1 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
3 <html xmlns="http://www.w3.org/1999/xhtml"><head><meta http-equiv="Content-Type" content="text/html; charset=UTF-8" /><title>bitmap_allocator</title><meta name="generator" content="DocBook XSL Stylesheets V1.75.2" /><meta name="keywords" content=" ISO C++ , allocator " /><meta name="keywords" content=" ISO C++ , library " /><link rel="home" href="../spine.html" title="The GNU C++ Library Documentation" /><link rel="up" href="ext_allocators.html" title="Chapter 20. Allocators" /><link rel="prev" href="ext_allocators.html" title="Chapter 20. Allocators" /><link rel="next" href="ext_containers.html" title="Chapter 21. Containers" /></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">bitmap_allocator</th></tr><tr><td width="20%" align="left"><a accesskey="p" href="ext_allocators.html">Prev</a> </td><th width="60%" align="center">Chapter 20. Allocators</th><td width="20%" align="right"> <a accesskey="n" href="ext_containers.html">Next</a></td></tr></table><hr /></div><div class="sect1" title="bitmap_allocator"><div class="titlepage"><div><div><h2 class="title" style="clear: both"><a id="manual.ext.allocator.bitmap"></a>bitmap_allocator</h2></div></div></div><p>
4 </p><div class="sect2" title="Design"><div class="titlepage"><div><div><h3 class="title"><a id="allocator.bitmap.design"></a>Design</h3></div></div></div><p>
5 As this name suggests, this allocator uses a bit-map to keep track
6 of the used and unused memory locations for it's book-keeping
9 This allocator will make use of 1 single bit to keep track of
10 whether it has been allocated or not. A bit 1 indicates free,
11 while 0 indicates allocated. This has been done so that you can
12 easily check a collection of bits for a free block. This kind of
13 Bitmapped strategy works best for single object allocations, and
14 with the STL type parameterized allocators, we do not need to
15 choose any size for the block which will be represented by a
16 single bit. This will be the size of the parameter around which
17 the allocator has been parameterized. Thus, close to optimal
18 performance will result. Hence, this should be used for node based
19 containers which call the allocate function with an argument of 1.
21 The bitmapped allocator's internal pool is exponentially growing.
22 Meaning that internally, the blocks acquired from the Free List
23 Store will double every time the bitmapped allocator runs out of
26 The macro <code class="literal">__GTHREADS</code> decides whether to use
27 Mutex Protection around every allocation/deallocation. The state
28 of the macro is picked up automatically from the gthr abstraction
30 </p></div><div class="sect2" title="Implementation"><div class="titlepage"><div><div><h3 class="title"><a id="allocator.bitmap.impl"></a>Implementation</h3></div></div></div><div class="sect3" title="Free List Store"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.free_list_store"></a>Free List Store</h4></div></div></div><p>
31 The Free List Store (referred to as FLS for the remaining part of this
32 document) is the Global memory pool that is shared by all instances of
33 the bitmapped allocator instantiated for any type. This maintains a
34 sorted order of all free memory blocks given back to it by the
35 bitmapped allocator, and is also responsible for giving memory to the
36 bitmapped allocator when it asks for more.
38 Internally, there is a Free List threshold which indicates the
39 Maximum number of free lists that the FLS can hold internally
40 (cache). Currently, this value is set at 64. So, if there are
41 more than 64 free lists coming in, then some of them will be given
42 back to the OS using operator delete so that at any given time the
43 Free List's size does not exceed 64 entries. This is done because
44 a Binary Search is used to locate an entry in a free list when a
45 request for memory comes along. Thus, the run-time complexity of
46 the search would go up given an increasing size, for 64 entries
47 however, lg(64) == 6 comparisons are enough to locate the correct
48 free list if it exists.
50 Suppose the free list size has reached it's threshold, then the
51 largest block from among those in the list and the new block will
52 be selected and given back to the OS. This is done because it
53 reduces external fragmentation, and allows the OS to use the
54 larger blocks later in an orderly fashion, possibly merging them
55 later. Also, on some systems, large blocks are obtained via calls
56 to mmap, so giving them back to free system resources becomes most
59 The function _S_should_i_give decides the policy that determines
60 whether the current block of memory should be given to the
61 allocator for the request that it has made. That's because we may
62 not always have exact fits for the memory size that the allocator
63 requests. We do this mainly to prevent external fragmentation at
64 the cost of a little internal fragmentation. Now, the value of
65 this internal fragmentation has to be decided by this function. I
66 can see 3 possibilities right now. Please add more as and when you
67 find better strategies.
