Pengjia did a cron job to query the scaler, we had this kind of issues with older version of vxworks where the TCP/IP sockets were not freed fast enough. If there were a lot of queries to the TCP server the operating system eventually runs out of sockets. What is the CPU and kernel you are using ?