HCI and Database servers

There are constant infrastructure technology changes. There are regular infrastructure changes - for various reasons , such as hardware refresh , hardware not supported by vendor any longer , data centre simplification , changing data centres , on-prem versus off prem.      

What  remains permanent is the database servers that remain on the OS. So it's incumbent upon the DBA to approach the problem  with a risk analysis mindset. One of the first is start to understand the underlying infrastructure. 

Hyper Converged Infrastructure (HCI) utilizes storage residing in servers and distributing the storage treating it like a storage array. This consolidation of storage and server layers is one the main HCI principles

HCI  is one of those technologies promising untold benefits. So what questions should a DBA ask the infrastructure guys?  This will also feed into what sort of performance\acceptance testing you'll need to perform. 

These are some starter questions for HCI and database server

1)  Latency - each block of  data is mirrored or copied across the network to other nodes. What is the latency penalty? How can we test the latency at different workloads , and also differentiating between read and write latencies ?

2) Predictability -  each IO operation  flows through the CPU scheduler - both IO and acknowledgement  phases. If the throughput is heavy how will the SHARED cpu cope - even if CPU reservation is established ? The scheduler will still need to wait for another vCPU to finish.  It's easy to see that if the SHARED spu is flooded, response times will depreciate

3) IOPS performance - Flash disk arrays are dependant on CPU horsepower. If cpu is limited per node , will this throttle performance? 

4) Licensing - On a node how many CPUS per node are dedicated to storage? If you have a licensing agreement based on cpus or cores per node and some of those cpus are being taken up by storage resource management - your licensing costs are still based on all the CPUs for the whole node. SQL Server licensing is based on the cpu core count for the whole node . If cpu count is used for storage management - will you need extra  nodes to fulfill the resource requirements - will extra cpu be required? 

5) How does HCI deal with  RAID-levels and data protection , usually  assign per LUN?  Different vendors have different processes . VMWare Storage Policy Based Management (SPBM)  allows you to  assign Failure to Tolerate (FTT) values to server objects. Options range across : Mirror (RAID-1), Dual Mirror (2x RAID-1), EC-1 (RAID-5 like) and EC-2 (RAID-6 like) . Read more for RAID levels overview SPBM can assign per component like vmdk , per VM or even per drive - the idea is to be storage efficient - whilst maintaining storage performance

6)Data at Rest Encryption - Confirm it works with dedupe and compression .vSphere encryption or vSan Encryption

This is just a quick overview of questions. In future - I'll add more discussion around compression, benchmarking and performance 

--