1DBM::Deep::Internals(3pUms)er Contributed Perl DocumentatDiBoMn::Deep::Internals(3pm)
2
6 DBM::Deep::Internals - Out of date documentation on DBM::Deep internals
7
9 This document is out-of-date. It describes an intermediate file format
10 used during the development from 0.983 to 1.0000. It will be rewritten
11 soon.
12 So far, the description of the header format has been updated.
14
16 This is a document describing the internal workings of DBM::Deep. It is
17 not necessary to read this document if you only intend to be a user.
18 This document is intended for people who either want a deeper
19 understanding of specifics of how DBM::Deep works or who wish to help
20 program DBM::Deep.
21
23 DBM::Deep is broken up into five classes in three inheritance
24 hierarchies.
25 • DBM::Deep is the parent of DBM::Deep::Array and DBM::Deep::Hash.
27 These classes form the immediate interface to the outside world.
28 They are the classes that provide the TIE mechanisms as well as the
29 OO methods.
30 • DBM::Deep::Engine is the layer that deals with the mechanics of
32 reading and writing to the file. This is where the logic of the
33 file layout is handled.
34 • DBM::Deep::File is the layer that deals with the physical file. As
36 a singleton that every other object has a reference to, it also
37 provides a place to handle datastructure-wide items, such as
38 transactions.
39
41 This describes the 1.0003 and 2.0000 formats, which internally are
42 numbered 3 and 4, respectively. The internal numbers are used in this
43 section. These two formats are almost identical.
44 DBM::Deep uses a tagged file layout. Every section has a tag, a size,
46 and then the data.
47 File header
49 The file header consists of two parts. The first part is a fixed length
50 of 13 bytes:
51 DPDB h VVVV SSSS
53 \ / | \ \
54 \/ '---. \ '--- size of the second part of the header
55 file \ '--- version
56 signature tag
57 • File Signature
59 The first four bytes are 'DPDB' in network byte order, signifying
61 that this is a DBM::Deep file.
62 • File tag
64 A literal ASCII 'h', indicating that this is the header. The file
66 used by versions prior to 1.00 had a different fifth byte, allowing
67 the difference to be determined.
68 • Version
70 This is four bytes containing the file version. This lets the file
72 format change over time.
73 It is packed in network order, so version 4 is stored as
75 "\0\0\0\cD".
76 • Header size
78 The size of the second part of the header, in bytes. This number is
80 also packed in network order.
81 The second part of the header is as follows:
83 S B S T T(TTTTTTTTT...) (SS SS SS SS ...) (continued...)
85 | | | | \ |
86 | | | '----------. \ staleness counters
87 | | '--------. \ txn bitfield
88 | '------. \ number of transactions
89 byte size \ data sector size
90 max buckets
91 (continuation...)
93 BB(BBBBBB) DD(DDDDDD) II(IIIIII)
94 | | |
95 | free data |
96 free blist free index
97 • Constants
99 These are the file-wide constants that determine how the file is
101 laid out. They can only be set upon file creation.
102 The byte size is the number of bytes used to point to an offset
104 elsewhere in the file. This corresponds to the "pack_size" option.
105 This and the next three values are stored as packed 8-bit integers
106 (chars), so 2 is represented by "\cB".
107 "max_buckets" and "data_sector_size" are documented in the main
109 DBM::Deep man page. The number stored is actually one less than
110 what is passed to the constructor, to allow for a range of 1-256.
111 The number of transactions corresponds to the "num_txns" value
113 passed to the constructor.
114 • Transaction information
116 The transaction bitfield consists of one bit for every available
118 transaction ID. It is therefore anywhere from 1 byte to 32 bytes
119 long.
120 The staleness counters each take two bytes (packed 32-bit
122 integers), one for each transaction, not including the so-called
123 HEAD (the main transaction that all processes share before calling
124 "begin_work"). So these take up 0 to 508 bytes.
125 Staleness is explained in DBM::Deep::Engine.
127 • Freespace information
129 Pointers into the first free sectors of the various sector sizes
131 (Index, Bucketlist, and Data) are stored here. These are called
132 chains internally, as each free sector points to the next one.
