DE10354712B4 - Bit-error-free equivalent circuit for Ethernet and Fiber Channel - Google Patents

Abstract

Method for bit error-free replacement circuit for Ethernet and Fiber Channel, wherein the same signal is transmitted via two different optical transmission channels in a ring to a receiver, characterized in that
The replacement circuit takes place at the frame level, with a CRC-32 transmission error detection being carried out for each of the two synchronized channels of the Ethernet or Fiber Channel signals,
Upon detection of CRC-32 transmission errors in both of the synchronized channels, further transmission error detection is performed for the two synchronized channels at 8B / 10B level and that upon detection of a faulty 8B / 10B code word, the corresponding 8B / 10B code word of the other channel erroneously detected 8B / 10B code word replaced.

Description

equivalent circuit Today, in Wide Area Networks (WAN), it is essentially based on SDH / Sonet Level. SDH / Sonet offers the possibility Hitless Protection Switching in rings.

hitless Protection switching means that the ring in case of failure of a segment or network element is closed again within 50ms and the transfer then trouble-free continues.

At the SDH BLSR will permanently transmit SDH signals in both directions. For example, in the SDH BLSR in picture 1 the SDH signal of station 1 transmit directly (clockwise) to station 2. Furthermore but it is also over Transfer station 4 and station 3 to station 2 (counterclockwise). The recipient (Station 2) uses the signal with the better quality. Falls for example the direct transfer of Station 1 at station 2 off, the receiver uses the signal that against clockwise. Because the receiver can decide directly which of the two signals he uses this replacement switching technology is interruption free. Uninterrupted here means that switching to the replacement channel within must be completed by 50ms after the error has occurred. Depending on However, the switching criterion can be switched much faster become. For simple bit errors caused by B1 and B2 bytes in the regenerator (RSOH) or Multiplex Section Overhead (MSOH) can be monitored, an error can already in the next transmission frame (frame) are detected. From the occurrence of the error to discovery so passes a maximum of a frame period of 125μs. In case of loss of the transmission frame, detected by loss of frame (LOF) or out of frame (OOF) will pass at least 4 transmission frames (500μs). During this Time all received errors are passed on.

The SDH / Sonet protocol was for transmission developed by language. Uncompressed speech comes at 64kbit / s transmitted per channel. 30 voice channels are used in the PDH technique with signaling in E1 signals at 2.048 Mbps (32 x 64kbps). The base rate for SDH (STM-1) is 155 Mbps and can transmit 63 E1 channels. Higher transfer rates have 4 times the transmission speed the previous transmission rate.

Tabelle 1: SDH Signale

Table 1: SDH signals

meanwhile is the transmission of data has become more important than the transmission of language. In the area Local Area Network (LAN) has become Ethernet enforced. In the field of storage networks (Storage Area Network, SAN), Fiber Channel is widely used. If now different LAN or SAN need to be interconnected, so the data on the WAN transferred become. The following table shows common LAN and SAN standards.

