There are arguments for hyper-converged architecture and arguments for the traditional, separate storage and compute devices architecture.
Advantages of hyper-converged infrastructure:
– provides slightly better total cost for small configurations (e.g up to 5-10 hypervisors), because you don’t need to buy 3 storage nodes (circa $7k) in addition to the hypervisors
– single building block concept – simpler to extend because there is just one type of server
– simpler for hardware/procurement
– preferred by customers with smaller implementations or use case which has small IT needs in many locations, think ROBO (Remote-Office/Branch Office)
Disadvantages of hyper-converged infrastructure:
– HCI makes sense at a smaller scale – the cost savings of HCI are good for smaller sized deployments. For larger deployments – tens of computer servers and tens of terabytes of storage – it may be better to have computed and storage sub-systems segregated, as they can be sub-optimized independently to unlock saving from the economy of scale, to simplify deployment or to allow for better fit to a given situation/limiting factors.
– Out-of-the-box HCI solutions are a black-box – this means increased vendor lock-in and lack of transparency of what is in the box. For the prime use cases of HCI this is not an issue but may become a limiting factor for larger deployments or more sophisticated IT users, which require control over what building blocks are being used.
A few words on the other approach “segregated” or traditional separate compute + storage:
– better total cost for large configurations, due to rack-level optimization and server specialization
– simpler for software deployments, configuration, and operation
– Fewer issues overall, because VM-specific issues are isolated from storage-specific issues
– Sometimes segregated approach is used due to “best practices” – e.g. if a particular cloud management solution has the tendency to restart servers often, then it makes sense to decouple compute and storage
– Preferred by customers for larger implementations, where a 3-5 server distributed storage cluster provides shared storage to 15+ hypervisors
In summary: both approaches have advantages and disadvantages, which depend on the use case and the size of the deployment. We see companies going with both hyper-converged infrastructure and the traditional architecture. From our customer base, the split is about 50/50.
In some cases (usually even larger deployments) the best approach is to go with a mixed deployment model, which we call “semi-converged”. In the semi-converged model, hypervisors hold SSDs and HDDs are in separate storage nodes. This further simplifies the deployment and provides an additional level of cost savings due to specialization.
How do you run your IT infrastructure – converged or segregated? Tell us in the comments.
If you have any questions feel free to contact us at [email protected]
Share this Post