After upgrading to 6.5.0, why is search head cluster skipping about 50% of scheduled searches?

[rbal_splunk]

On 10 Node SHC deployment – post upgrade from 6.2.5 to 6.5.0 system, instance is skipping about 50% of the scheduled searches.

[rbal_splunk · Accepted solution]

This issue has been resolved and the following steps were taken to debug and resolve the issue:

1)Observation 1: The following search showed that SHC members the delegatejob was taking up to 200seconds

Index=_internal source=*splunkd_access.log delegatejob | stats avg(spent) by host

2)Observation 2: It was seen that on SHC members ps -ef | grep splunk | search, a lot of Splunk launcher jobs were hanging.

To resolve the issue, the following changes were implemented:

On All SHC members implemented:

$SPLUNK_HOME/etc/system/local/limits.conf

[search] 
search_process_mode = traditional 

After above changes were made the skipping searches frequency has dropped significantly

[onthebay]

What is the reasoning behind why this resolves the issue? What has changed with the auto setting in 6.5?