133 The number of bytes is determined by the byte size, ranging from 2
135 to 8.
136 Index
138 The Index parts can be tagged either as Hash, Array, or Index. The
139 latter is if there was a reindexing due to a bucketlist growing too
140 large. The others are the root index for their respective datatypes.
141 The index consists of a tag, a size, and then 256 sections containing
142 file locations. Each section corresponds to each value representable in
143 a byte.
144 The index is used as follows - whenever a hashed key is being looked
146 up, the first byte is used to determine which location to go to from
147 the root index. Then, if that's also an index, the second byte is
148 used, and so forth until a bucketlist is found.
149 Bucketlist
151 This is the part that contains the link to the data section. A
152 bucketlist defaults to being 16 buckets long (modifiable by the
153 max_buckets parameter used when creating a new file). Each bucket
154 contains an MD5 and a location of the appropriate key section.
155 Key area
157 This is the part that handles transactional awareness. There are
158 max_buckets sections. Each section contains the location to the data
159 section, a transaction ID, and whether that transaction considers this
160 key to be deleted or not.
161 Data area
163 This is the part that actual stores the key, value, and class (if
164 appropriate). The layout is:
165 • tag
167 • length of the value
169 • the actual value
171 • keylength
173 • the actual key
175 • a byte indicating if this value has a classname
177 • the classname (if one is there)
179 The key is stored after the value because the value is requested more
181 often than the key.
182
184 DBM::Deep is written completely in Perl. It also is a multi-process DBM
185 that uses the datafile as a method of synchronizing between multiple
186 processes. This is unlike most RDBMSes like MySQL and Oracle.
187 Furthermore, unlike all RDBMSes, DBM::Deep stores both the data and the
188 structure of that data as it would appear in a Perl program.
189 CPU
191 DBM::Deep attempts to be CPU-light. As it stores all the data on disk,
192 DBM::Deep is I/O-bound, not CPU-bound.
193 RAM
195 DBM::Deep uses extremely little RAM relative to the amount of data you
196 can access. You can iterate through a million keys (using each())
197 without increasing your memory usage at all.
198 DISK
200 DBM::Deep is I/O-bound, pure and simple. The faster your disk, the
201 faster DBM::Deep will be. Currently, when performing "my $x =
202 $db->{foo}", there are a minimum of 4 seeks and 1332 + N bytes read
203 (where N is the length of your data). (All values assume a medium
204 filesize.) The actions taken are:
205 1 Lock the file
207 1 Perform a stat() to determine if the inode has changed
208 1 Go to the primary index for the $db (1 seek)
209 1 Read the tag/size of the primary index (5 bytes)
210 1 Read the body of the primary index (1024 bytes)
211 1 Go to the bucketlist for this MD5 (1 seek)
212 1 Read the tag/size of the bucketlist (5 bytes)
213 1 Read the body of the bucketlist (144 bytes)
214 1 Go to the keys location for this MD5 (1 seek)
215 1 Read the tag/size of the keys section (5 bytes)
216 1 Read the body of the keys location (144 bytes)
217 1 Go to the data section that corresponds to this transaction ID. (1
218 seek)
219 1 Read the tag/size of the data section (5 bytes)
220 1 Read the value for this data (N bytes)
221 1 Unlock the file
222 Every additional level of indexing (if there are enough keys) requires
224 an additional seek and the reading of 1029 additional bytes. If the
225 value is blessed, an additional 1 seek and 9 + M bytes are read (where
226 M is the length of the classname).
227 Arrays are (currently) even worse because they're considered "funny
229 hashes" with the length stored as just another key. This means that if
230 you do any sort of lookup with a negative index, this entire process is
231 performed twice - once for the length and once for the value.
