Patent Application
20060047891
This invention relates to computer systems, and, more particularly, to a system and method for transmitting data packets in a computer system having a memory hub architecture.
Computer systems use memory devices, such as dynamic random access memory (“DRAM”) devices, to store data that are accessed by a processor. These memory devices are normally used as system memory in a computer system. In a typical computer system, the processor communicates with the system memory through a processor bus and a memory controller. The processor issues a memory request, which includes a memory command, such as a read command, and an address designating the location from which data or instructions are to be read. The memory controller uses the command and address to generate appropriate command signals as well as row and column addresses, which are applied to the system memory. In response to the commands and addresses, data are transferred between the system memory and the processor.
Although the operating speed of memory devices has continuously increased, this increase in operating speed has not kept pace with increases in the operating speed of processors. Even slower has been the increase in operating speed of memory controllers coupling processors to memory devices. The relatively slow speed of memory controllers and memory devices limits the data bandwidth between the processor and the memory devices.
In addition to the limited bandwidth between processors and memory devices, the performance of computer systems is also limited by latency problems that increase the time required to read data from system memory devices. More specifically, when a memory device read command is coupled to a system memory device, such as a synchronous DRAM (“SDRAM”) device, the read data are output from the SDRAM device only after a delay of several clock periods. Therefore, although SDRAM devices can synchronously output burst data at a high data rate, the delay in initially providing the data can significantly slow the operating speed of a computer system using such SDRAM devices.
One approach to alleviating the memory latency problem is to use multiple memory devices coupled to the processor through a memory hub. In a memory hub architecture, a system controller or memory controller is coupled over a high speed link to several memory modules. Typically, the memory modules are coupled in a point-to-point or daisy chain architecture such that the memory modules are connected one to another in series. Thus, the memory controller is coupled to a first memory module over a first high speed link, with the first memory module connected to a second memory module through a second high speed link, and the second memory module coupled to a third memory module through a third high speed link, and so on in a daisy chain fashion.
Each memory module includes a memory hub that is coupled to the corresponding high speed links and a number of memory devices on the module, with the memory hubs efficiently routing memory requests and memory responses between the controller and the memory devices over the high speed links. Computer systems employing this architecture can have a higher bandwidth because a processor can access one memory device while another memory device is responding to a prior memory access. For example, the processor can output write data to one of the memory devices in the system while another memory device in the system is preparing to provide read data to the processor. Moreover, this architecture also provides for easy expansion of the system memory without concern for degradation in signal quality as more memory modules are added, such as occurs in conventional multi drop bus architectures.
The system controller 108 includes a downstream transmission port 140 coupled to the downstream link 128 to provide memory requests to the memory module 104a, and also includes an upstream reception port 144 coupled to the upstream link 136 to receive memory responses from the memory module 104a. The ports 124, 132, 140, 144 and other ports to be discussed below are designated “physical” interfaces or ports since these ports are in what is commonly termed the “physical layer” of a communications system. In this case, the physical layer corresponds to components providing the actual physical connection and communications between the system controller 108 and system memory 102 as will be understood by those skilled in the art.
The nature of the reception ports 124, 144 and transmission ports 132, 140 will depend upon the characteristics of the links 128, 136. For example, in the event the links 128, 136 are implemented using optical communications paths, the reception ports 124, 144 will convert optical signals received through the optical communications path into electrical signals and the transmission ports 140, 132 will convert electrical signals into optical signals that are then transmitted over the corresponding optical communications path.
In operation, the reception port 124 captures the downstream memory requests and provides the captured memory request to local hub circuitry 148, which includes control logic for processing the request and accessing the memory devices 156 over a bus system 152 to provide the corresponding data when the request packet is directed to the memory module 104a. The reception port 124 also provides the captured downstream memory request to a downstream transmission port 160 on a bypass bus 180. The downstream transmission port 160, in turn, provides the memory request over the corresponding downstream link 128 to a downstream reception port 124 in the adjacent downstream memory module 104b. The port 124 in module 104b operates in the same way as the corresponding port in the module 104a, namely to capture the memory request and provide the request to the local hub circuitry 148 for processing and to provide the request to a downstream transmission port 160. The port 160 in the module 104b then operates in the same way as the corresponding port in module 104a to provide the memory request over the corresponding downstream link 128 to the next downstream memory module (not shown in
The memory hub 112 in the module 104a further includes an upstream reception port 164 that receives memory responses over the corresponding upstream link 136 from an upstream transmission port 132 in the adjacent module 104b. An upstream transmission port 132, in turn, provides the response over the upstream link 136 to the upstream physical reception port 144 in the system controller 108. Each of the memory modules 112 includes a corresponding downstream reception port 124, upstream transmission port 132, downstream transmission port 160, and upstream reception port 164. Moreover, these ports 124, 132, 160, 164 in each module 104b operate in the same way as just described for the corresponding ports in the module 104a.
