Accouting lsf : problem with max memory

Wen Chao MH Hao asked me to create a RFE for a bug in the accounting lsf.
We have a hot-fix to calculate the maxRMem to a correct value when the value exceeds a value (put to 0). But we don't have anything when the value is negative.

From: Wen Chao MH Hao/China/IBM
To: EP HPC-N3
Cc: Jean-Remi PONTVIANNE
Subject: RE: [TS001253795] Question about accouting lsf
Date: Fri, Sep 7, 2018 8:15 PM

Hi Jerome,

In the hot fix, we read the original value of maxRMem which overflow and convert it to the correct value. Here is the logic in code level:

When maxRMem in KB < 0, firstly, change the int negative value to a double negative value;
secondly, change the negative value to a positive value by (double). For example, a is the negative value, change it to b, b=(double) 2147483647 + (double) (2147483647 + a) + (double) 2.
thirdly, change the double value from KB to MB(by (double) value / 1024);
fourthly, change double value to int value.

After investigation, if we want to log a positive value (no overflow here) in this situation, there is no good way to handle it except changing the data type. When we change the data type from int to double (or long), it will bring compatibility issues. Therefore, I suggest
1. You can read the current maxRMem value in lsb.acct and handle it to a correct value. The logic below is a reference.
2. For the negative value in lsb.acct, you can log a RFE. R&D will follow it and handle this project in main release.

Feel free to let me know your comments and concerns. Thank you.
Have a nice weekend!

Regards,