Hi Jean, Sorry for the late reply... I'm out of town for a while. Yes, this phenomenon occurs for all eight cores on both Xeon processors. The temperature reading is dumped every second by running "sensors" of lm-sensors. My lm-sensor is configured to use "coretemp" kernel module to get temperature readings. My workload is a repetition of heavy computation for 400ms and then idle for 600ms. So, overall the CPU utilization is 40%. The period of oscillation showing in the graph is on the order of 10seconds, definitely not because of my workload. Thanks. -Simon On Fri, May 8, 2009 at 6:13 AM, Jean Delvare <khali at linux-fr.org> wrote: > Hi Simon, > > On Mon, 4 May 2009 21:53:22 -0400, Simon Chen wrote: >> I am using lm-sensor on a dual-quad-core system with two Xeon 5460. I >> am running a periodic workload on one of the cores: 400ms computation >> and 600ms idle. >> >> >From the thermal side, the temp reading is shown in the figure >> attached. Notice that there is a very obvious oscillation in the >> temperature reading on that core. The oscillating period is around >> 20-30 seconds, which is definitely not sub-second. >> >> I've no idea what's going on here. Can anyone share some thoughts? > > I can't remember anything like that before. But you didn't provide > enough data for me to comment on it. How do you measure the > temperature? Can you reproduce the behavior on a different core? Does > the temperature reading oscillation depend on the period of the > workload? Does it depend on the duty cycle of the workload? Do you have > any form of frequency scaling in effect on these CPUs? > > -- > Jean Delvare > http://khali.linux-fr.org/wishlist.html >
