This invention relates generally to the field of communications and, more particularly, to a system and a method for providing an Ethernet interface.
Communication systems and architectures have become increasingly important in today's society. One aspect of such architectures relates to interfaces, which can maximize bandwidth and minimize delays associated with data and information exchanges. Many architectures for effectuating proper data exchanges can add significant overhead and cost in order to accommodate a large number of data streams. In addition, as the size of data streams increases, communication systems should be developed to meet these robust conditions. Performance and compatibility are two eminent concerns for any person who is relegated the difficult task of developing an interface standard.
Accordingly, the ability to provide a communications system that consumes few resources, optimizes bandwidth, achieves superior performance, and offers minimal delay presents a significant challenge for network operators, service providers, and system administrators alike.
For a more complete understanding of the present invention and its advantages, reference is now made to the following description, taken in conjunction with the accompanying drawings, in which:
In one embodiment of the invention, an apparatus is provided that includes n communication channels, m communication media interfaces, and v virtual lanes. V is a positive integer multiple of the least common multiple of m and n. An information stream is transferred into data and alignment blocks striped across all of the v virtual lanes, the blocks being communicated from the virtual lanes onto the communication channels. The blocks are received on the communication channels. Each of the communication channels transmits a different portion of the blocks striped across all of the v virtual lanes.
In more particular embodiments, v>=n>=m. The communication media interfaces can be electrical and optical. Each of the communication channels can include a SerDes interface operating at least 5 Gigabits per second. Furthermore, each of the m communication media interfaces is configured to transmit a different stream of information over a single optical fiber. The single fiber is only one case, other cases could include multiple fibers and/or copper could be used.
In still other embodiments, the transforming includes one or more encoding modules for encoding the information stream into the data blocks using 64B/66B encoding (or other types of encoding could certainly be used). Also, each particular alignment block of the alignment blocks uniquely identifies on which of the virtual lanes the particular alignment block was placed.
A technical advantage of certain embodiments includes enabling lane bonding/aggregation at electrical and optical levels with a single alignment mechanism. This solution could allow for one PCS for many PMDs. In addition, in the proposed architecture there is low overhead, which is independent of packet size. Such a solution also minimizes latency, and minimizes buffer sizes: both of which are advantageous. Embodiments that follow may enjoy some, all, or none of these advantages.
The present invention provides a new interface that is scaled to meet the needs for 100GE. Such an optimal interface incorporates the optical level encoding. In addition, it allows the electrical interface encoding to be reused on the optical side even if the number of optical lanes is less than the number or electrical lanes with just simple multiplexing. There exists a need to define a new 100-gigabit electrical interface to a 100-gigabit optical module, as part of the 100 Gigabit Ethernet Standard (this can also be applied to the 40 Gigabit Ethernet Standard). Note that the present invention is not limited to 100 G, but readily includes 40 G models, and can fit future rates. A new electrical interface to a 100-Gb optical module should be defined as part of the 100GE standard. Past interface examples include XGMII, XAUI, and XFI for 10GE.
Note that certain definitions should be understood before proceeding further with the proposed concept. The CTBI lane is an electrical path from the transmit PCS to the transmit PMA, and from the receive PMA to the receive PCS. The PMA lane is an optical path from the transmit PMD to the receive PMD. The virtual lane is not a physical path, but rather a virtual path that can be multiplexed and carried across either an electrical or an optical embodiment.
The proposed electrical interface is ten (or four in the case of 40 Gigabit Ethernet) SERDES lanes: each run at 10.3 GHz and transporting two, bit multiplexed, virtual lanes encoded using 64B/66B encoding along with special alignment words. [As explained more fully below, the encoding is not bounded to SERDES and the electrical line and instead is bounded to the virtual lines.] The tendered proposal covers both support for PMA with alignment capabilities and a simple bit muxing PMA. The virtual lanes concept and the architecture and implementation PCS and PMA that support that concept are described below.
One method of the present invention defines a new electrical interface for a 100-Gb optical module. Past interface examples are XGMII, XAUI, and XFI for 10GE. A new interface has been developed that builds on the past 10 G interfaces, but that is scaled to meet the needs for 100GE and, further, that incorporates the optical encoding. It allows the electrical interface encoding to be reused on the optical side even if the number of optical lanes is less than the number or electrical lanes with just simple multiplexing.
The proposed electrical interface is ten SERDES lanes, where each run at 10.3125 GHz each of them transporting two virtual lanes encoded using standard 64B/66B and bit multiplexed (along with special alignment words). The standard 64B/66B encoding provides for frame delineation, and a guarantee of at least one clock transition per 66 bits. In addition, scrambling is performed on the 64 bits of data in order to provide greater transition density.
