jhammond / lltop Goto Github PK

I've been running LLTOP on a production Lustre system, and I noticed that every once in a while there are extremely high values appearing, I wonder if there's some counter which overflows or something alike. This is an example output, we run LLTOP with an interval of 50 seconds:

CLIENT-or-JOBID WR_MB RD_MB CLOSE XRENAME GETA GETXA LINK MKDIR MKNOD OPEN RENAME RMDIR SRENAME SETA SETXA STAT SYNC UNLINK OWNER NODES
1495209 138110882 54 256 0 1538 80928 0 0 192 283331 0 0 0 0 0 0 657 1009 someuser 62
1495165 111456 2 310 0 1732 0 0 0 0 310 0 0 0 0 0 0 1939 0 someuser 62
1495208 93836 3 260 0 1254 0 0 0 0 8540 0 0 0 0 0 0 722 0 someuser 52

This would mean that with the job in the first row, the user was writing at 138110882/50= 2,3TB/s ,which is not possible on out system.

Do you have any idea of why this is happening?