索引组织表（IOT）和堆组织表的性能对比测试

索引组织表（index organized table，IOT）就是存储在一个索引结构中的表，存储在堆中的表是无组织的，IOT中的数据则按照主键存储和排序。
使用堆组织表，Oracle会为表和表主键索引分别留出空间，而IOT则不存在主键的空间开销，因为索引就是数据，数据就是索引，两者已经合而为一了。
为了比较IOT和堆组织表的效率，首先创建两张表：
create table iot
( username varchar2(30),
document_name varchar2(30),
other_data char(200),
constraint iot_pk
primary key (username,document_name)
)
organization index
/

create table heap
( username varchar2(30),
document_name varchar2(30),
other_data char(200),
constraint heap_pk
primary key (username,document_name)
)
/

先测试插入效率，使用runstats工具：
exec runstats_pkg.rs_start

begin
for i in 1 .. 100
loop
for x in ( select username
from all_users )
loop
insert into heap
(username,document_name,other_data)
values
( x.username, x.username || '_' || i, 'x' );
end loop;
end loop;
commit;
end;
/

exec runstats_pkg.rs_middle

begin
for i in 1 .. 100
loop
for x in ( select username
from all_users )
loop
insert into iot
(username,document_name,other_data)
values
( x.username, x.username || '_' || i, 'x' );
end loop;
end loop;
commit;
end;
/

exec runstats_pkg.rs_stop(100)

测试结果如下：
Run1 ran in 57 hsecs
Run2 ran in 39 hsecs
run 1 ran in 146.15% of the time

Name Run1 Run2 Diff
LATCH.enqueue hash chains 351 464 113
LATCH.session allocation 1,065 1,180 115
STAT...workarea memory allocat -68 53 121
STAT...redo ordering marks 76 273 197
LATCH.cache buffers lru chain 170 381 211
STAT...commit cleanouts succes 134 347 213
STAT...commit cleanouts 138 351 213
STAT...free buffer requested 167 381 214
LATCH.simulator hash latch 938 721 -217
STAT...cleanout - number of kt 75 322 247
STAT...active txn count during 75 322 247
LATCH.simulator lru latch 908 658 -250
STAT...calls to kcmgas 124 420 296
STAT...immediate (CURRENT) blo 48 345 297
STAT...consistent gets - exami 123 483 360
LATCH.object queue header oper 326 720 394
STAT...consistent gets from ca 5,964 6,417 453
STAT...consistent gets 5,964 6,417 453
LATCH.undo global data 232 860 628
STAT...IMU undo allocation siz 52 880 828
STAT...session logical reads 14,041 13,015 -1,026
STAT...db block gets from cach 8,077 6,598 -1,479
STAT...db block gets 8,077 6,598 -1,479
STAT...redo entries 5,397 3,905 -1,492
STAT...db block changes 10,604 7,548 -3,056
LATCH.cache buffers chains 46,495 40,265 -6,230
STAT...IMU Redo allocation siz 0 11,072 11,072
STAT...session uga memory 96 65,464 65,368
STAT...session pga memory 196,608 65,536 -131,072
STAT...session uga memory max 385,404 0 -385,404
STAT...session pga memory max 589,824 0 -589,824
STAT...undo change vector size 494,448 1,130,136 635,688
STAT...redo size 1,981,952 2,993,508 1,011,556

Run1 latches total versus runs -- difference and pct
Run1 Run2 Diff Pct
63,861 58,553 -5,308 109.07%

从测试结果可以看出，使用IOT插入更快，使用的闩更少。

接下来测试查询的效率：
alter session set sql_trace=true;

declare
type array is table of varchar2(100);
l_array1 array;
l_array2 array;
l_array3 array;
begin
for i in 1 .. 10
loop
for x in (select username from all_users)
loop
for y in ( select * from heap single_row
where username = x.username )
loop
null;
end loop;
for y in ( select * from iot single_row
where username = x.username )
loop
null;
end loop;
select * bulk collect
into l_array1, l_array2, l_array2
from heap bulk_collect
where username = x.username;
select * bulk collect
into l_array1, l_array2, l_array2
from iot bulk_collect
where username = x.username;
end loop;
end loop;
end;
/

Tkprof工具运行的结果如下：
SELECT *
FROM
HEAP SINGLE_ROW WHERE USERNAME = :B1

call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 240 0.00 0.04 0 2 0 0
Fetch 480 0.04 0.06 0 19190 0 24000
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 721 0.04 0.11 0 19192 0 24000

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: 5 (recursive depth: 1)

Rows Row Source Operation
------- ---------------------------------------------------
24000 TABLE ACCESS FULL HEAP (cr=19190 pr=0 pw=0 time=130794 us)

