1. Overall focus
It is usually not enough to observe the system energy, it is necessary to take account of the whole systems such as disk/controller, CPU, network usage.
2. Operating System Tools: Sar,vmstat,iostat, etc.
UNIX vendors offer a variety of system monitoring utilities, including ordinary command-line utilities such as Sar,vmstat,iostat, and many vendors provide graphical tools such as HP's Glanceplus.
These utilities should be used with great care, and they often report seemingly similar metrics, but the exact definition of specific metrics between different vendors and even different versions of some vendors is not exactly the same.
3. Disk and Controller
Using the operating system Monitoring tool to adjust disk I/O
Trying to balance all disk and controller I/O, not just the system disk
RAID5 does not have a large hardware cache, the performance of the write disk will be very low
Try to keep the disk dense ratio below 30%, and if you reach 60%, you will find that performance will be compromised.
Do not worry excessively about iowait time, this measure is not easy to explain its intrinsic meaning, often lead to a person astray.
Locate the original partition carefully, minimizing the movement of the head on each disk.
Operating system/Hardware mirroring can help improve performance and is generally more performance-enhancing than Informix mirroring, depending on the level of vendor support.
4. The Network
Different UNIX vendors provide different levels of control over TCP/IP parameters, and the following are parameters that can work on performance
Number of stream buffers for client/server high load, more buffers can improve performance
Size of the packet (Packet)
Increasing packet size may increase throughput for reliable networks where clients frequently send/receive large amounts of data. A smaller size might be better for a less reliable network where clients often send/receive small amounts of data. The packet size is associated with the buffer size specified in the sqlhosts file.
5. Connection Queue Length
When there are many connection requests, it affects the performance of the connection request, and increasing the value can reduce the number of retries for user connections.
6.CPU usage
CPU usage monitoring is an important aspect of system performance adjustment
High CPU utilization on the system or Informix does not necessarily mean that performance issues should make the most of the available resources, which can become a problem when there is a lack of resources that cause processing delays.
Can monitor the overall CPU usage through ONSTAT-P
CPU usage for each VP can be monitored by onstat-g glo
The overall CPU usage of the system can be monitored by sar,vmstat
What should I look for:
Thread ready queue (Onstat-g rea)
If you keep showing a large number of threads waiting to run (more than 10 to 15), the CPU VP resource is severely missing.
A large number of idle CPUs (SAR,VMSTAT, etc.)
If you find that the ready queue lacks CPU VP resources and has redundant system CPUs, try adding another CPU VP. Typically there should not be more CPU VP than the number of physical CPUs, but this is not always true. If there is no spare processor, do not increase the CPU VP. Adding CPU VP When CPU utilization is close to 100% will only make the situation worse.
CPU intensive use of other processes
Perhaps these processes should run at leisure on other hosts or systems
CPU VP number is less than or equal to physical CPU number
The dynamic Scalable structure (DSA) design goal is to make the number of CPUs less than the number of physical CPUs to achieve optimal performance, but in some environments, increasing the CPU VP so that it is larger than the number of physical CPUs may improve performance, indeed, this may be a system bug. Maintenance should record the problem and seek help from the technical support department.