Storage Optimization Strategies - Tom Bascom

Report
Storage
Optimization
Strategies
Techniques for configuring your Progress
OpenEdge Database in order to minimize
IO operations
Tom Bascom, White Star Software
[email protected]
A Few Words about the Speaker
• Tom Bascom; Progress user & roaming DBA
since 1987
• President, DBAppraise, LLC
– Remote database management service for OpenEdge.
– Simplifying the job of managing and monitoring the
world’s best business applications.
– [email protected]
• VP, White Star Software, LLC
– Expert consulting services related to all aspects of
Progress and OpenEdge.
– [email protected]
2
We Will NOT be Talking about:
•
•
•
•
•
SANs
Servers
Operating systems
RAID levels
… and so forth.
3
What Do We Mean by
“Storage Optimization”?
•
•
•
•
The trade press thinks it means BIG DISKS.
Your CFO thinks it means BIG SAVINGS.
Programmers think it means BIG DATABASES.
SAN vendors think it means BIG COMMISSIONS.
• DBAs seek the best possible reliability
and performance at a reasonable cost.
4
The Foundation of
OpenEdge Storage
Optimization
5
Type 2 Storage Areas
• Type 2 storage areas are the foundation for all
advanced features of the OpenEdge database.
• Type 2 areas have cluster sizes of 8, 64 or 512.
# misc32 storage area
d “misc32_dat":11,32;8 .
• Cluster sizes of 0 or 1 are Type 1 areas.
• Data blocks in Type 2 areas contain data from
just one table.
6
Only Read What You Need
• Because data blocks in Type 2 storage areas
are “asocial”:
– Locality of reference is leveraged more strongly.
– Table-oriented utilities such as index rebuild,
binary dump and so forth know exactly which
blocks they need to read and which blocks they do
not need to read.
– DB features, such as the SQL-92 fast table scan
and fast table drop, can operate much more
effectively.
7
MYTH
• Storage optimization is just for large tables.
• Type 2 storage areas are just for large tables.
8
Truth
• Very small, yet active tables often
dominate an application’s IO profile.
• And type 2 areas are a very powerful
tool for addressing this.
9
Case Study
A system with 30,000 record reads/sec:
– The bulk of the reads were from one 10,000 record
table.
– Coincidentally, Big B was set to 10,000.
– That table was in a Type 1 area, and its records were
widely scattered.
– Moving the table to a Type 2 Area patched the
problem. Only 2% of –B was now needed for this
table!
– Performance improved dramatically.
10
Type 2 Storage Area Usage
• Always use type 2 areas…
• … for areas that contain data, indexes or LOBS.
• The schema area is a type 1 area 
11
How to Define Your
Storage Areas
12
Use the Largest DB Block Size
• Large blocks reduce IO; fewer operations are
needed to move the same amount of data.
• More data can be packed into the same space
because there is proportionally less overhead.
• Because a large block can contain more data,
it has improved odds of being a cache “hit.”
• Large blocks enable HW features to be
leveraged: especially SAN HW.
13
What about Windows?
• There are those who would say “except for
Windows.”
• (Because Windows is a 4K-oriented OS.)
• I have had good success with Windows & 8k
blocks.
• NTFS can be changed to use an 8k block…
14
Use Many (Type 2) Storage Areas
• Do NOT assign tables to areas based on
“function.”
• Instead group objects by common “technical
attributes.”
• Create distinct storage areas for:
– Each very large table
– Tables with common Rows Per Block settings
– Indexes versus data
15
Record
Fragmentation
16
Fragmentation and Scatter
• “Fragmentation” is splitting records into
multiple pieces.
• “Scatter” is the distance between (logically)
adjacent records.
17
Fragmentation and Scatter
• “Fragmentation” is splitting records into
multiple pieces.
• “Scatter” is the distance between (logically)
adjacent records.
$ proutil dbname –C dbanalys > dbname.dba
…
RECORD BLOCK SUMMARY FOR AREA "APP_FLAGS_Dat" : 95
------------------------------------------------------Record Size (B) -FragmentsScatter
Table
Records
Size Min Max Mean
Count Factor Factor
PUB.APP_FLAGS 1676180 47.9M
28
58
29 1676190
1.0
1.9
…
18
Fragmentation and Scatter
• “Fragmentation” is splitting records into
multiple pieces.
• “Scatter” is the distance between (logically)
adjacent records.
$ proutil dbname –C dbanalys > dbname.dba
…
RECORD BLOCK SUMMARY FOR AREA "APP_FLAGS_Dat" : 95
------------------------------------------------------Record Size (B) -FragmentsScatter
Table
Records
Size Min Max Mean
Count Factor Factor
PUB.APP_FLAGS 1676180 47.9M
28
58
29 1676190
1.0
1.9
…
19
Create Limit
• The minimum free space in a block
• Provides room for routine record expansion
• OE10.2B default is 150 (4k & 8k blocks)
• Must be smaller than the toss limit
• Only rarely worth adjusting
20
Toss Limit
• The minimum free space required to be on the
“RM Chain”
• Avoids looking for space in blocks that don’t have
much
•
•
•
•
Must be set higher than Create Limit.
