Tài liệu KRONE - White paper - TrueNET - Improving Bit by Bit - 2003 docx

many installers and customers are finding that physical networks designed, installed and tested to the current Cat 5e or proposed Cat 6 standards simply won’t work properly at 100 Megabit Fast Ethernet, let alone at Gigabit speeds. Independent Management Consultant Philip Turtle explains some of the reasons why this is happening and how one manufacturer seems to have come up with the solution. There have been numerous reports during the latter half of 1999 about Cat 5e and “proposed” Cat 6 systems being installed and passing category testing. Yet when presented with real network traffic they have failed to pass data sensibly at all, even though the link lights on the network interface cards (NICs) are “on,” both on the PC and the hub or switch. The practical solution in many cases has been to switch the 100 Mb/s NICs from Full Duplex to Half Duplex, effectively halving the theoretical maximum bandwidth to 50 Mb/s and in fact, taking the practical bandwidth much lower. It’s all about radio waves. As a teenager, I was a keen ham radio practitioner and built a VHF radio station operating on the “two meter” band at frequencies around 144 MHz. Why do I tell you this? Because Cat 5 systems use 100 MHz rated components and cable, Cat 6 uses 200 MHz—both very close to my radio station frequency. To move signals of that frequency around efficiently within my transmitting station needed what can only be described as plumbing transmission lines made out of thin copper pipes and ceramic insulators—the alternative being expensive, specially constructed, but very high loss, coaxial cable. Yet here we are twenty years later happily shoving similar frequencies down twisted pairs without really thinking about the physics involved! Three factors about radio frequency signals are very important in understanding our problem: 1. They become attenuated very rapidly in a transmission line system such as our twisted pair, and the power loss increases with both frequency and line length. So the signal you get at the receiving end is much, much less than you started with, a mere 0.005 watts out for every one-watt you put in at Cat 5 limits. 2. They treat every wire as an antenna radiating energy into other conductors and receiving energy from them, too. Just think, you are trying to receive a signal which is only five thousandths of the high power signal in an adjacent wire, a fantastic recipe for the interference we know as “crosstalk.” 3. They reflect or bounce back quite dramatically off any “discontinuity” that gets in their way, and any poor connections and impedance change or “mismatches” along the way prove to be very good reflectors. Every bit of power that is reflected doesn’t get to the far end, reducing our power received to even less than five thousandths of what we put in. And if that’s not enough, the reflected signal interferes with and distorts the signal which is going in the correct direction and can often turn a “zero” into a “one” or vice versa. Attenuation, crosstalk, reflections and digital signal processing. Without the wizardry of digital signal processing, high speed data over twisted pairs just wouldn’t work. If you look at figure 1, you’ll see how the signal goes in and how it comes out, and that it is unrecognizable. The trick is that most of Improving Bit by Bit. AS DATA RATES RRIISSEE ,, KRONE: 800-775-KRONE www.kroneamericas.com www.truenet-system.com. No part of this document may be reproduced without permission ©2000 KRONE, Inc. NEXT NEXT Figure 1 the crosstalk or noise is from a known or predictable source. In fact, most of the noise (or unwanted signal on the receive pair of a Cat 5 cable) is generated by the adjacent transmit pair. So, since we’re generating the transmit signal in the NIC, it is a relatively easy job to subtract the interference signal (which may well be greater than the signal you are trying to receive) and recover the weak signal that we wanted. That’s great. At least it was when we were dealing with 10 Mb/s applications like 10Base-T Ethernet. Unfortunately as we get faster, 100Base-T and then 1000Base-T, the effects of unpredictable noise become worse. Alien crosstalk signals, induced from other Cat 5 cables, become significant; and since we didn’t generate them in the NIC, we cannot correct for them. We call this “alien crosstalk.” Likewise, reflected power from impedance mismatches is proving to be an absolute killer and in the specifications, Return Loss (the measurement for reflected power) is listed as “For further study,” meaning no one knows what figure is acceptable and so no one tests for it! And a further impedance mismatch problem occurs where two mismatch or reflection points are particular distances apart: standing waves are set up. (Remember holding one end of a rope still and waggling the other until it formed a constant sine wave pattern?) These standing waves can absorb lots of energy making sure hardly any arrives at the far end! Reflection points We have talked quite a lot about reflection points, and recent lab tests have shown just what a disastrous effect they can have, but what are they? In a Cat 5 or Cat 6 cabling system, the obvious impedance change points are: ■ RJ45 plug to RJ45 jack (particularly if provided from different manufacturers). ■ RJ45 jack to horizontal cable. ■ Horizontal cable where it is severely bent, kinked or squashed. ■ RJ45 jack to patch cord. ■ And the worst culprit of all, patch cords. The Cat 5/5e specifications require every component to have nominal impedance of 100 ohms, but they allow a tolerance of plus or minus 15%. This means that at any of the points we’ve mentioned, there can be an impedance mismatch of up to 30 ohms (or 30%) and the system is still within spec. Yet as we now know, reflections are a very serious problem at higher data rates and a 30% mismatch would have terrible results. Actually, if it wasn’t for the fact that the reputable manufacturers use closer tolerances and build significant “headroom” into their systems, many installations wouldn’t work at all! But even though the reputable manufacturers take these precautions, the result is only assured if you use their components throughout the system. If you mix and match, you are absolutely guaranteed to have mismatches. As an example, one manufacturer might produce cables at 110 ohms (±5) ohms because they’re easier to manufacture and they’re inside the specification. However, use the cables in a network with another manufacturer’s 95 ohm connectors, and you can just see those reflections building up. Recent lab experiments at KRONE’s Laboratories (with deliberately mismatched components that were all well within the Cat 5e spec) showed just how real these impedance mismatch effects are. Looked at in the time domain (Figure 2), the system falls within the specification limits. But you can see the mismatch effect and you can start to imagine the potential for reflections at the patch and horizontal cable connection point. Looked at in the frequency domain (Figure 3) you can see that the impedance varies wildly far outside the specified limits at particular frequencies. In fact, you can see some of the effects of these reflections in the associated attenuation versus frequency graph (Figure 4), a characteristic not tested on site, which takes the received signal well below the spec level at frequencies which correspond to those wild 130.6 121.1 111.7 Mismatch Patch Cable Horizontal Connection Point Horizontal Cable 102.2 Ohms Impedance vs. Distance 92.8 83.3 73.9 0.1 9.9 19.7 29.4 39.2 130.6 58.8 68.5 Meters BIOGRAPHY Philip Turtle is an independent management consultant in the Internet, communications and public transport sectors, as well as a freelance technology writer. He is immediate past national chairman of the Institution of Electrical Engineers’ (IEE) Group and is actively involved in the revitalization and restructuring of the IEE’s Knowledge Services Division. Figure 2: Impedance versus distance of mismatched channel configuration. Note the impedance mismatch occurs at the near end, approximately 4.2 meters into the channel. KRONE: 800-775-KRONE www.kroneamericas.com www.truenet-system.com. No part of this document may be reproduced without permission ©2000 KRONE, Inc. impedance swings. So it is no wonder installers and users are having problems on site! Bit errors The ultimate effect of all of the problems we’ve looked at is that bit errors occur. “Ones” arrive as “zeros,” “zeros” arrive as “ones,” and either way, the chunk of data that we were trying to send ends up as garbage. Ethernet and other protocols are thankfully not stupid and can detect when they receive a packet of garbage. Then they shout back to the other end, “Sorry, didn’t get that one. Can you send it again please?” This is okay for one or two garbled packets, but as the incidence of garbling increases, so do the requests for retransmissions—and the retransmissions themselves become a significant but invisible part of the network traffic. In fact, they quickly become the major part of network traffic (Table 1). And just for good measure, of course, in a system with lots of reflections, quite a few of the retransmitted packets will also get garbled and have to be retransmitted a second or third time. In practice it is quite possible to produce horizontal cable and connectors to the exact nominal impedance of 100 ohms (not some other nominal figure because it is easier to manufacture) and to control the manufactured products’ impedance to within a few percent of that 100 ohms. In separate tests, KRONE labs discovered that, in sending one million bits through a Cat 5 system with only six ohm variations, some 365,000 (or 36.5%) were actually being sent again. Of these 365,000 retransmissions, a further 36.5% had to be sent again and so on. At the end of the day, Ethernet got the data through perfectly, but it had to send and receive 1.6 million bits to get one million correct bits. That meant a reduction in network throughput of nearly 40%. KRONE engineers claim to have seen far worse on site. “Field experience has shown that some low end systems are operating at as little as 4% of supposed capacity, that’s only 4 Mb/s from a supposedly 100 Mb/s system,” KRONE Technical Services Manager Karl Tryner claims. Is there a solution? Yes, radio to the rescue! But you knew I would say that. The theory, at least, is quite simple. If you closely impedance match all of the elements of the system, then the amount of power reflected is reduced, so is the distortion it causes, and so are the standing waves that produced all those. The big problem is with the patch cords because the cable used is flexible, the geography within the cable changes every time the cable moves. And therefore so does its impedance. In fact, a 100 ohm patch cable can quite readily change from 85 ohms to 115 ohms in the space of a