Electrical computers and digital data processing systems: input/ – Input/output data processing – Peripheral adapting
Reexamination Certificate
2002-12-04
2004-09-28
Fleming, Fritz (Department: 2182)
Electrical computers and digital data processing systems: input/
Input/output data processing
Peripheral adapting
C710S011000, C710S016000, C710S022000, C710S023000, C710S024000, C710S062000, C710S072000, C710S105000, C710S308000, C710S315000, C709S230000, C712S300000, C341S050000
Reexamination Certificate
active
06799232
ABSTRACT:
FIELD OF THE INVENTION
The invention relates to processing data at data network nodes, and in particular to methods and apparatus for performing Direct Memory Access (DMA) data transfers with improved efficiency.
BACKGROUND OF THE INVENTION
In the field of data processing it is known to use data buses to enable data transfers between data bus connected devices. In the field of data networking, network nodes such as, but not limited to: data switching nodes, bridges, routers, hubs, aggregation devices, deaggregation devices, multiplexers, demultiplexers, etc. are computing platforms which include data bus connected devices such as: memory storage, processors, physical interface peripherals (cards), etc.
Typical network node designs have an architecture in which a single bus master, a Central Processing Unit (CPU), oversees and coordinates the overall operation thereof. For short periods of time, as directed by the CPU, other data bus connected devices may control the data bus through bus mastering techniques.
Recently data network node designs have an architecture in which multiple data bus connected devices, such as processors and physical interface peripherals, cooperate to access a central memory storage to process data, each such data bus connected device operating as a bus master.
Although network nodes can be implemented on a computing platform such as a ubiquitous microcomputer, high data throughput capacity high performance network nodes are implemented as special purpose data processing equipment.
Data is received and transmitted via a multitude of interconnecting links associated with physical interfaces on the network equipment. In processing data at network nodes, data is received, switched and transmitted either severally or in combination in accordance with a multitude of transfer protocols. Each combination of a transfer protocol and a physical interface conveys data having a data format. The data format specifies a data transfer order and an alignment. Furthermore the data bus connected devices such as: interface cards, memory blocks, mass storage devices, switching processors, functional processors, supervisory processors, etc. process and convey data in accordance to a data transfer order and alignment associated therewith. Due to a competitive rapid development of data transport equipment and transport protocols, a data transfer order and alignment mismatch may exist therebetween.
A variety of data ordering conventions are being used and relate to an order in which data bits constitutive of data bytes are received, processed and transmitted. Data bit ordering includes, but is not limited to: a little-endian bit order and a big-endian bit order. With the advent of wide data buses the data ordering also includes byte ordering. A byte is a unit of data comprised of a sequence of bits used to represent numerical values and typographical symbols such as: characters and decimals. Data byte ordering includes, but is not limited to: a little-endian byte order and a big-endian byte order. Bytes are used singly or in combination to specify numerical values representative of the data. Combinations of two bytes are referred to as words, whereas combinations of four bytes are referred to as double words or dwords.
Little-endian ordering refers to the conveyance and/or processing of Least Significant Bits (LSB) of bit streams or bytes, first. With respect to bytes, little-endian ordering refers to the conveyance and/or processing of Least Significant Bytes (LSB) of a word/dword, first. Big-endian ordering refers to the conveyance and/or processing of Most Significant Bits/Bytes (MSB) first.
Bytes may be ordered in a variety of ways, the ordering being a design choice and/or a necessity in efficiently conveying and/or processing data for a particular application such as a network node.
The conveyance of data over data links (including between dissimilar vendor equipment) is also subject to data ordering imposed by transfer protocols used. Therefore, there may exist a clash between a network node implementing a particular byte order used in processing the data, such as little-endian byte ordering, whereas the transfer protocol used specifies a different byte order used in conveying data, such as big-endian byte ordering. The clash can only be eliminated by reordering the bytes at the network node prior to transmission or post reception of data.
The recent explosion of technological advances in the telecommunications field has benefited from a competitive environment in which a relatively large number of vendors have produced and deployed a large number of network node models using a large number of interface types each of which adhering to at least one of a relatively large number of transfer protocols.
Another important factor to be considered in processing data concerns data alignment and in particular byte alignment. The data buses used in conveying data internally between data bus connected devices use parallel data transfers, with 2, 4, 8, 16, etc. bytes being transferred on each cycle of an associated data bus clock according to the data bus width. For example, 4 bytes of data are transferred on each clock cycle for a 32 bit data bus.
Data buses also make use of addressing lines, integral to each data bus, to select memory locations where to write to, and to read from. Although indirect addressing is used to access every byte of data in memory, it is common practice to address a number of bytes at the same time since the bytes are transferred over the data bus in groups. For the above example, typically when using a 32 bit data bus, data bus addressing locations are defined every 4 bytes. This is referred to as dword alignment of data. Indirect addressing of data, to access the individual bytes in each dword for example, is performed subsequent to the transfer of the dword from the memory for processing. A data processing overhead is incurred if only a fewer number of bytes from the transferred dword are used in processing.
Such overheads can be significant in data switching environments where the data to be processed includes Protocol Data Units (PDUs). Each PDU has a PDU header and a payload. The header specifies routing information in accordance with a particular transfer protocol and is updated or rewritten at each network node in a transport path. The header may be discarded altogether as the data conveyed undergoes protocol conversion. The header may be considered part of the payload as the PDU is encapsulated and conveyed using another transfer protocol. All these processes necessitate data re-alignment in the processing thereof.
Prior art data bus connected devices such as interface cards, mass storage devices, etc. rely on switching processors, functional processors, supervisory processors, etc. to provide correct data ordering and alignment. Depending on the byte order of the data processing equipment and the applications used, a processor takes the time-consuming job to re-organize and re-order the data using storage resources and processing cycles to copy the data, re-order the bytes, and re-align the bytes before and/or after processing.
Current network nodes support a variety of low capacity and high capacity interfaces as well as a variety of transfer protocols to provide aggregation/deaggregation functionality in addition to PDU switching therefore a significant amount of processing time is devoted to reordering and re-aligning data.
An attempt at relieving a CPU from performing byte re-ordering is presented in prior art U.S. Pat. No. 5,862,407 entitled “SYSTEM FOR PERFORMING DMA BYTE SWAPPING WITHIN EACH DATA ELEMENT IN ACCORDANCE TO SWAPPING INDICATION BITS WITHIN A DMA COMMAND” which issued on Jan. 19
th
, 1999 to Sriti describes a video card for a computer which conveys a data stream using multiplexers. Based on swapping indication bits, the video card uses a different multiplexer tap for read and/or write operations. Although inventive, the swapping indication bits are specified by a high level software application such as a video driver. The vide
Fleming Fritz
Laubscher, Jr. Lawrence E.
Nguyen Tanh
Zarlink Semiconductor V.N. Inc.
LandOfFree
Automatic byte swap and alignment for descriptor-based... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Automatic byte swap and alignment for descriptor-based..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Automatic byte swap and alignment for descriptor-based... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3242161