Lock Tuning

AOBD 2007/08 H. Galhardas Concurrency Control Goals Performance goals  Reduce blocking One transaction waits for another to release its locks  Avoid deadlocks Transactions are waiting for each other to release their locks Correctness goals  Serializability: each transaction appears to execute in isolation  The programmer ensures that serial execution is correct. Trade-off between correctness and performance

AOBD 2007/08 H. Galhardas Ideal Transaction Acquires few locks and favors shared locks over exclusive locks  Reduce the number of conflicts -- conflicts are due to exclusive locks Acquires locks with fine granularity  Reduce the scope of each conflict Holds locks for a short time  Reduce waiting

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Bottlenecks  Using system features to circumvent bottlenecks Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Bottlenecks  Using system features to circumvent bottlenecks Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Facilities for long reads In some systems (e.g., Oracle), read-only queries hold no locks yet appear to execute serializably Method used (snaphot isolation in Oracle):  Re-create an old version of any data item that is changed after the read query begins  Gives the effect that a read-only transaction R reads the database as the database appeared when R began

AOBD 2007/08 H. Galhardas Snapshot isolation T1 T2 T3 TIME R(Y) returns 1 R(Z) returns 0 R(X) returns 0 W(Y:=1) W(X:=2, Z:=3) X=Y=Z=0 Each transaction executes against the version of the data items that was committed when the transaction started:  No locks for read  Locks for writes  Costs space (old copy of data must be kept) Almost serializable level, when extended to read/write transactions  T1: x:=y  T2: y:= x  Initially x=3 and y =17  Serial execution: x,y=17 or x,y=3  Snapshot isolation: x=17, y=3 if both transactions start at the same time.

AOBD 2007/08 H. Galhardas Recommendation (by default) To use snapshot isolation for read-only transactions, but ensure that read operations hold locks for transactions that perform updates.

AOBD 2007/08 H. Galhardas Eliminate unnecessary locking Locking is not necessary in two situations: 1. Only one transaction runs at a time  E.g. When loading the DB 2. When all transactions are read-only  E.g., decision support queries on archival DB  Reduce overhead by suppressing the acquisition of locks  Memory consumption for lock control blocks + processor time to process lock requests

AOBD 2007/08 H. Galhardas A If object not found in hash table, it is unlocked Lock info for A A... H Lock Table

AOBD 2007/08 H. Galhardas Locking in SQL Server 7 syslockinfo dbidobjid lock granularity lock ownerlock waiter lock mode 1117RID X PAG IX TAB IX RID S LO1 LO2, LO3 LW2 LW3 spid 10 Lock – 32 bytes Lock owner block – 32 bytes Lock waiter block – 32 bytes LW1, LW4

AOBD 2007/08 H. Galhardas Locking in Oracle 8i Data page row Interested transaction list (fixed array - INITRANS – MAXTRANS) T1 T1 lock Enqueue resource structure (fixed array – default 4 entries per transaction) T1 Enqueued locks array T2 lock Enqueue wait (time out ~ 3sec) Process T3 lock Deadlock detection H

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Bottlenecks  Using system features to circumvent bottlenecks Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Isolation Levels Read Uncomitted (No lost update)  Exclusive locks for write operations are held for the duration of the transactions  Lock for writes until commit time. No locks for reads  Reads may read dirty data and will not be repeatable. Read Committed (No inconsistent retrieval)  Lock for writes until commit time.  Shared locks are released as soon as the read operation terminates.  Reads may access only committed data, still not repeatable. Repeatable Read (no unrepeatable reads)  Strict two phase locking: lock for writes and reads until commit time. Serializable (no phantoms)  Table locking or index locking to avoid phantoms

AOBD 2007/08 H. Galhardas Sacrificing Isolation for Performance A transaction that holds locks during a screen interaction is an invitation to bottlenecks  Airline Reservation 1. Retrieve list of seats available 2. Talk with customer regarding availability 3. Secure seat Single transaction is intolerable, because each customer would hold lock on seats available. Keep user interaction outside a transactional context Correctness is sacrificed: ask for a seat but then find it’s unavailable. Tolerable in this application.

