Overview of Storage and Indexing (based on slides from Wisconsin) - PDF document

Overview of Storage and Indexing (based on slides from Wisconsin) Murali Mani Data on External Storage � Disks: Can retrieve random page at fixed cost � But reading several consecutive pages is much cheaper than reading them in random order � Tapes: Can only read pages in sequence � Cheaper than disks; used for archival storage � File organization: Method of arranging a file of records on external storage. � Record id (rid) is sufficient to physically locate record � Indexes are data structures that allow us to find the record ids of records with given values in index search key fields � Architecture: Buffer manager stages pages from external storage to main memory buffer pool. File and index layers make calls to the buffer manager. Murali Mani 1

Why Not Store Everything in Main Memory? � Cost . Disks always cheaper than memory. � Main memory is volatile . We want data to be saved between runs !! � Typical storage hierarchy: � Main memory (RAM) for currently used data. � Disk for the main database (secondary storage). � Tapes for archiving older versions of the data (tertiary storage). Murali Mani Disks � Secondary storage device of choice. � Main advantage over tapes: random access vs. sequential . � Data is stored and retrieved in units called disk blocks or pages . � Unlike RAM, time to retrieve a disk page varies depending upon location on disk. � Therefore, relative placement of pages on disk has major impact on DBMS performance! Murali Mani 2

Components of a Disk � The platters spin (say, 90rps). Spindle Tracks Disk head � The arm assembly is moved in or out to position a head on Sector a desired track. Tracks under heads make a cylinder (imaginary!). � Only one head Platters Arm movement reads/writes at any one time. � Block size is a multiple of Arm assembly sector size (which is fixed). Murali Mani Accessing a Disk Page � Time to access (read/write) a disk block: � seek time ( moving arms to position disk head on track ) � rotational delay ( waiting for block to rotate under head ) � transfer time ( actually moving data to/from disk surface ) � Seek time and rotational delay dominate. � Seek time varies from about 1 to 20msec � Rotational delay varies from 0 to 10msec � Transfer rate is about 1msec per 4KB page � Key to lower I/O cost: reduce seek/rotation delays! Hardware vs. software solutions? Murali Mani 3

Arranging Pages on Disk � ` Next ’ block concept: � blocks on same track, followed by � blocks on same cylinder, followed by � blocks on adjacent cylinder � Blocks in a file should be arranged sequentially on disk (by `next’), to minimize seek and rotational delay. � For a sequential scan, pre-fetching several pages at a time is a big win! Murali Mani RAID � Disk Array: Arrangement of several disks that gives abstraction of a single, large disk. � Goals: Increase performance and reliability. � Two main techniques: � Data striping: Data is partitioned; size of a partition is called the striping unit. Partitions are distributed over several disks. � Redundancy: More disks => can handle more failures. Redundant information allows reconstruction of data if a disk fails. Murali Mani 4

Disk Space Management � Lowest layer of DBMS software manages space on disk. � Higher levels call upon this layer to: � allocate/de-allocate a page � read/write a page � Request for a sequence of pages must be satisfied by allocating the pages sequentially on disk! Higher levels don’t need to know how this is done, or how free space is managed. Murali Mani Buffer Management in a DBMS Page Requests from Higher Levels BUFFER POOL disk page free frame MAIN MEMORY DISK choice of frame dictated DB by replacement policy � Data must be in RAM for DBMS to operate on it! � Table of <frame#, pageid> pairs is maintained. Murali Mani 5

DBMS vs. OS File System OS does disk space & buffer mgmt: why not let OS manage these tasks? � Differences in OS support: portability issues � Some limitations, e.g., files can’t span disks. Murali Mani Record Formats: Fixed Length F1 F2 F3 F4 L1 L2 L3 L4 Base address (B) Address = B+L1+L2 � Information about field types same for all records in a file; stored in system catalogs. � Finding i’th field does not require scan of record. Murali Mani 6

