Five core elements of architecture design

Catalog

One 、 performance

1、web Front end performance optimization

2、 Application server performance optimization

3、 Database layer optimization

4、 An indicator of website performance

5、 High availability

Two 、 Security

1、 Security system architecture

2、 Level of safety protection

3、 ... and 、 Usability

Four 、 Extensibility

5、 ... and 、 Scalability

---

Five important core indicators in the architecture ： They are performance 、 Usability 、 Scalability 、 Scalability and security .

One 、 performance

Performance is one of the core elements , Otherwise, why do I design architecture ？ Casually, a very low It's good to have your system online . Therefore, performance optimization is a barrier that many small companies cannot overcome . Of course, there are many ways to optimize website performance

1、web Front end performance optimization

1） Browser access optimization

Including browser cache 、 Page compression transmission 、 Reasonable layout of pages 、 Reduce Cookie transmission .

• Reduce http request . Avoid setting up too many communication links . take js、css、 Picture files should be merged as much as possible . Avoid too many requests . meanwhile , The back-end request of the system should also be reasonably designed as far as possible , To avoid too much interaction .
• Use the browser's cache .http Head set Cache-Control and Expires.js For example, the file name can be time stamped . Once there is an update, update the timestamp , Otherwise cache ; At the same time, try to avoid updating a large number of static resources at the same time .
• Compress static resources .
• css Place at the top of the page ,js At the bottom . In advance css Rendering . Also avoid js Page blocking . But it needs specific analysis of specific problems . Like pages dom Nodes need to rely on js Generate , Then change the file location according to the situation .
• Reduce cookie transmission . At the same time, let static resources have independent domain names , Do not send static resource requests cookie. So as to reduce the transmission cost .cookie Can pass document.cookie obtain .

2）CDN Speed up

Cache pictures 、 file 、CSS as well as script Script . however pc Upper CDN The acceleration effect is better than that of the mobile terminal . After investigation, we found that ,last-mile The higher the delay ,CDN The worse the relative effectiveness of .

“ Last mile ” It is a term used by the communication and technology industry to describe the technology and steps of connecting end users to the communication network .

3） Reverse proxy

It can provide seven layers of load balancing （http Request balancing strategy ）, And it can provide the cache of static resources , Request forwarding , Prevent network attacks, etc . The more popular ones are nginx.

2、 Application server performance optimization

If the request static interface is not stuck , But dynamic data is still a card , explain MySQL Too many requests to process , You can use server local cache and distributed cache , You can also speed up the response through asynchronous operation , In the case of highly concurrent requests , Multiple application servers can form a cluster to serve the public , Improve the overall processing capacity , improve performance .

1） Distributed cache

Website performance optimization law ： Priority is given to using cache to optimize performance

1. Generally speaking , Deposit in cache The reading and writing ratio of data is 2:1 above , And it should be hot data .
2. It is necessary to consider the possible data inconsistency in the short term if caching is adopted , Or if the cache is updated in real time, it may bring performance and resource overhead .
3. You need to consider cache Once it fails , A large number of requests may bring an avalanche of service performance . So you can be right cache Adopt cluster deployment , In order to avoid the sudden increase of service pressure caused by the loss of too much data .
4. Consider warm-up loading of cache for hot data . For example, before the peak , First, store the hot data into the cache in advance . So as to improve the service performance during peak periods .
5. To avoid malicious attacks , always query Nonexistent data , Lead to cache Frequent visits DB, Non existing data can also be cached and cleaned up regularly . At the same time, there is a mechanism to identify and block malicious requests .
6. Distributed cache should be decentralized and centrally managed . Scalability is ensured through non communication and isomorphism between different instances , And reduce the complexity of the system .

2） Asynchronism

Anything that can be done later should be done later , It feels like lazy loading

Peak shaving is achieved through distributed message queues . Solve problems through business cooperation technology . such as 12306 The line up .

3） colony

Adopting cluster is also an embodiment of service virtualization . To avoid single point problems , At the same time, it provides higher availability , High performance services .

4） Code optimization

• In a multithreaded , If it is intensive computing , The number of threads should not exceed CPU Check the number . If it is IO Handle , Number of threads =[ Task execution time /( Task execution time -IO Waiting time )] * CPU Check the number . besides , We should design objects as stateless objects , More local objects , Refine the lock properly .
• Reuse resources . For example, using singleton mode , For example, use connection pool .
• Setting up reasonable JVM Parameters , Avoid unreasonable full gc.

5） Storage performance optimization

The index of relational database adopts B+ Tree implementation . And many nosql The database uses LSM Tree for storage .LSM Keep the data added, deleted, modified and queried in memory , Until the memory cannot be put down , And the next level of disk LSM The tree goes on merge. So there are many write operations , The read operation is more about querying the recently written data , Its performance is much higher than b+ Trees ; use HDFS combination map reduce Massive data storage and analysis . It can automatically perform concurrent access and redundant backup , With high reliability . It is equivalent to achieving RAID（ Independent redundant disk array ） The function of .

