2 December 2005 Introduction to Databases Storage Management Prof. Beat Signer Department of Computer Science Vrije Universiteit Brussel beatsigner.com

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 2 April 3, 2019 Context of Today's Lecture Access Methods System Buffers Authorisation Control Integrity Checker Command Processor Program Object Code DDL Compiler File Manager Buffer Manager Recovery Manager Scheduler Query Optimiser Transaction Manager Query Compiler Queries Catalogue Manager DML Preprocessor Database Schema Application Programs Database Manager Data Manager DBMS Programmers Users DB Admins Based on 'Components of a DBMS', Database Systems, T. Connolly and C. Begg, Addison-Wesley 2010 Data, Indices and System Catalogue

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 3 April 3, 2019 Storage Device Hierarchy ▪ Storage devices vary in ▪ data capacity ▪ access speed ▪ cost per byte ▪ Devices with fastest access time have highest costs and smallest capacity Cache Main Memory Flash Memory Magnetic Disk Optical Disk Magnetic Tapes

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 4 April 3, 2019 Cache ▪ On-board caches on same chip as the microprocessor ▪ level 1 (L1) cache (typical size of ~64 kB ) - temporary storage of instructions and data ▪ level 2 (L2) cache (~1 MB) and level 3 (L3) cache (~8 MB) ▪ Data items in the cache are copies of values in main memory locations ▪ If data in the cache has been updated, changes must be reflected in the corresponding memory locations

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 5 April 3, 2019 Main Memory ▪ Main memory can be several gigabytes large ▪ Normally too small and too expensive for storing the entire database ▪ content is lost during power failure or crash (volatile memory) ▪ in-memory databases (IMDB) primarily rely on main memory - note that IMDBs lack durability (D of the ACID properties) ▪ IMDB size limited by the maximal addressable memory space - e.g. maximal 4 GB for 32-bit address space ▪ Random access memory (RAM) ▪ time to access data is more or less independent of its location (different from magnetic tapes) ▪ Typical access time of ~10 nanoseconds (10-8 seconds)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 6 April 3, 2019 Secondary Storage (Hard Disk) ▪ Essentially random access ▪ Files are moved between a hard disk and main memory (disk I/O) by the operating system (OS) or the DBMS ▪ the transfer units are blocks ▪ tendency for larger block sizes ▪ Parts of the main memory are used to buffer blocks ▪ the buffer manager of the DBMS manages the loading and unloading of blocks for specific DBMS operations ▪ Typical block I/O time (seek time) ~10 milliseconds ▪ 1'000'000 times slower than main memory access ▪ Capacity of multiple terabytes and a system can use many disk units

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 7 April 3, 2019 Hard Disk ▪ A hard disk contains one or more platters and one or more heads ▪ The platters were originally addressed in terms of cylinders, heads and sectors ( block) ▪ cylinder-head-sector (CHS) scheme ▪ max of 1024 cylinders, 16 heads and 63 sectors ▪ Current hard disks offer logical block addressing (LBA) ▪ hides the physical disk geometry

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 8 April 3, 2019 Solid-State Drives (SSD) ▪ Storage device that uses solid-state memory (flash memory) to persistently store data ▪ Offers a hard disk interface with a storage capacity of up to a few hundred gigabytes ▪ Typical block I/O time (seek time) ~0.1 milliseconds ▪ SSDs might help to reduce the gap between primary and secondary storage in DBMS systems ▪ Currently there are still some limitations of SSDs ▪ the limited number of SSD write operations before failure can be a problem for DBs with a lot of update operations ▪ write operations are often still much slower than read operations

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 9 April 3, 2019 Tertiary Storage ▪ No random access ▪ access time depends on data location ▪ Different devices ▪ tapes ▪ optical disk jukeboxes - racks of CD-ROMs (read only) ▪ tape silos - room-sized devices holding racks of tapes operated by tape robots - e.g. StorageTek PowderHorn with up to 28.8 petabytes

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 10 April 3, 2019 Models of Computation ▪ RAM model of computation ▪ assumes that all data is held in main memory ▪ DBMS model of computation ▪ assumes that data does not fit into main memory ▪ efficient algorithms must take into account secondary and even tertiary storage ▪ best algorithms for processing large amounts of data often differ from those for the RAM model of computation ▪ minimising disk accesses plays a major role - I/O model of computation ▪ I/O model of computation ▪ the time to move a block between disk and memory is much higher than the time for the corresponding computation

