Linux Cluster Visualization Monitoring Tools And Technology

The challenge of massive computation problem in many modern fields need parallel computers of high performance, and the PC cluster, for the advantage of high performance /price ratio, extensibility, software inheritable and easy built, is widely used. So the performance of PC cluster is very important for the user. With the size of cluster becoming bigger and bigger, more and more hardware and software are involved, an effective cluster monitor system is very necessary.In my paper, by studying the existent cluster monitor systems, analyzing the common advantage and disadvantage about them, basing on lm_sensors for hardware information, /proc file system for node load information and MPE for log file recording the program running, We develop a visual tool for parallel computing monitor and evaluation tools on InfiniBand 3 rd generation 20PCs cluster, which is capable of showing the hardware information, node load information all at the same time and recurring the log file after the running of parallel program. This system is of great value for the increasement of cluster performance, the improvement of parallel program effective and easy use for the user.As we all known that some factors such as the temperature of CPU, the speed of system fan, the usage of CPU and memory, the average load of system and the transportation of network play an important role for cluster system performance, so we choose them as the main parameter.We use /proc file system to get nodes information, for its advantage of getting update information including CPU, memory, I/O, network, etc with little cost. Meanwhile all the user procedures information indicated by the PID can get from there, lm-sensor can work with the sensors on the main board, providing information of CPU temperature, CPU fan speed, and other hardware parameter. MPE is the extend for MPI as a performance scale. First we define some events we are interested, then we insert MPE sentence into the MPI parallel program. After running the MPI program, we will get a log file. By viewing the log file, we can learn how much time have spent on different evens.We choose the MySQL as the database to store all the date we get from the nodes, and the Kylix to form the GUI.After all that work, we make a test on the InfiniBand 3rd generation cluster and get some useful information.