In addition to the memory responses from the downstream hubs, the local hub circuitry 148 also receives memory responses from a local memory 156. The local memory 156 may be a DRAM type memory device or other suitable memory devices as will be appreciated by those skilled in the art. The local hub circuitry 148 provides the memory responses from the local memory 156 to the upstream transmission port 132 for transmission over the upstream link 136 to the upstream reception port 144 of the controller 108. Thus, the local hub circuitry 148 must monitor and control transmission of memory responses to the system controller 108 from the downstream memory module 104b and from the local memory 156. Since the hub circuitry 148 must monitor and control transmission of memory responses to the system controller 108 from the downstream memory module 104b and the local memory 156, the hub circuitry 148 must determine the priority of transmission of the memory responses. The hub circuitry 148 also must efficiently switch the transmission of memory responses from one source to another source. The hub circuitry 148 also must switch transmission of memory responses from one source to another source at an appropriate time.
The system controller 108 can control the timing of the memory responses inside the memory hubs 112. However, if there are a large number of memory hubs 112 coupled to the system controller 108, it becomes complicated for the system controller 108 to efficiently determine the priority of transmission of memory responses and to do the scheduling in all the memory hubs 112. Also when the system controller 108 controls the scheduling of memory responses inside the memory hubs 112, the bandwidth available for data transmission is reduced.
Accordingly, there is a need for a system and method for efficiently determining the priority of transmission of the memory responses inside the memory hub 112. There is a need for a system and method for efficiently switching transmission of the memory responses from one source to another source inside the memory hub 112. There is a need for a system and method for efficiently switching transmission of the memory responses from one source to another source at an appropriate point.
The present invention is directed to a system and method for transmitting data packets from a memory hub to a memory controller. In one embodiment, the system includes an upstream reception port coupled to an upstream link. The upstream reception port receives the data packets from downstream memory hubs. The system further includes a bypass bus coupled to the upstream reception port. The bypass bus transports the data packets from the upstream reception port. The system further includes a temporary storage coupled to the upstream reception port and configured to receive the data packets from the upstream reception port. The system further includes a bypass multiplexer for selectively coupling an upstream transmission port to either one of a core logic circuit, the temporary storage, or the bypass bus. The system further includes a breakpoint logic circuit coupled to the bypass multiplexer and configured to switch the bypass multiplexer to selectively connect the upstream transmission port to either one of the core logic circuit, the bypass bus, or the temporary storage. The system further includes a local memory coupled to the core logic circuit and operable to receive and send the data packets to the core logic circuit. The bypass bus transports data packets from the downstream hubs to the upstream link when the bypass multiplexer is switched to the bypass bus. The upstream temporary storage stores the data packets from the downstream hubs when the bypass multiplexer is switched to the core logic circuit. The core logic circuit transmits the data packets from the local memory when the bypass bus is switched to the core logic circuit. The data packets from the temporary storage are transported to the upstream link when the bypass multiplexer is switched to the temporary storage.
The downstream reception port 124 receives read and write requests from the system controller 108 (not shown in
The memory hub 200 further includes the upstream transmission port 132 that is linked to the system controller 108 by the upstream link 136. As will be discussed further, read and write responses from the core logic circuit 204 and the downstream hubs (not shown in
The memory hub 200 further includes a bypass multiplexer 212 coupled to the core logic 204 and a temporary storage 216. The bypass multiplexer 212 is also connected to the upstream reception port 164 via a bypass bus 220. The bypass multiplexer 212 selectively couples either the core logic 204, the bypass bus 220 or the temporary storage 216 to the upstream transmission port 132.