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 11 April 3, 2019 Accelerating Secondary Storage Access ▪ Various possible strategies to improve secondary storage access ▪ placement of blocks that are often accessed together on the same disk cylinder ▪ distribute data across multiple disks to profit from parallel disk accesses (e.g. RAID) ▪ mirroring of data ▪ use of disk scheduling algorithms in OS, DBMS or disk controller to determine order of requested block read/writes - e.g. elevator algorithm ▪ prefetching of disk blocks ▪ efficient caching - main memory - disk controllers

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 12 April 3, 2019 Redundant Array of Independent Disks ▪ The redundant array of independent disks (RAID) organisation technique provides a single disk view for a number (array) of disks ▪ divide and replicate data across multiple hard disks ▪ introduced in 1987 by D.A. Patterson, G.A. Gibson and R. Katz ▪ The main goals of a RAID solution are ▪ higher capacity by grouping multiple disks - originally a RAID was also a cheaper alternative to expensive large disks • original name: Redundant Array of Inexpensive Disks ▪ higher performance due to parallel disk access - multiple parallel read/write operations ▪ increased reliability since data might be stored redundantly - data can be restored if a disk fails

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 13 April 3, 2019 RAID ... ▪ There are three main concepts in RAID systems ▪ identical data is written to more than one disk (mirroring) ▪ data is split across multiple disks (striping) ▪ redundant parity data is stored on separated disks and used to detect and fix problems (error correction)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 14 April 3, 2019 RAID Reliability ▪ The mean time between failures (MTBF) is the average time until a disk failure occurs ▪ e.g. a hard disk might have a MTBF of 200'000 hours (22.8 years) - note that the MTBF decreases as disks get older ▪ If a DBMS uses an array of disks, then the overall system's MTBF can be much lower ▪ e.g. the MTBF for a disk array of 100 of the disks mentioned above is 200'000 hours/100 = 2'000 hours (83 days) ▪ By storing information redundantly, data can be restored in the case of a disk failure

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 15 April 3, 2019 RAID Reliability ... ▪ The mean time to data loss (MTTDL) depends on the MTBF and the mean time to repair (MTTR) ▪ for mirroring data on two disks the MTTDL is defined by ▪ if we mirror the information on two disks with a MTBF of 200'000 hours and a mean time to repair of 10 hours then the MTTDL is 200'0002/(2*10) hours = 228'000 years ▪ of course in reality it is more likely that an error occurs on multiple disks around the same time - drives have the same age - power failure, earthquake, fire, ... MTTR MTBF MTTDL  = 2 2

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 16 April 3, 2019 RAID Levels ▪ The different RAID levels offer different cost-performance trade-offs ▪ RAID 0 ▪ block level striping without any redundancy ▪ RAID 1 ▪ mirroring without striping ▪ RAID 2 ▪ bit level striping ▪ multiple parity disks ▪ RAID 3 ▪ byte level striping ▪ one parity disk [http://en.wikipedia.org/wiki/RAID]

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 17 April 3, 2019 RAID Levels ... ▪ RAID 4 ▪ block level striping ▪ one parity disk ▪ similar to RAID 3 ▪ RAID 5 ▪ block level striping with distributed parity ▪ no dedicated parity disk ▪ RAID 6 ▪ block level striping with dual distributed parity ▪ no dedicated parity disk ▪ similar to RAID 5

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 18 April 3, 2019 Data Representation ▪ A DBMS has to define how the elements of its data model (e.g. relational model) are mapped to secondary storage ▪ a field contains a fixed- or variable-length sequence of bytes and represents an attribute ▪ a record contains a fixed- or variable-length sequence of fields and represents a tuple ▪ records are stored in fixed-length physical block storage units representing a set of tuples - the blocks also represent the units of data transfer ▪ a file contains a collection of blocks and represents a relation ▪ A database is finally mapped to a number of files managed by the underlying operating system ▪ index structures are stored in separate files

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 19 April 3, 2019 Relational Model Representation ▪ A number of issues have to be addressed when mapping the basic elements of the relational model to secondary storage ▪ how to map the SQL datatypes to fields? ▪ how to represent tuples as records? ▪ how to represent records in blocks? ▪ how to represent a relation as a collection of blocks? ▪ how to deal with record sizes that do not fit into blocks? ▪ how to deal with variable-length records? ▪ how to deal with schema updates and growing record lengths? ▪ ...

