|
|
| [back to "Optimizing server performance: Semaphores (Part 1)"]
Troubleshooting recommendations for UNIX (sidebar)
For additional monitoring of internal (platform-level) information, UNIX systems can use the suggestions presented below along with Domino statistics.
Note that the following tips on gathering additional information apply to all platforms, but especially have relevance on the UNIX platform.
- After you compile system metrics and response time numbers and see, for example, that the Context Switches, the end user response time, and the time spent in the kernel space and/or system CPU time are all increasing, you have an indication that the system needs a closer review. Generally, you should collect this information on a continuous basis. This information is accessible from VMSTAT or SAR. We recommend you set the collection interval rate at 30 or 45 seconds.
- The next step in the process is to use a lock monitoring utility, such as lockstat for Solaris systems, as it reveals whether the lock latency rate is increasing. This type of monitoring provides information on semaphores that exist at the kernel level. Although these semaphores may be difficult to map into the Domino semaphores, they may provide additional information when putting together the big picture. If the lock latency rate increases, you should ensure the next level of information is revealed for analysis and possible troubleshooting.
Note: You should only use lock monitoring if you are very familiar with the UNIX OS and have previously used the lock monitoring information in other situations. Use of this type of information requires advanced knowledge of the system.
- Enable the NOTES.INI setting DEBUG_SHOW_TIMEOUT, as described within the main article. Output from this variable alerts you if the system is reporting any blocked activities through the semaphore timeout messages.
- Perform periodic executions of the NSD tool. This tool, developed by Lotus and shipped with Domino for UNIX platforms, makes use of kernel-level debugging information and helps to analyze a problem situation. When a possible problem needs further analyzing, run NSD every 3 to 5 minutes. From the stack trace that NSD generates, you will likely see many Domino server threads locked on the same semaphore.
- Collect and analyze Domino server console messages that reveal which semaphores are blocked (as described in main article) or pass the issue on to Lotus Customer Support.
Note that when comparing UNIX information display to NT information display that UNIX information is reported in decimal format, from Release 4.x onwards.
|
 | |
|