Tabelle 2: LAN/SAN Standards

Table 2: LAN / SAN standards

• 1. In DE 101 61 186 A1 ist ein System beschrieben, das Ringstrukturen erlaubt, aber weder Fehler reduziert, noch bitfehlerfrei umschaltet. Das Umschalten erfolgt im Millisekundenbereich. Während dieser Zeit wird nicht übertragen. Es ist deshalb für Anwendungen, die hohe Ansprüche an Echtzeitfähigkeit (Video, Sprache etc.) weniger geeignet als dies hier beschrieben Verfahren, die direkt ein Ersatzsignal zur Verfügung stellt. Durch dieses direkte Bereitstellen wird erreicht, dass sowohl kontinuierlich übertragen wird als auch mögliche Übertragungsfehler, die vor dem Ausfall auftreten ausgefiltert werden.
• 2. In DE 100 55 066 A1 wird ein Verfahren beschrieben, das lange Ethernet Rahmen in kurze 44 Byte lange Segmente beschreibt. Zur Ersatzschaltung, gar zur bitfehlerfreien Ersatzschaltung wird keine Aussage getroffen.
• 3. DE 199 59 375 A1 beschreibt einen Umschaltvorgang, der neben Übertragungsfehlern die Laufzeit als Qualitätskriterium heranzieht. Kurze Laufzeit wird als hohe Qualität angesehen, was bei Routern und elektrischen Switches sinnvoll ist. Das von mir beschriebene Verfahren arbeitet jedoch ausschließlich mit optischen Switches.
• 4. In DE 100 04 432 A1 wird ein Verfahren beschrieben, das eine Ringbildung zulässt aber den Ringschluss durch Verbinden der Linienenden aufgrund des Fehlens von Testtelegrammen auslöst. Vom Auftreten des Fehlers bis zur Detektion des fehlenden Testtelegramms wird jedoch nicht reagiert. Dadurch, dass immer nur ein Telegramm empfangen wird, kann keine Kopie dieses Telegramms zur Fehlerkorrektur herangezogen werden,
• 5. In DE 102 34 149 A1 ein Verfahren beschrieben, das Ethernet Telegramme über verschiedene Wege überträgt und die empfangenen Daten auf einer höheren Ebene im OSI Protokoll Stack vergleicht und falsch empfangene oder nicht empfangene Telegramme überschreibt. Neben der unterschiedlichen Arbeitsweise ist dieses Verfahren nicht in der Lage aus zwei fehlerhafter Telegrammen ein fehlerfreies Telegramm herzustellen, wie dies in meinem Verfahren beschieben ist.
• 6. In DE 198 46 353 A1 wird ein Verfahren zur Zeitkompensation beschrieben, das beim "softer handover" in UMTS verwendet wird. Das hier beschriebene Verfahren besitzt ebenfalls eine Zeitkompensation. Das in DE 198 46 353 A1 beschrieben Verfahren beinhaltet keinen Hinweis auf bitfehlerfreie Ersatzschaltung.

The transfer rate is 25% higher than the capacity, because of the 8B / 10B encoding. This encoding converts 8bit data into a 10bit code that is transmitted. When comparing the transmission rates of the LAN / SAN protocols with the available bandwidths of the WAN protocols, it is noticeable that the available bandwidth with the exception of 2 Gigabit Fiber Channel is only used very inefficiently. On LAN / SAN protocols, the following equivalent switching mechanisms can be used:

• 1. In DE 101 61 186 A1 a system is described that allows ring structures but does not reduce errors ziert, still switching without bit error. Switching takes place in the millisecond range. During this time is not transferred. It is therefore less suitable for applications that require high demands on real-time capability (video, voice, etc.) than the method described here, which directly provides a replacement signal. This direct provision ensures that both continuous transmission and possible transmission errors that occur before the failure are filtered out.
• 2. In DE 100 55 066 A1 describes a method that describes long Ethernet frames into short 44 byte long segments. For replacement circuit, even to the bit error-free equivalent circuit no statement is made.
• Third DE 199 59 375 A1 describes a switching process, which uses transit time as a quality criterion in addition to transmission errors. Short runtime is considered high quality, which is useful for routers and electrical switches. However, the method I have described works exclusively with optical switches.
• 4. In DE 100 04 432 A1 describes a method that allows ringing but triggers the ring closure by connecting the line ends due to the absence of test telegrams. However, the occurrence of the error until the detection of the missing test telegram is not reacted. Because only one telegram is ever received, no copy of this telegram can be used for error correction.
• 5. In DE 102 34 149 A1 A method is described which transmits Ethernet telegrams via different paths and compares the received data at a higher level in the OSI protocol stack and overwrites incorrectly received or not received telegrams. In addition to the different methods of operation, this method is not able to produce a fault-free telegram from two faulty telegrams, as described in my method.
• 6. In DE 198 46 353 A1 describes a method for time compensation, which is used in the "soft handover" in UMTS. The method described here also has a time compensation. This in DE 198 46 353 A1 described method includes no reference to bit error-free equivalent circuit.

It There is a mechanism in SDH / Sonet, Virtual Concatenation, the various C-4 (C-3) containers bundles. A C-4 container can Transport 149.76Mbps. Virtual Concatenation offers the like the possibility, e.g. Gigabit Ethernet without 8B / 10B coding, the so-called MAC Frame (MAC, Media Access Controller) in 7 virtual bundled C-4 containers to transport a C-4-7vc with a capacity of 1048,32Mbit / s. Virtual Concatenation is technically quite complicated and only brings then useful if the remaining capacity of the WAN signal (e.g. STM 16) actually used becomes.