68 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>Equal size check. Return true only when the 2 blocks are of equal
69 size.</p></li><li class="listitem"><p>Difference Threshold: Return true only when the _block_size is
70 greater than or equal to the _required_size, and if the _BS is > _RS
71 by a difference of less than some THRESHOLD value, then return true,
72 else return false. </p></li><li class="listitem"><p>Percentage Threshold. Return true only when the _block_size is
73 greater than or equal to the _required_size, and if the _BS is > _RS
74 by a percentage of less than some THRESHOLD value, then return true,
75 else return false.</p></li></ol></div><p>
76 Currently, (3) is being used with a value of 36% Maximum wastage per
78 </p></div><div class="sect3" title="Super Block"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.super_block"></a>Super Block</h4></div></div></div><p>
79 A super block is the block of memory acquired from the FLS from
80 which the bitmap allocator carves out memory for single objects
81 and satisfies the user's requests. These super blocks come in
82 sizes that are powers of 2 and multiples of 32
83 (_Bits_Per_Block). Yes both at the same time! That's because the
84 next super block acquired will be 2 times the previous one, and
85 also all super blocks have to be multiples of the _Bits_Per_Block
88 How does it interact with the free list store?
90 The super block is contained in the FLS, and the FLS is responsible for
91 getting / returning Super Bocks to and from the OS using operator new
92 as defined by the C++ standard.
93 </p></div><div class="sect3" title="Super Block Data Layout"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.super_block_data"></a>Super Block Data Layout</h4></div></div></div><p>
94 Each Super Block will be of some size that is a multiple of the
95 number of Bits Per Block. Typically, this value is chosen as
96 Bits_Per_Byte x sizeof(size_t). On an x86 system, this gives the
97 figure 8 x 4 = 32. Thus, each Super Block will be of size 32
98 x Some_Value. This Some_Value is sizeof(value_type). For now, let
99 it be called 'K'. Thus, finally, Super Block size is 32 x K bytes.
101 This value of 32 has been chosen because each size_t has 32-bits
102 and Maximum use of these can be made with such a figure.
104 Consider a block of size 64 ints. In memory, it would look like this:
105 (assume a 32-bit system where, size_t is a 32-bit entity).
106 </p><div class="table"><a id="id545173"></a><p class="title"><b>Table 20.1. Bitmap Allocator Memory Map</b></p><div class="table-contents"><table summary="Bitmap Allocator Memory Map" border="1"><colgroup><col align="left" /><col align="left" /><col align="left" /><col align="left" /><col align="left" /></colgroup><tbody><tr><td align="left">268</td><td align="left">0</td><td align="left">4294967295</td><td align="left">4294967295</td><td align="left">Data -> Space for 64 ints</td></tr></tbody></table></div></div><br class="table-break" /><p>
107 The first Column(268) represents the size of the Block in bytes as
108 seen by the Bitmap Allocator. Internally, a global free list is
109 used to keep track of the free blocks used and given back by the
110 bitmap allocator. It is this Free List Store that is responsible
111 for writing and managing this information. Actually the number of
112 bytes allocated in this case would be: 4 + 4 + (4x2) + (64x4) =
113 272 bytes, but the first 4 bytes are an addition by the Free List
114 Store, so the Bitmap Allocator sees only 268 bytes. These first 4
115 bytes about which the bitmapped allocator is not aware hold the
118 What do the remaining values represent?</p><p>
119 The 2nd 4 in the expression is the sizeof(size_t) because the
120 Bitmapped Allocator maintains a used count for each Super Block,
121 which is initially set to 0 (as indicated in the diagram). This is
122 incremented every time a block is removed from this super block
123 (allocated), and decremented whenever it is given back. So, when
124 the used count falls to 0, the whole super block will be given
125 back to the Free List Store.
127 The value 4294967295 represents the integer corresponding to the bit
128 representation of all bits set: 11111111111111111111111111111111.
130 The 3rd 4x2 is size of the bitmap itself, which is the size of 32-bits
132 which is 8-bytes, or 2 x sizeof(size_t).
133 </p></div><div class="sect3" title="Maximum Wasted Percentage"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.max_wasted"></a>Maximum Wasted Percentage</h4></div></div></div><p>
134 This has nothing to do with the algorithm per-se,
135 only with some vales that must be chosen correctly to ensure that the
136 allocator performs well in a real word scenario, and maintains a good
137 balance between the memory consumption and the allocation/deallocation
140 The formula for calculating the maximum wastage as a percentage:
142 (32 x k + 1) / (2 x (32 x k + 1 + 32 x c)) x 100.
144 where k is the constant overhead per node (e.g., for list, it is
145 8 bytes, and for map it is 12 bytes) and c is the size of the
146 base type on which the map/list is instantiated. Thus, suppose the
147 type1 is int and type2 is double, they are related by the relation
148 sizeof(double) == 2*sizeof(int). Thus, all types must have this
149 double size relation for this formula to work properly.