In operation, read and write responses from the downstream hubs are received by the upstream reception port 164 over the upstream link 136 and are passed on to the upstream transmission port 132 over the bypass bus 220 and through bypass multiplexer 212. Read responses are received by the core logic 204 from the local memory 156 and are passed on to the upstream transmission port 132 through the bypass multiplexer 212. Write responses are generated in the core logic 204 and are also passed on to the upstream transmission port 132 through the bypass multiplexer 212. As will be discussed further, when the bypass multiplexer 212 couples the core logic 204 to the upstream transmission port 132, the temporary storage 216 is used to temporarily store read and write responses from the downstream hubs. In the following description, write and read responses from the core logic 204, the downstream hubs and the temporary storage 216 will be referred to simply as “data.”
As described above, the upstream transmission port 132 transmits data, over the upstream link 136, originating from one of several sources: (1) the local memory 156; (2) downstream hubs; and the temporary storage 216. The multiplexer 212 selectively couples the upstream link 136, through the transmission port 132, to either the core logic 204, the bypass bus 220 or the temporary storage 216. The multiplexer 212 is switched so that data originating from either the core logic 204, the bypass bus 220 or the temporary storage 216 are transmitted over the upstream link 136 to the system controller 108. A breakpoint logic 208 coupled to the bypass multiplexer 212 provides the switching algorithm to the bypass multiplexer 212. The switching algorithm locates switch points (also referred to as breakpoints) when a switch may occur. If the switching algorithm locates a breakpoint and it is determined that a switch should be made to another data source that has data available, the bypass multiplexer is switched so that the new data source is coupled to the upstream link 136 through the upstream transmission port 132.
In general, data is transferred among the memory hub 200, the system controller 108 and downstream hubs in a fixed data packet format. A data packet includes a beginning and an end. The breakpoint logic 208 determines the beginning or end of a data packet, and a switch is made at the beginning or end of a data packet.
In one embodiment, the core logic 204 operates at 400 MHz. The reception ports 124, 164, and the transmission ports 132, 160 operate at 1.6 GHz. The upstream link 136 and the downstream link 128 operate at 6.4 GHz.
The operating speed of these devices are selected due to design requirements. The upstream and downstream links are operated at very high speed (6.4 GHz) in order to provide a large bandwidth. However, the transmission ports 136, 160, the reception ports 124, 164, and the core logic 204 cannot be operated at such high speed using current technology. Thus, as data is transferred from the downstream link to the reception port, the transfer speed is reduced. As data is moved to the core logic, the speed is reduced further.
In
In one embodiment, a data packet moves from the upstream reception port 164 to the upstream transmission port 132 in one 1.6 GHz clock period. However, the breakpoint logic 208, which switches the bypass multiplexer 212, requires three clock periods to complete the switch because of the time required to process a decode and drive logic to switch the bypass multiplexer 212. Thus, the beginning of the data packet is located as it enters the memory hub 200, and then switching is initiated three clock cycles prior to the breakpoint so that the bypass multiplexer 212 is switched in time as the data packet arrives.
As described before, the determination that the bypass multiplexer 212 will be switched is made three clock cycles before the arrival of a data packet. By looking ahead three clock cycles before the data arrives, the switching process of the bypass multiplexer 212 can begin so that the switch coincides with the data arrival. The write response data packet in
When the bypass multiplexer 212 is switched to the core logic 204, incoming data packets from the downstream hubs are first stored in the bypass FIFO 516. Since the bypass FIFO 516 operates at high speed (4X clock speed), the bypass FIFO 516 can transfer data packets from its input to its output very quickly. Thus, if the core logic 204 completes sending data packet and the bypass multiplexer switches to the temporary storages, the data from the bypass FIFO 516 is available immediately.
However, if the bypass multiplexer 212 remains switched to the core logic 204, incoming data packets from the downstream hubs fill up the bypass FIFO 516. When the bypass FIFO 516 is filled up, the upstream buffer 512 is used to store data packets. As will be understood by those skilled in the art, the bypass FIFO 516 is fast, but is expensive to implement. Thus a small bypass FIFO 516 is typically used. The upstream buffer 512 is slower, but is less expensive to implement. Thus, a large upstream buffer 516 is used.
