Solved

VMWare Backups and Maximising Disk Performance with a HSX Storage Pool

  • 17 May 2024
  • 2 replies
  • 34 views

Badge +1

Hi All,

We have a HSX Storage Pool that completes backups for an OnPrem VMWare Environment. In this scenario, our HSX Environment completes backups through SAN and HotAdd Proxies.

When running a Subclient backup, VMs will use their perspective proxies, though 1 HSX node is always the Media Agent for all child VMs jobs.

Does the above imply that data written to the Hedvig Library is only using that HSX Node’s physical disks for the first point write? I may be fundamentally misunderstanding the storage technology, though my assumption is that each node would write to its local disk with the hblock process handling it from there. In this scenario, are we therefore only using a % of the total spindles for direct writes, or is that simply not how Hedvig works that this is not a concern?

In the same line-of-thinking, would this also apply to the Network data flow and is all traffic is being routed through the 1 node, and therefore limiting the maximum throughput for a single subclient.

If the above is correct, is the solution to scale on a subclient-level (instead of say the # of data readers) to improve parallel operations?

icon

Best answer by Emils 23 May 2024, 03:59

View original

2 replies

Badge +1

Hi @Emils,

Thanks for taking the time in your detailed response. Most appreciated!

Userlevel 4
Badge +12

Hi @Gareth Faichnie 

Thanks for reaching out.

I’ll break this question down to 2x parts and answer them individually

  1. When writing data to a single MA in HSX cluster is the data distributed?

Yes, when the MediaAgent sends the data to the HSX library, it written via the NFS mount and not directly to the individual disks, the NFS mounts talks to its hblock neighbours to see where to write the data, the data is then sent to the hblock nodes and each node will commit the data in the most efficient way possible, ensuring the data is distributed.

When the data is written it will be sent to the nodes via the Storage Pool network on all nodes.

  1. Why does a subclient only use a single MediaAgent during backup even though there are multiple VMs/Streams?

This is by design and recommend running VM backups across multiple subclients for alternate MAs to be used .

 

Reply