Performance specifications Reference Index

customer satisfaction

参考Apdex( Application Performance Index)。

Apdex standard from the user's point of view, the application will show the response time for the user into measurable application performance is in the range of 0-1 Satisfaction.

Apdex optimal threshold defines application response time is T, additionally defines three different performance depending on the application response time binding T:

  • Satisfied (satisfied): The response time of less than or equal to T, such as T 1.5s, in response to the result of a time-consuming 1s is satisfied may be considered the.
  • Tolerating (tolerable): application response time is greater than T, but less than or equal to 4T. Assume that the application is set T 1s, the 4 * 1 = 4 seconds very tolerable limit application response time.
  • Frustrated (irritability period): The response time is greater than 4T.

Calculated:
Satisfied the Count is specified response time of the sampling time to meet the number of applications in response Satisfied requirements; and Tolerating Count is a response within a specified sampling time the application time meet Tolerating the required response times; final Total Samples is the total of the total number of sampling times. As can be seen from the formula, regardless of the application Apdex score sample duration, associated with a target response time T (in the case of using a fixed total number, T indirectly affect the final score, and by affecting the value Satisfied Count of Tolerating Count)

Calculation example: assume that the application can be expected in the response time 1000 ms, 100 samples, there are 50 application response time is less than 1000 ms, 30 times between the application response time is 1000 ms to 4000 ms (4 * 1000ms) , the remaining response time is longer than 20 times 4000 ms, then the app T = Apdex value in a case of 1000ms: (50 + 30/2) /100=0.65
recommendations: Apdex core functionality of not less than 0.9, the edge function Apdex not less than 0.8

Response time

Reference Industry 2-5-10 principle

  • When the user can obtain a response within two seconds, you will feel fast response of the system;
  • When the user gets responses between 2-5 seconds, the system will feel the response speed may also be;
  • When the user get a response within 5-10 seconds, you will feel very slow response speed, but it is also acceptable;
  • When more than 10 seconds after the user still can not get a response, the system will feel bad, or that the system is already unresponsive, and choose to leave this Web site, or initiate a second request. Recommendation: The core function within the response time 2S, 5S edge function within the response time

Error rate

Error rate refers to the probability of the system under load, failed requests. Error rate = (Number of failed requests / total requests) * 100%.
Recommendation: The core function of the error rate of no more than six thousandths, ie less than 0.6%, the edge function error rate does not exceed ten thousandth, that is less than 1%

System capacity

System capacity refers to the ability of information processing systems using system hardware platform and software platform.

Unit: TPS (Transaction per Second): number of transactions per second system, the unit is a pen / sec

Calculated: Throughput = (number of requests) / (total time)
Throughput - Load correspondence relationship:
① rising phase: throughput increases as the load increases, and throughput is proportional to the load;
② stationary phase: Throughput with increase of the load while maintaining stable without much variation or fluctuation;
③ the descending phase: throughput decreases as the load increases, and throughput is inversely proportional to the load;

larger area a1, the stronger the performance capabilities of the system, a2 larger the area, the better, the larger the area a3 system stability, the better the system fault tolerance

Software Index

Level indicators Secondary indicators unit Explanation
GC GC frequency Times / second java virtual machine garbage recycling frequency part
GC long second For a long time garbage collection
Thread Bad Request Times / second Error number of requests
request request Times / second The total request requests
Busy thread More The number of threads busy
JDBC More JDBC connections
Call complexity Floor Call complexity

Recommended: GC frequency can not be frequent, especially FULL GC can not be more frequent; the number of threads that are currently running, can not exceed the set; the number of JDBC connections currently running, can not exceed the set; the complexity of the 10 calls or less.

Indicators database

Level indicators Secondary indicators unit Explanation
SQL time consuming Microsecond SQL execution time-consuming
Shooting Key Buffer hit rate % Index buffer hit ratio
Query Cache hit rate % Query cache hit rate
Table Cache hit rate % Table cache hit rate

mysql Query Cache mechanism: QueryCache is based on the SQL statement to the cache. If you start with a SQL query select, then MySQL server will attempt to use its QC, if the cache hit, will return results immediately, skipping the parsing, optimization and implementation phases. Each Cache SQL text is to be stored as a key. Is a frequently updated list if applied QC, may increase the burden of database, involving frequently updated list of SQL statements plus SQL_NO_CACHE keywords to disable its CACHE, so you can avoid unnecessary memory operations as much as possible.

Extended index

指应用软件以群集方式部署,增加的硬件资源与增加的处理能力之间的关系。正常情况下扩展指标应该是线性或者接近线性。

可靠性指标

节点切换、故障恢复耗时1S内,并且无业务中断(失败请求)

硬件资源指标

序号 指标 解释 标准
1 CPU使用率 指用户进程与系统进程消耗的CPU时间百分比 长时间情况下,一般可接受上限不超过85%
2 内存利用率 内存利用率=(1-空闲内存/总内存大小)*100% 一般至少有10%可用内存,内存使用率可接受上限为85%;
3 磁盘I/O 磁盘主要用于存取数据,对应的是写IO操作与读IO操作 一般使用% Disk Time(磁盘用于读写操作所占用的时间百分比)度量磁盘读写性能
4 网络带宽 使用计数器Bytes Total/sec来度量,其表示为发送和接收字节的速率; 判断网络连接速度是否是瓶颈,可以用该计数器的值和目前网络的带宽比较

系统按照最大容量的80%或标准压力(系统的预期日常压力)情况下运行,能够稳定运行的时间长短。

稳定性指标

TPS曲线稳定,没有大幅度的波动;其余各项指标正常。

备注

Minor GC ,Full GC 触发条件

Minor GC触发条件:当Eden区满时,触发Minor GC。

Full GC触发条件:

(1)调用System.gc时,系统建议执行Full GC,但是不必然执行

(2)老年代空间不足

(3)方法区空间不足

(4)通过Minor GC后进入老年代的平均大小大于老年代的可用内存

(5)由Eden区、From Space区向To Space区复制时,对象大小大于To Space可用内存,则把该对象转存到老年代,且老年代的可用内存小于该对象大小

Guess you like

Origin www.cnblogs.com/aresxin/p/p35435.html