Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
1997-11-19
2001-07-31
Gossage, Glenn (Department: 2187)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C711S155000, C711S171000, C714S006130, C714S801000, C714S779000
Reexamination Certificate
active
06269424
ABSTRACT:
FIELD OF THE INVENTION
The present invention relates to a technique for controlling a disk array, and more particularly to a technique for enhancing the efficiency and reliability of a process for generating and writing redundant data to a disk array storage device.
BACKGROUND OF THE INVENTION
In 1987, David A. Patterson, et al. reported a technique for saving redundant data in storage devices. See David A. Patterson, “A Case for Redundant Arrays of Inexpensive Disks (RAID),” University of California at Berkeley, Computer Science Division Report UCB:CSD-87-391 (December 1987); see also David A. Patterson, “A Case for Redundant Arrays of Inexpensive Disks (RAID),” ACM SIGMOD conference proceeding, Chicago, Ill., Jun. 1-3, 1988, pp. 109-116. This technique is based on a generalized method of concatenating multiple physical storage devices into one logical storage unit. When a high level device (e.g., host computer or dedicated controller) writes user data to this type of logical storage unit, the data may be divided into a number of parts corresponding to the number of physical storage devices. At the same time, redundant data may also be generated and stored among the several physical storage devices. In the event that one of the physical storage devices fails, the stored redundant data can facilitate the recovery of stored user data. The Patterson document described the following two methods for generating redundant data.
The first method for generating redundant data, referred to herein as the “method of read and modify,” generates redundant data from the combination of three sources: (1) write data from a high level device, (2) previous data stored in a storage device where the write data is to be stored, and (3) previous redundant data stored in a storage device where newly generated redundant data is to be stored.
Assuming that the write data is divided into a partitions, this method of generating redundant data requires a read operations to access previously stored data, a write operations to store the updated write data, one read operation to retrieve the previous redundant data, and one write operation to store the updated redundant data. To generate and store redundant data in this example, (&agr;+1) read operations and (&agr;+1) write operations are required, totaling (2&agr;+2) input and output operations. If the use of redundant data is unnecessary, only &agr; (write) operations would be required to store the data. This means that, using this method of generating redundant data, an additional (&agr;+2) input and output operations are required.
The second method for generating redundant data, referred to herein as the “method of all stripes,” generates redundant data from partitioned write data received from the high level device and from previous data read from the storage devices. In this method, however, the devices which store redundant data do not read previously-stored redundant data.
With this method, assume again that the write data is divided into &agr; partitions. Assume also that the number of storage devices, except those for saving redundant data, is &bgr;, and further that &agr;<=&bgr;. In this method, then, the total number of input operations to and output operations from the storage devices is (&bgr;+1), wherein the number of input operations is (&bgr;−&agr;) and the number of output operations, including those containing redundant data, is &agr;+1. If redundant data is not necessary, only &agr; (write) operations would be required to store the data. This means that the additional number of input and output operations required is (&bgr;+1−&agr;), when redundant data is generated by this second “method of all stripes.”
Apart from the foregoing methods, a method for generating redundant data in storage devices has been disclosed in U.S. Pat. No. 5,613,088, wherein the storage device uses two heads for simultaneous reading and writing. Specifically, the read head and the write head are fixed on a common actuator. During the data update process, the read head reads existing parity data and then the write head follows, updating the same area with new parity data generated, in part, from the old parity data.
The foregoing two methods for generating redundant data, that is, the “method of read and modify” and the “method of all stripes,” increase the number of input and output (I/O) operations associated with storing data on a disk. This means that the disk control device with redundancy is inferior in performance to the disk control device without redundancy. Hence, according to the present invention, a conventional disk control device with redundancy is made to selectively employ the redundant data generation method that results in a smaller number of I/O operations to and from the storage device. This selection makes it possible to reduce the burden on the storage device and thereby improve the processing speed. Specifically, in the case of (&agr;>=(&bgr;−1)/2), the “method of all stripes” will use a smaller number of storage device I/O operations than the “method of read and modify,” while in the case of (&agr;<(&bgr;−1)/2), the “method of read and modify” will use a smaller number. Therefore, if the length of the write data received from the high level device is in the range of (&agr;<(&bgr;−1)/2), for example, in the case of a short transaction process, a disk control device that is configured to use the present invention will select the “method of read and modify” to generate redundant data.
The number of I/O operations using the “method of read and modify” is minimized at four (4) when &agr;=1. This means that when &agr;=1, performance cannot be improved further unless the method of processing is reconsidered. The problem with the “method of read and modify” is essentially based on the fact that two I/O operations must be issued to the storage device for each partition of data. With each I/O operation there is significant overhead associated with such factors as movement of the head and rotational latency. This mechanical overhead is a great bottleneck on disk control devices.
The method disclosed in U.S. Pat. No. 5,613,088 makes it possible to generate redundant data in a storage device configured with a read head and a write head mounted on a single actuator. Expanding this method to a general storage device provided with a single read-write head, the resulting method, referred to herein as the method of “generation in a drive,” employs the following steps. First, the data to be written to the disk drive device (the “write data”) and the existing data that will eventually be updated with the “write data” (the “data before update”) are transferred to the actual physical storage device that is responsible for generating and storing the redundant data. Within this redundant data physical storage device, the existing redundant data that will be updated (the “redundant data before update”) is read, and the new redundant data is generated from the combination of the “write data,” the “data before update,” and the “redundant data before update.” In this method of “generation in a drive,” the head is positioned to read the “redundant data before update,” and the updated redundant data is calculated, and when the disk reaches the next writing position, the write operation is started and the updated redundant data is stored. This operation makes it possible to avoid the spinning on standby that normally occurs during the interval between reading and writing, and merely requires one movement of the head and one standby spin. As a result, if the length of the data from the high level device is short, the processing speed of the control device can be improved further.
However, the method of “generation in a drive” cannot always calculate and store redundant data in the most efficient manner. If the length of the generated redundant data is longer than can be stored within one spin of the disk, the method of “generation in a drive” will require the disk to
Katsuragi Eiju
Kurano Akira
Ogata Mikito
Takahashi Naoya
Tamiya Toshihiko
Gossage Glenn
Hitachi , Ltd.
Kenyon & Kenyon
LandOfFree
Disk array device with selectable method for generating... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Disk array device with selectable method for generating..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Disk array device with selectable method for generating... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2558835