splunk tuning (tuned-adm recommendations for profiles/settings)

[t9445]

Hi, first (in-case anyone goes there, we are aware of the Transparent Huge Pages (THP) issue; and have insured they are disabled)

We have a large-scale deployment (currently 14 indexers, 2 search-heads, 4 distribution servers) and running CentOS 6.7, won't go into the hardware details (they are "big")

We are in the process of using tuned/ktune and are trying multiple configurations of tuned.conf, the available profiles as well as some of our own derivations.

However, before we spin our wheels much more - just wondering if anyone has a recommending tuned custom-configurations/default-profiles that you might recommend.

Indexers vs Search-Heads (dataless), etc (also, for a couple reasons we have not yet gone to clustered search-heads, unrelated at this point)

Any recommendations greatly appreciated

thanks

-tom

[seunomosowon]

Hi Tom,

I was just looking at tuned as well, but do share if you find some optimal values for this.

I'd recommend looking at the the sap-hana as a base, and adjusting any other configs you might find that makes a difference.
"tuned-profiles-sap-hana" takes some of its settings from the enterprise storage profile with the addition of some settings to disable THP.
Swappiness could definitely go lower as recommended in the comments - 10.

You should also turn off THP defray, as i've noticed a warning in mongod logs even if THP is turned off.
Also, look out for network errors, packets being pruned, which may signify that you also have to tune the network settings and increase the buffer a bit.

It'll definitely help if there was a profile for indexers, and another for search heads or other Splunk components.

• Seun