Influence of off-line crosstalk on data center networks

This article mainly describes the influence of off-line crosstalk on data center cabling. The out-of-Line crosstalk of twisted pair wires is not a new thing. As early as 10 years ago, some people proposed this concept. At that time, due to the constraints of testing methods and the fact that the network was not sensitive to off-line crosstalk, therefore, it is not paid much attention. However, with the widespread application of 10G Ethernet 10Gbase-T in data centers in recent years, the influence of off-line crosstalk has become more and more obvious.

1. What is out-of-Line crosstalk?

The out-of-Line Crosstalk is Alien Crosstalk, also known as the off-line Crosstalk. As the name suggests, it mainly refers to the term used to describe the signal energy coupling between cables in a bundle. Its concept is very similar to the near-end crosstalk NEXT in the cable. The out-of-Line crosstalk has the greatest impact on the so-called 6 packet 1 situation. The twisted pair cable in the middle is affected by the energy coupling from the six cables. In particular, the line with a relatively large twist distance has a greater impact on the line.

The out-of-Line crosstalk can be divided into ANEXT and AFEXT, as shown in. ANEXT is on two adjacent channels, one receiving end RX2 receives the same end and the other cable transmitting end (TX1) coupled energy. AFEXT is the energy that RX4 is coupled by TX1.

In the past, we did not pay much attention to off-line crosstalk because the NEXT and FEXT influences inside the cable are much greater than off-line crosstalk. However, as DSP technology advances. Whether it is NEXT, FEXT, or RL, you can get a considerable filter. Therefore, out-of-Line crosstalk, which cannot be actively eliminated by active means, becomes the biggest trouble limiting the signal-to-noise ratio.

Ii. Common Ethernet transmission coding and signal-to-noise ratio in data centers

To understand the impact of off-line crosstalk on Ethernet transmission, we need to first understand the relationship between the encoding method, signal-to-noise ratio, and bit error rate. Generally, the HDA to PC Server of a data center uses an Ethernet based on twisted pair wires. The most common Ethernet standard in a data center is 1000Base-T or 1000Base-TX, some data centers have started to adopt 10G Base-T. Therefore, this article will mainly study the relationship between these three standards and off-line crosstalk. This article will begin with the three standard modulation methods, analyze their ultimate signal-to-noise ratio, and then compare the signal-to-noise ratio that can be achieved by the transmission media, finally, let's take a look at the influence of off-line Crosstalk on these three standards.

The three Ethernet standards use PAM pulse amplitude modulation. According to the nequest sampling Law, the pulse sign rate is at least twice the frequency, for example, a positive/electrical flat pulse symbol and a negative pulse symbol can be combined into a sine wave. Conversely, a single sine wave can carry at least two pulse symbols. One pulse symbol can carry multiple bits (0 and 1 in a computer), but how much can it take, it depends on the modulation and encoding method, which can be measured by the pulse symbol bit rate ). We can use the pulse symbol bit rate * pulse symbol rate * to obtain the bps of the entire transmission system.

We know Shannon formula) can be used to describe the maximum transmission rate of a given transmission bandwidth and signal-to-noise ratio. In this way, we know the bit rate of the pulse symbol, and we can calculate the required ultimate signal-to-noise ratio. The formula is as follows: C indicates the bit rate of the pulse symbol, and S/N indicates the signal-to-noise ratio.

Let's look at the modulation method, 1000Base-T (IEEE 802.3ab) uses 4D-PAM5 modulation method 1000Base-TX, that is, TIA/EIA-854 standard is an upgraded version of 1000Base-T, the modulation method is the same ). 10Gbase-T uses PAM16 + DSQ + LDPC2048, 1723) modulation and encoding methods. In this way, we can calculate the data in the following table based on the modulation method and the above formula.

Table 1

The maximum signal-to-noise ratio in the above table does not mean that we can achieve satisfactory transmission performance in such an environment. If the maximum signal-to-noise ratio is used, the error rate calculated based on the theory is about 31%, such a network is obviously unacceptable. Therefore, IEEE specifies the error rate. The error rate of 1000Base-T/TX must be 10-10, and that of 10GBase-T must be 10-12. Therefore, it is called Gap to increase the signal-to-noise ratio to a higher level ), increase by DB based on 10-10 and dB by 10-12. At the same time, efficient error correction coding can reduce the signal-to-noise ratio requirements ,. Therefore, the calculation formula for the minimum signal-to-noise ratio required for Ethernet transmission is as follows:

The system allows the lowest signal-to-noise ratio = ultimate signal-to-noise ratio + Gap-encoding gain.

