tiero - Fotolia
The amount of flash that is recommended for use with hyper-converged systems is largely dependent on how the architecture will use that storage. It also depends on how sensitive the organization is to a decrease in performance if data needs to be recalled from disk instead of flash -- also referred to as a cache miss.
Almost every hyper-converged architecture uses flash in some fashion. Some create an aggregated pool of flash from internal server resources and present it as a virtual volume. The downside to this is that latency is introduced as the individual drives are stitched together across the virtual clusters network. Other hyper-converged products have a dedicated flash resource in each server. Part of their intelligence is making sure that a copy of the virtual machine’s (VM's) data is on the flash storage inside the server it is running on. This allows for very fast read performance because there is no network latency. It creates a more complex software offering, but the user is not typically impacted by that.
As described above, flash can be aggregated and used across all the servers in the environment or it can be dedicated to particular servers within the cluster. In most cases, the dedicated use of flash inside the server is almost always temporal in nature. In other words, it acts as a read cache. All writes and any non-cached read requests are sent to an aggregated volume that typically consists of hard disk drives. Hyper-converged systems that aggregate the flash capacity inside of servers can use that as a static tier or as a cache.
If flash is used as a static tier, the math is simple. The flash capacity has to match the capacity requirements of the VMs that will use that tier. If flash is temporal in nature, meaning that it is used as a cache or a storage tier, the math is more complicated. In general, the flash capacity will be a small percentage of the overall capacity, typically 5% to 10%. The downside to the temporal use case is what happens to performance when there is a cache or tier missing and data has to be delivered from the hard disk tier. For many applications, that occasional loss in performance is not a problem; for a few, though, it could be a disaster. Knowing the applications' sensitivity to performance loss is a critical determining factor to deployment.
If performance is of concern and flash is being dedicated to each server so VMs can have read requests responded to from local flash, the size of that server flash should be equal to the size of the VMs that are to be stored on that server.
How hyper-converged systems affect storage provisioning
Hyper-converged options get appliance-like feel with reference architectures
Dig Deeper on Hyper-Converged Infrastructure Implementation
Related Q&A from George Crump
Shadow IT means enterprises are at increasing risk of cloud data loss, but providing employees with comparable file sharing apps can help. Continue Reading
According to analyst George Crump, you might want to think about going with a non-traditional Hadoop architecture. Continue Reading
Cloud storage doesn't just have to be for backup. According to George Crump, cloud services can make deploying a new application or disaster recovery... Continue Reading
Have a question for an expert?
Please add a title for your question
Get answers from a TechTarget expert on whatever's puzzling you.