Present situation
According to the request memory type, the networking saves possibly contains several kind of technologies. What is most common is the optical fiber channel saves confined network (SAN), the majority Major industries have deployed this kind of process good confirmation, the reliable high performance block data level memory network. Since many years, the optical fiber channel technology has experienced several evolutions, the link speed enhanced from 1 Gbps to 2G; Now, some promotes for the first time the testing version product may support the 4G link. Besides uses to make connection between the storage device, the switchboard and server’s, the optical fiber channel link may also use for the bit server system’s in bit server continually in together and connects the storage device. In these bit server system, has the second set of link, they use the ethernet, the load bearing non-block data level transmission capacity, like document level transmission capacity and system administration transmission capacity. Uses two kind of connection technologies as well as the system redundancy requests to mean in the bit system to have two redundancy exchange structure. In 1G to the 4G link speed range, from economical angle embarking, HBA or the controller chip as well as the optical fiber switchboard port’s cost may support this kind of architecture. Certainly, with uses one kind of connection technology to compare, this kind of architecture cannot simplify the multi-structure system’s supervisory work.
Another kind used many kinds of technical the storage schemes to use the network attachment to save the (NAS) equipment, also called the external memory storage device; These equipment (more suitable) to process the document level memory transmission capacity, but non-block data level. One year ago, all external memory storage device also was only has disposed the ethernet connection in front end. The external memory storage device’s internal floppy disk saves may carry on the expansion through the rear end optical fiber channel port. Is the same with the external memory storage device’s all internal memory, all expansions are (stored datum which the block data level external memory storage device unfolds are document level view), therefore the external memory storage device manufacturer increased front end the block data level turning on function through the front end optical fiber channel port, or the iSCSI goal pattern software which the ethernet port supports through the external memory storage device front end increased the iSCSI turning on function. Is similar to the bit server is the same, when link speed when 1G to 4G scope, the external memory storage device can through many HBA support integration block data/file memory.
The next generation link speed will have high? 10G? 8G?
The optical fiber channel 10G port appeared for the first time in 2002 main engine bus adapter prototype demonstration; in September, 2003, has demonstrated the first optical fiber channel 10G switchboard port. In the demonstration, the switchboard port may move under two kind of patterns - acts as between large-scale optical fiber channel switchboard’s ISL (between switchboard link), or realizes the exchange interoperability together with FC HBA under the main engine connection pattern. Tentative plan was, after the 2G FC next migration will achieve 10G. But had discovered several questions, caused this kind of supposition to fall two times, namely 4G. First was the FC driver side has the change, the driver connection migrated 4G. Developed 4G the SERDES chip essence, caused the switchboard and the HBA other 4G infrastructure function module has achieved 4G. The second question is, 10G is unable backward compatible 1G and 2G. From 1G to 2G and in the 4G transition process, the light infrastructure maintains invariable, the use already installed 50/125 and more than 62.5/125 micron mold fiber optics implementation short distance application (most length 550 meters). But, when the speed achieves 10G, this kind of fiber optics’s distance cannot surpass 100 meters, has 33 meters short. In order to achieve the best 10G multi-mold performance, the people have formulated the new fiber optics standard, 850 nm, the laser optimize, 50/125 micron optical fiber, the effective pattern band width is 2000 MHz*km. This kind of optical fiber may use the 10Gbase-S connection to achieve 300 meters distances. A more detailed discussion, please see also the 10G ethernet alliance website (www.10gea.org). In this website discussed deploys time 10G with traditional 1G and the 100M link some very important fiber optics infrastructure matters needing attention.
Therefore, although the 10G optical fiber channel technology may succeed the movement, but the 10G link’s high cost (the 10G light transceiver present price about 1000 US dollars) adds the present to be very few some people to need this kind of band width, will enable FC SAN 10G only to limit using the near future in ISL; This kind of situation even possibly continues many years.
10G ethernet’s what state? In the ethernet aspect, the 10GbE port on a liang year, mostly had already served as ISL. If FC SAN will not need the 10G band width in the near future, the ethernet/IP memory domain’s trend of development will promote the application which the 10G main engine will connect? Yes, definitely meeting. If we have discussed that zai guoqu di ji nian zhong, “the fusion saved” obtains the preliminary deployment - mainly to include the equipment which the union block data level and the document level saved. Considered the 10G light mouth the cost, presses one kind of current capacity type very much (block data or document) uses a 10G link. But, if may support many kinds of current capacity types on sole HBA, permission use sole 10G link connection fusion storage device, then may solve the 10G cost problem. We may be called this kind of sole current capacity pipeline “Unified Wire”. The first possibility uses this kind of pipeline’s equipment possibly has extra block data memory support function NAS Filer. This method’s obvious superiority is, it eliminated the independent document and the block data memory network. If has used the iSCSI block data connection, then it may eliminate the independent optical fiber channel and the ethernet infrastructure. Why fuses the memory to use the ethernet, but does not use the optical fiber channel? Because the optical fiber channel can only support the block data current capacity. The optical fiber channel may support the TCP/IP current capacity the tunnel technology, but cannot accelerate in HAB the IP storehouse.
