15.7 BUFFER MANAGEMENT
15.7.1 Buffer Management Architecture
• The buffer manager controls main memory
directly, as in many relational DBMS’s
• The buffer manager allocates buffers in virtual
memory, allowing the operating system to decide
which buffers are actually in main memory at any
time and which are in the “swap space” on disk
that the operating system manages. Many “mainmemory” DBMS’s and “object-oriented” DBMS’s
operate this way.
15.7.1 Buffer Management Architecture
(cont.)
• Buffer manager controls main memory
• If request exceed available space, it has to select a
buffer to empty by returning its contents to disk.
• If buffered block has not been changed, buffer manager
erased the block from main memory.
• Write the buffered block to disk if changed.
• Buffer manager controls virtual memory
• Buffer manager has the option to allocate more
buffers than can fit in main memory.
• Thrashing problem occurs, many blocks are moved in and
out of the disk’s swap space
15.7.1 Buffer Management Architecture
(cont.)
• Normally, the number of buffers is a
parameter set when the DBMS is initialized.
• We will assumed that there is a fixed-sized buffer
pool.
15.7.2 Buffer Management Strategies
• The critical choice that the buffer manager must
make is what block to throw out of the buff pool
when the buffer is needed for newly requested
block. There are some common strategies, such
as in operating systems. These include:
•
•
•
•
LRU
FIFO
Clock
System Control
15.7.2 Buffer Management Strategies
(cont.)
• LRU – Least recently used
• Throw out block that has not been read or written for
the longest time.
• FIFO – First in first out
• The oldest block in the buffer is emptied for the new
block
• Clock – Let blocks in buffer have second chance
to live.
• Clock wisely, buffer manager loops through blocks in
buffer and looks for block that hasn’t been accessed
for two rounds. Then, replace this block with new
block.
15.7.2 Buffer Management Strategies
(cont.)
0
1
0
1
the buffer with
a 0 flag will
be replaced
0
0
1
1
Start point to
search a 0 flag
The flag will
be set to 0
By next time the hand
reaches it, if the content of
this buffer is not accessed,
i.e. flag=0, this buffer will
be replaced.
That’s “Second Chance”.
15.7.2 Buffer Management Strategies
(cont.)
• System Control
• Strict used of LRU, FIFO, or Clock can have
disadvantage
• Recall from sect 13.6.5, there are sometimes technical
reasons that a block can not be moved to disk without first
modifying certain blocks that point to it. These blocks are
called “pinned”. (b-tree index)
• FIFO pin root block of B-tree
• Similarly, for one-pass hash-join, the query processor may
“pin” the blocks of the smaller relation in order to assure
that it will remain in main memory during the entire time.
• Buffer manager has to modify its buffer-replacement
strategy to avoid expelling pinned blocks
15.7.3 The Relationship Between Physical
Operator Selection and Buffer
Management
• The buffer manager may not be willing or able to guarantee
the availability of M buffers when the query is executed
• I.e multiple queries
• Two questions arise
• Can the algorithm adapt to the changes in the value of M, the
number of main-memory buffers available?
• When the expected M buffers are not available, and some
blocks that are expected to be in the memory have actually
been moved to disk by the buffer manager, how does the
buffer-replacement strategy used by the buffer manager impact
the number of additional I/O’s that must be performed?
15.7.3 The Relationship Between Physical
Operator Selection and Buffer
Management (cont.)
FOR each chunk of M(available)-1 blocks of S DO BEGIN
read these blocks into main-memory buffers;
organize their tuples into a search structure whose
search key is the common attributes of R and S;
FOR each block b of R DO BEGIN
read b into main memory;
FOR each tuple t of b DO BEGIN
find the tuples of S in main memory that
join with t ;
output the join of t with each of these tuples;
END ;
END ;
END ;
I1 = 5 blocks
4 blocks of S
1 block of R
I2 = 11 blocks
10 blocks of S
1 block of R
15.7.3 The Relationship Between Physical
Operator Selection and Buffer
Management (cont.)
FOR each chunk of M(available)-1 blocks of S DO BEGIN
read these blocks into main-memory buffers;
organize their tuples into a search structure whose
search key is the common attributes of R and S;
alternate the order of blocks R (rocking)
FOR each block b of R DO BEGIN
read b into main memory;
FOR each tuple t of b DO BEGIN
find the tuples of S in main memory that
join with t ;
output the join of t with each of these tuples;
END ;
END ;
END ;
LRU improvement –
alternate the order of
blocks R
Save I/O for first or last
blocks