How to analyze the statistics data | Summary


The best practice for AIX performance analysis


+

Search Tips   |   Advanced Search

 

  1. Identify the level of your AIX operating system.

    Use oslevel -r to verify operating system level.

    Verify all filesets used for those products exist on the system.

  2. Identify the workload of your system.

    You can use the command uptime to indicate the number of programs that are being executed at the same time, and how long your system is running.



    Figure 15-16 Output for command uptime

  3. Identify the workload of CPU.

    By using the command vmstat 5 20 (or the nmon and nmon analyzer tool to generate the analysis result), you can get an indication of how busy your CPU is, and whether there are any free CPU resources to handle specific and additional requirements. If the result indicates that the CPU is near 100% utilization, you can safely assume that the CPU has become a performance issue. Specific to the WebSphere Commerce scenario, let us state that there is a significant number of users accessing and shopping on the Commerce server concurrently. For example, there are greater than 20 concurrent users. If the server is not powerful enough to manage the activities above and beyond the 20 concurrent users, the CPU of the server can easily be driven to 100% utilization. Therefore, reconfigure your server to have a CPU or multiple CPUs that can support the number of concurrent users and reduce the utilization of the CPUs to a more acceptable utilization rate and where performance is satisfactory.

    CPU: Breakdown of percentage usage of processor time.

      us User time. If the current physical processor consumption of the uncapped partitions exceeds the entitled capacity, the percentage becomes relative to the number of physical processor consumed (pc).
      sy System time. If the current physical processor consumption of the uncapped partitions exceeds the entitled capacity, the percentage becomes relative to the number of physical processor consumed (pc).
      id Processor idle time. If the current physical processor consumption of the uncapped partitions exceeds the entitled capacity, the percentage becomes relative to the number of physical processor consumed (pc).
      wa Processor idle time during which the system had outstanding disk/NFS I/O request. If the current physical processor consumption of the uncapped partitions exceeds the entitled capacity, the percentage becomes relative to the number of physical processor consumed (pc).
      pc Number of physical processors used. Displayed only if the partition is running with shared processor.
      ec The percentage of entitled capacity that is consumed. Displayed only if the partition is running with shared processor. Because the time base over which this data is computed can vary, the entitled capacity percentage can sometimes exceed 100%. This excess is noticeable only with small sampling intervals.
      rc The percentage of processor resource that is used. This information is displayed only for the WPARs with enforced processor resource limit.

  4. Identify the memory and swap status in your system.

    The primary memory is the most important resource a computer has. Since CPUs are only made with instructions for reading and writing to memory, no programs would be able to run without it. If we identify that the CPU is working in a stable/reasonable status, it is the time to consider whether there is a problem in your memory. You can use the command vmstat or nmon to analyze the status of memory.

    WebSphere Commerce is a complex eBusiness solution based on J2EE techniques. Generally, a large number of Java objects will be created and loaded into JVM in Commerce runtime, which is really memory consumed in the Commerce infrastructure. Although you want to enlarge the JVM to support Commerce server runtime to contain more capacity to handle business requirements, sometimes this does not work, since it is up to the status of your memory. If memory is already 100% consumed for JVM and system usage, there should be a bottleneck in the system memory.

  5. Identify the disk I/O status in your system.

    Disk I/O may be another bottleneck in your system when it is under peak transactions. Sometimes we can see that the status of CPU and memory is good, but the system suffers serious and long response times. Or we can see that the rate of CPU idle is high. In that case, we should convert to disk I/O. From the result from iostat or nmon tooling, we can identify the frequency of I/O read and I/O write. If there is always high I/O access to the disk, this might be a problem of disk bottleneck in the system.

  6. Match the symptom with one of the monitoring results that you got from previous steps, then locate the root cause of this problem.

  7. If you cannot find the root cause from the previous steps, it should be some type of product-specific problem. You can continue reading this book to learn more about how to monitor an individual product in the WebSphere Commerce environment separately.