********************************************************************************

SELECT *
FROM
IOT SINGLE_ROW WHERE USERNAME = :B1

call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 240 0.00 0.00 0 2 0 0
Fetch 480 0.03 0.03 0 1672 0 24000
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 721 0.03 0.03 0 1674 0 24000

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: 5 (recursive depth: 1)

Rows Row Source Operation
------- ---------------------------------------------------
24000 INDEX RANGE SCAN IOT_PK (cr=1672 pr=0 pw=0 time=102198 us)(object id 53339)

********************************************************************************

对于单行获取：Oracle为heap表选择了全表扫描，而对于IOT则使用了索引范围扫描。逻辑I/O的次数相差很多倍，对于heap表，使用了19190个逻辑I/O，而IOT只使用了1672个逻辑I/O。

对于批量获取：
SELECT *
FROM
HEAP BULK_COLLECT WHERE USERNAME = :B1

call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 240 0.00 0.00 0 2 0 0
Fetch 240 0.10 0.11 0 18960 0 24000
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 481 0.10 0.12 0 18962 0 24000

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: 5 (recursive depth: 1)

Rows Row Source Operation
------- ---------------------------------------------------
24000 TABLE ACCESS FULL HEAP (cr=18960 pr=0 pw=0 time=128099 us)

********************************************************************************

SELECT *
FROM
IOT BULK_COLLECT WHERE USERNAME = :B1

call count cpu elapsed disk query current rows
------- ------ -------- ---------- ---------- ---------- ---------- ----------
Parse 1 0.00 0.00 0 0 0 0
Execute 240 0.00 0.00 0 2 0 0
Fetch 240 0.07 0.08 0 1540 0 24000
------- ------ -------- ---------- ---------- ---------- ---------- ----------
total 481 0.07 0.09 0 1542 0 24000

Misses in library cache during parse: 1
Optimizer mode: ALL_ROWS
Parsing user id: 5 (recursive depth: 1)

Rows Row Source Operation
------- ---------------------------------------------------
24000 INDEX RANGE SCAN IOT_PK (cr=1540 pr=0 pw=0 time=95013 us)(object id 53339)

********************************************************************************

对于批量获取：heap表用了18960个逻辑I/O，IOT用了1540个了逻辑I/O。

这个测试结果显示heap表没有使用索引，我原以为没有使用索引是因为没有对表进行分析导致的，后来经测试，发现不是，即使对表进行了分析，并收集了统计信息：
analyze table heap compute statistics
for table
for all indexes
for all indexed columns;

begin
dbms_stats.gather_table_stats( user, 'heap', cascade=>true );
end;
/

Oracle依然会对heap表选择全表扫描 。

测试结论：不管是插入，单行查询还是批量获取，IOT都要优于heap表。

最后我们看一下，空间使用情况：
analyze table heap compute statistics
for table
for all indexes
for all indexed columns;

analyze index HEAP_PK validate structure;
select height,btree_space,used_space,opt_cmpr_count,opt_cmpr_pctsave from index_stats;

HEIGHT BTREE_SPACE USED_SPACE OPT_CMPR_COUNT OPT_CMPR_PCTSAVE
---------- ----------- ---------- -------------- ----------------
2 128032 68713 1 26

analyze table iot compute statistics
for table
for all indexes
for all indexed columns;

analyze index IOT_PK validate structure;
select height,btree_space,used_space,opt_cmpr_count,opt_cmpr_pctsave from index_stats;

HEIGHT BTREE_SPACE USED_SPACE OPT_CMPR_COUNT OPT_CMPR_PCTSAVE
---------- ----------- ---------- -------------- ----------------
2 864032 545892 1 3

IOT因为索引包含数据，所以空间一定比heap的索引空间占用大。

select segment_name,sum(bytes) from dba_extents
where segment_name ='HEAP' group by segment_name；

SEGMENT_NAME SUM(BYTES)
------------------------------ ----------
HEAP 655360

select segment_name,sum(bytes) from dba_extents
where segment_name ='HEAP_PK' group by segment_name

SEGMENT_NAME SUM(BYTES)
------------------------------ ----------
HEAP_PK 196608

select segment_name,sum(bytes) from dba_extents
where segment_name ='IOT_PK' group by segment_name

SEGMENT_NAME SUM(BYTES)
------------------------------ ----------
IOT_PK 917504

从结果来看，还是IOT占用空间大一点，不过可能情况没有这么简单，这里只是计算使用的区段的大小总和，但这些区段不一定会被完全使用，因此这样算得的结果不一定准确。
按照《Oracle高效设计》的说法，IOT会节约空间。

参考文献：
《Oracle9i&10g编程艺术》
《Oracle高效设计》