Default is 300 (4k & 8k blocks)
Ideally should be less than average row size
Only rarely worth adjusting
21
Fragmentation, Create
& Toss
Summary
22
Create and Toss Limit Usage
Symptom
Action
Fragmentation occurs on updates to existing
records.
You anticipated one fragment, but two were
created.
Increase Create Limit
- or Decrease Rows Per Block
There is limited (or no) fragmentation, but database Decrease Create Limit
block space is being used inefficiently, and records
- or are not expected to grow beyond their original size. Increase Rows Per Block
You have many (thousands, not hundreds) of blocks Increase Toss Limit
on the RM chain with insufficient space to create
new records.
* Create and Toss limits are per area for Type 1 areas and per table for Type 2 areas.
23
Rows
Per
Block
24
Why not “One Size Fits All”?
• A universal setting such as 128 rows per block
seems simple.
• And for many situations it is adequate.
• But…
• Too large a value may lead to fragmentation and
too small to wasted space.
• It also makes advanced data analysis more
difficult.
• And it really isn’t that hard to pick good values for
RPB.
25
Set Rows Per Block Optimally
# misc32 storage area
d “misc32_dat":11,32;8 .
• Use the largest Rows Per Block that:
– Fills the block
– But does not unnecessarily fragment it
• Rough Guideline:
– Next power of 2 after BlockSize / (AvgRecSize + 20)
– Example: 8192 / (220 + 20) = 34, next power of 2 = 64
• Caveat: there are far more complex rules that can
be used and a great deal depends on the
application’s record creation & update behavior.
26
RPB
Example
27
Set Rows Per Block Optimally
Original
BlkSz
RPB
Blocks
Disk
(KB)
Waste/
Blk
%Used
Actual IO/1,000
RPB
Recs
1
4
3,015
3,015
124
86%
3
333
4
4
2,500
10,000
2,965
23%
4
250
4
8
1,250
5,000
2,075
46%
8
125
4
16
627
2,508
295
92%
16
62
4
32
596
2,384
112
97%
17
59
8
4
2,500
20,000
7,060
11%
4
250
8
16
625
5,000
4,383
44.76
16
62
8
32
313
2,504
806
90%
32
31
8
64
286
2,288
114
98%
35
29
8
128
285
2,280
109
98%
35
29
28
Set Rows Per Block Optimally
Original
Oops!
BlkSz
RPB
Blocks
Disk
(KB)
Waste/
Blk
%Used
Actual IO/1,000
RPB
Recs
1
4
3,015
3,015
124
86%
3
333
4
4
2,500
10,000
2,965
23%
4
250
4
8
1,250
5,000
2,075
46%
8
125
4
16
627
2,508
295
92%
16
62
4
32
596
2,384
112
97%
17
59
8
4
2,500
20,000
7,060
11%
4
250
8
16
625
5,000
4,383
44.76
16
62
8
32
313
2,504
806
90%
32
31
8
64
286
2,288
114
98%
35
29
8
128
285
2,280
109
98%
35
29
29
Set Rows Per Block Optimally
Original
Oops!
Suggested
BlkSz
RPB
Blocks
Disk
(KB)
Waste/
Blk
%Used
Actual IO/1,000
RPB
Recs
1
4
3,015
3,015
124
86%
3
333
4
4
2,500
10,000
2,965
23%
4
250
4
8
1,250
5,000
2,075
46%
8
125
4
16
627
2,508
295
92%
16
62
4
32
596
2,384
112
97%
17
59
8
4
2,500
20,000
7,060
11%
4
250
8
16
625
5,000
4,383
44.76
16
62
8
32
313
2,504
806
90%
32
31
8
64
286
2,288
114
98%
35
29
8
128
285
2,280
109
98%
35
29
30
Rows Per Block Caveats
• Blocks have overhead, which varies by storage area
type, block size, Progress version and by tweaking
the create and toss limits.
• Not all data behaves the same:
– Records that are created small and that grow frequently
may tend to fragment if RPB is too high.
– Record size distribution is not always Gaussian.
• If you’re unsure – round up!
31
Cluster
Size
32
Blocks Per Cluster
# misc32 storage area
d “misc32_dat":11,32;8 .
• When a type 2 area expands it will do so a
cluster at a time.
• Larger clusters are more efficient:
– Expansion occurs less frequently.
– Disk space is more likely to be contiguously
arranged.
33
Why not “One Size Fits All”?
• A universal setting such as 512 blocks per
cluster seems simple…
34
Set Cluster Size Optimally
• There is no advantage to having a cluster more
than twice the size of the table.
• Except that you need a cluster size of at least 8
to be Type 2.
