Electrical computers and digital data processing systems: input/ – Input/output data processing – Direct memory accessing
Reexamination Certificate
1996-04-24
2002-02-26
Luu, Le Hien (Department: 2152)
Electrical computers and digital data processing systems: input/
Input/output data processing
Direct memory accessing
C710S052000, C710S057000
Reexamination Certificate
active
06351780
ABSTRACT:
TECHNICAL FIELD
This invention relates generally to data communication systems and, more particularly, to a data network system and method wherein a controller that receives and stores data in its internal memory buffer can automatically invoke a Direct Memory Access (DMA) mode of data transfer to prevent overflowed data from being lost.
BACKGROUND
The task of accommodating increasing bus traffic in data communication networks continues to pose as a challenge. The primary bottle-neck in most data communication networks appears to be the system bus. The system bus is a bottle-neck primarily because many devices share the same bus and must contend for its use in transmitting or receiving data. For this reason, the system bus may be unavailable at crucial times for reasons such as operating priorities, system interrupts, and other bus traffic. Thus, in a computer system, it is necessary to buffer data received to accommodate bus traffic with other data devices.
Examples of devices buffering data received to accommodate bus traffic with other data sources and sinks include data conversion devices. Data conversion devices are basically front end communications processors that act as the interface between the host CPU and external terminals. Data conversions devices provide limited internal storage to continue receiving data from data sources while waiting for converted data to be moved to data sinks via the system bus.
Since the prior art solution is to provide storage capability while the system bus is busy, there exists an inherent problem in the prior art solution. This problem manifests when the period during which the system bus remains busy is long enough so that the amount of data accumulated in the temporary storage exceeds its storing capacity. When that occurs, data may be lost. To prevent data loss, data must be transferred to another location.
Generally, the data transport techniques most commonly used are Programmed I/O and Direct Memory Access (DMA). With Programmed I/O, the Host CPU is involved in every aspect of the data transport process. Even if the network adapter's control and data port parameters are defined and mapped in system memory, the performance of the network/system interface is, in most cases, limited by the CPU's input/output bandwidth and system bus utilization. Moreover, because Programmed I/O utilizes the CPU in every aspect of transferring data, when used as the transport mechanism in preventing data loss due to other bus traffic, this technique or those using the Host CPU as the main driving force in the data transport process is subject to an additional possibility of data loss when CPU interrupt latency rises due to CPU workload.
Direct Memory Access (DMA), on the other hand, is designed to transfer large “blocks or units of data” with little or no Host CPU intervention. Most DMA transport mechanisms are based on the Master-Slave transport model. With this type of transport mechanism, the Host CPU initiates the data move or transfer, and once started, the Host CPU is not involved again until the entire “block or unit of data” has been transferred. For this reason, DMA is much more efficient in term of CPU time required and is a preferred method to transfer large blocks of data than Programmed I/O or other CPU-intensive data transfer methods on the whole. On the other hand, DMA data transfer is slower than Programmed I/O and other CPU-intensive data transfer methods due to the extra moving steps involved in DMA transfer. Moreover, DMA techniques, such as DMA masters and DMA slaves, have their limitation in that additional processing is normally required to move data from where the DMA process puts it to the data sink (e.g., a specified application memory location). This is because the DMA controller is not able to determine the type and size of data, and the CPU must examine the data to determine the proper data sink.
For the above reasons, in devising a data transport technique to use in moving data from the temporary storage buffer to another location, a dual approach, which utilizes non-DMA data transfer (e.g., Programmed I/O) to move data to the proper sink in most instances but switches to DMA transfer if the internal storage limit is approached, is most desirable. Current commercially available data conversion devices that are DMA masters (e.g., Cirrus Logic CD-2400 Serial Interface Controller) or DMA slaves (e.g., Advanced Micro Devices Am7990 Lance Ethernet Controller) do not allow the CPU to move data from internal storage. On the other hand, commercially available data conversion devices with internal storage that rely on the CPU to move data (e.g., National Semiconductor 16550 AFN Serial Interface Controller) have the aforementioned additional possibility of data loss when CPU latency rises due to its workload.
Thus, it is desirable to provide the capability for automatic switching between a non-DMA data transfer mode and a DMA transfer mode to move data from the buffer storage to another location before the buffer storage is full.
Furthermore, it is desirable to provide an internal storage manager to monitor the fill and drain streams of data frames together with the current buffer capacity for storage to decide whether data loss is anticipated before switching to a DMA data transfer mode.
DISCLOSURE OF THE INVENTION
Accordingly, one advantage of the invention is in providing a network controller that has the ability to accumulate data received in its internal buffer storage and to selectively engage between a non-DMA data transfer mode and a DMA transfer mode to transfer data from the internal buffer storage to another location before it is full.
Another advantage of the invention is in providing an internal storage manager that monitors the fill and drain streams of data frames together with the current buffer storage capacity of a network controller to determine whether data loss is anticipated before switching to a DMA data transfer mode.
The above and other advantages of the invention are achieved, at least, in part, by providing a system for receiving from a plurality of terminals data frames to be processed by a host processor that comprises a controller responsive to the data frames for providing an interface between the terminals and the host processor and having a memory buffer for holding the data frames received from the terminals for processing by the host processor. Also, a system memory having a capacity larger than the memory buffer is available for storing the data frames received from the plurality of terminals. While a non-DMA data transfer mode is normally used to transfer data from the memory buffer to any desired location, a DMA means responsive to the host processor is used for writing the data frames received from the terminals into the system memory. The controller includes a buffer manager for monitoring the data frames supplied to the memory buffer and switching means responsive to the buffer manager for automatically engaging the DMA means. While the controller is preferably provided on a chip external with respect to the system memory, the DMA means may be provided either on the chip or externally.
In accordance with one feature of the invention, the DMA means is engaged to transfer data frames received from the terminals to the system memory when the number of data frames received from the terminals exceeds a predetermined number.
In accordance with another feature of the invention, the DMA means is engaged to transfer data frames received from the terminals to the system memory when the time period during which the frames have been received from the terminals exceeds a predetermined period.
In accordance with a further feature of the invention, the DMA means is engaged to transfer data frames received from the terminals to the system memory when the capacity of the memory buffer occupied by the data frames exceeds a predetermined value.
Preferably, the buffer manager makes a decision whether to engage the DMA means based on the rate of loading data frames into the memory buffer and the rate o
Cirrus Logic Inc.
Luu Le Hien
Skjerven Morrill & MacPherson LLP
LandOfFree
Network controller using held data frame monitor and... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Network controller using held data frame monitor and..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Network controller using held data frame monitor and... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2947711