There is a realm called feeling (2)

There is a realm called feeling (2)

Wu Yu

Taiyan Network Studio

 

In fact, we don't need to discuss whether to compress this matter. We wanted to solve this problem technically six months ago, but we had never found any relevant algorithms; we should discuss how to compress to achieve the expected results. My denial of my colleagues is that we have come up with no good solutions. In other words, if your goal is to go from Beijing to Shenzhen for a maximum of 24 hours, you have no hope if you choose to walk, run, ride a bicycle, or even drive a car to take the highway.

If I can theoretically understand what is not feasible, I will prove it again in practice, which is a blow to me. But there is one exception, that is, a considerable number of people think they can only know if they try.

 

When I put this matter in an email, please participate in the discussion and return two comments. One is that I lack evidence and no data; the other is that every m of bandwidth is precious and is worth a lot of money.

To this end, I organized a meeting of related technical personnel to discuss this matter. I think this result is caused by poor communication with everyone: I think we can come to a conclusion that we don't need to be able to know through empirical evidence, some colleagues think that I must come up with evidence. Obviously, this is an attitude issue. After all, "attitude determines everything, and Details determine success or failure ".

There are also two opinions on the results of the meeting. One is to propose more algorithms (these algorithms are estimated to be somewhat improved, but they are still too far away from our goal and complicated ); second, we must come up with empirical data.

I decided to do it myself.

 

It is easy to verify that message compression is not feasible. Because there is only one large packet message, and the rest are messages smaller than 50 bytes. The conclusion is that the traffic is slightly higher than the original traffic.

Classification compression is also very easy to verify. The compression ratio of a single large packet is about 50%, but the amount of such messages is too small (other large messages have been filtered out ), basically, the total traffic is reduced by 10%. That is to say, it is feasible to apply for a bandwidth of 8 Mbps (but it is estimated that the network service provider will not agree ).

 

I didn't continue to verify those complex algorithms, and I knew what their results were. Just like repairing the Yangtze River dike, our goal is not to face the average traffic, but to ensure the safety of peak traffic. If we choose the wrong direction, we will only get ourselves killed and get tired of others.

The key to the problem is how to resolve small messages to form large data packets, and how to resolve the delay in the packet group process without increasing the complexity. I have seen a lot of people introduce a bunch of problems to solve a problem. For example, the latest new real estate policy in Hainan was originally intended to curb the rapid growth of house prices, the real result is that house prices have risen faster and are more outrageous. The bubble is bigger and the economy is less secure! (I really suspect that policy makers want this effect, or else it is impossible to make this decision !)

This is like a person's toothache is a little serious, the doctor said, give you a dose of fierce medicine, you are not so painful. After the patient finishes eating, it does not hurt so badly; but it takes a few days to discover that his kidney, liver, and heart are abnormal. What's worse, there will indeed be people who do not know that this medicine will hurt the kidney, liver, and heart.

 

I was hoping to have an expert committee to handle this algorithm. This item is too professional and can be tried without any idea.

I feel the danger is getting closer and closer to our program.

In his "the world is different from you", Lee Kai-Fu said,VistaAfter two years of development, I had to re-develop it.I believeDuring the development over the past two years, too many people have known from the very beginning that there is a dead end. In this case, there must be no talents with excellent intelligence and ability, however, they can only wait for the project to die.Or,They may have made some effort, but it does not affect the project's death.

Will these people feel lonely at the time?

 

It is better to use empirical evidence to prove authenticity than to use empirical evidence to prove authenticity.

I hope there will be good algorithms. I once again think of the way the operating system sends network messages. It is a method of both size and time. I think this method is too complicated and difficult for our programs, is the next policy. This conclusion has been drawn countless times, and you do not need to prove it again. How can I remove a condition?

I suddenly thought of 4 K, which is the size of the socket window. Inspired by this, if I use a 4 K large group package, Its latency is allowed. Because we use streaming data, there will be a fixed minimum traffic per second! That is to say, within the range permitted by the latency, I must be able to make up at least one 4 K packet and ensure the integrity of the original data packet. At the same time, the pressure and solution time of 4 K packages are acceptable! What makes me most satisfied is that it has no complexity!

This is a perfect ending. The best result is 80% less traffic, and the worst result is 40% less traffic. Of course, this is an actual test.

 

More than one expert told me that when you think there is a problem with one thing, 80% of it will be a problem.

I would like to say that if you feel that one thing is okay, and 80% has problems, please train your feelings.