Overview of the XAUI, XLAUI and CAUI: Part2

Part2 of the “Overview of the XAUI, XLAU and CAUI” blog focuses on the details of the XLAUI and CAUI interfaces and how the new MLD function works. Rather than re-inventing the wheel we suggest you grab copies of the top class presentations, from the IEEE P802.3ba Task Force Meeting Munich May 2008, listed in the text below.

The “hidden” XLGMII and CGMII interface
(XL/CGMII and RS proposal :

The XLGMII/CGMII is specified as a nominal 64bits wide data bus, separated into 8 lanes, with control/data signal per lane (not to be mixed up with the virtual lanes of the MLD function in the PCS). At 64bits the clock rate must be 625MHz for 40G and 1.5625GHz for 100G, though scaling this down to a typical 256width bus of 32 lanes would results in more reasonable values of 156MHz and 390Mhz.

The control lines, one per 8bit lane, indicate if the associated lanes contain data or control and help to identify the start/stop of the packet and indicate when IPG or Ordered Set control data is present. Unlike 10G, data is always assumed to be byte aligned which helps make the PCS a little less complicated (unless your build a 4×10G / 40G Ethernet MAC!)

XLAUI/CAUI – The PCS and its new function the MLD
(100/40Ge PCS proposal :

The XLGMII (and CGMII) provides the interface to the PCS block, sometimes referred to as the PCS/MLD block (technically the MLD is just a function inside the PCS). In the transmit direction the first stage is the 64b to 66b encoding where the information in the outband control/data lines is turned into a 2bit sync header and, if a control word is present, a block type field which resides in the first byte of the 64bit payload. The 64bit payload is then scrambled with the standard 1+ X^39+X^58 and the 2bit sync header added to make a 66bit block.

Upto this point it has been the same as serial 10G Ethernet and if the optical link were serial the data could have been passed straight away to the PMA function, however to pass over the multi 10G serial interface of the XLAUI and CAUI the data needs the help of the MLD block (Multi Lane Distribution) that maps the data into Virtual Lanes (VL’s) so it can pass over the 4×10G XLAUI and 10×10G CAUI. For the XLAUI, the number VL’s is 4 and corresponds to both the number of 10G channels in the XLAUI and the 4 channels of the PMA for 40GBase-xR4. The CAUI however has 20 VL’s! the reason for that will be explained later.

Assuming a 40G Ethernet link, the transmit path of the MLD is as follows; The data is transmitted by sending 64/66b code words on each lane in a round robin fashion, this basically means there are four 64/66B codes words being sent at the same time, one code word per phyiscal link (just like XAUI) . In order to keep alignment a 64/66b alignment word is sent simultaneously in each VL every 16384 66b blocks. Unlike the XAUI were the /A/ code is only sent in the IPG the MLD alignment 66b word can be sent in the middle of the payload so it needs to be removed by the MLD function in the receiver before the 64/66b decoding. The four VL’s are then passed to four separate SERDES which each produce a 10.3125Gbps serial signal giving us the transmit XLAUI interface. The receive path is pretty much the reverse but with the need to perform 64/66b block alignment on each of the physical lanes of the XLAUI, to detect (and delete) the alignment word in each VL and to finally buffer the data on each VL so that the received 64/66b blocks are re-aligned so they can be decoded and passed to the XLGMII interface.

The process for the CAUI is identical apart from the fact that there are 20 VL’s. This is due to the fact that whilst the CAUI is 10bits the PMA service layer can be either ten (100GBase-SR10) or four (100GBase-LR4). The lowest common dividable number for 10 and 4 is 20, hence 20VL’s for 100G.

Available implementations from Sarance Technologies and MorethanIP
OK first point; this is not a competitive analysis and we think it’s fantastic that these two companies have invested their time and resources in producing pre-standard building blocks to help 40/100G Ethernet hit the ground running.

Sarance Technologies is based in Ottowa,Canada and along with their 40/100G FPGA IP Core also provide FPGA IP for the 120G Interlaken interface as well as Core for Classification and Traffic Management.

The Sarance 40/100G Core is in three sub blocks; MAC, PCS and MLD and supports either a single channel 40 or 100G implementation. The data path of the Core can be selected to be either 256bits/200Mhz (for 40G) or 512bits/250MHz (for 100G). There is no internal XLGMII/CGMII interface so it seems it is intended to use all three block together. On the XLAUI/CAUI interface the core relies on the internal SERDES on the FPGA, whilst the diagram on the product brief implies that it will be a 10G SERDES but it can probably also support 2×5.12Gbps as they are working closley with Xilinx but its not stated in available info. The Sarance core is available for both Altera and Xilinx, but they seem to have more connection to Xilinx, they were part of a demo with Xilinx, Netlogic, Avago and IXIA at NFOEC in March 2009 and provide the 100G Ethernet core on the Xilinx Virtex5 TXT 100G Platform.

Visit the Sarance Technologies website at http://www.sarance.com

MorethanIP is based in Munich,Germany and has been going for nearly 10 years. They have a good portfolio of Ethernet (GigE/10GE) and SONET FPGA IP Cores.

The MorethanIP 40/100G IP Core comprises of a PCS Core and a MAC core, thus enabling then to be used separately. The interface between the MAC and PCS is a XLGMII or CGMII interface as described above. The MAC core supports the standard necessary features but they also added some support for per channel flow control and time stamping to help with applications such as FCoE (nice!). On the PCS side both 4 and 20VL’s are supported with again the SERDES being provided by the FPGA, the SERDES can be either of the 10.3G or 5.15Gbps(XSBI) variety, the support for 5.15Gbps enable lower cost FPGA’s to be used. The MorethanIP 40/100G Ethernet solution is available as both generic synthesizable cores, in both VHDL and Verilog, and  a targeted version optimized for Altera StratixIV GT.

Visit the More than IP website at http://www.morethanip.com

Best of luck to both companies!

The 40G Ethernet Resource Center

One Response to “Overview of the XAUI, XLAUI and CAUI: Part2”

  1. Alex Lubivy Says:

    Good stuff, thanks

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: