Transfer: https://blog.csdn.net/varyall/article/details/79747657
The PV
the PV is the abbreviation for the Page View. Users access the page through a browser, an application server for each request generated,
referred to as a PV. Taobao performance test environment, this concept extends made, a request for handling a real system, depending on
a PV. I.e., PV concepts apply to the interface.
PV statistics can generally be obtained by monitoring Buried statistics or access logs statistics.
When it comes to PV there is a special case, called PeakPV, refers to the number of day peak PV value PV to achieve.
Through some monitoring systems, you can visually see the statistics.
QPS / TPS
QPS / TPS original meaning: the system can handle the number of requests per second / transaction, or throughput. We are more concerned about the application of the web is the number of request web application can be processed per second. This is an important measure of system performance.
QPS (TPS) = number of concurrent / average response time.
PV QPS amount of statistics can correspond to time by the access log statistics obtained by dividing the corresponding time. It can be obtained by testing the performance test tool.
General statistics are often the peak of PV corresponding QPS.
ResponseTime response times
Response time (RT) refers to from the client a request start time, returned to the client receives a response result from the server end of the elapsed time, the response time from the request transmission time, transmission time and the network server processing time of three parts.
Load LOAD
average system load is defined as the average number of processes running within a specific time interval in the queue. If a full process
, the following conditions it will be positioned in the run queue:
- it is not waiting for the results of I / O operations
- it is not actively enter the wait state (i.e. not called 'wait')
- is not stopped (for example: waiting for termination) 1
index value of this ideal load value is the number of cpu core number * 0.7 *, if more than this value over the long term than the system needs to be careful.
CPU resources,
CPU resources here refers to the CPU resource usage application service system. CPU system resources is an important parameter to determine whether the processing power and applications running stability.
JVM GC and FullGC
for performance java applications must ultimately related indicators of the GC. In general, our application should try to avoid FGC. Because FGC will be complete garbage, the application will run very slowly, so it is necessary to avoid FGC by setting the appropriate parameters of JVM and GC policy. There are usually indicators to monitor the response time and the number of GC.
Commonly used performance indicators as well as memory usage, disk io and some other indicators, not to list here.
The above describes some of the concepts and statistical performance indicators, here are some important links and differences between a few of them.
1. capacity forecast
for the design of our systems, we can receive access to the user how much of the amount in the affirmative on the front lines need to be tested. That wish to evaluate the maximum daily PV comes, our system is able to support. But how to assess it, do you want to make a maximum daily pv scenarios to test? In fact, based on existing experience and data, it can be summed up the relationship between the peak and QPS day of pv.
We can see daily by QPS and PV charts, daily curves are basically the same. By mathematical modeling, we can see the peak of each server QPS = ((total PV * 80%) / (24 * 60 * 60 * 40%)) / number of servers 1. 80% and 40% of these two figures is not a fixed parameter, represented by this formula means that produce 80% of the mean QPS pv is within 40% of the time (12 hours). For different scenarios have different parameters.
So that we can obtain the pressure sensing applications by QPS its peak, then calculate the specific date PV peak QPS formula according to the capacity predicted by this.
Namely: the estimated PV = pressure measurement QPS * (24 * 60 * 60 * percentage of time) /0.8 * number of machines
2. CPU resource utilization and LOAD
by many people's minds cpu occupancy rate and are using the current cpu load statistical rates. But in fact these two indexes still very different.
cpu occupancy rate is well understood, is the use of cpu time share ratio. Load the cpu is based on the average length of the waiting period of time cpu processing task queue. This index has a higher value than the reference cpu occupancy rate at high loads. Because the high-load period, cpu occupancy rate are substantially close to 100%, it can not reflect the extent of the engine load. Instead, the system can reflect the current load is serious by the length of the task queue statistics, whether controllable.
Can be well understood by the concept of the relationship between road load of the vehicle in the following figure:
(a single processor system)
when the load is equal to 1, the full system load, but can meet the current requirements of the system;
when the load is less than 1 the system easy to run;
when the load is greater than 1, there was a lot of vehicles waiting to enter the highway, waiting for just such tasks as handling cpu, cpu occupancy rate at this time simply can not distinguish load = 1 and load> 1 these two cases.
So read the load for the current running load is helpful to understand the system.
Performance indicators still have a lot of information you can dig up, this article from the perspective of application monitoring conducted a simple introduction. Undeniably, however, read each performance indicator is the person in charge of the health system to understand the application of the necessary conditions, but also the content of each development should be concerned about.
References:
Evaluation system throughput http://blog.csdn.net/fenglibing/article/details/6223197
LINUX understanding of processor load mean http://www.gracecode.com/archives/2973/