BLSR with Ethernet and Fiber Channel

One BLSR lets can also be implemented directly with Ethernet or Fiber Channel. basis is the bit-error-free equivalent circuit as described in "Transmission Error Reduction by bit error-free equivalent circuit "(file number: 103 09 089.4-31 at German Patent and Trademark Office) is described.

• 1. Bessere Signalqualität durch Reduzierung der Übertragungsfehler
• 2. Höherwertiger Service durch bitfehlerfreie Ersatzschaltung (interessant z.B. für TV Live Übertragungen)
• 3. Niedrigere Kosten durch einfachere Systeme
• 4. Höhere Reichweite in optischen Ringen durch niedrigere Übertragungsrate und Ersatzschaltgewinn (so lassen sich die Kosten weiter reduzieren, wenn auf Verstärker bzw. Regeneratoren verzichtet werden kann)

The main advantages are:

• 1. Better signal quality by reducing transmission errors
• 2. High quality service by bit error-free replacement circuit (interesting eg for TV live transmissions)
• 3. Lower costs through simpler systems
• 4. Higher range in optical rings due to lower transmission rate and replacement switching gain (thus the costs can be further reduced if amplifiers or regenerators can be dispensed with)

One BLSR with Ethernet / Fiber Channel transmits the same optical signal in both directions. Will now be in the receiver error detected in one of the two received signals, the other signal used. Both Ethernet and Fiber Channel offer for error detection both 8B10B codes and cyclic redundancy exam (CRC, cyclic redundancy check) via the transmission frame. Thus bitfehterfrei can be switched back and forth the be both signals in sync with each other. This will be buffered by the signal with the shorter one Runtime reached. To find the signal with the shorter duration, have to evaluated the overheads of the Fiber Channel or the Ethernet frames become.

synchronization

Fiber Channel provides information in the overhead, which is the confusion make the frame very unlikely.

Tabelle 3: Fibre Channel Overhead

Table 3: Fiber Channel overhead

It is not enough, only the sequence identification (SEQ_ID) and the sequence counter (SEQ_CNT) to consider to synchronize the incoming frames in the receiver. There depending on the class frames in any order from a fiber Channel fabric can be sent is not guaranteed that the sequence counter the received frame increases in consecutive order.

The synchronization mechanism works as follows:
A receiver stores the received 10bit codes of a channel (channel A) and searches for the 10bit coded start of frame (SOF). If a Start of Frame (SOF) is found, the following 20 bytes up to and including SEQ_CNT are decoded and loaded into the comparator. Now, the second channel (channel B) is examined for an overhead.

Becomes now the same overhead found in the comparator are the two channels pre-synchronized and the first channel (channel A) is now out of the Memory read out. In the presynchronization states Channel A is read from memory and the read out overheads compared with the received in the channel B overheads. Is that true? Overhead n times in a row, are the channels synchronized. If the same overhead is not found before the Memory overflows, lies it is obvious that the stored channel the greater runtime Has.

Now the other channel (channel B) is stored and the overhead in loaded the comparator and compared with the overheads of channel A. If the overheads match, so are the channels pre-synchronized. In the presynchronization states Channel B is read from memory and the read out overheads compared with the received in the channel A overheads. Is that true? Overhead n times, are the channels synchronized.

vote the error-free received 10bit codes m times in a row do not match, So the synchronization machine goes back to the state Loss of synchronization.

at Ethernet frame, the overhead is lower. This increases the probability confusing the frame. Confusing the frame means that the receiver is faulty synchronized.

Tabelle 4: Ethernet Rahmen

Table 4: Ethernet frame

To reduce the likelihood of a mis-synchronization, it is necessary to compare more than just an overhead. A synchronization mechanism that runs through several states until it is synchronized or desynchronized offers:
Because Ethernet often transfers large files in a series of Ethernet frames with the same sender and destination address and maximum frame size, the overhead for the series of Ethernet frames is the same. In order to keep the synchronization mechanism independent of the higher protocol layers, it makes sense to increase the number of bytes compared for synchronization (eg 20 as in Fiber Channel) and to trigger the synchronization mechanism on overhead transitions. The synchronization mechanism is otherwise the same as the synchronization mechanism in Fiber Channel.

