�ݺ�ߣ

RDBMS vs. Other Data Stores forScalabilityramki.g@directi.comTechTalk 2009, IIIT Hyderabad

ScalabilityIncrease Resources  Increase Performance (Linearly)Performance?Latency, Capacity, ThroughputVertical Scalability (Scaling Up)Divide the functionalityHorizontal Scalability (Scaling Out)Divide the data

Relational DatabaseTable, Row, ColumnSet, Item, Property

Relational TheorySelection: SELECTFilter: WHEREJoin: JOIN, LEFT JOIN,RIGHT JOINCorrelation: SELECT a FROM A WHERE A.b IN (SELECT b FROM B WHERE b.a > a)

Relational TheoryAggregationSet OperatorsUnion, Intersection, MinusGroup ByMAX, MIN, SUM, AVG

Transactions: AtomicityTransaction LevelEntire Logical operations is a transactionMultiple statementsStatement levelEach statement is either successful or not, no partial successMultiple recordsRecord LevelAll modifications to a record are successful or not

Transactions: ConsistencyIntegrity ConstraintsReferential Integrity

Transactions: Isolation LevelsSerializableA definite order of mutations/transactions is possible to arrive to state B from state ARepeatable ReadAny data read by a transaction will remain so till transaction is completeNon Repeatable Read aka Read CommittedTwo reads within a transaction may give different resultsDirty ReadA transaction might read data which might then be rolledback

RDBMS LuxuriesMultiple IndexesAuto Increments/SequencesTriggers

Scalability in RDBMSReplicationRead Replication (Master-Slave)Read Write Replication (Master-Master)ClusterDistributed TransactionTwo-phase commits

Scalability ImpedimentsPerformanceSub-Queries/Correlation, Joins, Aggregates, Referential Integrity constraintsBasic GuaranteeConsistencyAvailability

CAP?Conjecture: Distributed systems cannot ensure all three of the following properties at onceConsistency The client perceives that a set of operations has occurred all at once.Availability Every operation must terminate in an intended response.Partition tolerance Operations will complete, even if individual components are unavailable.

ACID to BASEBasically Available - system seems to work all the timeSoft State - it doesn't have to be consistent all the timeEventually Consistent - becomes consistent at some later time

BASE: An ExampleBEGIN TransactionINSERT INTO ORDER( oid, timestamp, customer)FOREACH item IN itemList INSERT INTO ORDER_ITEM ( oid, item.id, item.quantity, item.unitprice) //UPDATE INVENTORY SET quantity=quantity- item.quantityWHERE item = item.idCOMMITEND TransactionAssume Each statement is queued for execution You will get COMMIT success

Alternate ImplementationsBigTable – Google – CPHbase – Apache – CP HyperTable – Community - CPDynamo – Amazon – APSimpleDB– Amazon - APVoldemort – LinkedIn – APCassandra – Facebook– APMemcacheDB - community – CP/AP

Data ModelsKey/Value Pairs Dynamo, MemcacheDB, VoldemortRow-ColumnBigTable, Casandra, SimpleDB, Hypertable, Hbase

Programming Models// Open the tableTable *T = OpenOrDie("/bigtable/web/webtable");// Write a new anchor and delete an old anchorRowMutation r1(T, "com.cnn.www");r1.Set("anchor:www.c-span.org", "CNN");r1.Delete("anchor:www.abc.com");Operation op;Apply(&op, &r1);

BigTable: Consistent yet Infinitely ScalableSingle MasterB+ tree based data distribution

BigTable: TransactionsEnities and Entity GroupsInvoiceInvoice ItemDelivery Note

Dynamo: Highly available and Infinitely ScalableConsistent HashingPeer to Peer DistributedGossip based member discovery

RDBMS or Other?Nature of BusinessMaturity of the ProductCost of AdoptionMaturity of the alternative Datastores

�ݺ�ߣ

Scalability: Rdbms Vs Other Data Stores

More Related Content

Scalability: Rdbms Vs Other Data Stores

Editor's Notes