Chap 5. Disk IO Distribution Chap 6. Index Architecture Written by Yong-soon Kwon Summerized By Sungchan IDS Lab Copyright 2008 by CEBT Overview Chap 5. Disk IO Distribution Oracle Storage Architecture Tuning by Disk IO Distribution Chap 6. Index Architecture Introduction to Index B+-Tree Index Bitmap Index Reverse Key Index Function Base Index Random Access Tuning Center for E-Business Technology Chap 5. Disk IO Distribution Copyright 2008 by CEBT Disk IO Distribution Pros Improving Performance Cons Probability of error increase Problem on one disk affects whole table Hard to manage Center for E-Business Technology Copyright 2008 by CEBT Disk IO Distribution Rule Disk IO Distribution Rule We should increase the probability that all disks are accessed. Data in each table should be distributed to all disks How to satisfy the rule? 1.Optimizing Disk Organization 2.Segment Extent Striping 3.Applying Table Archtecture Partition table, IOT table Center for E-Business Technology Copyright 2008 by CEBT Optimizing Disk Organization: Volume Group Volume Groups(VG) Groups of Physical Disks One physical disk cannot belong to multiple VG Checking VG(HP UX) vgdisplay v vg01 Important Info Cur PV : 4 : This VG consists of four physical disks Act PV : 4 : Four physical disks are active in this VG Center for E-Business Technology Copyright 2008 by CEBT Optimizing Disk Organization: Logical Volume Concept of Logical Volume Space for file system in VG Logical Volume Usage File System Raw Device Center for E-Business Technology Copyright 2008 by CEBT Optimizing Disk Organization: Logical Volume Striping Logical Volume Striping Center for E-Business Technology Copyright 2008 by CEBT Optimizing Disk Organization: Logical Volume Striping Good Point of Logical Volume Striping Achieve Disk IO distribution rule for each VG Checking Logical Volume ldisplay v /dev/vg01/lvol1 Important Info Stripes 4 : This logical volume is striped with 4 disks Stripe Size(Kbytes) 64 : The striping size is 64KB Center for E-Business Technology Copyright 2008 by CEBT Oracle Storage Architecture Center for E-Business Technology Copyright 2008 by CEBT Oracle Storage Architecture OS Block OS Level IO unit DB Block n OS Block. DB IO unit Extent Contiguous DB Blocks. Building unit of Segment Segment Group of Extents(table, index) Tablespace Group of Segments Database Group of Tablespaces Datafile Stores Segments. n Datafile make one Tablespace Center for E-Business Technology Copyright 2008 by CEBT Segment Extent Striping: Create Tablespace Center for E-Business Technology Copyright 2008 by CEBT Segment Extent Striping: Create Table on Tablespace Center for E-Business Technology Copyright 2008 by CEBT Segment Extent Striping: Create Table on Tablespace Center for E-Business Technology Copyright 2008 by CEBT Segment Extent Striping Good Points Storing data evenly over all disks for all tables Disk IO distribution rule optimization Checking Exntent Striping Center for E-Business Technology Copyright 2008 by CEBT Partition Table Example Data about 20~30 year old people is frequently accessed If we just store all data in a table, there can be some disks which do not contain any data about 20~30s Data are not properly distributed! Partition table Create partition w.r.t. age! Center for E-Business Technology Copyright 2008 by CEBT Partition Table: Example Center for E-Business Technology Result Data about 20~30s are distributed to all 96 disks! Copyright 2008 by CEBT Wrong IO Distrubution #1 Center for E-Business Technology Copyright 2008 by CEBT Wrong IO Distribution #2 Incrementally adding datafile to tablespace Center for E-Business Technology Chap 6. Index Architecture Copyright 2008 by CEBT Pros. and Cons. of Indexing Pros. Improving Retrieving Performance Cons. Degrading Data Manipulation Performance Insert, Delete, Storage Center for E-Business Technology Copyright 2008 by CEBT ROWID ROWID ID given to all rows in database Center for E-Business Technology Copyright 2008 by CEBT Example of ROWID Center for E-Business Technology Copyright 2008 by CEBT Usage of ROWID Index Creation Data Access SQL Tuning Center for E-Business Technology Copyright 2008 by CEBT Index B+-Tree Index Bitmap Index Reverse Key Index Function Based Index Center for E-Business Technology Copyright 2008 by CEBT B+-Tree Index B+-Tree Index Support efficient search, delete, and insert Center for E-Business Technology Copyright 2008 by CEBT B+-Tree Index Access Center for E-Business Technology Copyright 2008 by CEBT B+-Tree Index: Pros. and Cons. Pros. Balanced Range scan using linked list Good for OLTP small number row access Cons. Bad for column with low cardinality Index scanning 3%~5%of Table row: Worse than table full scan Possibility of full scan on OR operation Center for E-Business Technology Copyright 2008 by CEBT Bitmap Index Bitmap Index Good for column with low cardinality Center for E-Business Technology Copyright 2008 by CEBT Bitmap Index structure Center for E-Business Technology Copyright 2008 by CEBT Bitmap Index: Pros. and Cons. Pros. Small space Good for column with low cardinality Cons. Bad for table with frequent DML Bitmap is need to be decompressed before DML operation Need block level lock Bad for column with high cardinality Center for E-Business Technology Copyright 2008 by CEBT Reversed Key Index Reversed Key Index B+-Tree is bad for sequentially increasing index key column All insert operation need to access left block Competetion for left block occur! Reversed Key Index using reversed key to avoid above situation Center for E-Business Technology Copyright 2008 by CEBT Reversed Key Index: Pros. and Cons. Pros. Avoid competetion on left block Cons. Range scan is impossible Center for E-Business Technology Copyright 2008 by CEBT Function Based Index Function Based Index Index for transformed value Example SELECT employee_id, ceil(salary/12)*(grade-5) FROM emp WHERE mod(grade, 10)*1.2 > 10 Creating Function Based Index SQL> CREATE INDEX bonus_index on emp(mod(grade, 10)*1.2); Center for E-Business Technology Copyright 2008 by CEBT Function Based Index: Pros. and Cons. Pros. Index scan for transformed value is possible Cons. DML overhead Flexibility of index is low Only used for where cluases in predefined form Center for E-Business Technology Copyright 2008 by CEBT Random Access Random Access IO single Block Should be reduced Table Full Scan Multi Block IO Not Random Access Index Scan Random Access occurs Center for E-Business Technology Copyright 2008 by CEBT Reducing Random Access Reducing Random Access Optimizing Cluster Factor Using Optimized Index Ex)SELECT name, date, item FROM trans WHERE id = 10 AND date BETWEEN 200401 AND Center for E-Business Technology Copyright 2008 by CEBT Random Access and Index Scan Random Access and Index Scan Table Access via Index Scan is Random Access Index Scan itself is NOT Random Access Index Scan below 3% of data in table is effective Over 3%, Full table Scan is effective Center for E-Business Technology