The memory hub 500 includes clock domain change circuits 520, 524, 508. As noted before, since the downstream ports 124, 160 operate at different clock frequency than the core logic 204, the downstream ports 124, 160 are not synchronous with the core logic 204. Thus, data packets cannot be directly transferred between the core logic and the downstream ports 124, 160. The clock domain change circuit 520 allows transfer of data packets from the downstream port 124 to the core logic 204, and the clock domain change circuit 524 allow the transfer of data packets from the core logic 204 to the downstream port 160. The core logic 204 is synchronous with the bypass multiplexer 212, and the clock domain change circuit 508 allows the transfer of data packets from the core logic 204 to the bypass multiplexer 212.
In one embodiment, after power up, the breakpoint control logic 208 initially switches the bypass multiplexer 212 to the bypass bus 220, thus connecting the bypass bus 220 to the upstream link 136. The bypass bus 220 remains connected to the upstream link 136 until the core logic 204 has data to be sent and a breakpoint is available on the bypass bus 220. If the core logic 204 has data available and a breakpoint is available, the bypass multiplexer 212 is switched to the core logic 212.
When the bypass multiplexer 212 is switched to the bypass bus 220, data on the bypass bus 220 is sent to upstream link 136. When the bypass multiplexer 212 is switched to the core logic 204, data from the core logic 204 is sent to the upstream link 136. While the bypass multiplexer 212 remains switched to the core logic 204, incoming data on the bypass bus 220 is sent first to the bypass FIFO 516. When the bypass FIFO 316 is filled up, data is next to the upstream buffer 512.
In one embodiment, the bypass multiplexer 212 remains switched to the core logic 204 until the core logic 204 is empty or if a higher priority requires a switch. A higher priority is determined if the temporary storages, i.e., the bypass FIFO 516 or the upstream buffer 512, have available data. When the bypass multiplexer 212 is switched away from the core logic 204, the multiplexer 212 is first switched to the bypass FIFO 516. The data in the bypass FIFO 516 is sent upstream over the upstream link 136 until the bypass FIFO is exhausted. In general, after the bypass FIFO 516 is exhausted, the bypass multiplexer 212 is next switched to the upstream buffer 512, which is then emptied.
If the core logic 204 has data available, a switch can be made from the bypass FIFO 516 to the core logic 204 even though the bypass FIFO has not been exhausted. If a switch is made from the bypass FIFO 516 to the core logic 204, the next switch is made back to the bypass FIFO 516 in order to send the upstream data in the order it was received. When the bypass FIFO 516 empties, data is next taken from the upstream buffer 512. A switch to the core logic 204 can be made from the upstream buffer 512 even though the upstream buffer has not been exhausted. However, the next switch is made back to the upstream buffer 512 in order to send the upstream data in the order it was received.
After the bypass FIFO 516 and the upstream buffer 512 are cleared, the multiplexer 212 is normally switched to the bypass buss 220. If, however, the core logic 204 has available data, the multiplexer 212 is switched to the core logic 204. As discussed before, while the bypass multiplexer 212 is switched to the core logic 204, upstream data is first loaded into the bypass FIFO 516 and then into the upstream buffer 512. When the bypass multiplexer 212 is switched to the temporary storages, the bypass FIFO 516 is emptied first and then the upstream buffer 512 is emptied next. After the bypass FIFO 516 is emptied, it is not loaded again until the upstream buffer 512 has been emptied.
In the preceding description, certain details were set forth to provide a sufficient understanding of the present invention. One skilled in the art will appreciate, however, that the invention may be practiced without these particular details. Furthermore, one skilled in the art will appreciate that the example embodiments described above do not limit the scope of the present invention, and will also understand that various equivalent embodiments or combinations of the disclosed example embodiments are within the scope of the present invention. Illustrative examples set forth above are intended only to further illustrate certain details of the various embodiments, and should not be interpreted as limiting the scope of the present invention. Also, in the description above the operation of well known components has not been shown or described in detail to avoid unnecessarily obscuring the present invention. Finally, the invention is to be limited only by the appended claims, and is not limited to the described examples or embodiments of the invention.