3、 Database layer optimization

The database layer is actually the most vulnerable one , Generally in the application design, the request needs to be intercepted upstream , The database layer only undertakes “ In one's power ” Access requests for , therefore , We introduce queues and caches in the service layer , Let the underlying database rest easy . But if requests surge , There is still a lot of query pressure to MySQL, At this time, we need to find a way to solve it MySQL It's the bottleneck of the market , At this time, you can use the index 、 cache 、SQL Performance optimization and other means , You can also use NoSQL Database to optimize the data model 、 Storage structure, etc .

4、 An indicator of website performance

The important thing is response time 、TPS、 System performance counters, etc , Through these indicators to determine whether the system design has achieved the goal .

1） response time .

2） Concurrency number . If there is no corresponding accurate monitoring for the time being , For different business models , There can be different estimates of concurrency . If our system estimates the peak concurrency , There is a rough calculation , That is, the average number of concurrent requests per second throughout the day * 3. But it also needs specific analysis of specific problems .

3） throughput . The more common ones are QPS（ Queries per second ）、HPS（ Per second http Number of requests ） as well as TPS（ Number of transactions processed per second ）.

4） Performance counters . Including system load 、 Number of threads 、cpu、 Memory usage, etc .

It can be used top、free、cat /proc/cpuinfo Wait for the order to see . The definition of system load is currently CPU Number of threads executed / Waiting to be CPU Number of bus passes executed . When its value and logic cpu The same number is in good condition , It means that all resources are used to a great extent . But some people think that when the load is 0.7 Multiple logic CPU A few hours is better .

5、 High availability

Including highly available applications 、 Highly available services 、 Highly available data and services and highly available monitoring .

Two 、 Security

The Internet is open , Anyone can visit the website anywhere . The security architecture of the website is to protect the website from malicious access and attack , Protect the important data of the website from being stolen .

Safe 5 Elements ： Confidentiality 、 integrity 、 Usability 、 Controllability and auditability .

1、 Security system architecture

• Security service refers to the security protection measures provided by computer network , Including certification services 、 Access control 、 Data confidentiality Services 、 Data integrity services and undeniable Services .
• A specific security mechanism is a mechanism used to implement security services , Including encryption mechanism 、 Data signature mechanism 、 Access control mechanism 、 Data integrity mechanism 、 Authentication exchange mechanism 、 Flow filling mechanism 、 Routing control mechanism and notarization mechanism .
• Universal security mechanisms are not ad hoc for any particular service , It belongs to safety management , Divided into trusted functionality 、 Safety signs 、 Event detection 、 Security audit trail and security recovery .

2、 Level of safety protection

• User independent protection level
• System audit protection level
• Safety mark protection level
• Structured protection level
• Access authentication protection level

The standard to measure the website security architecture is to target all kinds of existing and potential attacks and secret stealing methods , Is there a reliable response strategy .

3、 ... and 、 Usability

High availability , High availability , It refers to shortening the time when the server cannot provide services normally due to failure or other reasons . It is often seen by major cloud manufacturers 6 individual 9,11 individual 9 That's what I'm saying . such as 6 individual 9 Namely 99.9999%, Then the time when the server is unavailable throughout the year is (1-99.9999%)*365*86400=31.5 second . That is to say, in a whole year , Once the server crashes , The recovery time will not exceed 31.5 second .

High availability has high fault tolerance and recoverability .

Measure whether a system architecture design meets the goal of high availability , If one or more servers are down , And all kinds of unexpected problems , Is the overall system still available .

Generally, there are three means 、 redundancy 、 clustering 、 Distributed .

The main means of high availability of websites is redundancy , Applications are deployed on multiple servers to provide services at the same time , Data is stored on multiple servers and backed up with each other , Any server will not affect the overall performance of the application , The usual implementation means is to form a cluster of multiple servers through load balancing devices .

Four 、 Extensibility

Extensibility （Extensibility） When the impact on the existing system is minimal , The ability of a system to continuously expand or enhance its functions . The performance is that the system infrastructure is stable and does not need frequent changes , Less dependency and coupling between applications , When the system adds new functions , No need to modify the structure and code of the existing system . Scalability depends on the good architecture design in the early stage . Reasonable business logic abstraction , level / Vertical cutting, distributed and so on .

The main means of website extensible architecture are event driven architecture and distributed services .

Event driven is usually implemented by message queue , In this way, message production and processing logic are separated .

Server service is to separate business and reusable services , Call through the distributed service framework . The newly added products can realize their own business logic by calling reusable services , Without any impact on existing products .

Regarding this ,《 Scalable art 》 The book proposes a more systematic and extensible model —— AKF Expandable Cube （Scalability Cube）. This cube is set to... Along three axes ：X、Y、Z.

• X Axis extension —— Pay attention to the level of data and service cloning , That's what I mentioned earlier “ Add machines to solve problems ”
• Y Axis extension —— Focus on the division of responsibilities in application , For example, data types , Division of transaction execution types
• Z Axis extension —— Focus on prioritizing services and data , Such as the division of regions

The whole extended model , Use the figure below to show , Where the origin represents the state of complete no expansion .

5、 ... and 、 Scalability

Services should be isomorphic .DB、cache When considering distribution, try to design the expansion scheme in advance . You can also use some mainstream ones that support horizontal scaling better nosql、memcached、hbase etc. .

（1） Lateral separation ： Separate and deploy different business modules , Realize the scalability of the system ;

 

（2） Vertical separation ： Separate and deploy different parts of the business process , Realize the scalability of the system ;