Performance improvement at the cost of other components is not good

In the development of operating system , We must have an overall view of the system performance . The goal of the system designer is ： Make the whole system run faster , At the same time, a certain balance is achieved in the running speed of each application .

On the other hand , Applications tend to take a selfish view of performance ：” I will try my best to run faster . The impact on the rest of the system is not my concern .”

Some applications put themselves in the boot group , For faster loading . This doesn't really make the system run faster . It is only in the stage of transfer and startup . By putting some application startup costs into operating system startup , Although the time between the user double clicking the application icon and the application being ready to run is reduced , But the total time has not changed .

for example , Consider the following time chart . “*” Mark the time when the user turns on the computer ,”+” Mark the point in time when the Explorer is ready and the user double clicks the application icon ,”!” Mark the point in time when the application is ready .

Then the application developer says ：“ Oh my god , The pink one ‘ Application startup ’ Part is very large . What can we do to make it smaller ？ That's true. , Let's split the application startup into two parts ……

And put some of them in Startup In the group .

“ wow , see , The size of the pink bar （ Represents the time required for our application to be ready after the user double clicks the icon ） It's much shorter now ！” The team then puts this new, shorter value into their performance status report , Everyone gets a raise , Maybe they go to have a big dinner to celebrate .

Of course , Take a look at the whole situation , From the asterisk to the exclamation point , Nothing has changed . It still takes the same amount of time for the application to be ready from a cold start . All these performance improvements are just robbing Peter to pay Paul . perform “ Application startup 1” The time spent is now counted in the operating system and not in the application . You're messing around with numbers , But end users get nothing .

For the chart above , Suppose the user wants to run your application ！ If the user doesn't want to run your application , I just want to check their email , Then they will be “ Application startup 1” pay , Even if they don't get any benefit .

Another example of an application that has a selfish view of performance comes from a company that develops icon overlay handlers (icon overlay handler) The company . The shell treats the coverage calculation as a low priority item , Because it is more important to display icons on the screen , So users can start doing whatever they want .

The company wants to know if there is a way to improve their performance , Overlay the icon on the screen even before it appears , Show right “ performance ” A very selfish explanation of .

Performance is about getting users to complete their tasks faster . If the task does not involve running your program , So your “ Performance improvements ” It's really a performance barrier . I believe your program is very good , But if you expect everyone who installs your program to think it should take precedence over what they do , That is quite presumptuous .

summary

Or that sentence ： people oriented , user-centric .
Before we mess around with the code , Let's put ourselves in another position ： If I am a user of this software , What do I feel ？ Can you swear at your mother ？
Do not drill into the tip of a technical bull's horn ！

Last

Raymond Chen Of 《The Old New Thing》 It's one of my favorite blogs , There's a lot about Windows Little knowledge , For the vast Windows For platform developers , It's really helpful .
This article is from ：《Performance gains at the cost of other components》