The frame preemption protocol still misses the era when carriages and horses were slow, letters were far away, and the bandwidth was only tens of megabytes.
▎The splendor of the old times
In the autumn of 1983, the IEEE 802.3 University was formally established. This university followed strict standardized management. The food, clothing, housing and transportation of the data packets must comply with the school’s regulations, such as the completely uniform dress at first glance.
Figure 1 802.3 frame format
Frame Preemption is an old man who was born in the 1990s. I don’t know who he is. He has been in the field of real-time systems for more than ten years. In 2016, he officially got the teaching position of 802.3 University. Everyone calls him Mr. br. The old man loves to chat with others in his free time, and whenever he talks about the research of those years, he always takes care of his surroundings with bright eyes, and his pride is beyond words.
The old man's research focuses on the scheduling of data packets, and the research topic is called "On How to Give Certain Privileges to High-Priority Data Packets to Avoid the Occurrence of Low-Priority Inversion Phenomenon". The low-priority inversion in it means that the high-priority data packets need to wait for the low-priority data packets to be transmitted before they can be transmitted. This is also a classic problem, and there are not many solutions. The plan that the old man researched, in simple terms, is actually jumping the queue, and jumping the queue is an art.
Figure 2 High priority forwarding sequence in different situations
We must know that the conditions of the school in the past were not good, and the available bandwidth was only a dozen or twenty megabytes at most, and the efficiency of work was low. When the reversal phenomenon occurred, the high priority data packets should not waste a few milliseconds. If you come a few more times, it will go down in a few seconds, tsk, the data packets will make a fuss.
Figure 3 Single hop saves delay
Let's go far, let's talk about the old gentleman, he was thinking about how to solve the problem of slow work, after much deliberation, he finally realized this method of jumping in the queue, and carried out experiments in the school. First of all, let’s talk about the outfit he changed. The Ethernet school uniform of more than 40 years was changed as soon as he said it. He directly changed two styles of the school uniform in the first picture, saying that he was afraid of losing the data packets. Forget it. Let the low-priority data packets wear the same suit on the left as before when there is nothing to do, and change to the suit on the right if they are cut in the queue. See if the CRC on the right is gone, wear the CRC when you have finished your work, and wear the mCRC if you have not finished it. There is also the Frag_count, which is used to see how many times you have been inserted into the team. This record cannot be lost, and if it is lost, it will not do anything for you.
Figure 4 Two formats of mPacket
To say that the old Mr. br was very high-spirited at that time, and even the MAC layer was divided into two for him, a preemptable MAC specialized in handling those low-priority data packets, and an express MAC specialized in dealing with those high-priority matters. If the person in charge is undecided, just call and ask. There is a standardized process, which is called a state machine.
Figure 5 pMAC and eMAC
The above changes were changed, and the Ethernet bureau at that time was alarmed, and they came to ask several questions. I still remember the question, but it was funny.
Q: What are you doing?
Answer: Drop a little delay...
Q: So drop? ?
Answer: Well...
Q: Hey, it’s okay to let those data packets do less work at a time...
The old man boasted and took out a table: Yes, it can, but according to my analysis, the efficiency of the work is still around 1500 bytes. High point, you see, the data segment accounts for 98%.
The old man was already prepared.
Figure 6 The proportion of data segments under different frame lengths
It's a pity for the old man. Do you know what the people in the Ethernet office said at the end? They said: Hey, Mr. br, it's hard work, but don't worry, everyone's work efficiency will be improved soon.
Who is the old man, the people in the Ethernet bureau don't know, how can he still not know. After his work efficiency increased, his research lost the meaning of continuing. Ten trillion one millisecond, one hundred trillion one hundred microseconds, one thousand trillion ten microseconds, the old man has his own ruler in mind.
▎ Ending the curtain, pursuing the journey of redemption
The vision of the Ethernet station was soon realized, and 100M and 1000M networks became common, and 10G and 100G networks were no longer uncommon. From then on, people often saw him sitting against the wall outside his MAC Merge laboratory. Chatting with him is still full of mCRC, SMD, state machine and so on, which makes people half understand. Although the content is still the same, it is indeed not the same as before.
I haven't seen the old man at the bottom of the laboratory wall for a long time, and I learned a little about the situation after asking the data packets that the gang just recruited. It is said that the old man no longer intends to go deeper into the subject. Recently, he is working on a horizontal project with the very popular Qbv laboratory of 802.1 University. The project seems to be called "On how to deepen the 802.3br laboratory and 802.1 The cooperation of Qbv laboratory: the wonderful use of Hold and Release", I also heard that based on this project, a Qbu laboratory was built in 802.1 University.
The old man is really a god.