Substitute circuit at frame level

If both channels are synchronized, bit error-free switching between them can be performed. In order to reliably detect transmission errors, the CRC-32 of the Ethernet / Fiber Channel frame must be evaluated. The likelihood that a CRC-32 will not detect transmission errors in a frame is 1 to 2 ³² , or 1 to 4 294 967 296 in other words. Thus, Fiber Channel frames will be all 1-bit errors, 2-bit errors, and all errors with odd number of bit errors recognized.

Become now the frames of both channels evaluated with the help of the CRC-32 and in case that a channel one CRC-32 has errors, the other channel, which has no CRC-32 error has forwarded, so reduce the forwarded transmission errors considerably. Recognizable errors are only forwarded if they are in both Copies of the same frame occur. For evaluation of the CRC-32 the entire frame must be cached.

Starting from known bit error rates BER _in (BER, bit error rate) in both independent channels results in a bit error rate of the forwarded bit error, which is dependent on the length of the frame. BER _outCRC = BER _inCanalA · BER _inKanalB · _Frame length (in bits) where BER _in · frame length (in bits) is the probability that a transmission error will occur in one frame, resulting in that errors in the other channel can not be corrected.

Tabelle 5: maximale Bitfehlerraten in Abhängigkeit von den Eingangsfehlerraten BER_in For Ethernet, the maximum frame length is 1624Byte, for Fiber Channel 2212Byte. If one now assumes the maximum frame length and the same bit error rates BER _in both channels, the following values result for BER _out :

Table 5: maximum bit error rates as a function of the input error rates BER _in

Replacement circuit on 8B / 10B Code level

A further error reduction leaves reach if you recognize errors by the 8B / 10B code are reduced. This error reduction is used when both received frames have CRC errors.

• • in beiden Rahmen im gleichen Kodewort ein Fehler auftritt
• • ein Fehler in einem 8B/10B Kode nicht erkannt wird

The equivalent circuit at 8B / 10B level occurs when in one channel a bad detected frame in an 8B / 10B code word an error occurs while the corresponding 8B / 10B code word of the other channel is error free. The equivalent circuit does not work when

• • An error occurs in both frames in the same codeword
• • an error in an 8B / 10B code is not detected

The simplest and thus most likely error pattern that of the 8B / 10B code is not detected is a complementary error, consisting of a 1 error and a 0 error in the same 10 bit code.

The probability that a complementary error will occur in a frame that has already been erroneously detected by CRC is approximately: 9 · BER in · 0.5 = 4.5 · BER in 1 is the probability that an error is in an 8B / 10B code,
9 · BER _in is the probability that an error will occur in the remaining 9 bits of the 8B / 10B code
0.5 is the probability that this error is complementary

The Probability that in both detected as faulty frame the error occurring in the same 8B / 10B code is approximately: 1 / frame length (in Byte)

This means that the error with the approximate probability: 1 - (4.5 * BER in + 1 / frame length (in bytes)) can be corrected by additional equivalent circuit on 8B / 10B code level.

As the output error rate of the combined equivalent circuit of the system thus results BER OUTSYS = (4.5 · BER in + 1 / frame length (in bytes)) · BER outCRC

Tabelle 6: maximale Ausgangsfehlerraten in Abhängigkeit der Rahmenlänge und Eingangsfehlerraten und sich daraus ergebender Ersatzschaltgewinn (NPG, Net Protection Gain) The following table shows the equivalent switching gain Net Protection Gain. NPG):

Table 6: maximum output error rates as a function of the frame length and input error rates and resulting replacement switching gain (NPG, Net Protection Gain)

Claims (1)

Method for bit error-free equivalent circuit for Ethernet and Fiber Channel, wherein the same signal is transmitted via two different optical transmission channels in a ring to a receiver, characterized in that - the equivalent circuit is done at the frame level, wherein for the two synchronized channels of the Ethernet or Fiber Channel CRC-32 transmission error detection is carried out in each case; upon detection of CRC-32 transmission errors in both of the synchronized channels, another transmission error detection for the two synchronized channels takes place at 8B / 10B level and that upon detection of a faulty 8B / 10B code word corresponding 8B / 10B code word of the other channel replaces the erroneously detected 8B / 10B code word.