MSA 2060 best practice - Volumes

[asahi]

We have a new MSA 2060 10GB DAC, with (24) 2.4 TB 10K drives., configured for iSCSI.

And we have (3) Esxi 8 servers configured in a cluster. Each server uses (3) 10GB ethernet ports configured for iSCSI to connect to the MSA. Everything is set to Jumbo.

The Best Practice Guide says use MSA-DP+ and create only one large Disk Group in Pool A. (The fewer Disk Groups the better. ) And use Pool B only as backup (just in case something goes horribly wrong with Pool A).

This kind of course contradicts what the MSA Storage Best practice says if you consider iops .... our current 2050 does over 300 iops, spread across A and B.

Should there be one large volume? Neither best practice document covers in-depth the best way to present volumes to ESXI hosts. Would one big volume of 42TB from the one Disk Pool be okay? Shared across three servers in a cluster? Or are multiple volumes more efficient?

[Vinky_99]

When it comes to setting up volumes on the HPE MSA 2060, the best practice is to follow the manufacturer's recommendations. While the Best Practice Guide suggests creating only one large Disk Group in Pool A, it's important to consider your specific requirements and workload characteristics.

Creating one large volume from the entire Disk Pool A can be a viable option, especially if you have a consolidated workload across your ESXi cluster. This approach simplifies management and reduces potential performance bottlenecks associated with multiple volumes. Sharing the large volume across the three servers in your cluster can help maximize storage utilization and streamline resource allocation.

However, it's important to note that workload characteristics can vary, and there may be scenarios where using multiple volumes can offer benefits. For example, if you have distinct workload profiles with different performance requirements, separating them into separate volumes can help ensure optimal performance and isolation between workloads. Additionally, if you have specific data protection or data management requirements that necessitate different volume configurations, multiple volumes can be useful.

Consider the following factors when deciding between one large volume or multiple volumes:

1. Workload characteristics: Evaluate the performance and capacity requirements of your workloads. Determine if they can be effectively managed and meet performance goals within a single large volume or if separation into multiple volumes is necessary.

2. Data management and protection: Consider any specific data management or data protection requirements. For example, if you require different backup or replication strategies for different subsets of data, multiple volumes may be beneficial.

3. Resource isolation: If you have critical workloads that demand dedicated resources or need to be isolated from other workloads, using separate volumes can provide better resource isolation.

4. Ease of management: Creating one large volume simplifies management, as there are fewer entities to manage. However, multiple volumes allow for more granular control and flexibility.

In summary, while the Best Practice Guide suggests creating only one large Disk Group in Pool A, it's essential to assess your specific requirements and workload characteristics to determine the most suitable approach. Consider factors like workload profiles, data management needs, resource isolation, and ease of management to make an informed decision.

Hope this give some insights. Let me know 

[ArunKKR]

Hi, I agree with Vinky_99's suggestions. As best practices guide says utilize Ninja online sizer. There is no significant performance improvement when single pool with 24 drives (DP+ disk group) is compared with 2 pools (12 drives DP+ disk group in each pool).

[ArunKKR]

Hi, We would like to know if the information shared addressed your query. Request you to please respond with a Yes/No You may also click on "Accept as a Solution" button if the issue is resolved.

[asahi]

Hello ArunKKR,

Neither response was a good answer. The first answer is basically impractical. The only true way to test the workload is to move everything over and see the results. There isn't a practicle way to simulate the workload.  If you move your VMs over and then it turns out bad, well then your stuck.  

And you can't use the Nija sizing tool because it doesn't give real world solutions. In fact if you play aroound with it and select the what appear to be the basic, or out of the box settings, it doesn't recomend the use of one pool.    it contradicts the "best practices".  I am still playing around with it.

In anycase, neither answers the question to use 1 large volume or not.

[Theo_dor]

what I have , and I guess what you want , is a single large volume with good performance
I use two drivegroups on controller A for a single volume
each dg is raid6 with 10 disks + 2 global spare drives + 1 ssd read cache

spreading drivegroups over both controllers may be faster but then you have two volumes , which is not handy

[JonPaul]

@asahi 
There are 2 parts to your question which seemed to get confused in the answers.
1. What will give the best performance from the MSA to a host given 24x 10k drives?
In this case the overall performance of the system will be limited by the 10k drives.  A single controller will be able to use ALL the performance of the spinning drives.  So the best option will be to put all the drives into the MSA-DP+ disk-group in a single controller/Pool.  This also gives you all the benefits of the MSA-DP+ RAID level and limits the capacity lost to overhead (SPARE and Parity) with a single disk-group vs multiple.
2. What number of volumes will give the best performance in VMware?
Frankly I don't know.  You may be able to find more information in a VMware forum.  With a random workload, which VMs tend to be, it should not make a difference on the MSA if you carve 1 volume or multiple volumes from a single Pool. It may be better performance to make 3 volumes with the intent that each server be a 'primary' on a specific volume and limit sharing a single volume/LUN.  But it may be better or the same performance using a single volume where all servers are accessing the same volume all the time.  VMware's locking mechanism does use ATS (Atomic Test and Set) so the locking should have limited interference, unlike older locking mechanisms which can time slice the entire LUN.
Many configurations look for the best performance in an easy to understand configuration and in that case, the single volume/LUN would be the way to go.