DB2 LUW : Why load failed with SQL0958C even though the free space is enough

Body

I have a DPF server with partitions on multiple machines.  I find load from one machine always failed, but not from other machines.

Checking db2diag.log, I see error saying disk full:

2017-01-19-12.06.00.653983+480 E709034575E1105       LEVEL: Error (OS)
 PID     : 8645                 TID : 46955996636928  PROC : db2sysc 999
 INSTANCE: db2inst1             NODE : 999            DB   : SAMPLE
 APPHDL  : 999-66               APPID: xxx.xx.xx.xx.39597.170119040925
 AUTHID  : DB2INST1             HOSTNAME: myhost1
 EDUID   : 142                  EDUNAME: db2agntp (SAMPLE) 999
 FUNCTION: DB2 UDB, oper system services, sqloopenp, probe:80
 MESSAGE : ZRC=0x850F000C=-2062614516=SQLO_DISK "Disk full."
           DIA8312C Disk was full.
 CALLED  : OS, -, open                             OSERR: ENOSPC (28)
 DATA #1 : Codepath, 8 bytes
 5:11:18:26:40
 DATA #2 : File name, 44 bytes
 /db2loadtmp/DB200007.PID/DB201691.OID/999-66
 DATA #3 : SQO Open File Options, PD_TYPE_SQO_FILE_OPEN_OPTIONS, 4 bytes
 SQLO_CRTREV, SQLO_READWRITE, SQLO_SHAREWRITE
 DATA #4 : Hex integer, 4 bytes
 0x000001B6
 DATA #5 : signed integer, 4 bytes
 0
 DATA #6 : Hex integer, 4 bytes
 0x00000000
 DATA #7 : String, 105 bytes
 Search for ossError*Analysis probe point after this log entry for further
 self-diagnosis of this problem.

However "df -lk" shows the disk should have enough space left:

     Filesystem                        1K-blocks       Used  Available Use% Mounted on
     ...
     /dev/mapper/rootvg-lvdb2loadtmp    51606140     396704   48587996   1% /db2loadtmp

After investigating for a long time, I checked "df -li" output,  I can seeinode is used up

         Filesystem                         Inodes   IUsed    IFree IUse% Mounted on
     ...
         /dev/mapper/rootvg-lvdb2loadtmp   3276800 3276800       0  100% /db2loadtmp

Looks like there are many small files in this path, I clean them up and try again, the load completes successfully!!