Record Formats: Variable Length Two alternative formats (# fields is fixed): F1 F2 F3 F4 4 $ $ $ $ Fields Delimited by Special Symbols Field Count F1 F2 F3 F4 Array of Field Offsets � Second offers direct access to i’th field, efficient storage of nulls (special don’t know value); small directory overhead. Murali Mani Page Formats: Fixed Length Records Slot 1 Slot 1 Slot 2 Slot 2 Free . . . . . . Space Slot N Slot N Slot M N 1 . . . 0 1 1 M M ... 3 2 1 number number PACKED of records UNPACKED, BITMAP of slots � Record id = <page id, slot #>. In first alternative, moving records for free space management changes rid; may not be acceptable. Murali Mani 7

Page Formats: Variable Length Records Rid = (i,N) Page i Rid = (i,2) Rid = (i,1) N 20 16 24 Pointer to start N . . . 2 1 # slots of free space SLOT DIRECTORY � Can move records on page without changing rid; so, attractive for fixed-length records too . Murali Mani Files of Records � Page or block is OK when doing I/O, but higher levels of DBMS operate on records , and files of records . � FILE : A collection of pages, each containing a collection of records. Must support: � insert/delete/modify record � read a particular record (specified using record id ) � scan all records (possibly with some conditions on the records to be retrieved) Murali Mani 8

Alternative File Organizations Many alternatives exist, each ideal for some situations, and not so good in others: � Heap (random order) files: Suitable when typical access is a file scan retrieving all records. � Sorted Files: Best if records must be retrieved in some order, or only a `range’ of records is needed. � Indexes: Data structures to organize records via trees or hashing. � Like sorted files, they speed up searches for a subset of records, based on values in certain (“search key”) fields � Updates are much faster than in sorted files. Murali Mani Indexes � An index on a file speeds up selections on the search key fields for the index. � Any subset of the fields of a relation can be the search key for an index on the relation. � Search key is not the same as key (minimal set of fields that uniquely identify a record in a relation). � An index contains a collection of data entries , and supports efficient retrieval of all data entries k* with a given key value k . Murali Mani 9

B+ Tree Indexes Non-leaf Pages Leaf Pages (Sorted by search key) � Leaf pages contain data entries , and are chained (prev & next) � Non-leaf pages have index entries; only used to direct searches: index entry P0 K 1 P 1 K 2 P m P 2 K m Murali Mani Example B+ Tree Note how data entries in leaf level are sorted Root 17 Entries <= 17 Entries > 17 27 5 13 30 33* 34*38* 39* 2* 3* 5* 7* 8* 14* 16* 22*24* 27* 29* � Find 28*? 29*? All > 15* and < 30* � Insert/delete: Find data entry in leaf, then change it. Need to adjust parent sometimes. � And change sometimes bubbles up the tree Murali Mani 10

Hash-Based Indexes � Good for equality selections. � Index is a collection of buckets. � Bucket = primary page plus zero or more overflow pages. � Buckets contain data entries. � Hashing function h : h ( r ) = bucket in which (data entry for) record r belongs. h looks at the search key fields of r. � No need for “index entries” in this scheme. Murali Mani Choice of Indexes � What indexes should we create? � Which relations should have indexes? What field(s) should be the search key? Should we build several indexes? � For each index, what kind of an index should it be? � Hash/tree? Murali Mani 11

Choice of Indexes (Contd.) � One approach: Consider the most important queries in turn. Consider the best plan using the current indexes, and see if a better plan is possible with an additional index. If so, create it. � Obviously, this implies that we must understand how a DBMS evaluates queries and creates query evaluation plans! � For now, we discuss simple 1-table queries. � Before creating an index, must also consider the impact on updates in the workload! � Trade-off: Indexes can make queries go faster, updates slower. Require disk space, too. Murali Mani Index Selection Guidelines � Attributes in WHERE clause are candidates for index keys. � Exact match condition suggests hash index. � Range query suggests tree index. � Multi-attribute search keys should be considered when a WHERE clause contains several conditions. � Order of attributes is important for range queries. � Such indexes can sometimes enable index-only strategies for important queries. � Try to choose indexes that benefit as many queries as possible. Since only one index can be clustered per relation, choose it based on important queries that would benefit the most from clustering. Murali Mani 12