服务器性能监控
Performance Testing is not just testing. As mentioned earlier, it is the process of identifying the performance issues. Well, when you identify that performance is poor, what will be the next step? Obviously it is to find out the root cause of the performance problem. Here is where Performance Monitoring helps. In other words, it helps to gather more data from the server side to narrow down the root cause of the performance problem. More details on different techniques to identify root cause can be seen in Techniques for Performance Bottleneck Analysis page
What to monitor?
Before determining what needs to be monitored, it is important to understand different type of resources involved in a software deployment and level of monitoring required for each type of resources . Let us consider the case of Web application which has a typical 3 tier architecture – Web, App & DB. Let us also assume Web, App and DB servers are deployed in one physical server(which will not be the case in most of the system). With that assumption, following diagram shows different type of resources involved in an application.
As shown in the above diagram, following are the different type of resources involved in an application deployed in a server
- Hardware resources like CPU, Memory, Disk IO etc..
- Software Server resources
- Web Server – Number of Http connections
- Application Server – JVM Heap, Web Container thread pool, Database Connection pool etc..
- DB Server – SGA Memory
- Application Code – Memory used by an object, CPU time of a Method etc..
Hardware resource monitoring:
The objective here is to monitor the hardware resources such as CPU utilization, Memory utilization, IO usage. The monitoring result will tell us the used capacity of the system and how much more capacity is remaining to serve more request.
The hardware resource can be monitored using Operating System specific tools such as Perfmon in Windows or vmstat, top in Unix flavors.
Software resource monitoring:
Web Server resources like number of http connections can be monitoring using utilities like server-status in Apache/IBM Http server which has a web interface to monitor
Most of the common Application servers in industry are J2EE based servers. The most common ones are Oracle WebLogic and IBM Webpshere. WebLogic server resources like JVM Heap, Thread pool etc..can be monitoring using JMX monitors built in to WebLogic admin console or part of SiteScope/LoadRunner tools. Similar to WebbLogic, Websphere can be monitored by using Tivoli performance viewer which is part of Websphere or using performance servlet option through LoadRunner
From database monitoring stand point, Oracle awr report is the best way to monitor Oracle resources like Top SQLs, I/O rate, transaction rate, contentions etc..
Application code monitoring/profiling:
Application profiling is technique to find out the piece of code which is taking more time or more memory. There are lot of tools available to do application profiling. Some example for J2EE applications are JProfiler and JProbe. Other option in Java based application is VisualVM shipped with Sun JDK.
Here is quick table to summarize monitors in a Linux and Java based application
Following are the Windows specific OS counters for monitoring.
Perfmon is a single point solution for almost all the resource monitoring in Windows servers including application servers like Internet Information Server (IIS). It can be used to view and log the server statistics. It is part of Windows OS and need not be installed separately. That means in all the Windows computers including PCs, you can find this tool.
Following are the some of the counters and how they help to identify the performance bottlenecks. Please refer the perfmon help page in windows for setup and more details.
%Processor Time – It shows the total CPU utilization of the Windows server. If there are multiple CPUs in the server, this will give an average of all the processor utilization. CPU Utilization consistently above 85 % can be considered as CPU bottleneck. Again, many people confuse with CPU spikes. Intermittent CPU spikes may not be a concern, but if it is with too short interval or showing consistencies, it will be a problem.
Processor Queue Length – It is the number of threads waiting to get CPU time. As name says if the there are more threads waiting, it shows CPU is more busy. The value of this counter more than 2 times the number of CPUs in the server is an indicator of CPU bottleneck. But always try combine with CPU utilization to conclude the CPU issues.
Available MBytes – It will give the measure of available physical memory in Mega Bytes. During long running endurance tests, if the available memory going down over a period of time even with constant user load, it is good indicator of memory leak in the server.
Pages/sec – Along with Available Mbytes, this counter will give useful information on systems memory usage. If pages/sec is more, it is an indicator of lack of physical memory. There can be situations where Pages/sec shows huge values even with lot of physical memory. These issues can be seen when new processes are creating in the server. One of the classic examples is when users are logging into the Citrix servers. It can not be an indicator of memory shortage and can be due to memory mapped files. Please refer Microsoft documentation for more details.
Private bytes – It shows the dedicate memory of a windows process. It should be monitored to find out the memory usage of a process. Increase in memory usage over a period of time is a good indicator of memory leak. This can be combined with memory utilization to come to final conclusion.
