当前位置:网站首页>How to evaluate load balancing performance parameters?
How to evaluate load balancing performance parameters?
2022-07-07 01:13:00 【Don't bug, please】
How to evaluate load balancing performance parameters ?
At present , In the demand environment dominated by cloud computing and big data , Load balancing and application delivery have been developed for large enterprise data centers “ Escort ” It's a must-have for you . however , Load balancing as a comparison “ novel ” The equipment , Although its main functions are well known to relevant demand persons , But its various performance indicators are “ The flowers are getting more and more attractive ”, Most people think for a long time, so .
that , From what aspects should the performance indicators of load balancing products be evaluated ? The editor specially interviewed Mr. Yu Zhenbo, product director of Taiyi Xingchen, a domestic emerging application delivery enterprise . meanwhile , Not long ago, a batch of self-developed application delivery products passed the centralized purchase test of Unicom . As a top domestic operator , Unicom is very professional in testing the main functions and performance indicators of load balancing products . thus , Yu Zhenbo also made an in-depth interpretation of the performance parameters of load balancing in combination with Unicom's centralized purchase test .
Yu Zhenbo said , There are three main dimensions to evaluate the performance parameters of load balancing equipment , They are four / Layer 7 throughput 、 The rate of the maximum number of concurrent connections and the number of new connections in the fourth layer , The specific indicators of these three dimensions determine the performance effect of load balancing products .
One 、 Four / Layer 7 throughput , No more watching throughput, It depends on HTTPGoodput
1. What is throughput
Measuring the performance of load balancing products is the most common , The most basic indicator is “ throughput ”, This concept applies to routing devices 、 Firewall and other network equipment .
What is? “ throughput ”? Simply speaking , We can think of the load equipment as a pipe , The maximum flow that can pass through the pipeline in unit time is the throughput . therefore , The larger the flow that can pass per unit time , It means that the processing capacity of the load balancing device is higher . The so-called “ Four / Seven layer throughput ”, That is, when you turn on the four layer function and the seven layer function , Processing capacity of the equipment .
2. Standard measurement method of throughput
There are standard test methods for throughput , Traditional router 、 Firewalls are tested in different sizes per unit time UDP Data package to measure product performance , That's what we usually say “throughput”. However, this indicator has been applied to load balancing products “ Out of style ” 了 .
In the Unicom test , Operators use a more “ Strictly ” To consider the performance of load balancing products ——“HTTP Effective throughput (HTTPGoodput)”. because UDP The transmission of is too simple , It is far from the actual processing capacity of the equipment , In the real network , Most of the application protocols are through TCP Transmission of ,TCP The agreement has a more complex negotiation process , As a load balancing device closely combined with Applications , Traditional firewall UDP Throughput is not critical , The real ability to deal with application layer protocols is to see “ Zhenzhang ” The place of .
In the actual test , Operators will use professional test instruments to simulate real customers to launch HTTP request , After passing the load equipment , The simulation server will respond , In the process , Continue to increase the number of simulation clients , Until the load equipment appears and cannot be handled in time 、 Connection failure or response timeout 、 The maximum that can be handled without failure recorded on the test instrument HTTP Until the transmission traffic statistics .
With the traditional UDP Compared with throughput ,HTTP Effective throughput is closer to the actual environment in measuring the performance of load devices , Also more strict .
Two 、 Maximum number of concurrent connections
if , The throughput index measures a pipeline , Then the maximum number of concurrent connections is equivalent to one “ A bottle of water ”, The so-called maximum concurrent connection refers to what this bottle can hold “ water ” The capacity of , That is, the maximum number of connections that the load equipment can maintain .
Testing for concurrent connections , The tester needs to simulate HTTP The client sends connection requests to the load device at a certain rate , And keep the connection uninterrupted . In this process, gradually increase the number of connections , Until the load device fails to process , Then count the maximum number of connections that the load equipment can carry .
Although this parameter is important , But it was not deliberately emphasized in the Unicom test . The reason is simple , This indicator only depends on the memory capacity of the device , To put it bluntly , The number of concurrent connections is determined by the size of memory , And now the number of memory devices can carry connections has already exceeded most practical scenarios .
3、 ... and 、 The number and rate of new connections on the fourth floor (L4 CPS)
In the Unicom test , The new connection rate per second of the fourth layer is highly valued , The concept of connection is also strictly defined : There needs to be three complete handshakes (SYN,ACK/SYN,ACK), once HTTP The request and response of , as well as TCP Connection closed (FIN,ACK,FIN,ACK) The process —— This operation is to simulate the actual application as much as possible .
Four layer connection rate , It is to send connection requests to devices at the highest possible rate , After completing the connection data transmission , And quickly close the connection , To avoid the impact of packet size on the final result , The data transmitted by the connection is set to 64 byte , Unicom system fully tests the number of connections that the equipment can handle per second in this way . For operator users , This indicator is very important for application delivery products to cope with sudden large number of connections . It's a bit like the subway entrance around us , Although the subway transportation capacity is very strong , But if the entrance repair is too small , During rush hours, many passengers still can't walk in smoothly !
边栏推荐
- 深入探索编译插桩技术(四、ASM 探秘)
- golang中的WaitGroup实现原理
- NEON优化:性能优化常见问题QA
- Address information parsing in one line of code
- ZABBIX 5.0: automatically monitor Alibaba cloud RDS through LLD
- A brief history of deep learning (II)
- Eventbus source code analysis
- Grc: personal information protection law, personal privacy, corporate risk compliance governance
- 身体质量指数程序,入门写死的小程序项目
- 「精致店主理人」青年创业孵化营·首期顺德场圆满结束!
猜你喜欢
![[Niuke] b-complete square](/img/bd/0812b4fb1c4f6217ad5a0f3f3b8d5e.png)
[Niuke] b-complete square
![[100 cases of JVM tuning practice] 04 - Method area tuning practice (Part 1)](/img/7a/bd03943c39d3f731afb51fe2e0f898.png)
[100 cases of JVM tuning practice] 04 - Method area tuning practice (Part 1)

pyflink的安装和测试
![[HFCTF2020]BabyUpload session解析引擎](/img/db/6003129bc16f943ad9868561a2d5dc.png)
[HFCTF2020]BabyUpload session解析引擎

Batch obtain the latitude coordinates of all administrative regions in China (to the county level)
深入探索编译插桩技术(四、ASM 探秘)

力扣1037. 有效的回旋镖

城联优品入股浩柏国际进军国际资本市场,已完成第一步

Maidong Internet won the bid of Beijing life insurance to boost customers' brand value

【JVM调优实战100例】05——方法区调优实战(下)
随机推荐
[force buckle]41 Missing first positive number
The difference between spin and sleep
Realize incremental data synchronization between MySQL and ES
Oracle: Practice of CDB restricting PDB resources
[100 cases of JVM tuning practice] 04 - Method area tuning practice (Part 1)
Telerik UI 2022 R2 SP1 Retail-Not Crack
做微服务研发工程师的一年来的总结
Chapter 5 DML data operation
【js】获取当前时间的前后n天或前后n个月(时分秒年月日都可)
Niuke cold training camp 6B (Freund has no green name level)
[hfctf2020]babyupload session parsing engine
第六篇,STM32脉冲宽度调制(PWM)编程
Zabbix 5.0:通过LLD方式自动化监控阿里云RDS
[Niuke] b-complete square
深入探索编译插桩技术(四、ASM 探秘)
ARM裸板调试之JTAG原理
C# 计算农历日期方法 2022
NEON优化:关于交叉存取与反向交叉存取
pyflink的安装和测试
[牛客] B-完全平方数