Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
2000-08-31
2003-04-29
Kim, Matthew (Department: 2186)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C711S114000, C711S137000
Reexamination Certificate
active
06557075
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates generally to data storage systems associated with computer systems and, in particular, to a method and system for improving the transfer of data to and from a storage system that has been configured to act as a RAID-
1
system by storing two sets of duplicated data and which uses application-specific input/output characteristics.
2. Description of the Related Art
During the last decade, the amount of data to be processed, stored and accessed by certain industries, such as banks, financial and insurance institutions, automobile manufacturers and airlines as everyday, normal business operations, in particular the number of data accessing requests, have exploded. These vastly increased data processing needs have spurred the creation of new systems for storing and accessing data, for example, Redundant Arrays of Inexpensive Disks (RAID) and Storage Area Networks (SAN), as well as the development of faster computer-to-storage device interface technology and protocol standards, such as Fibre Channel standards and Small Computer System Interface (SCSI) by which to improve the rate of data transfer, i.e., data throughput.
Also accompanying this explosion in data processing needs has been a concomitant increase in the number of patents related to improving the performance of storing and accessing data using the new storage system technologies. For example, a simple search for United States patents shows that there have been at least 50 patents this year alone that relate to arrayed memory in a computer. For example, U.S. Pat. No. 6,076,143 to Blumenau which discloses a method that manages the writing of data blocks to be accessed on a disc drive storage system whereby a data block is written to one disc at a different physical sector address than the address to which the same data block is written to on a redundant disc. Also, U.S. Pat. No. 5,974,502 to DeKoning et al. provides a method for improving throughput between the host computer and an array of disk drives by splitting up large I/O requests from the computer into smaller, more manageable pieces and processing these requests as if they were individual I/O requests. Further, U.S. Pat. No. 5,787,463 to Gajjar introduces a dual-ported staging memory between the RAID engine and the host bus so that both the host and the RAID engine can concurrently access such stage memory, thereby allowing asynchronous memory operation.
These patents seek to improve data transfer or throughput in a disc drive storage system largely by focusing on upgrading the performance, sequencing, or timing of the storage hardware. However, there is also another approach to improving data throughput, by taking into consideration the kinds of data accessed, the kinds of application software used to input the data as well as the kinds of data requests processed. These are especially important considerations in the industries named above, inasmuch as in these and many other industrial contexts, data is processed in two very different yet predictable ways: first, decision support system processing and second, transaction processing.
The conflicting behaviors of Transaction Processing (TP) Applications and Decision Support System Applications (DSS) have caused the replication of data to flourish and created huge costs and latencies in order to speed up the storing and accessing of data. For example, users of the DSS Applications in a banking context are usually requesting either reports or performing complex arithmetic operations that involve reading out from storage disks a long and huge stream of data, which typically requires the disk head to move sequentially around the platter from sector to sector that are more or less adjacent to each other. On the other hand, users of the TS Applications are usually writing in or requesting short blocks of data that are not written in or read out sequentially but are stored or accessed across platter tracks in a manner that typically requires the disk head to “skip” all over the platter.
Conflicts inevitably arise when one disk head is called upon both to read out long streams of sequentially-stored data and to read and write short bursts of non-sequentially-stored data. In short, in responding to requests for processing from both a TP software and a DSS software, the disk heads will be working at cross purposes, which implicates that the physical data path from disk to storage cannot be shared for processing requests from these two kinds of software.
Because DSS software typically read data sequentially, TP software generally does not allow real time access to the DSS system, in order not to negatively impact business performance. Due to the disparity between how DSS applications and TP applications are stored and accessed, users in the data warehouse, datamart and data mining lines of a business, those who typically use the DSS software have had to create copies of the “real-time” data in order to crunch or report on them. This need to duplicate data within an enterprise in order to have them available for different processing needs has in turn created a massive sub-industry of copy management as well as fostered data bandwidth and CPU capacity obstacles.
To solve the problems inherent in required data duplication due to different processing needs, an enterprise can rely on a storage system that either has two sets of disk heads or that has been configured to operate as if there are two sets of disk heads. A storage system that stores duplicate data is a RAID-1 engine, which is an array of paired storage devices. A storage system that does not actually comprise a RAID-1 engine may nevertheless be configured to store duplicate sets of mirrored data and so operate as if it were a RAID-1 configuration.
SUMMARY OF THE INVENTION
The present invention provides a method of accessing and storing data in a memory system communicating with one or more computers generating read and write requests. The memory system comprises a controller, a memory cache for temporarily storing data. The memory cache comprises an A-cache and a B-cache, and a pairwise-redundant direct access storage device comprising an A-DASD and a B-DASD. The B-cache is a read-ahead cache of data read from B-DASD.
One embodiment of a method of the present invention comprises the steps of providing an A-interface and a B-interface to the memory system, configuring transaction processing applications on a computer communicating with the, memory system to direct read and write requests to the A-interface, configuring decision support system applications on a computer communicating with the memory system to direct read and write requests to the B-interface, fulfilling write requests received at the A-interface by writing data to-the A-cache, fulfilling write requests received at the B-interface by writing data to the A-cache.
The method also comprises the steps of fulfilling read requests received at the A-interface by reading data from the A-cache whenever it contains the requested data or else reading data from the A-DASD, fulfilling read requests received at the B-interface by reading data from the B-cache whenever it contains the requested data or else reading data from the B-DASD, writing data, not yet been committed to A-DASD, from the A-cache to the A-DASD whenever the A-DASD is not fulfilling a read request, and writing data that has not yet been committed to B-DASD, from the A-cache to the B-DASD whenever the B-DASD is not fulfilling a read request. The average time for fulfilling read requests is improved over that of a corresponding memory system using a RAID-1 controller.
Another embodiment of a method of the present invention further comprises the steps of interrupting, whenever A-cache becomes full, the flow of data at A-interface and B-interface, including any read operation from B-DASD, writing to B-DASD records in A-cache that are changed but not yet committed to B-DASD, in the preferential sequence of those records which are logically in read sequence before the current reading
Choi Woo H.
Elman Gerry J.
Elman Technology Law P.C.
Kim Matthew
LandOfFree
Maximizing throughput in a pairwise-redundant storage system does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Maximizing throughput in a pairwise-redundant storage system, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Maximizing throughput in a pairwise-redundant storage system will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3064970