Concurrency control in XML databases

Ali Abbasi

Concurrency in XML Bases

XML : standard format of data exchange on the internet

XML docs is stored in databases Concurrency control is an important function of a DB Traditional CC protocols are not tailored to XML

database systems New protocols designed based on different access

methods : DOM, XPath, ...

XPath

XPath models an XML document as a tree of nodes. XPath expressions are used to indicate the requested

nodes in the XML tree An Xpath exp : /S1,j /S2,j /S3,j /.../Sm,j

Si,j : Axis::NodeTest[Predicate]

Non-conflicting conditions A transaction T is a sequence of pairs of operations Oj(x)

and location path Lj Operations are : pass-by, read, write, insert, delete

For any x,y, <Pi(x),Pj(y)>, <Pi(x),Rj(y)> and <Ri(x),Rj(y)>,

For any x,y, <Ri(x),Ij(y)>, <Wi(x),Ij(y)>

For any y, if there exists a location step S in location path L in Ti such that x is not a member of R(S) and RI(S), <Pi(x),Wj(y)>, <Pi(x),Ij(y)>, <Pi(x),Dj(y)>

Conflicting conditions <Ri(x),Wj(x)>, <Ri(x),Dj(x)> <Wi(x),Wj(x)>, <Wi(x),Dj(x)> <Ii(x),Ij(x)>, <Ii(x),Dj(x)>, <Di(x),Dj(x)> For any x, if there exists a location step S in location

path L such that x is a member of R(S) or RI(S), <Pi(x),Wj(x)>, <Pi(x),Ij(x)>, <Pi(x),Dj(x)>

XLP XLP is a lock-based protocol with five lock modes,

denoted by P-, R-, W-, I- and D-locks XLP is 2PL Nodes in the M-set of Si,j (M(Si,j) and MI(Si,j)) are all locked

by P-locks before performing the Node-Test and Predicate

Granularity Rules : (1) Lock granularity of P-, R-, I-, or W-locks on a node is

only the node itself. (2) Lock granularity of D-locks on a node includes the whole

subtree rooted at the node.

XLP Compatibility Rule.

A particular type of lock on location step Si can be granted as long as the compatibility matrix is respected.

Release Rules R-, W-, I- or D-locks on Nd(Lj) can only be released in the

shrinking phase of a transaction. P-locks on nodes in the set are released only in the

shrinking phase; Other P-locks on are released after location step Si,j

finishes

Properties

XLP ensures conflict serializability. XLP outperforms 2PL owing to the earlier release of P-

locks and lower lock conflicts according to its lock compatibility matrix.

XLP fully supports most XPath’s axis operations

phantom problem may exist in some schedules under XLP

LWMGL A lock-based CC protocol XML documents have a hierarchical structure Two hierarchical locking protocol: TL, MGL None of them a is an appropriate solution LWMGL is a hybrid of MGL and TL It’s goal is preventing phantoms and providing high

concurrency

pseudo-conflict:

T1 retrieves all the /description elements

T2 inserts price element

Pseudo-conflict

phantom

T1 retrieves all the elements in /Books/Book

T2 inserts Discount

XIET data model The element table schema E has a set of element

records E = {indexing group, elements, rowID} The data table schema D has a set of data records

D = {rowID, TextString} indexing group I = {G,S,D}

LWMGL Protocol

Find the target element by using GSD indices Acquire S_locks for reading each element where the

target element traverses subelements Acquire X_locks on the leaf element in order to update

the value in the data table Acquire IX_lock on the parent for inserting a node for

prohibiting phantom

LWMGL Protocol

Acquire IX locks from the root element to target element in order to delete an element in the element table so as to prohibit parent element deletion from other transactions during deletion.

Follow the order in locking from the parent element of the subelements to their child. In an unlocking fashion, follow it in the opposite direction, i.e., from the child element to the parent of the subelements.

References Kuen-Fang Jea, Shih-Ying Chen, A High concurrency XPath-based

locking protocol for XML Databases, In: Elsevier Science ,2005.

Yonggoo Choi, Songchum Moon, Lightweight multigranularity locking for transaction management in XML database systems, In: Elsevier Science ,2005.

Grabs, T., Bo¨hm, K., Schek, H.-J., XMLTM: efficient transaction management for XML documents. In: Proceedings of the ACM International Conference on Information and Knowledge Management, CIKM 2002

L :/child::a/descendant::d[position()=2] M(S1,1)=R (S1,1)= {a} , MI(S1,1) and RI(S1,1) are empty MI(S2,1)={b2, b3, c1, c2} , M(S2,1)={d1, d2, d3} RI(S2,1)={b2, c1} , R(S2,1)={d2}.