Why can by the 10G link speed, process the TCP/IP current capacity in the hardware? If we pay attention to the present implementation iSCSI, in which software iSCSI storehouse might use in the majority operating systems as well as has implemented all iSCSI in the card function hardware HBA. If the link speed is 1G, then the software and the hardware implement the performance will be very close - the difference to lie in the movement storehouse consumption main engine CPU. If uses the hardware, the movement storehouse only needs 5% CPU, but the software possibly consumes about 40%, will be higher - is decided by the block data size. In the majority situations, used the software already to be enough.
At present, the development used for to process the block data and the document acceleration current capacity driver storehouse says that easily not likely. Processes many current capacity types the driver question to be complex. But, some emerging enterprises were already beginning to develop the chip, and starts to design is suitable for this kind of HAB driver. Various standard organization is also exploring several kind of models, seeks can realize this goal model well. Below is each method schematic drawing:
Ethernet sole line HBA and SW construction
Today, several manufacturers may provide iSCSI HBA. at the beginning of 2005, in the market will present the first section of combination integrated line HBA and ASIC. In this kind of situation, along with the different method’s discussion and the test, solution’s flexibility will become very important. In the above system structure drawing has introduced RDMA. Using RDMA (long range data storage access), may put in directly the data the application in the buffer, thus eliminates the RDMA data transmission request many data copy. IETF will be discussing, processes all current capacities on RDMA the method to arouse people’s interest? Or, uses in the document current capacity (non-block data current capacity) RDMA will win? Only the time can answer this question. Here will also have one to have the dispute view - only then traditional large-scale optical fiber channel SAN only then to shift the 10G optical fiber channel ISL. the 8G optical fiber channel link is possibly the last optical fiber channel high speed main engine connection. the 10G main engine connection possibly only uses in the ethernet domain.
Regardless of the result, several years will be how thrilling recently; All these new technologies will appear, to strive for to obtain the acceptance. The memory networking domain cannot be lonely!
the 10G ethernet has surmounted copper cable sidebar?
In the market has many kinds of technologies to be possible to realize based on the copper cable carries on the 10G signal the transmission. And one kind is Infiniband develops, it is called CX4.
1. this technology has used the XAUI encoder/decoder which same 10G bit MAC, in the XGMII connection and 802.3 assigns, take the 3.125g baudrate the signal decomposition as 4 different ways.
2. the transmission pre-emphasis compensates the PC module’s loss emphatically in the high-frequency component. Coupling and cable assembly.
3. although the coupling and the cable assembly is the design uses in infinBand, but is actually 802.3ak to adapt the balancer request to stipulate the use.
4. is weakened the signal by the cable assembly to carry on last pressure-rise by the receive balancer.
CX4 is possibly suitable for the rack between the system; Because it supports the distance is too short, but cannot use in the more widespread data central. Another shortcoming is, the CX4 shielded cable’s volume is too big, causes the wiring difficulty to enlarge. Because many switchboard manufacturers have not made the widespread pledge, therefore supports CX4 in the 10G ethernet switchboard still at to discuss.
Realizes the 10G prospect how on the unshielded twisted pair line (UTP) electric cable? Until now, each ethernet 10X has come under the speed influence, widely uses this technical the key to lie in UTP the support ability. Some emerging enterprises were already beginning to research and develop UTP the 10G technology, and has demonstrated several kind of prototypes. Looked on the present situation that as if impossible to support 10G on 5 kind of UTP; But the people generally believe that in 35 meters distances, kind or on 7 kind of electric cables supports 10G in 6E is possible.
10G light sidebar
The present 10G solution uses XPAK or the X2 light module. These equipment support the XAUI 4 narrow connections, may transmit the 10G signal to has in the good shield light module. But these modules are expensive. In the future, we will see in the smaller XFP module will use the 10G light mouth; This kind of module is ready to be used the LC size coupling, is similar with today’s 2G and the 4G light module. These light mouth’s input device is XFI connection - PC on the board sole 10G serial signal. Yes, does this is possible, moreover already had the related demonstration. According to the estimate, the XFP light mouth’s cost will drop greatly.
