Abstract: A method for determining a single reference residency time of a cache comprises the steps of causing test data to be staged to the cache and measuring a response time after a wait time has elapsed. The measuring step is repeated for a plurality of values of wait time. The method also includes the step of determining a boundary value of the wait time. A wait time of less than or equal to the boundary value yields a corresponding response time representing a cache hit and a wait time of greater than the boundary value yields a corresponding response time representing a cache miss. The boundary value is an estimate of the single reference residency time of the cache.

Abstract: A computer system has a host computer that runs applications that require data objects from a storage system. The storage system has a storage device, such as a disk storage device, and a cache memory. Data objects that are frequently used by the host computer are stored in the cache memory. The data objects are also stored in the storage device logically arranged in segments of data objects, and groups of segments. The cache memory is logically partitioned into a segment cache and a group cache. Data of a small granular nature namely segments, are stored in the segment cache. Data of a larger granular nature, namely groups, is stored in the group cache. When the host computer requests accesses to read or write data objects, groups of segments are staged or destaged between the storage device and the group cache, with at least one segment of each staged group containing one of the requested data objects.

Abstract: A method and apparatus for defining a random miss list and producing a desired hit ratio in a benchmark test of computer disk storage are disclosed. The disk area to be tested is divided into N tracks. A random miss list of L tracks is defined as specified by the invention. The tracks in the miss list are subdivided into groups, with each group having G tracks. Each one of a plurality of concurrent processes is assigned a different group of G tracks from the miss list. A pointer for each process is initialized to point to the first track in the corresponding group of G tracks. A plurality of input/output (I/O) operations are performed for each process, with each I/O operation being performed on the track pointed to by the pointer for the corresponding process. For each process, prior to each I/O operation that is performed after the first I/O operation, a random decision is made whether or not to increment the pointer belonging to the process.

Abstract: A method and means for emulating realistic access requests used in static or dynamic performance testing of a disk-based storage subsystem. The method and means are based on the fact that a test driver can substitute for an actual application if an executing process generates a pattern of accesses to disk subsystem addresses as a prescribed random walk function among a cluster of contiguous tracks associated with the process. The access pattern will emulate both locality of referencing and the fact that the likelihood of rereferencing the same track varies inversely over time.

Abstract: A system and method are provided for managing persistent and non-persistent data in a log structured disk array. In a log structured array, data are first collected in a log stored in a buffer. When the buffer is full, the data are written to a disk drive. Persistent data are continually accessed over a long period of time such that, most disk activity is associated with persistent data. Preferably, the persistent data are stored in the part of the disk where access times are minimal (active region), such as in the middle cylinders of a disk. When writing data (including during garbage collection procedure) persistent data is written to the active region and non-persistent data is written to the other regions (inactive region). The persistence of the data is monitored. Persistent data is migrated to the active region and non-persistent is migrated to the inactive region.

Abstract: A cache management system for dynamically switching between record caching and track caching. Statistics are collected within the cache management system to determine the type and amount of locality associated with a specific data set currently within the cache. Using these statistics, metrics of the actual hit ratio, whole-track hit ratio, and other-record-hit ratio are generated. The metrics are periodically reviewed for a given band, and the band is placed into one of three cache modes accordingly. The mode assignment determines what type of staging will be performed for data within the band in the event of a miss. The possible modes are: track mode for track-based staging, record mode for staging of a single record only, or uncached mode if no data is to be staged. If a band has been assigned to either record or track mode, the assignment is re-evaluated after for a predetermined number of I/O's (128).

Abstract: A cache having dynamic cache memory allocation is provided. A cache memory stores a plurality of data blocks, each block belonging to one of a plurality of data sets. A cache directory maintains a list of entries associated with the data blocks stored in the cache memory, wherein each entry corresponds to an individual data block and has fields for storing information including a designation of the data set to which the corresponding data block belongs. A directory controller generates each entry when the corresponding data block is loaded in the cache. The directory controller inserts the generated entry into the list at the optimal insertion point for the data block's data set, which is derived from a calculated optimal single-reference residency time for that data set. Further, the directory controller moves an entry in the list to the insertion point for the given data set of a corresponding data block when the corresponding data block is referenced in the cache.

Abstract: Dynamic allocation of read cache space is allocated among bands of DASD cylinders rather than to data sets or processes as a function of a weighted average hit ratio to the counterpart cache space. Upon the hit ratio falling below a predetermined threshold, the bands are disabled for a defined interval as measured by cache accesses and then rebound to cache space again.