Jonathan found a correlation between the number of NDS requests around 15:15 this afternoon and the crash of nds1-daqd. Plot show number of processes and cpu usage against nds1-daqd uptime. Logs suggest a nds client rapidly asked for data, possibly on cdsws37.