Skip to main content
Question

AuxCopy to tape performance

  • July 15, 2026
  • 2 replies
  • 25 views

Forum|alt.badge.img+4

I am currently questioning our AuxCopy to tape performance and want to get some input on how to identify the bottleneck.

We have multiple plans which all have two copies to tape. One is synchronous and one is selective.
Each of the two is consolidated onto another Global Secondary Copy Policy.
The storage target are six LTO9 drives with a mixed LTO8 and LTO9 media inventory.
The source for the copy is a NetApp StorageGRID S3 backed storage pool with deduplication connected on a 100G fiber network.
We have two Media Agents which are both connected to all LTO drives and the NetApp (same VLAN).

Looking at the reported speed of the drives in Commvault I get around 130MB/s all of the time.
I could confirm this looking at SFP speeds on the FC switch which also stay around this value with rare peaks to around 200MB/s.
What I tested already to look for the bottleneck:
- only run one AuxCopy Job
- run AuxCopy when no other jobs where running
- validate drive > 300MB/s throughput with LTO8 media
- different multiplexing settings: none, 5, 10, 20
- no Hardware compression

Total throughput with all drives is > 1GB/s but an individual drive always stays in the mentioned speed range.
All the testing in my opinion points to a bottleneck in the AuxCopy process itself.

Furthermore I wanted to look into the Performance Analysis Tool but unfortunately every job where I looked at the PerfAnalysis.log it was empty.

Is it really just a limitation of the software or could there be more to it?
I know this is a complex topic but maybe there are some settings I am missing out on here.

2 replies

Forum|alt.badge.img+6
  • Vaulter
  • July 16, 2026

Hi ​@Marcel_RE ,

 

First, let's verify that the Performance Analysis Tool is being used correctly, as it is the primary and documented method for identifying AuxCopy performance bottlenecks.

Please note the following prerequisites:

The AuxCopy job must be completed or suspended (it should not be in a running state).

The job must have run for at least 10 minutes before the analysis can be performed.

The Performance Analysis Tool can be executed in either of the following ways:

Automatically by selecting "Send Logs for a Job" from the CommCell Console.

Manually from the MediaAgent command line.

Once the analysis is complete, the generated report will be available in the MediaAgent log directory.

For detailed instructions on using the tool, please refer to the following documentation:

 

https://documentation.commvault.com/2024e/commcell-console/performance_analysis_tool.html


And we need the review the performance analysis log to conclude the exact cause of bottleneck.

Regards,

Karthik


Forum|alt.badge.img+4
  • Author
  • Apprentice
  • July 16, 2026

Hi ​@kkarthikk ,

thanks for the explanation. I did end up with output in some PerfAnalysis logs at least after suspending the jobs.
The other ones only have the phrase “Could not read log  file or Performance counters for JobID X not found in the file” in it.

The feedback so far was “Reads from the source are slow” and “Network transfers are slow”.
Recommendations where to increase DataMoverLookAheadLinkReaderSlots,  which were already at 2048, but I increased them to 4096 and noticed a bit of a speed bump but need to confirm this with more examples.

For network transfer “nNumPipelineBuffers” was mentioned which already is set to 900 in my case. Could it make any difference increasing it further to 1020?

 

Are there any other parameters which make sense tuning for better tape throughput in our scenario besides  multiplexing?

Does chunk size make a difference here?