Linux/Unix systems using malloc or SAP’s extended memory (EM) can suffer fragmentation. The OS reports “free” memory, but there is no contiguous block large enough to satisfy a work process request. SAP’s dispatching logic hangs as it repeatedly fails to allocate a memory segment.
Identify the PID from SM50 (if accessible) or dev_disp. Then:
kill -9 <PID>
Do not kill the main dispatcher. After killing the hung process, the dispatcher often reinitializes a new work process and clears the free/hang paradox.
lsof -p 12345 | grep -E "pipe|socket|NFS"
The EWPROD application server may be waiting for a pooled database connection. The DB listener queue fills up; SAP instances see no error (so they don’t release the process) but also receive no data. The work processes hang, and because no computation occurs, the CPU remains free.