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 20 April 3, 2019 Representation of SQL Datatypes ▪ Fixed-length character string (CHAR(n)) ▪ represented as a field which is an array of n bytes ▪ strings that are shorter than n bytes are filled up with a special "pad" character ▪ Variable-length character string (VARCHAR(n)) ▪ two common representations (non-fixed length version later) ▪ length plus content - allocate an array of n + 1 bytes - the first byte represents the length of the string (8-bit integer) followed by the string content - limited to a maximal string length of 255 characters ▪ null-terminated string - allocate an array of n + 1 bytes - terminate the string with a special null character (like in C)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 21 April 3, 2019 Representation of SQL Datatypes ... ▪ Dates (DATE) ▪ fixed-length character string ▪ Time (TIME(n)) ▪ the precision n leads to strings of variable length and two possible representations ▪ fixed-precision - limit the precision to a fixed value and store as VARCHAR(m) ▪ true-variable length - store the time as true variable length value ▪ Bits (BIT(n)) ▪ bit values of size n can be packed into single bytes ▪ packing of multiple bit values into a single byte is not recommended - makes the retrieval and updating of a value more complex and error-prone

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 22 April 3, 2019 Storage Access ▪ A part of the system's main memory is used as a buffer to store copies of disk blocks ▪ The buffer manager is responsible to move data from secondary disk storage into memory ▪ the number of block transfers between disk and memory should be minimised ▪ as many blocks a possible should be kept in memory ▪ The buffer manager is called by the DMBS every time a disk block has to be accessed ▪ the buffer manager has to check whether the block is already allocated in the buffer (main memory)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 23 April 3, 2019 Buffer Manager ▪ If the requested block is already in the buffer, the buffer manager returns the corresponding address ▪ If the block is not yet in the buffer, the buffer manager performs the following steps ▪ allocate buffer space - if no space is available, remove an existing block from the buffer (based on a buffer replacement strategy) and write it back to the disk if it has been modified since it was last fetched/written to disk ▪ read the block from the disk, add it to the buffer and return the corresponding memory address ▪ Note the similarities to a virtual memory manager

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 24 April 3, 2019 Buffer Replacement Strategies ▪ Most operating systems use a least recently used (LRU) strategy where the block that was least recently used is moved back from memory to disk ▪ use past access pattern to predict future block access ▪ A DBMS is able to predict future access patterns more accurately than an operating system ▪ a request to the DBMS involves multiple steps and the DBMS might be able to determine which blocks will be needed by analysing the different steps of the operation ▪ note that LRU might not always be the best replacement strategy for a DBMS

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 25 April 3, 2019 Buffer Replacement Strategies ... ▪ Let us have a look at the procedure to compute the following natural join query: order ⋈ customer ▪ note that we will see more efficient solutions for this problem when discussing query optimisation for each tuple o of order { for each tuple c of customer { if o.customerID = c.customerID { create a new tuple r with: r.customerID := c.customerID r.name := c.name ... r.orderID := o.orderID ... add tuple r to the result set of the join operation } } }

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 26 April 3, 2019 Buffer Replacement Strategies ... ▪ We further assume that the two relations order and customer are stored in separate files ▪ From the pseudocode we can see that ▪ once an order tuple has been processed, it is not needed anymore - if a whole block of order tuples has been processed, that block is no longer required in memory (but an LRU strategy might keep it) - as soon as the last tuple of an order block has been processed, the buffer manager should free the memory space → toss-immediate strategy ▪ once a customer tuple has been processed, it is not accessed again until all the other customer tuples have been accessed - when the processing of a customer block has been finished, the least recently used customer block will be requested next - we should replace the block that has been most recently used (MRU)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 27 April 3, 2019 Buffer Replacement Strategies ... ▪ A memory block can be marked to indicate that this block is not allowed to be written back to disk (pinned block) ▪ note that if we want to use an MRU strategy for the inner loop of the previous example, the block has to be pinned - the block has to be unpinned after the last tuple in the block has be processed ▪ the pinning of blocks provides some control to restrict the time when blocks can be written back to disk - important for crash recovery - blocks that are currently updated should not be written to disk ▪ The prefetching of blocks might be used to further increase the performance of the overall system ▪ e.g. for serial scans (relation scans)