AOBD 2007/08 H. Galhardas Value of Serializability -- Data Settings: accounts( number, branchnum, balance); create clustered index c on accounts(number);  rows  Cold buffer; same buffer size on all systems.  Row level locking  Isolation level (SERIALIZABLE or READ COMMITTED)  SQL Server 7, DB2 v7.1 and Oracle 8i on Windows 2000  Dual Xeon (550MHz,512Kb), 1Gb RAM, Internal RAID controller from Adaptec (80Mb), 4x18Gb drives (10000RPM), Windows 2000.

AOBD 2007/08 H. Galhardas Value of Serializability -- transactions Concurrent Transactions:  T1: summation query [1 thread] select sum(balance) from accounts;  T2: swap balance between two account numbers (in order of scan to avoid deadlocks) [N threads] valX:=select balance from accounts where number=X; valY:=select balance from accounts where number=Y; update accounts set balance=valX where number=Y; update accounts set balance=valY where number=X;

AOBD 2007/08 H. Galhardas 18 Value of Serializability -- results With SQL Server and DB2 the scan returns incorrect answers if the read committed isolation level is used (default setting) With Oracle correct answers are returned (snapshot isolation), but beware of swapping

AOBD 2007/08 H. Galhardas 19 Cost of Serializability Because the update conflicts with the scan, correct answers are obtained at the cost of decreased concurrency and thus decreased throughput.

AOBD 2007/08 H. Galhardas Recommendation Begin with the highest degree of isolation (serializable)  If a transaction either suffers extensive deadlock or causes significant blocking, consider weakening the degree of isolation  Be aware that the answers may not be correct.

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Bottlenecks  Using system features to circumvent bottlenecks Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Select appropriate granularity of locking RDBMSs support different granularities of locking: row or record-level (default), page- level, table-level Record-level locking is the best for online transaction environments where each transaction accesses only a few records spread on different pages

AOBD 2007/08 H. Galhardas Locking Overhead -- data Settings: accounts( number, branchnum, balance); create clustered index c on accounts(number);  rows  Cold buffer  SQL Server 7, DB2 v7.1 and Oracle 8i on Windows 2000  No lock escalation on Oracle; Parameter set so that there is no lock escalation on DB2; no control on SQL Server.  Dual Xeon (550MHz,512Kb), 1Gb RAM, Internal RAID controller from Adaptec (80Mb), 4x18Gb drives (10000RPM), Windows 2000.

AOBD 2007/08 H. Galhardas Locking Overhead -- transactions Nb Concurrent Transactions:  Update [ updates] update accounts set balance = Val;  Insert [ inserts], e.g. typical one: insert into accounts values(664366,72255, );

AOBD 2007/08 H. Galhardas 25 Locking Overhead Row locking is barely more expensive than table locking because recovery overhead is higher than locking overhead  Exception is updates on DB2 where table locking is distinctly less expensive than row locking.

AOBD 2007/08 H. Galhardas Why choosing table-level locking? Used to avoid blocking long transactions Can be used to avoid deadlocks Reduce locking overhead in the case that there is no concurrency Recommendation: Long transactions (accesses almost all pages of the table) should use table locks mostly to avoid deadlocks Short transactions should use record locks to enhance concurrency

AOBD 2007/08 H. Galhardas How the user can control the granule size? Explicit control of the granularity  Within a transaction (Oracle, DB2): statement within a transaction explicitly requests a table-level lock  Across transactions (SQL Server): command defines the lock granularity for a table or group of tables. All transactions accessing these tables use the same lock granularity Setting the escalation point (SQL Server or DB2): systems acquire the default (finest) granularity lock until the nb of acquired locks exceeds some threshold set by the DBA. At that point, the next coarser granularity lock will be acquired.  The rule of thumb is to set the threshold high enough so that escalation never takes place in an online environment of relatively short transactions Size of the lock table: If the administrator selects a small lock table size, the system will be forced to escalate the lock granularity even if all transactions are short.

