Hewlett-packard Company

Andrew J Wilkes

Robert M English

Chia-Chiang Chao

Alexander A Stepanov

David M Jacobson

Stefan Savage

There is provided a method and apparatus for effectuating data updates on a fault-tolerant storage device having a plurality of disk-drive units forming a disk array, with parity-block updates temporarily suspended to provide increased performance. Parity-block updates are typically deferred until the array is idle. Thus, data is written onto the array with no parity error correction information being present for a short amount of time. Data updates are effectuated by identifying the blocks in the array where data is to be written. New data is then written directly to the identified blocks without reading the old data present therein. A 'needs parity rebuild bit' (NPRB) is written to non-volatile memory, identifying the stripes not having parity information on the array. The number of NPRBs present is monitored to allow continuous calculation of a mean-time-to-data-loss (MTTDL). If the MTTDL is acceptable, parity updates are deferred until the array is idle. To that end, the I/O patterns to and from the array are monitored to determine idleness. If the I/O patterns indicate that the disk array is idle, new data is read and parity information is generated which is then written to the array.

Frequently-redundant array of independent disks

An electronic data storage system including a memory, a plurality of magnetic disk units, and a controller. The memory contains an index cross-referencing logical address with physical addresses, an obsolete list and a free list. In response to a 'write ' command, the controller selects a physical address according to which segment can be used the most quickly, appends a tag to the data to be written, and writes the data to the selected segment. Appropriate entries are made in the index and the free list. The system recovers from memory loss by using a checkpoint log and a set of checkpoint segments on the disk that together contain backups of the index and other critical information needed to restore the system. Group indices are used for roll-back groups; operations on data in a group are invisible outside the group until after a 'commit ' command is issued.

High performance multiple-unit electronic data storage system with

An interconnect interface for connecting a host computer to an interconnect fabric in a multi-computer data processing system including a plurality of host computers. A sending process having permission to send messages to a receiving process on another computer in the system controls a portion of the computer memory on the computer running the receiving process. The location of the memory in question together with a protection key value are stored in a table in the receiving terminus of the interconnect interface on the recipient. The sending terminus of the interconnect interface on the computer running the sending process incorporates information in each message packet which specifies an entry in this table. Each message also includes a protection key that must match the corresponding entry in the table on the receiving terminus.

Growing community of inventors

Palo Alto, CA, United States of America

Andrew J Wilkes