few seconds as you re-patch it or run it around the cable management. The only solution is a new design of flexible cable, which does not exhibit these wild impedance changes. KRONE has done this, indeed it is one of the main advances in KRONE’s “TrueNet™ Technology,” which has been demonstrated at trade shows around the world and with which they are claiming they can guarantee zero bit error rates on both Cat 5e and Cat 6 solutions. % of Retransmissions Data Rate 100Mbps 20Mbps 4Mbps 800Mbps 160Mbps 32Mbps (modem speed) 0% 1% 2% 3% 4% 5% 1.0 28.6 56.3 84.0 111.7 139.3 167.0 194.7 121.1 111.7 102.2 Ohms Frequency (MHz) Impedance vs. Frequency 92.8 83.3 73.9 64.4 55.0 130.6 SKIN EFFECT At radio frequencies, electrical energy or power does not flow along the inside of the conductor, but actually travels around the outside as an electromagnetic field. This is why crosstalk is such a problem, because every conductor in the system is a radio transmission antenna. Table 1 KRONE: 800-775-KRONE www.kroneamericas.com www.truenet-system.com. No part of this document may be reproduced without permission ©2000 KRONE, Inc. Figure 3: Mismatched channel configuration frequency domain. -0.0 -4.4 1.0 28.6 56.3 84.0 111.7 139.3 167.0 194.7 -8.9 -13.3 Insertion Loss Deviation -17.8 dB Frequency (MHz) Attenuation vs. Frequency -22.2 -26.7 -31.1 Figure 4: Mismatched channel configuration (attenuation versus frequency) shows insertion loss deviation greater than the standard. TrueNet technology KRONE has applied my beloved radio theory and produced a patented TrueNet Technology patch cable wherein the individual copper strands are somehow glued to each other to maintain an overall cylindrical shape and the internal geography of the cable is very closely constrained while maintaining flexibility. The end result is a patch cable that has an impedance of 100 (±3) ohms — a factor of ten times better than the spec. But you cannot buy KRONE’s patch cable except as made-up patch leads because they believe that the quality of termination, strain relief, and testing as a finished assembly is the only way to be sure of total system performance. They make them only to specific lengths so as to avoid known standing-wave wavelengths and their associated problems. KRONE connectors have always been manufactured to 100 (±3) ohms so they match the patch cords well. KRONE’s cable operation has also developed TrueNet Cat 5e and Cat 6 horizontal cables that are not only 100 (±3) ohms, they also include the neat device of a helically round central core which significantly reduces the effects of alien crosstalk. The overall result is a physical layer solution, closely impedance matched throughout, both in the time domain (Figure 5) and the frequency domain (Figure 6), which exhibits such low levels of reflection, attenuation deviation (Figure 7) and alien crosstalk, that KRONE is prepared to guarantee the network to be bit error free for five years. They even test your as-installed network free of charge to verify the performance warranty. 121.1 111.7 Mismatch Patch Cable Horizontal Connection Point Horizontal Cable 102.2 Ohms Meters Impedance vs. Distance 92.8 83.3 73.9 0.1 9.9 19.7 29.4 39.2 130.6 58.8 140.0 130.6 1.0 28.6 56.3 84.0 111.7 139.3 167.0 194.7 222.3 121.1 111.7 102.2 Ohms Frequency (MHz) Impedance vs. Frequency 92.8 83.3 73.9 64.4 -0.0 -4.4 1.0 28.6 56.3 84.0 111.7 139.3 167.0 194.7 222.3 250.0 -8.9 -13.3 No Deviation -17.8 dB Frequency (MHz) Attenuation vs. Frequency -22.2 -26.7 -31.1 -35.6 -40.0 Impedance* Impedance is a measure of the total opposition that a circuit or a part of a circuit presents to electric current. Impedance includes both resistance and reactance. The resistance component arises from the collisions of the current-carrying charged particles with the internal structure of the conductor. The reactance component is an additional opposition to the movement of electric charge that arises from the changing magnetic and electric fields in circuits carrying alternating current. The magnitude of the impedance Z of a circuit is equal to the maximum value of the potential difference, or voltage, V (volts) across the circuit, divided by the maximum value of the current I (amperes) through the circuit, or simply Z=VII. The unit of impedance, like that of resistance, is the ohm. Depending on the nature of the reactance component of the impedance (whether predominantly inductive or capacitive), the alternating current either lags or leads the voltage. *“electrical impedance” Encyclopaedia Britannica Online KRONE, Inc. North America Headquarters 6950 South Tucson Way Englewood, CO 80112-3922 Telephone: (303) 790.2619 Toll-Free: (800) 775.KRONE Facsimile: (303) 790.2117 www.kroneamericas.com www.truenet-system.com Figure 6: Well balanced channel configuration (frequency domain). Figure 7: Well balanced channel configuration (attenuation versus frequency). No insertion deviation. Figure 5: Well balanced channel configuration (impedance versus distance). KRONE: 800-775-KRONE www.kroneamericas.com www.truenet-system.com. No part of this document may be reproduced without permission ©2000 KRONE, Inc. . The trick is that most of Improving Bit by Bit. AS DATA RATES RRIISSEE ,, KRONE: 80 0-7 75 -KRONE www.kroneamericas.com www .truenet- system.com. No part of. channel. KRONE: 80 0-7 75 -KRONE www.kroneamericas.com www .truenet- system.com. No part of this document may be reproduced without permission ©2000 KRONE, Inc. impedance