AOBD 2007/08 H. Galhardas Avoid DDL statements Data Definition Language: language used to access and manipulate information about table names, column widths,.... that is metadata stored in the system catalog Metadata must be accessed by every transaction that:  Performs a compilation  Adds or removes a table or index  Changes an attribute description Can easily become a bottleneck Recommendation: avoid updates to the system catalog during heavy system activity

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Circumventing bottlenecks (due to hot spots) Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Circumventing bottlenecks Hot spot: piece of data that is accessed by many transactions and is updated by some.  Causes bottlenecks because each updating transaction must complete before any other transaction can obtain a lock on the hot data item Techniques to circumvent:  Use partitioning to eliminate it  Access the hot spot as late as possible in the transaction  Use special database management facilities

AOBD 2007/08 H. Galhardas 31 Example of hot spot: Sequential key generation Consider an application in which one needs a sequential number to act as a key in a table, e.g. invoice numbers for bills.  Transactions that insert data associate a unique identifier with each new data item  Concurrent transactions must coordinate to avoid giving the same identifier to different data items Ad hoc approach: a separate table holding the last invoice number. Fetch and update that number on each insert transaction.  Problem: This table becomes a bottleneck, because a transaction will release its lock on the counter only when the transaction commits Counter approach: use facility such as Sequence (Oracle)/Identity(SQL Server), that enables transactions to hold a latch on the counter.

AOBD 2007/08 H. Galhardas Latches and Locks Locks are used for concurrency control  Requests for locks are queued Priority queue  Lock data structure Locking mode, lock granularity, transaction id. Lock table Latches are used for mutual exclusion  Requests for latch succeeds or fails  Released immediately after access rather than being held till the end of the transaction  Does not allow shared mode  Does not provide support for queuing waiting threads  Single location in memory Test and set for latch manipulation

AOBD 2007/08 H. Galhardas Counter Facility -- data Settings:  default isolation level: READ COMMITTED; Empty tables  Dual Xeon (550MHz,512Kb), 1Gb RAM, Internal RAID controller from Adaptec (80Mb), 4x18Gb drives (10000RPM), Windows accounts( number, branchnum, balance); create clustered index c on accounts(number); counter ( nextkey ); insert into counter values (1);

AOBD 2007/08 H. Galhardas Counter Facility -- transactions Concurrent Transactions:  System [ inserts, N threads] SQL Server 7 (uses Identity column) insert into accounts values (94496,2789); Oracle 8i insert into accounts values (seq.nextval,94496,2789);  Ad-hoc [ inserts, N threads] begin transaction NextKey:=select nextkey from counter; update counter set nextkey = NextKey+1; commit transaction begin transaction insert into accounts values(NextKey,?,?); commit transaction

AOBD 2007/08 H. Galhardas Avoid Bottlenecks: Counters System generated counter (system) much better than a counter managed as an attribute value within a table (ad hoc). The Oracle counter can become a bottleneck if every update is logged to disk, but caching many counter numbers is possible. Counters may miss ids.

AOBD 2007/08 H. Galhardas Lock Tuning Use special facilities for long reads Eliminate unnecessary locking Weaken isolation levels when application allows  Relaxing correctness to improve performance Select appropriate granularity of locking Avoid DDL statements Bottlenecks  Using system features to circumvent bottlenecks Transaction Chopping  Rewriting applications to obtain best locking performance

AOBD 2007/08 H. Galhardas Transaction chopping How long should a transaction be? Transaction length influence performance:  The more locks a transaction requests, the more likely it will have to wait for some other transaction to release a lock.  The longer a transaction T executes, the more time another transaction will have to wait if it is blocked by T In situations in which blocking occurs, short transactions are better than longer ones  Sometimes, we can “chop” transactions in shorter ones without losing isolation guarantees

AOBD 2007/08 H. Galhardas Example: Simple Purchases Purchase item I for price P 1. If cash < P then roll back transaction (constraint) 2. Inventory(I) := inventory(I)+P 3. Cash := Cash – P Two purchase transaction P1 and P2  P1 has item I for price 50  P2 has item I for price 75  Cash is 100

AOBD 2007/08 H. Galhardas Example: Simple Purchases If as one transaction then one of P1, P2 rolls back. If 1, 2, 3 as three distinct transactions:  P1 checks that cash > 50. It is.  P2 checks that cash > 75. It is.  P1 completes. Cash = 50.  P2 completes. Cash = - 25.

