Since the dawn of the storage, Application IOPS requirement discussion has been very critical and heated up. May be it is time to end IOPS discussion for all applications expect for mission critical. Enterprise should be consuming resources whether it is public or private cloud than worry about whether they are catering for enough IOPS or not. This article will focus on how to end IOPS Discussion in Private Cloud with the help of SSDs.
Each year compute resources capabilities are being increased. This is enabling higher consolidation ratio in single node. But storage has not seen that kind of consolidation nor performance increase. IT administrators are quickly able to provision additional CPU and RAM but almost always get stuck with storage provisioning. One of the main reasons for delay is the calculating IOPS and expansion of the storage. Different application characteristics and requirements only makes it more difficult.
What is the solution for this kind of heterogeneous IT workload environment?
In my opinion, Hyperconvergence is the answer to address various requirements of heterogeneous environment.
But does hyperconvergence really works well without SSD in a heterogeneous IT environment?
May be there are some good reference there. But as more and more applications start to be deployed – we again start to bring same old discussion of disk performance and IOPS available.
Some of the benefits of SSD in Hyperconverged Private Cloud are:
1) Easier to deploy and configure (no need to do complicated cache / Tier configurations)
2) Easier to manage (single tier for all the VMs)
3) Remove any latency from the environment which could have been there because of disk performance.
4) Better availability as rebuild time is much shorter compare to HDD.
With 100% SSD, majority of the workload can be easily migrated into Hyperconverged farm. With many benefits there are also some drawbacks. If there is requirements for increasing a particular resource from CPU, Memory and Storage, it may not be possible without increasing others.
A mix approach can be taken to resolve the inter-dependency of compute and storage resources.
a) Hyperconverged nodes for virtual machines and its associated storage.
b) A Network Attached Storage (NAS) to address the demand of urgent storage requests and large data files. (This NAS can consists of both SSD and Magnetic Disk. )
Network Recommendations on High Performance Hyperconverged SSD Farm:
With so much IOPS available in a single node, network throughput and latency becomes very important. Don’t ignore the network switches. 10GbE switches are must. If have a chance for creating a new Hyperconverged farm from scratch – go for 40GbE or 100 GbE Switches. Each Hyperconverged nodes must have minimum 4 units of 10GbE ports, more the merrier.