The Data Center Infrastructure of 40G and 100G

Once the server upgrade, top of rack switch uplink require a higher speed. However, during the transition from 1G to 10G full of frustrations. In the past, server vendors comes 1GbE RJ-45 LAN-on-motherboard is free of charge. But now a dual-port 10GBASE-T expensive. Cat5e is almost free, Interconnect never in the past is not a serious cost problem, and now it is. The server companies sale 10G ports, can be mounted on pluggable “daughter card”, cut off the posterior of the subsequent market competitiors and monopoly high prices. Daughter card can and 1G 10GBASE-T, 2-4 SFP+ port and dual-port QSFP combination. With 100G CXP and CFP /2 will be collocation is used. Due to server company in 10G/40G upgrades made a large sums of money, we will ask such a question: “server company will return to the LOM mode, buyers will not need to pay?” Our answer: Yes. But just before the transition to 40G! 10GBASE-T has a problem in high power consumption, size and cost, therefore, when the market of the 10GBASE-T in the development of 28-nm version, so that the SFP+ DAC take advantage. This makes the entire industry landscape has changed greatly. DAC also has its problems. Because it is electrically connected to two different systems, not all of the SFP+ port is the same.

The server-switch link update from 1G to 10G, and the switch uplink increased to 40G, TOR (Top of Rack) switch to connect to the EOR (End of Row) switch until the aggregation switching layer. Data center operators just the economic gloom stood up, still tight budget. “Incremental upgrade” is the investment strategy of the operators. Increase “necessary” 10G/40G link is the current investment. 100G seems to be the exhibition and media attention, but 40G is to make money in the next 2-3 years. Data center began to need 4 ~ 6g, not even to 10G. Therefore, many data center is still in excess, will need to upgrade “. Google, Facebook, Microsoft and other so-called $ 1 billion super data center so that people stare, but they do not represent the mainstream of the data center.

To chase fiber transceiver opportunities, multiple transceiver suppliers is the first to provide less than 50 meters transmission distance of 40G QSFP SR transceiver and Ethernet AOC. 40G QSFP MSA blessed with multi-mode fiber can support short-range (SR) – 100 meters, with the dual-core single-mode fiber with 10km-all in the same QSFP switch port can support. QSFP can plug 36-44 ports per line card, while the CFP can only be inserted 2 in 32W. Although it is very popular in the telecommunications, but not in data communication! OEM prices ranging from $2000 to $3000, it depending on the needs of the data center or telecommunications.

The urgent need for the data center to support tens of thousands of 100G medium transmission link length from the information explosion in demand. Industry conference clamoring. These flow requirements from server virtualization, big data, smart phones, tablet PCs and even software defined network (SDN). Large core switch companies mainly 10-Channel CXP for multi-mode, working together with the transceiver and the AOC. In 4x25G the 25G transmission, multi-mode noise spikes may pose a threat to the multimode transceiver down to 25-50 meters FEC and / or equalization may be required to reach 125 meters. This will make 25-125 meters away from the transceiver higher prices. With 2km Single mode Fiber Optic Transceiverprices narrowed.

At present, the transmission distance of 100m – 600m, there is no economically feasible 100G solutions (unless it can be described with two 40G and 10G transceivers). When the data center becomes greater, which is a hot spot, the IEEE focus of debate. Each additional add 1m, causes the transceiver OEM Price from CXP $ 1,000 he went to the telecentres CFP $ 16,000! Usually claimed transmission 2 km, in fact, can only transmit 400-600 meters, in a loss of data center environments, patch panel and dirty connector can only get 4-5 dB, 10 km link you need to 6dB. Next-generation lasers and instead of SiGe CMOS electronic devices are being developed, but, CMOS electronic harder to develop.

The 40G and 100G are two main data center “form”. Short-range transceiver (SR4), the use of multimode optical fiber can transmit about 100 meters. Using single-mode fiber can transmit 100 meters to 10 kilometers long range transceiver (LR4). This so-called, no formal terminology NR4 aspirations 2km 4dB. SR transceivers typically used to connect a computer cluster and switch layer in the data center. SR transceiver and OM4 fiber combination, can transmit about 300 meters. 125-200 meters, the conversion using single-mode fiber, transceivers and fiber can bring benefit return; even in 25G transmission, also can bring benefits.

The 40G usually in the QSFP or QSFP MSA, usually in four of the 10G channels laying. SR transceiver uses eight multimode fiber (corresponding to one direction), VCSEL lasers and QSFP MSA. LR transceiver uses edge-emitting lasers, multiplex four 10G channels to two single-mode fiber, the single-mode fiber in the transmission of the CFP module MSA 10km, soon also be reached in 28 CFP / 2 and QSFP MSA on this distance. 40G, SR4 and LR4 can be used for the same QSFP switch interface, no problem – just plug in, you can run – you can reach one meter -10 km without any problems. (But still does not work in the 100G)

The 100G SR10 use 20 multimode fiber, VCSELs and CXP MSA. 100G LR4 CFP and two single-mode fiber. Though he promised transmission 100 meters, but the SR10 CXP transceiver typically used to connect large-scale aggregation and core switches from less than 50 meters, when the distance becomes longer, the more than 20-mode fiber will be very expensive, because the multi-mode the optical fiber is about 3 times more expensive than the single-mode fiber. Only in 2012, a number of transceiver companies have announced the development of the CXP 100G SR transceiver. The the 40G QSFP transceiver and the AOCs since 2008 come out. Later on, 4x25G QSFP SR transceiver may to appear CXP transceiver market 10x10G.