Server memory is also memory (RAM), it and ordinary PC (PC) machine There is no obvious difference in appearance and structure, mainly in the memory of the introduction of some new unique technologies, such as ECC, Chipkill, hot plug technology, with high stability and error correction performance.
Server memory Main technology:
In normal memory, a technique that is often used, i.e. Parity, where the same bit check code (Parity check codes) is widely used on the debug code (Error detectioncodes), adds a character (or byte) that checks the bits to each data. and can detect all the odd (even) same-bit errors in a character, but parity has a disadvantage, when the computer detects a byte error, and can not determine the error in which bit, it can not fix the error. Based on the above situation, a new memory error correction technology is produced, that is, the ECC,ECC itself is not a memory model, nor is it a special memory technology, it is a kind of computer instruction widely used in various fields, and it is a kind of instruction error correcting technology. ECC's English full name is "error Checking and Correcting", the corresponding Chinese name is called "Bug Check and correction", from which we can see that its main function is "to find and correct errors", It is more advanced than parity correction technology mainly because it can not only detect errors, but also to correct these errors, these errors after the computer can correctly perform the following tasks, to ensure the normal operation of the server. It's not a memory model. That is because it is not a technology that affects the memory structure and storage speed, it can be applied to different memory types, like the previous "parity correction" memory, it is not a memory, the first application of this technology is Edo memory, now the SD also has applications, and ECC memory is mainly from the SD memory began to be widely used, and the new DDR, RDRAM also has a corresponding application, the current mainstream ECC memory is actually a kind of SD memory.
Chipkill Technology is a new ECC memory protection standard developed by IBM to solve the current shortcomings of ECC technology in server memory. We know that ECC memory can only detect and correct single bit errors at the same time, but it is generally powerless to detect errors in more than two bits of data at the same time. The current ECC technology is widely used in server memory, one because before this other new memory technology is immature, and in the current server system speed is still very high, in this frequency in general the simultaneous occurrence of multiple-bit error rarely occurs, Because of this, ECC technology is fully recognized and applied, which makes ECC memory technology become the memory standard on almost all servers.
However, with the CPU performance of the intel-based processor-based server increasing at a geometric level, while the performance of the hard drive has increased only a few multiples in the same period, the server requires a large amount of memory to temporarily save the data needed to be read on the CPU in order to achieve sufficient performance. Such large data traffic results in a single memory chip that typically provides 4 (32-bit) or 8 (64-bit) bits of data per visit, read so much data at once, the likelihood of multiple-bit data errors will be greatly improved, and ECC can not correct the errors above the double bit, This is likely to result in the loss of all bit data, and the system crashes quickly. IBM's Chipkill technology is the use of memory substructure method to solve this problem. The memory subsystem is designed so that a single chip, regardless of the width of the data, can affect up to one bit only for a given ECC identifier. For example, if the parity of each bit in a 4-bit-wide dram,4 bit is composed of different ECC identifiers, the ECC identifier is saved with a single data bit, that is, stored in a different memory space address. Therefore, even if the entire memory chip out of the fault, each ECC identification code will appear up to a bit of bad data, and this situation can be fixed through ECC logic, so as to ensure the fault tolerance of the memory subsystem, the server in the event of failure, there is a strong ability to recover. With this memory technology, memory can simultaneously check and repair 4 error data bits, the server's reliability and stability are more fully protected.
Register is register or directory register, the role in memory we can take it to understand the catalog of the book, with it, when the memory received read and write instructions, the directory will be retrieved before the read and write operation, which will greatly improve the server memory efficiency. Memory with register must have buffer (buffer), and the current register memory can also have ECC function, its main application in high-end server and graphics workstation, such as IBM Netfinity 5000.
Fb-dimm (fully BUFFERED-DIMM, fully buffered memory module) is a new type of memory module and interconnect architecture developed by Intel on DDR2 and DDR3, which can match the current DDR2 memory chip and the future DDR3 memory chip. Fb-dimm can greatly increase the system memory bandwidth and greatly enhance the maximum memory capacity.
FB-DIMM technology is intel to solve the memory performance constraints on the overall performance of the system developed, on the basis of existing technology to achieve a leapfrog performance improvement, while the cost is relatively low. In the whole computer system, memory is the key factor to determine the performance of the machine, the light has a fast CPU, no good memory system with it, CPU performance is no good can not play. This situation is determined by the principle of the computer, the CPU in the operation of the data required are obtained from the memory, if the memory system can not provide data to the CPU in a timely manner, the CPU has long been in a waiting state, the hardware resources are idle, the performance of nature can not play. For ordinary PCs, because of the single processor system, the current memory bandwidth has been able to meet its performance requirements, and for multiple servers, because it is a multiprocessor system, its memory bandwidth and memory capacity is extremely hungry, the traditional memory technology has been unable to meet its needs. This is because the current common DIMM is a "short-term connection" (Stub-bus) topology, in this structure, each chip and the memory controller of the data bus has a short line connected, which will cause resistance to the instability, thus affecting the stability and integrity of the signal, The higher the frequency or the more chip data, the greater the impact. Although new memory technologies such as XDR memory introduced by the Rambus company have high performance, there is a high cost problem that makes it less popular. The advent of FB-DIMM technology to solve this problem, both to provide greater memory capacity and ideal memory bandwidth, but also to maintain a relatively low cost. Fb-dimm compared to XDR, the cost is much cheaper than XDR, although the performance is less than the new schema of XDR.
Compared with the existing common DDR2 memory, FB-DIMM technology has a great advantage: in the same memory frequency of the current can provide four times times the bandwidth of ordinary memory, and can support the maximum memory capacity is also up to 24 times times the normal memory, the system can support the largest 192GB memory. The biggest feature of FB-DIMM is the use of existing DDR2 memory chip (DDR3 memory chip will be used later), but it uses a buffer chip on the memory PCB Amb (Advanced Memory buffer, advanced memory buffer) to convert parallel data into serial data streams, The data is transmitted to the processor via a point-to-point high speed serial bus similar to PCI Express.
Compared with the common DIMM module technology, the data and command transmission between FB-DIMM and memory controller is no longer a traditional parallel line, but the design of parallel serial interface is similar to that of pci-express, which is transmitted in serial way. In this new architecture, the buffers on each DIMM are concatenated, the point-to-point connection, and the data is passed down a buffer after the first buffer so that the connection impedance between the first buffer and the memory controller is always stable, thus contributing to increased capacity and frequency.
Typical server memory type
At present the server commonly used memory has SDRAM and DDR,DDR2 three kinds of memory.
(2) DDR SDRAM
(3) DDR2 SDRAM
Because the server has a variety of technical relative compatible machines to be much stricter, it emphasizes not only the speed of memory, but its inherent error-correcting technology and stability. So it's only after the compatible or normal desktop memory is currently on the FSB. At present, the FSB of desktop is generally more than 150MHz times, but 133 FSB is still the mainstream. and the server due to the entire accessories FSB and high stability requirements, the mainstream FSB or 100MHz, but the 133MHz FSB has been gradually implemented in various grades of servers in the choice of the server of course, it is best to choose 133MHz FSB! Memory, other accessories are the same, to try to synchronize, Otherwise, the performance of the entire server will be affected. At present, the main server memory brands are mainly Kingmax, Kinghorse, Hyundai, Samsung, Kingstone, IBM, VIKING, NEC, etc., but mainly in the previous several in the market is more common, and the quality can be better protected.