#Whether FlexGroup volumes are the right fit for our workloads/environment

1 messages · Page 1 of 1 (latest)

pearl oak
#

We have total of ~1, 000 NFS volumes, 400 of them are used for VMWare datastores. 40 volumes out of total are with the size of 10TB-40TB as the maximum, and among those, a large chunk of space has been used by SnapShots. Sometime we hit performance issues without knowing obvious causes. We are just starting to look into FG volume. My questions:
What volumes/workloads should be converted to FG?
Should we convert those whenever they hit the performance issues?
If we adopt a policy to convert all those 40 volumes with the size of 10TB-40TB to FG's, would that be a good one?

frank depot
#

Few Reasons I see people using Flexgroups -

high file counts
Going above 100T or 300T flexvols.
high performace (scalling across multiple nodes)

I'd do some investigating as to what the perf issue is, if it's maxing a single volumes throughput, then yes, it's worth looking in to FlexGroups.

tepid coyote
#

Same.

#

There is a great Perf resolution guide which can guide you through troubleshooting the exact same way a perf TSE would.

fallen prism
#

In VMware I prefer using storage clusters rather than flexgroups. Why? File placement. You have no control for file placement. If your groups get lopsided, that doesn’t help. Plus your members are already payin possibly smaller. I just put one or two volumes on an aggregate per node and then add them to the storage cluster and disable auto storage drs. Then when you migrate into the storage cluster, VMware decides which volume to put things on and you know where everything is living. Need to move it to another volume? Do it, just check the disable storage drs box and pick where.

pearl oak
#

I am not looking for a solution on troubleshooting the performance.
@frank depot A couple of follow-ups:

  1. For NFS volumes or datastores with the size of several TB's or even 10-40TB's, if there have been no performance concerns, should we convert them to FG volumes and why? I am just not sure of how much performance gain that we can have.
  2. For those with an unknown cause in performance issues, but only with several TB's, should we convert them to FG volume in order to improve?
  3. A FG volume can utilize resources on multiple nodes, but, it is not multipathing, and I/O processing would still be one at a time. Correct?
  4. How high file counts are too high and need to convert to FG?
#

@fallen prism Not quite understand you. But, how FG volumes get lopsided? What chances are when one volume/member gets too hot, and the others get too idled since it is evenly distributed among multiple aggrs/nodes? If that was what you meant

fallen prism
#

Whole files get placed into FGs. It’s possible that one FG will fill up before orders. There is some logic in ONTAP to prevent or reduce this but it does happen.

Think of this: I place a 50g VM on the FG. It’s placed in fg member 1. I then expand the vm to 2t for whatever reason. I’ve possibly just filled the FG.

Don’t get me wrong. I love FlexGroups. I just think the use inside VMware is not a real good case especially if you have the ability to use storage datastore clusters.

#

And VMware will likely never hit the counts for inodes. It just doesn’t use a lot of files. So placement will be on capacity and to some extent I/O. But if VMs grow, members will fill up unevenly causing ONTAP to possibly auto grow and/or use the elastic sizing to adjust members.

pearl oak
#

@fallen prism I am sorry, but there some basics I don't understand here, since I am new to FG, and don't know how "storage datastore clusters" works.

  1. If you place a 50g VM on the FG, won't it distribute across all members in the FG? If I have 5 members, then 10g on each, not 50g in member1, right?

  2. But if VMs grow, members will fill up unevenly
    I understand members may not fill up exactly evenly, but, in general, the way it works should be almost evenly, right?

  3. I don't know how "storage datasgtore clusters" works, but would a VM get migrated to a different NetApp storage cluster node or datastore?

fallen prism
#

A single file is never split.

#

If a single 50g file is placed in the fg, it will be in exactly one member. If I grow that fine it will be expanded in that member only

#

Storage data store clusters are a VMware construct. I would NOT suggest placing fkexgrouos into the storage cluster. Only single volumes. Volumes can be in any node in any cluster.

pearl oak
#

I am starting to understand you now. So, it sounds not recommended to use FG for VMWare datastore.

Please confirm one more thing with me. Can "storage datastore cluster" allow you to migrate VM's to different NetApp cluster node or different datastores by using DRS? I previously thought it can only allow you to place VM's among ESXi hosts?