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 28 April 3, 2019 Buffer Replacement Strategies ... ▪ The buffer manager can also use statistical information about the probability that a request will reference a particular relation (and its related blocks) ▪ the system catalogue (data dictionary) with its metadata is one of the most frequently accessed parts of the database - if possible, system catalogue blocks should always be in the buffer ▪ index files might be accessed more often than the corresponding files themselves - do not remove index files from the buffer if not necessary ▪ the crash recovery manager can also provide constraints for the buffer manager - the recovery manager might demand that other blocks have to be written first (force-output) before a specific block can be written to disk

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 29 April 3, 2019 System Catalogue / Data Dictionary ▪ Stores metadata about the database ▪ names of the relations ▪ names, domain and lengths of the attributes of each relation ▪ names of views ▪ names of indices - name of relation that is indexed - name of attributes - type of index ▪ integrity constraints ▪ users and their authorisations ▪ statistical data - number of tuples in relation, storage method, ... ▪ ...

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 30 April 3, 2019 File Organisation ▪ A file is a logically organised as a sequence of records ▪ each record contains a sequence of fields ▪ name, datatype and offset of record fields are defined by the schema ▪ record types (schema) might change over time ▪ The records are mapped to disk blocks ▪ the block size is fixed and defined by the physical properties of the disk and the operating system ▪ the record size might vary for different relations and even between tuples of the same relation (variable field size) ▪ There are different possible mappings of records to files ▪ use multiple files and only store fixed-length records in each file ▪ store variable-length records in a file

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 31 April 3, 2019 Fixed-Length Records ▪ If we assume that an integer requires 2 bytes and characters are represented by one byte, then the customer record is 64 bytes long type customer = record cID int; name varchar(30) street varchar(30) end cID name street ... ... 0 2 33 64 Block

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 32 April 3, 2019 Fixed-Length Records ... ▪ Often a record header is added to each record for managing metadata about ▪ the record schema (pointer s to the DBMS schema information) ▪ timestamp t about the last access or modification time ▪ the length l of the record - could be computed from the schema but the information is convenient if we want to quickly access the next record without having to consult the schema ▪ ... 0 12 48 80 cID name street ... ... s t l 16 Block

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 33 April 3, 2019 Fixed-Length Records in Blocks/Files ▪ Problems with this fixed length representation ▪ after a record has been deleted, its space has to be filled with another record - could move all records after the deleted one but that is too expensive - can move the last record to the deleted record's position but also that might require an additional block access ▪ if the block size is not a multiple of the record size, some records will cross block boundaries and we need two block accesses to read/write such a record 1 Max Frisch Bahnhofstrasse 7 h 2 Eddy Merckx Pleinlaan 25 h 5 Claude Debussy 12 Rue Louise h 53 Albert Einstein Bergstrasse 18 h 8 Max Frisch ETH Zentrum h record 0 record 1 record 2 record 3 record 4

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 34 April 3, 2019 Fixed-Length Records in Blocks/Files ... ▪ Since insert operations tend to be more frequent that delete operations, it might be acceptable to leave the space of the deleted record open until a new record is inserted ▪ we cannot just add an additional boolean flag ("free") to the record since it will be hard to find the free records ▪ allocate a certain amount of bytes for a file header containing metadata about the file ▪ The block/file header contains a pointer (address) to the first deleted record ▪ each deleted record contains a pointer (address) to the next deleted record ▪ the linked list of deleted records is called a free list

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 35 April 3, 2019 record 0 record 1 record 2 record 3 record 4 header Fixed-Length Records in Blocks/Files ... ▪ To insert a new record, the first free record pointed to by the header is used and the address in the header is updated to the free record that the used record was pointing to ▪ to save some space, the pointers of the free list can also be stored in the unused space of deleted records (no additional field) 1 Max Frisch Bahnhofstrasse 7 h 5 Claude Debussy 12 Rue Louise h 8 Max Frisch ETH Zentrum h

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 36 April 3, 2019 Address Space ▪ There are several ways how the database address space (blocks and block offsets) can be represented ▪ physical addresses consisting of byte strings (up to 16 bytes) that address - host - storage device identifier (e.g. hard disk ID) - cylinder number of the disk - track within the cylinder (for multi-surface disks) - block within the track - potential offset of record within the block ▪ logical addresses consisting of an arbitrary string of length n