• Indexes are usually much smaller than data.
• There may be dramatic differences in the size
of indexes even on the same table.
35
Different Index Sizes
$ proutil dbname –C dbanalys > dbname.dba
…
RECORD BLOCK SUMMARY FOR AREA "APP_FLAGS_Dat" : 95
------------------------------------------------------Record Size (B) -Fragments- Scatter
Table
Records Size Min Max Mean
Count Factor Factor
PUB.APP_FLAGS 1676180 47.9M 28 58
29 1676190
1.0
1.9
…
INDEX BLOCK SUMMARY FOR AREA "APP_FLAGS_Idx" : 96
------------------------------------------------------Table
Index Flds Lvls Blks
Size %Util Factor
PUB.APP_FLAGS
AppNo
183
1
3 4764
37.1M 99.9
1.0
FaxDateTime
184
2
2
45 259.8K 72.4
1.6
FaxUserNotified
185
2
2
86 450.1K 65.6
1.7
36
Logical
Scatter
37
Logical Scatter Case Study
• A process reading approximately 1,000,000
records.
• An initial run time of 2 hours.
– 139 records/sec.
• Un-optimized database.
38
Perform IO in the Optimal Order
4k DB Block Table
Table1
Table2
Index
%Sequential %Idx Used
Density
t1_idx1*
0%
100%
0.09
t1_idx2
0%
0%
0.09
t2_idx1
69%
99%
0.51
t2_idx2*
98%
1%
0.51
t2_idx3
74%
0%
0.51
39
Perform IO in the Optimal Order
4k DB Block Table
Table1
Table2
8k DB Block Table
Table1
Table2
Index
%Sequential %Idx Used
Density
t1_idx1*
0%
100%
0.09
t1_idx2
0%
0%
0.09
t2_idx1
69%
99%
0.51
t2_idx2*
98%
1%
0.51
t2_idx3
74%
0%
0.51
Index
%Sequential %Idx Used
Density
t1_idx1*
71%
100%
0.10
t1_idx2
63%
0%
0.10
t2_idx1
85%
99%
1.00
t2_idx2*
100%
1%
1.00
t2_idx3
83%
0%
0.99
40
Perform IO in the Optimal Order
4k DB Block Table
Table1
Table2
8k DB Block Table
Table1
Oops!
Table2
Index
%Sequential %Idx Used
Density
t1_idx1*
0%
100%
0.09
t1_idx2
0%
0%
0.09
t2_idx1
69%
99%
0.51
t2_idx2*
98%
1%
0.51
t2_idx3
74%
0%
0.51
Index
%Sequential %Idx Used
Density
t1_idx1*
71%
100%
0.10
t1_idx2
63%
0%
0.10
t2_idx1
85%
99%
1.00
t2_idx2*
100%
1%
1.00
t2_idx3
83%
0%
0.99
41
Logical Scatter Case Study
Block
Size
Hit
Ratio
%Sequential Block References
IO Ops
Time
4k
95
69
319,719
19,208
120
4k
98
69
320,149
9,816
60
4k
99
69
320,350
6,416
40
8k
95
85
160,026
9,417
55
8k
98
85
159,805
4,746
30
8k
99
85
160,008
3,192
20
The process was improved from an initial runtime of roughly 2 hours
(top line, in red) to approximately 20 minutes (bottom) by moving from
4k blocks and 69% sequential access at a hit ratio of approximately 95%
to 8k blocks, 85% sequential access and a hit ratio of 99%.
42
Avoid IO,
But If You Must…
43
… in Big B You Should Trust!
Layer
Time
# of
Recs
# of Ops
Cost per
Op
Relative
Progress to –B
0.96
100,000
203,473
0.000005
1
-B to FS Cache
10.24
100,000
26,711
0.000383
75
FS Cache to SAN
5.93
100,000
26,711
0.000222
45
-B to SAN Cache*
11.17
100,000
26,711
0.000605
120
SAN Cache to Disk
200.35 100,000
26,711
0.007500
1500
-B to Disk
211.52 100,000
26,711
0.007919
1585
* Used concurrent IO to eliminate FS cache
44
New Feature!
• 10.2B supports a new feature called “Alternate Buffer
Pool.”
• This can be used to isolate specified database objects
(tables and/or indexes).
• The alternate buffer pool has its own distinct –B.
• If the database objects are smaller than –B, there is
no need for the LRU algorithm.
• This can result in major performance improvements
for small, but very active, tables.
• proutil dbname –C enableB2 areaname
• Table and Index level selection is for Type 2 only!
45
Conclusion
• Always use Type 2 storage areas.
• Define your storage areas based on technical
attributes of the data.
• Static analysis isn’t enough – you need to also
monitor runtime behaviors.
• White Star Software has a great deal of experience
in optimizing storage. We would be happy to
engage with any customer that would like our
help!
46
Thank You!
47
Questions?
48

similar documents