AOBD 2007/08 H. Galhardas Example: Simple Purchases Orthodox solution  Make whole program a single transaction Cash becomes a bottleneck! Chopping solution  Find a way to rearrange and then chop up the transactions without violating serializable isolation level.

AOBD 2007/08 H. Galhardas Example: Simple Purchases Chopping solution: 1. If Cash < P then roll back. Cash := Cash – P. 2. Inventory(I) := inventory(I) + P Chopping execution:  P11: 100 > 50. Cash := 50.  P21: 75 > 50. Rollback.  P12: inventory := inventory + 50.

AOBD 2007/08 H. Galhardas Transaction Chopping Execution rules:  When pieces execute, they follow the partial order defined by the transactions.  If a piece is aborted because of a conflict, it will be resubmitted until it commits  If a piece is aborted because of an abort, no other pieces for that transaction will execute.

AOBD 2007/08 H. Galhardas When is transaction chopping possible? Two important questions must be answered:  Will the transactions that are concurrent with T cause T to produce an inconsistent state or to observe an inconsistent value if T is broken up?  Will the transactions that are concurrent with T be made inconsistent if T is broken up? Rule of thumb: Suppose T accesses data X and Y, but any other transaction T’ accesses at most one of X or Y and nothing else. Then, T can be divided into two transactions, one of which accesses X and the other accesses Y Caution: Adding a new transaction to a set of existing transactions may invalidate all previously established choppings.

AOBD 2007/08 H. Galhardas Transaction Chopping Let T1, T2, …, Tn be a set of transactions. A chopping partitions each Ti into pieces ci1, ci2, …, cik. A chopping of T is rollback-safe if (a)T does not contain any abort commands or (b) if the abort commands are in the first piece.

AOBD 2007/08 H. Galhardas Correct Chopping Chopping graph (variation of the serialization graph):  Nodes are pieces  Edges: C-edges: C stands for conflict. There is a C-edge between two pieces from different transactions if they contain operations that access the same data item and one operation is a write. S-edges: S stands for siblings. There is an S-edge between two pieces, iff they come from the same transaction. A chopping graph contains an S-C cycle if it contains a cycle that includes at least one S-edge and one C-edge.

AOBD 2007/08 H. Galhardas Correct Chopping A chopping is correct if it is rollback safe and its chopping graph contains no SC-cycle. T1: r(x) w(x) r(y) w(y) T2: r(x) w(x) T3: r(y) w(y) T1 T2T3 T11: r(x) w(x) T12: r(y) w(y) T11T12 T3T2 S C C T11: r(x) T12: w(x) T13: r(y) w(y) T12T13 T3T2 S C C T11 S C CORRECT NOT CORRECT

AOBD 2007/08 H. Galhardas Chopping Example T1: RW(A) RW (B) T2: RW(D) RW(B) T3: RW(E) RW(C) T4: R(F) T5: R(E) T6: R(A) R(F) R(D) R(B) R(E) R(G) R(C)

AOBD 2007/08 H. Galhardas Chopping Example T1T2 T4 T5 T3 T62T61 T61: R(A) R(F) R(D) R(B) T62: R(E) R(G) R(C) CC C C C S

AOBD 2007/08 H. Galhardas Finest Chopping A private chopping of transaction Ti, denoted private(Ti) is a set of pieces {ci1, ci2, …, cik} such that:  {ci1, ci2, …, cik} is a rollback safe chopping  There is no SC-cycle in the graph whose nodes are {T1, …, Ti-1, ci1, ci2, …, cik, Ti+1, … Tn} The chopping consisting of {private(T1), private(T2), …, private(T2)} is rollback-safe and has no SC- cycles.

AOBD 2007/08 H. Galhardas Finest Chopping In: T, {T1,.. Tn-1} Initialization  If there are abort commands then p_1 := all writes of T (and all non swappable reads)that may occur before or concurrently with any abort command in T else p_1 := first database access  P := {x | x is a database operation not in p_1}  P := P {p_1}

AOBD 2007/08 H. Galhardas Finest Chopping Merging pieces  Construct the connected components of the graph induced by C edges alone on all transactions {T1, …, Tn-1} and on the pieces in P.  Update P based on the following rule: If p_j and p_k are in the same connected component and j < k, then  add the accesses from p_k to p_j  delete p_k from P