In operation, the data coming from a 100 G MAC can be striped 8 bytes at a time across each 20 virtual lanes and encoded in 64B/66B encoding. A special alignment code can be sent on each virtual lanes periodically. The alignment code will use reserved values of the standard 64B/66B encoding values, and can take 80 bytes of overhead to send. This will allow the receiving side to re-align the lanes of data, even after sending it over the 10 electrical lanes and a smaller number of optical lanes. For instance a typical application is twenty virtual lanes bit multiplexed and transmitted over a ten lane electrical interface to an optical module, and within the optical module it can be bit multiplexed down again to five lanes (or two, or four, etc.) in order to go over five separate fibers or wavelengths.
On the receiver side, the lanes can be first bit demultiplexed down to 10 lanes in the optical module, sent again across a ten-lane electrical interface and finally, back up to the MAC layer, demultiplexed to twenty virtual lanes, decoded and realigned using the alignment code, before extracting the transmitted data.
The bandwidth that is used for the alignment mechanism can be regained by defining an Alignment Deficit Counter (ADC), which over time will borrow Inter-Packet Gaps bytes (IPG) so that the overall line rate of 100 G is 10x10GE. The plan is to ‘steal’ four bytes each packet until the ADC is at 80, and then the alignment words can be sent.
There are some keys to this proposed concept, including reusing familiar 64B/66B encoding, while adding a multi-lane alignment mechanism. Furthermore, this alignment mechanism can be reused for both the electrical interface and at the optical interface without re-encoding the data. Such a concept provides an efficient 100 G electrical and optical encoding method and protocol. Additionally, a deficit counter mechanism allows for a full line rate.
Thus, features of the proposed concept include alignment words: one alignment word per virtual lane per 16K words sums to 0.006%, and virtual lanes that provide a mechanism to support alignment only on the receive PCS (none necessarily in the optical module).
Note that there is little work in the area of 100GE interfaces since the standard is in its infancy. Other proposals for bonding lanes together are fixed fragment-based, which has a high overhead (10-40%) that is problematic.
Note that in taking a step back from the details of the proposed idea, considerations for the interface include: supporting a single rate MAC speed (100 G or 40 G); optical and electrical technologies requiring a multi-channel/lane approach; channels needing to be bonded into one flow; scheme needing to be robust to technology advances; allowing future PMDs with reduced numbers of channels; minimizing complexity of PMDs/Optical Modules; low overhead that is independent of packet size; enabling small buffers; allowing for differential delay due to wavelengths/fibers; and no auto-negotiation required between end points.
Non-considerations for the interface include support for a scalable MAC; one MAC rate for the MAC client is preferred; resiliency to single lane failures (not considered necessary since a system level); redundancy would be required in any event (e.g. cable break, card-power failure); and working/protection links more than likely to be used.
In terms of a transmit overview, the interface consists of ten SerDes per-direction: each running at 10.3125 Gbps raw rate. In the transmit direction, the following is done in the PCS. The aggregate 100 Gbps stream is first encoded into 64B/66B blocks using the same rules as 10GBASE-R. This allows frames to be delineated, in addition to allowing control blocks to be sent, idles being identified, etc. 64 bits of each 66-bit block are scrambled by a self-synchronous scrambler in order to provide transition density and control DC balance.
Note that the definition of CTBI allows simplified implementation of muxing function to enable support of many variants of PMDs that may be defined. Note also that each of the communication media interfaces can be operating at a different wavelength multiplexed over a same optical fiber in a wave division multiplex (WDM) application.
Then the 66-bit blocks are inverse multiplexed to virtual lanes. Virtual lanes allow the ability to only have alignment in the receive PCS. Added to each virtual lane is a periodic alignment block, which identifies the virtual lane and allows the receiver to get everything back in alignment (this interrupts packet transmission). Block 1 goes to VL 1, block two goes to VL 2, block n goes to VL n, block n+1 goes to VL 1, etc.
Subsequently, the virtual lanes are bit multiplexed onto each electrical lane. If there are twenty VLs and ten electrical lanes, this means that two VLs are bit multiplexed together onto each electrical lane. An interesting note, it does not matter which VL goes to which electrical interface, in fact, this interface is completely tolerant of misconnections of the electrical interface SerDes lanes. Each 10.3125 G stream is then sent to a SerDes for transmission to the optical module.
In terms of the receive overview, the interface consists of ten SerDes per-direction: each running at 10.3125 Gbps raw rate. In the receive direction, the following is done in the PCS. Each 10.3125 G stream is received from a SerDes from the optical module. Then the virtual lanes are bit demultiplexed from each electrical lane into VLs. If there are twenty VLs and ten electrical lanes, this means that two VLs are bit demultiplexed from the each electrical lane. Now the receiver will find a 66-bit alignment on each VL independently. This follows the same process as 10GBASE-R uses. Once a 66-bit alignment is discovered on all VLs, then the VLs should be aligned. This is done via some amount of buffering and pointers within the receive PCS. The alignment blocks were not scrambled so those can be found, and then the delay between the VLs can be compensated for.
