ERROR in the RFC communication when parabatch is set to more 3 (default)

Hi experts,

We have a 10 blades BIA system connecting to our BW Production. When we set the parabatch to more than the default 3 (we set to 4 or 5), We keep getting 'Batchpara Error' and 'Error in the RFC Communication: Error when opening an RFC Connection' during the initial index build. Here are what we do and what we see:

1. set parabatch to 5 and build 1 initial index (cube is 46 mil rows, 22 GB in size) - Failed with the above errors

2. set parabatch back to 3 and build 1 initial index (same cube as above)- successful but slow

3. set parbatch back to 3 and build 2 initial indexes simultaneously (16 mil rows/4GB and 6 mil rows/2 GB) - Failed with the above errors.

In the BIA index log I see the following error:

Error in the RFC communication: Error when opening an RFC connection (TREX_BIP)

Message no. STREX005

In SM37 in one of the logs I see the followings:

Parabatch Error

Any light you can shed on why we are encountering this issue will be much appreciated. Also, I observed that there is a long idle time in all systems (BIA, DB2, BW) and seemed like their respective resources were not utilized at all. Once in a while I do see surge of cpu/memory usage in all systems but not too terrible. I'm puzzled by why there is such long interval of idle time in all systems while the initial index build is going on.

BTW...we did the same thing (set parabatch to 5, and run parallel initial builds) in our non-prod BIA (2 blades) and QA BW landscape and do not have any issue at all.