure (HCI) systems.
One of the first issues you will need to consider when evaluating how solid-state storage characteristics can affect your workload is the capacity requirements of your HCI system. While it is true that multi-terabyte SSDs are available, high-capacity drives such as those in the 100 TB range still have a high price tag, which means SSDs are not the best choice for high-volume storage.
Even if multi-terabyte SSDs fit your budget and meet your capacity requirements, durability is a critical issue. Unlike hard disk drives, SSDs allow for a limited number of write cycles before individual memory cells begin to fail. As such, all-flash storage probably isn't going to be the best choice for write-heavy workloads, such as continuous data storage from internet-of-things sensors.
If you do install high-capacity SSDs into your hyper-converged systems, then make sure those SSDs are enterprise-rated. One benefit of HCI is the hardware adheres to a strict reference architecture set forth by the vendor. Even so, some vendors may allow SSDs based on triple-level cell or even quad-level cell drives, sacrificing durability in order to allow for greater storage capacities.
Performance one of the primary solid-state storage characteristics
The primary reason for installing SSDs in HCI systems is performance. Generally speaking, SSDs offer far higher levels of IOPS than HDDs, especially when it comes to nonlinear read and write operations. In spite of SSD's reputation for being fast, those who require their hyper-converged systems to deliver high-performance storage I/O should consider whether SSDs are the best choice.
One advantage of solid-state storage characteristics is SSDs commonly use SATA or SAS controllers, just like HDDs. This makes it easy to replace an HDD with an SSD whenever it is time for an upgrade. However, this backward compatibility is something of a double-edged sword.
Some SSDs are theoretically capable of delivering far more IOPS than storage controllers such as SATA and SAS can handle. As such, the disk controller itself can become a bottleneck.
One way of improving the system's overall storage performance is to use an alternative form of flash storage, such as nonvolatile memory express. NVMe storage does not attach to the server through a legacy disk controller, but rather through the PCI Express (PCIe) bus. This allows NVMe storage devices to deliver data more quickly than would be possible through an ordinary storage controller.
Of all the solid-state storage characteristics specific to NVMe, two things stand out that you must consider prior to using NVMe storage in an HCI system. First, you will have to determine whether the vendor's reference architecture allows you to use NVMe storage. Second, you will need to determine how many PCIe slots are available within your nodes and if the available slots will be able to accommodate an adequate number of NVMe storage devices.
Taking advantage of tiered storage
Each vendor has its own approach to storage tiering, but in many cases, the high-speed tier acts as a read and write cache. Recently accessed data is copied to the high-speed tier as a way of improving the performance of subsequent reads. Likewise, the high-speed tier may act as a temporary repository for newly written data, allowing the system to write data quickly, and then transfer it later to HDD storage at a more convenient time.
Although not every vendor allows it, consider creating a fast tier that consists of several SSDs. One way to take advantage of solid-state storage characteristics is by striping data across multiple SSDs. This further improves performance, while also distributing wear across multiple drives, increasing the usable life of the SSDs.