232
234 SPEED
235 Obviously, DBM::Deep isn't going to be as fast as some C-based DBMs,
236 such as the almighty BerkeleyDB. But it makes up for it in features
237 like true multi-level hash/array support, and cross-platform FTPable
238 files. Even so, DBM::Deep is still pretty fast, and the speed stays
239 fairly consistent, even with huge databases. Here is some test data:
240 Adding 1,000,000 keys to new DB file...
242 At 100 keys, avg. speed is 2,703 keys/sec
244 At 200 keys, avg. speed is 2,642 keys/sec
245 At 300 keys, avg. speed is 2,598 keys/sec
246 At 400 keys, avg. speed is 2,578 keys/sec
247 At 500 keys, avg. speed is 2,722 keys/sec
248 At 600 keys, avg. speed is 2,628 keys/sec
249 At 700 keys, avg. speed is 2,700 keys/sec
250 At 800 keys, avg. speed is 2,607 keys/sec
251 At 900 keys, avg. speed is 2,190 keys/sec
252 At 1,000 keys, avg. speed is 2,570 keys/sec
253 At 2,000 keys, avg. speed is 2,417 keys/sec
254 At 3,000 keys, avg. speed is 1,982 keys/sec
255 At 4,000 keys, avg. speed is 1,568 keys/sec
256 At 5,000 keys, avg. speed is 1,533 keys/sec
257 At 6,000 keys, avg. speed is 1,787 keys/sec
258 At 7,000 keys, avg. speed is 1,977 keys/sec
259 At 8,000 keys, avg. speed is 2,028 keys/sec
260 At 9,000 keys, avg. speed is 2,077 keys/sec
261 At 10,000 keys, avg. speed is 2,031 keys/sec
262 At 20,000 keys, avg. speed is 1,970 keys/sec
263 At 30,000 keys, avg. speed is 2,050 keys/sec
264 At 40,000 keys, avg. speed is 2,073 keys/sec
265 At 50,000 keys, avg. speed is 1,973 keys/sec
266 At 60,000 keys, avg. speed is 1,914 keys/sec
267 At 70,000 keys, avg. speed is 2,091 keys/sec
268 At 80,000 keys, avg. speed is 2,103 keys/sec
269 At 90,000 keys, avg. speed is 1,886 keys/sec
270 At 100,000 keys, avg. speed is 1,970 keys/sec
271 At 200,000 keys, avg. speed is 2,053 keys/sec
272 At 300,000 keys, avg. speed is 1,697 keys/sec
273 At 400,000 keys, avg. speed is 1,838 keys/sec
274 At 500,000 keys, avg. speed is 1,941 keys/sec
275 At 600,000 keys, avg. speed is 1,930 keys/sec
276 At 700,000 keys, avg. speed is 1,735 keys/sec
277 At 800,000 keys, avg. speed is 1,795 keys/sec
278 At 900,000 keys, avg. speed is 1,221 keys/sec
279 At 1,000,000 keys, avg. speed is 1,077 keys/sec
280 This test was performed on a PowerMac G4 1gHz running Mac OS X 10.3.2 &
282 Perl 5.8.1, with an 80GB Ultra ATA/100 HD spinning at 7200RPM. The
283 hash keys and values were between 6 - 12 chars in length. The DB file
284 ended up at 210MB. Run time was 12 min 3 sec.
285 MEMORY USAGE
287 One of the great things about DBM::Deep is that it uses very little
288 memory. Even with huge databases (1,000,000+ keys) you will not see
289 much increased memory on your process. DBM::Deep relies solely on the
290 filesystem for storing and fetching data. Here is output from top
291 before even opening a database handle:
292 PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
294 22831 root 11 0 2716 2716 1296 R 0.0 0.2 0:07 perl
295 Basically the process is taking 2,716K of memory. And here is the same
297 process after storing and fetching 1,000,000 keys:
298 PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND
300 22831 root 14 0 2772 2772 1328 R 0.0 0.2 13:32 perl
301 Notice the memory usage increased by only 56K. Test was performed on a
303 700mHz x86 box running Linux RedHat 7.2 & Perl 5.6.1.
304perl v5.38.0 2023-09-03 DBM::Deep::Internals(3pm)