After calculation, we have obtained the minimum signal-to-noise ratio allowed for the three Ethernet standard systems commonly used in data centers, as shown in the following table:

Table 2

Iii. Effect of cabling system and out-of-Line crosstalk on Signal-to-Noise Ratio in data centers

The above table shows the minimum signal-to-noise ratio required for Ethernet transmission. We can see whether the signal-to-noise ratio of the transmission cable in our data center can meet the system transmission requirements. Currently, Cat6 is commonly used for data transmission in integrated cabling of data centers. Cat6A is also used for Cat5E, and almost no data center is used for data transmission, so it is not in the scope of this article ). The signal-to-noise ratio can be calculated based on the data specified in the standard according to the following formula ).

SNR (dB) = signal-insertion loss-integrated near-end crosstalk-Integrated Remote crosstalk-Return Loss-Inter-Line crosstalk (Cat 6A) -ambient noise + near-end crosstalk processing gain + Remote crosstalk processing gain + Return Loss Processing gain.

Let's take a look at each of these parts. We use the maximum PSD defined in the IEEE Standard; IL, PSNEXT, PSFEXT, RL, PSANEXT, PSAFEXT, and PSAFEXT take TIA568C. 2. The minimum value specified in the standard. After studying environmental noise, generally the data center is in-150dbm, so it is ignored here. At present, the DSP's processing gain for internal interference can at least achieve SOLARFLARE's data)

Table 3

Based on the above formula and data, we can calculate that the remaining amount of the twisted pair testing results in the extreme condition is 0), and the performance of the three Ethernet standards on different cabling systems.

Table 4

From the above calculation results, we can see that, without considering out-of-Line crosstalk, The Gbit/s network is more than 10 DBS higher than the signal-to-noise ratio of 10 Gbit/s, this means that the interference margin of 1000Base-T is more than 10 times that of 10Gbase-T. The calculation result shows that the 1000Base-T running on Cat6 is not sensitive to off-line crosstalk, and even FEXT Canceller is not used. Cat6 cables do not have high requirements. Even if the margin of the cable test results is not large, the network can be properly operated. The 10Gbase-T running on the Cat6A cable is not optimistic, because the ISIinter-sybmol interfernce interference must be considered and the remaining 3-4 DBS must be retained. Therefore, if the test results of Cat6A cables only pass the pass, the margin will be very tight. Of course, if you do not consider off-line crosstalk when using shielded cables, the margin is absolutely sufficient.

4. Test the transmission of a simulated data center network in Alien Crosstalk

As you know, the cabling of twisted pair wires in the data center is very intensive. I have encountered a project with 960 copper cables in a cabinet. All the conclusions above are the results of theoretical calculations. What if we simulate an experiment in the actual environment? The IEEE Committee submitted a report when preparing the 10Gbase-T, simulating a 6-Pack 1 link of a data center. 1000Base-T can pass 100% on Cat 5E and Cat6 cables.

 

Simulation Test of Gigabit Ethernet

10Gbase-T was not simulated at the time, but now a research institute is testing the actual transmission performance of 10G Ethernet Over UTP and STP. The results have not yet been fully published, but the general conclusion is that 10G Ethernet is very sensitive to off-line crosstalk on CAT6A UTP cable. If you are not careful, problems will occur. From the above theoretical calculations and actual test results, we can draw such a conclusion that, currently, no matter 1000Base-T or 1000Base-TX is used on six types of online data centers, the problem of out-of-Line crosstalk is not considered, because their performance margin is enough to avoid this problem. For 10 gbase-t 10G Ethernet, whether UTP or STP, if it is well installed, it can theoretically avoid the problem caused by off-line crosstalk. However, considering that the current testing method is not able to quickly detect out-of-Line crosstalk indicators and whether the installation is good), it is recommended to use STP-type shielded cabling systems. Tested. The test results of STP out-of-Line crosstalk are at least better than that of UTP 15 dB, so the out-Line crosstalk is basically immune.