Once VL alignment is found, then the data is again a coherent 100 G stream. It can now be descrambled (just the 64 bits, not necessarily the two overhead bits). The aggregate 100 Gbps stream is then decoded back into frame streams, which can be sent up to the MAC.
After the VLs are created, alignment blocks are added to each VL periodically and at the same time. This will allow the receiver to re-align everything. VLs are created to ensure that all of the bits for a particular stream of 66-bit words are transmitted on the same CTBI and PMD lane combination. This will ensure that all of the bits for any particular virtual lane do not experience any differential (bit-to-bit) skew and are, therefore, always received in the correct order at the receive PCS.
If the bits for a virtual lane were not constrained to be transmitted on the same CTBI and PMD lane combination, then the bits would arrive out of order at the receiving end of the link (after each of the bits having experienced different delays) and there would be no way to piece together the information.
There will still be a differential delay between the different virtual streams, as these will have passed over different CTBI and PMD lanes, but it is a simple task to use the markers added to each virtual lane to detect and compensate for this skew. This allows the receive PCS to be able to demultiplex all of the virtual streams back into one single continuous 100 G aggregate stream with all of the individual 66-bit words in the correct order.
In addressing how many VLs are needed, this depends on which optical and electrical lane combinations are sought to be supported. What should ideally be chosen is the least common multiple of the number of electrical lanes and the number of optical lanes (be they wavelengths, fibers etc). Having the LCM of these two numbers will ensure that the bits for a VL stay on one combination of the interfaces.
In terms of the virtual lane creation, virtual lanes can be created simply by a round robin function that plays out the 66-bit blocks in order to each virtual lane. If there are twenty VLs, then after twenty VLs, the user is back to the beginning (VL#0).
If there are greater than twenty VLs, then potentially the run length could become a problem. In that case, different patterns (that look more random) could be sent on each VL for the alignment block, which would reduce the chance of interference.
The following are the worst-case Optical Skew numbers:
For up to 300 m of MM, 30 nsec;
For 10 km SM, 5 channels, 17 ns; and
For 1000 km, 10 channels, 543 ns.
For electrical skew, XAUI's specification for electrical skew is 41 UI, or about 13 ns. Assume that there are many places for electrical skew: Tx CTBI, in the Tx optical module, in the Rx optical module, and in the Rx CTBI. That is at least four different skew sets, so lets go with 4*the XAUI budget for now, or 52 ns for electrical skew. Therefore, the worst-case total skew is ˜600 ns. This equates to 60 k bits of memory needed for this function. Rounding up to 1 u-sec, then it is 100 k bits of memory (˜0.2 mm square per 1 uS in 65 nm) needed for the deskew/alignment function.
As the virtual lane data is stored into each VL buffer, the alignment word is searched for. This is done simply by looking for the alignment pattern (as shown in section 3.1.5 above). Once a candidate control block matches the alignment pattern (excluding the VL fields), the receiver will look again 16 k blocks later for the next alignment block. If four are found in a row error free, then alignment on that VL is declared. In one example, if four in a row are found to be in error, then out of alignment is declared for that VL.
Once the alignment block is found on all VLs, then the data can be read out in order to re-create the original 100 G data stream. Data is re-aligned and reordered by reading it out in from the appropriate buffers based on the alignment blocks. Now that the 100 G stream is recreated, descrambling can take place.
In summary, the CTBI is proposed as a potential interface for HSSG. It enables lane bonding/aggregation at electrical and optical levels with a single alignment mechanism. This solution allows for one PCS for many PMDs. There is low overhead, which is independent of packet size. It minimizes latency, and minimizes buffer sizes: both of which are advantageous.
Some of the steps discussed with reference to the preceding FIGURES may be changed or deleted where appropriate and additional steps may also be added to these process flows. These changes may be based on specific communication architectures or particular interfacing arrangements and configurations of associated elements and do not depart from the scope or the teachings of the present invention.
Additionally, although described in specific environments and contexts, the present invention could be used in countless applications. Numerous other changes, substitutions, variations, alterations, and modifications may be ascertained by those skilled in the art and it is intended that the present invention encompass all such changes, substitutions, variations, alterations, and modifications as falling within the spirit and scope of the appended claims. Moreover, the present invention is not intended to be limited in any way by any statement in the specification that is not otherwise reflected in the appended claims.
Although the present invention has been described with several embodiments, a myriad of changes, variations, alterations, transformations, and modifications may be suggested to one skilled in the art, and it is intended that the present invention encompass such changes, variations, alterations, transformations, and modifications as fall within the scope of the appended claims.
This Application claims priority under 35 U.S.C. §119 of provisional application No. 60/869,505 filed Dec. 11, 2006, entitled SYSTEM AND METHOD FOR PROVIDING AN ETHERNET INTERFACE. This Application also incorporates by reference that Provisional in its entirety.
Number | Date | Country | |
---|---|---|---|
60869505 | Dec 2006 | US |