Electrical computers and digital processing systems: memory – Storage accessing and control – Specific memory composition
Reexamination Certificate
2001-03-15
2003-04-08
Nguyen, Hiep T. (Department: 2187)
Electrical computers and digital processing systems: memory
Storage accessing and control
Specific memory composition
C710S039000, C710S040000, C711S158000, C711S162000, C714S006130
Reexamination Certificate
active
06546459
ABSTRACT:
FIELD OF THE INVENTION
The present invention provides redundant data storage systems and methods of operating redundant data storage systems.
BACKGROUND OF THE INVENTION
One complicating factor in conventional computer electronics is the increasing use of parallel operations. Further, transaction operations (e.g., read and write memory operations to a large DRAM array) are frequently reordered so as to present a minimal latency to transaction masters (e.g., a processor waiting for read data from DRAM).
For a given RAID subsystem application, there are certain commitments the subsystem typically makes to a coupled host system before the aggregate system can operate properly. Consider a case when the host system writes data to the RAID subsystem. The RAID subsystem stores the data in a fault tolerant condition. For the sake of performance, the RAID subsystem will store the host data into a low latency, high bandwidth storage device. This device is commonly implemented using DRAM technology. Only after the data has been safely stored can the RAID subsystem return a “good status” to the host. However, the longer the RAID subsystem waits to return “good status” to the host, the worse the performance will be. On the other hand, a premature “good status” may result in data loss if an interim failure is able to cause a situation where all of the data is lost.
The processors in a RAID subsystem are usually responsible for providing the “good status” message to the host system. In order to achieve this function, the processor determines that the data written by the host is completely safe from failure before it can make the “good status” commitment to the host system.
The processor will originate several read and write transactions in the system for each host operation. In some configurations, transactions may be reordered at some point along the data path to minimize the time a processor spends handling the host operation. In such instances, some transactions are set aside while higher priority transactions are completed. An example of this would be when a memory write transaction to some control structure is stored in a write buffer. The write buffer operates as a transaction processing device. A subsequent read request generally has a higher priority since components (such as processors) are susceptible to data starvation. Depending on the design characteristics, this could delay the processing of a transaction indefinitely. Thus, the control structure in main memory will be stale until the write buffer is flushed.
Consider again the case where the host writes data to the RAID subsystem. One of the issues for processors in a RAID subsystem is their need to know when the host data is stored redundantly and is safe from failure. Combining the technology that allows for reordered parallel operations and the need to have a point of commitment directly conflict. For example, if the data that has been written from the host is sitting inside a write buffer for a relatively long period of time and the processor sends “good status” before the data in the write buffer is flushed to the redundant storage component (DRAM), then the commitment could be presented to the host before the data is transferred to DRAM. The host will continue its operation knowing the RAID subsystem will not lose the data. However, without the present invention, data loss may occur if a failure disables the flushing of the data in the write buffer. The data loss scenario is as simple as finding a parity corruption as the data is drained from the write buffer.
It follows that a write buffer may hold data for a period of time that allows ill-advised commitments from the RAID subsystem including indicating “good status” before the data is actually redundantly stored. Reordering and parallel paths are good for performance, but bad for proper operations in conventional RAID subsystems.
Accordingly, there exists a need to provide improved data storage systems and methods which utilize the benefits of parallel paths without compromising the integrity of redundantly stored data.
SUMMARY OF THE INVENTION
The invention provides redundant data storage systems and methods of operating redundant data storage systems.
In one aspect of the invention, a redundant data storage system comprises: a plurality of storage devices configured to redundantly store digital data; a plurality of transaction originating devices configured to originate a plurality of transactions to control operations of the storage devices; a plurality of parallel data buses configured to communicate data relative to the respective transaction originating devices; and a plurality of transaction processing devices coupled with the parallel data buses and configured to process the transactions in an order according to a transaction ordering protocol common to at least some of the transaction processing devices.
In another aspect of the invention, a method of operating a redundant data storage system comprises: redundantly storing data using a plurality of storage devices; originating a plurality of transactions to control operations of the storage devices using a plurality of transaction originating devices; communicating data with respect to the transaction originating devices using a plurality of parallel data buses; and processing the transactions using a plurality of processing transaction devices coupled with the parallel data buses, wherein the processing comprises processing the transactions in an order according to a transaction ordering protocol common to at least some of the processing transaction devices.
Another aspect provides a method of operating a redundant data storage system comprising: redundantly storing data using a plurality of storage devices of a redundant array of independent disks (RAID) storage system; originating a plurality of transactions to control operations of the storage devices using a plurality of transaction originating devices, the originating comprising originating a write transaction using an input/output processor of the storage system and originating a read transaction using a controller of the storage system; communicating data with respect to the transaction originating devices using a plurality of parallel data buses coupled with respective ones of the processor and the controller; and processing the transactions using a plurality of processing transaction devices coupled with the parallel data buses, wherein the processing comprises processing the transactions in an order according to a transaction ordering protocol common to at least some of the processing transaction devices which defines that the write transaction from the input/output processor precedes the read transaction from the controller.
Other features and advantages of the invention will become apparent to those of ordinary skill in the art upon review of the following detailed description, claims, and drawings.
REFERENCES:
patent: 5905876 (1999-05-01), Pawlowski et al.
patent: 6230240 (2001-05-01), Shrader et al.
patent: 6282610 (2001-08-01), Bergsten
patent: 6301643 (2001-10-01), Crockett et al.
Grund Christine
Johansson Christopher W
Oldfield Barry J
Rust Robert A.
Hewlett Packard Development Company, L. P.
Nguyen Hiep T.
LandOfFree
Redundant data storage systems and methods of operating a... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Redundant data storage systems and methods of operating a..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Redundant data storage systems and methods of operating a... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3029484