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 37 April 3, 2019 Address Space Mapping ▪ A map table is stored at a known disk location and provides a mapping between the logical and physical address spaces ▪ introduces some indirection since the map table has to be consulted to get the physical address ▪ flexibility to rearrange records within blocks or move them to other blocks without affecting the record's logical address ▪ different combinations of logical and physical addresses are possible (structured address schemes) ... ... logical physical logical address physical address map table

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 38 April 3, 2019 Variable-Length Data ▪ Records of the same type may have different lengths ▪ We may want to represent ▪ record fields with varying size (e.g. VARCHAR(n)) ▪ large fields (e.g. images) ▪ ... ▪ We need an alternative data representation to deal with these requirements

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 39 April 3, 2019 Variable-Length Record Fields ▪ Scheme for records with variable-length fields ▪ put all fixed-length fields first (e.g. cID) ▪ add the length of the record to the record header ▪ add the offsets of the variable-length fields to the record header ▪ Note that if the order of the variable-length fields is always the same, we do not have to store an offset for the first variable-length field (e.g. name) cID name street record length

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 40 April 3, 2019 Variable-Length Records ▪ There are different reasons why we might have to use variable-length records ▪ to store records that have at least one field with a variable length ▪ to store different record types in a single block/file ▪ Structured address scheme (slotted-page structure) ▪ address of a record consists of the block address in combination with an offset table index ▪ records can be moved around record3 record2 record1 ... free ... offset table

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 41 April 3, 2019 Large Records ▪ Sometimes we have to deal with values that do not fit into a single block (e.g. audio or movie clips) ▪ a record that is split across two or more blocks is called a spanned record ▪ spanned records can also be used to pack blocks more efficiently ▪ Extra header information ▪ each record header carries a bit to indicate if it is a fragment - fragments have some more bits; telling whether first or last fragment of record ▪ potential pointers to previous and next fragment block header record header record2b record3 record2a record1 block 1 block 2

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 42 April 3, 2019 Storage of Binary Large Objects (BLOBS) ▪ BLOB is stored as a sequence of blocks ▪ often blocks allocated successively on a disk cylinder ▪ BLOB might be striped across multiple disks for more efficient retrieval ▪ BLOB field might not be automatically fetched into memory ▪ user has to explicitly load parts of the BLOB ▪ possibly index structures to retrieve parts of a BLOB

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 43 April 3, 2019 Insertion of Records ▪ If the records are not kept in a particular order, we can just find a block with some empty space or create a new block if there is no such space ▪ If the record has to be inserted in a particular order, but there is no space in the block, there are two alternatives ▪ find space in a nearby block and rearrange some records ▪ create an overflow block and link it from the header of the original block - note that an overflow block might point to another overflow block and so on record3 record2 record1 ... free ... offset table

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 44 April 3, 2019 Deletion of Records ▪ If we use an offset table, we may compact the free space in the block (slide around the records) ▪ If the records cannot be moved, we might have a free list in the header ▪ We might also be able to remove an overflow block after a delete operation record3 record2 record1 ... free ... offset table

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 45 April 3, 2019 Update of Records ▪ If we have to update a fixed-length record there is no problem since we will still use the same space ▪ If the updated record is larger than the original version, then we might have to create more space ▪ same options as discussed for insert operation ▪ If the updated record is smaller, then we may compact some free space or remove overflow blocks ▪ similar to delete operation record3 record2 record1 ... free ... offset table

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 46 April 3, 2019 Homework ▪ Study the following chapter of the Database System Concepts book ▪ chapter 10 - sections 10.1-10.9 - Storage and File Structure

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 47 April 3, 2019 Exercise 8 ▪ Functional Dependencies and Normalisation

Beat Signer - Department of Computer Science - bsigner@vub.ac.be 48 April 3, 2019 References ▪ H. Garcia-Molina, J.D. Ullman and J. Widom, Database Systems – The Complete Book, Prentice Hall, 2002 ▪ A. Silberschatz, H. Korth and S. Sudarshan, Database System Concepts (Sixth Edition), McGraw-Hill, 2010

2 December 2005 Next Lecture Access Methods