Why you don’t want to use Storage DRS

Why you don't want to use Storage DRS Dell EMC FAST Fully Automated Storage Tiering
Dell EMC FAST Fully Automated Storage Tiering

Well maybe you should still use it but here is an explanation of what it does and why you may want to re-consider enabling it for the sake of using all the features.

Intro

Storage DRS or SDRS is a feature which is enabled on a datastore cluster to automatically, distribute VM’s between datastores in the cluster based on either: datastore I/O latency, and capacity utilization. It will also provide an initial placement recommendation for new VM’s.

Now I know what you are thinking, all three of those sound great and you want them all.

Most of the customers I work with use Dell EMC Enterprise Hyrbid Cloud in which datastore clusters are currently not supported, so I often end up going through the use cases with customers and determine that there is little to no benefit of using SDRS, and sometimes even a detriment.

Use Case 1: Initial Placement

The initial placement of a VM is one feature that is quoted as being useful in SDRS, however if you are using vRealize Automation (an integral component of EHC) then vRA will take care of this for you.

If you are still using vSphere, I would strongly urge you to look at automating your environment.

Use Case 2: Datastore I/O Latency Balancing

The datastore I/O latency re-balancing feature will move workloads that are I/O intensive based on I/O latency to another datastore within the cluster.

This is a great feature if you have dumb storage, but most of the storage enterprise customers I work with have uses tier’s of disks to ensure that “hot” data (data that is frequently used) is moved to the most expensive flash storage, and “cold” data (data that is less frequently accessed) is moved down to SAS, or SATA storage.

Dell EMC FAST: Fully Automated Storage Tiering

Dell EMC FAST: Fully Automated Storage Tiering

The problem you introduce by having vSphere move disks is that as the disk is moved from one datastore to another it will appear as though this data is hot, and therefore will live in your more expensive disk till the array realises it is cold and starts to move it from cache, to SSD, to SAS, to SATA.

All the while this process will be using up the more expensive and better performing disks that should be used by applications that need them.

While this needs to be reviewed with the storage that you use, it is worth keeping in mind before blindly enabling.

Use Case 3: Capacity Utilisation Balancing

The capacity utilisation balancing feature will move workloads from datastores that are nearing the capacity limit set to a datastore with more free space.

This is one feature that is worhwhile enabling SDRS for if you are over allocating storage at the vShere level. However vCenter server has a number of alarms out of the box, and proper monitoring and capacity management of your environment should prevent a datastore from ever getting close to 90%.

With the smallest datastores I often being 2TB and the average being higher, there are bigger problems if you are consuming storage at a rate that could fill 200GB without anybody noticing.

Conclusion

So maybe you don’t want to use Storage DRS after reading this, or maybe you have a number of very valid reasons to use it, but please don’t enable it without doing your homework.

1 Comment

  • Sijith Maniyappan says:

    Ben, Good article. I had the same question this week. If we enable SDRS, you may also have to align SDRS clusters matching your Business Group reservation policies in vRA, which could end up having many SDRS clusters if the business group count is large.

    Also, there are restrictions when it comes to SDRS and VPLEX (Metro or local) like the manual mode or forming clusters using site affinity.

    More I think about it, it makes little sense to use SDRS in a Cloud environment using vRA.

Leave a Reply

Your email address will not be published. Required fields are marked *