what is the server memory
server memory and memory (RAM), it has to do with ordinary PC (personal computer) machine appearance and structure exists inside there is no clear difference in the material, mainly on the memory introduces some new technologies, such as ECC, ChipKill, hot-swappable technology, high stability and performance of error correction.
server memory technology,
on common memory, often use a technique, namely the Parity, Parity check code (Parity check codes) is widely used in the wrong words.if budgeting (error detectioncodes), they add a check for each data of character (or Byte), and the ability to detect all odd (even) in a character with a mistake, but Parity have a disadvantage, when a computer to a Byte has a mistake, not sure where a bit error, also won't be able to correct mistakes.Based on the above situation, leads to a new memory error correction technology, ECC, ECC itself is not a memory model, nor a special memory technology, it is a computer instruction is widely applied in various fields, error correction technology is a kind of instructions.ECC English full name is The Error Checking and Correcting”, the corresponding Chinese name is called Error checking and correcting ", from the name we can see that its main function is Find and correct errors ", it is more advanced than odd-even correction technology aspect mainly is that it can not only find errors, and to correct these errors and correct these errors after the computer to perform the following tasks correctly, ensure the normal operation of the server.Say it is not a memory model, it is because is not an influence on the speed of memory structure and storage technology, it can be applied to different types of memory, like said before Parity correction "Memory, it is not a kind of memory, the first application of this technique is the EDO memory, SD is used now, and ECC memory mainly starts from the SD memory is widely used, and the new DDR, RDRAM has corresponding application, the current mainstream of ECC memory is actually a SD memory.
Chipkill technology is IBM server memory in order to solve the current lack of ECC technology development, is a kind of new ECC memory protection standards.We know that the ECC memory can only detect and correct single bit error at the same time, but if detected more than two bits of data at the same time there is an error, is powerless in general.The ECC technology is widely used in server memory, partly because before this other new memory technology is not yet mature, moreover in the present server system speed is very high, generally on the frequency at the same time, the phenomenon of the multiple bit errors are rare, this makes ECC technology has been fully recognized and application, makes the ECC memory technology became the standard almost all the memory on the server.
but with the CPU performance of the server based on Intel processor architecture in increase in multiples of geometry level, and the performance of the hard drive in the same period increased only a handful of times, so in order to obtain adequate performance, the server needs large amount of memory to temporarily hold on the CPU to read the data, data traffic has become such a single memory chips on each time you visit often to provide 4 (32-bit) or more than 8 (64) bits of data, a one-time read so much data, the possibility that a number of data errors will be improved greatly and the ECC couldn't correct more than double bit error, so it is likely to cause the loss of all the bits of data, the system will collapse soon.IBM Chipkill technology is to use the memory of the substructure method to solve this problem.Memory subsystem design principle is that the single chip, regardless of the number of data width is only for a given ECC identification number, up to the influence of it a bit.For example to illustrate that if using 4 bits wide DRAM, the parity of the 4 bits of each will form different ECC identification number, the ECC identification code is to use a single data bits to save, that is stored in different memory address space.Therefore, even if the whole memory chips out of the fault, per ECC identification number will also appear a bit bad data, and this kind of situation can repair by ECC logic, so as to ensure the fault tolerance of memory subsystem, ensure the server in the event of a failure, has a strong resilience.Using this memory technology of memory can check and repair at the same time four wrong data, the reliability of the server got more adequate security and stability.
Register or directory registers, the action on the memory we can understand a book catalog it, with it, when memory been instructed to read and write, will first retrieve the directory, and then to read and write operations, which will greatly improve the working efficiency of server memory.With the Register must take Buffer memory (Buffer), and now can see the Register also have ECC memory function, its main application in the high-end server and graphics workstations, such as IBM Netfinity 5000.
(4) the FB - DIMM
FB - DIMM (Fully Buffered - DIMM, full buffer memory modules), Intel is in DDR2 and DDR3 developed on the basis of a new type of memory modules and interconnection architecture, can match already now DDR2 memory chips, also can match DDR3 memory chips in the future.FB - DIMM can greatly improve the system memory bandwidth and greatly increase the memory capacity.Fb -
DIMM technology is Intel in order to solve the memory performance of overall system performance and development, on the basis of existing technology realized leap-forward performance improvement, and the cost is relatively low.In the entire computer system, memory is a key factor of the machine performance, light has a faster CPU, no good with the memory system, CPU performance again good also don't play.This kind of situation is determined by the principle of the computer, when the CPU in computing the required data from memory, if memory system can't supply the CPU data in time, the CPU has to for a long time in a wait state, hardware resources are idle, the performance of natural don't play.For ordinary PC, because it is a single processor system, the current memory bandwidth have been able to meet its performance requirements;For multiple servers, as is the multiprocessor systems, the memory bandwidth and memory capacity are desperate for, traditional memory technologies have been unable to meet their needs.This is because the current regular DIMM USES is a kind of Short line connection "(Stub - bus) topology structure, this kind of structure, the data bus of each chip and memory controller has a short circuit is linked together, this will cause electrical impedance does not continue to sex, thus affecting the stability of the signal with the complete, the higher the frequency or chip data, the more the greater the effect.Although Rambus XDR launched by the new memory technology such as memory has a very high performance, but there is the problem of high cost, making it not popular.And the emergence of FB - DIMM technology can better solve the problem, can provide greater memory capacity and ideal memory bandwidth, can maintain a relatively low cost.FB - DIMM compared with XDR, although not as good as new architecture XDR performance, but it is much cheaper than XDR cost.
compared with the existing ordinary DDR2 memory, FB - DIMM technology has a great advantage: in the case of memory with the same frequency can provide four times the normal memory bandwidth, and can support one of the biggest memory capacity 24 times also reached the common memory system can support 192 gb of memory.FB - DIMM the biggest characteristic is to use existing DDR2 Memory chips (later also will use DDR3 Memory chips), but it USES a Buffer Memory on the PCB chip AMB (Advanced Buffer Memory, Advanced Memory Buffer) will parallel data into serial data stream, and through the PCI Express similar point-to-point high-speed serial bus to transmit data to the processor.
compared with the ordinary DIMM module technology, FB - DIMM and data between the memory controller and command transmission is no longer the traditional design of parallel lines, and using the serial interface is similar to the PCI Express multichannel parallel design, in the form of serial data transfer.In this new architecture, each DIMM on the buffer is in series with each other, is a point-to-point connection between data after the first buffer and send down a buffer, so, the first buffer and memory controller can keep stable, the connection between the impedance to help with the new capacity of the frequency.Typical type
server memory currently used memory server has the SDRAM and DDR, DDR2 memory of three.DDR SDRAM
(1) (2) (3) DDR2 SDRAM
server what is memory?Due to server within various technical compatibles relatively is much more strictly, it emphasizes not only the speed of memory, but its inherent error correction capability and stability.So for now on outer frequency can only be compatibles or ordinary desktop memory heels.Outside the current desktop frequency in general has reached more than 150 MHZ era, but 133 frequently is the mainstream.The server due to the parts outside the frequency and the requirement of high stability, outside the mainstream frequency is 100 MHZ, 133 MHZ frequency has gradually in the server implementation of each class, the choose and buy the server had better choose 133 MHZ frequency, of course!Memory, and other accessories, too, to synchronize as far as possible, otherwise it will affect the performance of the entire server.The main server memory brand mainly Kingmax, kinghorse, hyundai, samsung, kingstone, IBM, VIKING, NEC, etc., but mainly in the front several is common on the market, but also can get good quality guarantee.