HP 3000 Manuals HP 3000 Manuals
Internal Techniques Although it is not necessary to know the following techniques to use TurboIMAGE/XL, an understanding of them can help you design a more efficient database. Primary Address Calculation [REV BEG] TurboIMAGE/XL employs two distinct methods of calculating primary addresses in master data sets. The intent of the first method is to spread master entries as uniformly as possible throughout the record space of the data file. This method is referred to as "hashing" and applies only to master data sets with key items of type U, X, Z, or P. In this case, the entire key item value regardless of its length is folded into a positive 32-bit value. This value is reduced modulo the data set capacity and then incremented by one to form a primary address.
NOTE Because values are folded from right to left, applications should place values in a type U, X, Z, or P field as follows: * Values that change should be placed first in the field. * Values that are static should be placed last in the field.
The intent of the second method is to permit you to control the primary address assignment. This method applies only to master data sets with key items of type I, J, K, R, or E. In this case, the right-most block of 32 bits of the key value is treated as a double integer. (If its sign bit is not zero, it is set to zero. If the key item is only 16 bits long, a 32-bit value is created by prefacing these 16 bits with 16 zero bits.) This 32-bit integer is decremented by 1 and reduced modulo the data set capacity after which 1 is added to the result to form the primary address. If the application provides key values whose right-most 32 bits take on values between 1 and N (where N is no greater than the data set capacity), the corresponding entries will be assigned primary addresses 1 through N which is identical to the Direct Access Method (DAM). In this event, there are no secondaries and performance is outstanding. However, if the application has no control of the key value assignment and/or if N exceeds the capacity, secondaries will occur; this, along with the clustering typical of this method, may result in poor DBPUT performance. This method should be used only if you have determined that the potential poor performance consequences cannot occur.[REV END] The intent of the two primary address algorithms is to spread master entries as uniformly as possible throughout the record space of the data file. This uniform spread reduces the number of synonyms occurring in the master data set.
NOTE In general, a master data set with a capacity equal to a prime number or to the product of two or three primes can yield fewer synonyms than master data sets with capacities consisting of many factors. Refer to appendix C for a list of prime numbers.
Migrating Secondaries In some cases, secondary entries of master data sets are automatically moved to storage locations other than the one originally assigned. This most often occurs when a new master data entry is assigned a primary address occupied by a secondary entry. By definition, the secondary entry is a synonym to some other primary entry resident at their common primary address. Thus, the new entry represents the beginning of a new synonym chain. To accommodate this new chain, the secondary entry is moved to an alternate secondary address and the new entry is added to the data set as a new primary entry. This move and the necessary linkage and chain head maintenance is done automatically. A move can also occur when the primary entry of a synonym chain that has one or more secondary entries is deleted. Because retrieval of each entry occurs through a synonym chain, each synonym chain must have a primary entry. To maintain the integrity of a synonym chain, TurboIMAGE/XL always moves the first secondary entry to the primary address of the deleted primary entry. Space Allocation for Master Data Sets Space allocation for each master data set is controlled by a free space counter resident in the user label of the data set, and by each bit map that monitors each block of the data set. When a new entry is added, TurboIMAGE/XL decrements the free space counter and sets the bit corresponding to the newly assigned record address to a one. If the bit is a zero before the record is added, the assigned record address is the primary address. If the bit is a one before the record is added, it indicates that an entry already exists. A secondary address is identified by a serial search of the bit maps for a zero indicating an unused record, and this address is assigned to the entry being added. Space Allocation for Detail Data Sets Space allocation for each detail data set is controlled by a free space counter, an end-of-file pointer and a pointer to a delete chain. The end-of-file pointer contains the record address of the highest-numbered entry which has existed so far in the data set. The delete chain pointer contains the record address of the entry which was most recently deleted. When each detail data set is first created, the end-of-file pointer and delete chain pointer are both zero. [REV BEG] When a new entry is added to a detail data set, TurboIMAGE/XL assigns to it the record address referenced by the delete chain pointer, unless the pointer is zero or HWMPUT has been enabled either using DBCONTROL or using DBUTIL. If the chain pointer is zero or if the HWMPUT flag is enabled, the end-of-file pointer is incremented and then used as the assigned record address. The free space counter is decremented in either case. When an existing entry is deleted, its media record is zeroed, the first word is replaced with the current delete chain pointer, and the block is written to disk. The delete chain pointer is set to the address of the newly deleted entry and the free space counter is incremented. The delete chain is, in effect, a "last-in-first-out" linked list of reusable media record space. Reusable space is allocated in preference to the unused space represented by record addresses beyond the end-of-file pointer except when HWMPUT is enabled. [REV END] Addition and deletion of data entries also requires data chain maintenance and turning on or turning off the corresponding bit of the appropriate bit map. Both of these are necessary for retrieval integrity but neither play a role in space allocation for detail data sets. Buffer Management TurboIMAGE/XL maintains a set of buffer partitions in the DBB for all users of an open database. DBFIND, DBGET, DBUPDATE, DBPUT, and DBDELETE locate a buffer header from one of these partitions. Each partition is allocated its own buffer header pool, hash table, and free list. The buffer header pool is a set of buffer headers allocated for the accessors of its corresponding partition. The hash table consists of linked lists of buffer header addresses either in use or ready to be released. The free list is a linked list of available buffer headers. Initially, when the DBB is created, all of the buffer headers belonging to a partition are linked to a free list and all the hash table chains are empty. TurboIMAGE/XL uses a two-level hashing algorithm based on the block number of the data set to determine the partition number as well as the hash table entry to be used. When an intrinsic issues a request for a data set block, the buffer manager starts the search from its hash table entry. If the hash table chain is empty, it acquires a buffer header from the free list. The buffer header is first allocated from the free list to build the buffer header for the data set block and link it to its appropriate hash table chain. When the hash chain, as well as the free list search, is exhausted, the process pauses to wait for other processes to release buffers then retries the buffer header pool scan. Locking Internals Within the DBG is a large lock area that provides space for the entries described below. Accessor Entries. One of these is created for each successful call to DBOPEN (each access path). Although located in the lock area, each accessor entry is the link with which TurboIMAGE/XL controls access to the database. An accessor entry is deleted when DBCLOSE is called for the access path, and the space is reused. Set Entries. One of these is created for every data set that is specified in a lock request. Therefore, the maximum number of set entries is equal to the number of data sets in the database. These entries are never deleted. Descriptor Entries. These entries contain the internal form of the lock descriptors specified in locking mode 5 or 6. They disappear when the locks are released (when DBUNLOCK is called) and the space is reused.