151 Plugging-in: For List: k = 8 and c = 4 (int and double), we get:
154 For map/multimap: k = 12, and c = 4 (int and double), we get: 37.524%
156 Thus, knowing these values, and based on the sizeof(value_type), we may
157 create a function that returns the Max_Wastage_Percentage for us to use.
158 </p></div><div class="sect3" title="allocate"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.allocate"></a><code class="function">allocate</code></h4></div></div></div><p>
159 The allocate function is specialized for single object allocation
160 ONLY. Thus, ONLY if n == 1, will the bitmap_allocator's
161 specialized algorithm be used. Otherwise, the request is satisfied
162 directly by calling operator new.
164 Suppose n == 1, then the allocator does the following:
165 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
166 Checks to see whether a free block exists somewhere in a region
167 of memory close to the last satisfied request. If so, then that
168 block is marked as allocated in the bit map and given to the
169 user. If not, then (2) is executed.
170 </p></li><li class="listitem"><p>
171 Is there a free block anywhere after the current block right
172 up to the end of the memory that we have? If so, that block is
173 found, and the same procedure is applied as above, and
174 returned to the user. If not, then (3) is executed.
175 </p></li><li class="listitem"><p>
176 Is there any block in whatever region of memory that we own
177 free? This is done by checking
178 </p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
179 The use count for each super block, and if that fails then
180 </p></li><li class="listitem"><p>
181 The individual bit-maps for each super block.
182 </p></li></ul></div><p>
183 Note: Here we are never touching any of the memory that the
184 user will be given, and we are confining all memory accesses
185 to a small region of memory! This helps reduce cache
186 misses. If this succeeds then we apply the same procedure on
187 that bit-map as (1), and return that block of memory to the
188 user. However, if this process fails, then we resort to (4).
189 </p></li><li class="listitem"><p>
190 This process involves Refilling the internal exponentially
191 growing memory pool. The said effect is achieved by calling
192 _S_refill_pool which does the following:
193 </p><div class="itemizedlist"><ul class="itemizedlist" type="disc"><li class="listitem"><p>
194 Gets more memory from the Global Free List of the Required
196 </p></li><li class="listitem"><p>
197 Adjusts the size for the next call to itself.
198 </p></li><li class="listitem"><p>
199 Writes the appropriate headers in the bit-maps.
200 </p></li><li class="listitem"><p>
201 Sets the use count for that super-block just allocated to 0
203 </p></li><li class="listitem"><p>
204 All of the above accounts to maintaining the basic invariant
205 for the allocator. If the invariant is maintained, we are
206 sure that all is well. Now, the same process is applied on
207 the newly acquired free blocks, which are dispatched
209 </p></li></ul></div></li></ol></div><p>
210 Thus, you can clearly see that the allocate function is nothing but a
211 combination of the next-fit and first-fit algorithm optimized ONLY for
212 single object allocations.
213 </p></div><div class="sect3" title="deallocate"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.deallocate"></a><code class="function">deallocate</code></h4></div></div></div><p>
214 The deallocate function again is specialized for single objects ONLY.
215 For all n belonging to > 1, the operator delete is called without
216 further ado, and the deallocate function returns.
218 However for n == 1, a series of steps are performed:
219 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>
220 We first need to locate that super-block which holds the memory
221 location given to us by the user. For that purpose, we maintain
222 a static variable _S_last_dealloc_index, which holds the index
223 into the vector of block pairs which indicates the index of the
224 last super-block from which memory was freed. We use this
225 strategy in the hope that the user will deallocate memory in a
226 region close to what he/she deallocated the last time around. If
227 the check for belongs_to succeeds, then we determine the bit-map
228 for the given pointer, and locate the index into that bit-map,
229 and mark that bit as free by setting it.
230 </p></li><li class="listitem"><p>
231 If the _S_last_dealloc_index does not point to the memory block
232 that we're looking for, then we do a linear search on the block
233 stored in the vector of Block Pairs. This vector in code is
234 called _S_mem_blocks. When the corresponding super-block is
235 found, we apply the same procedure as we did for (1) to mark the
236 block as free in the bit-map.
237 </p></li></ol></div><p>
238 Now, whenever a block is freed, the use count of that particular
239 super block goes down by 1. When this use count hits 0, we remove
240 that super block from the list of all valid super blocks stored in
241 the vector. While doing this, we also make sure that the basic
242 invariant is maintained by making sure that _S_last_request and
243 _S_last_dealloc_index point to valid locations within the vector.
244 </p></div><div class="sect3" title="Questions"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.questions"></a>Questions</h4></div></div></div><div class="sect4" title="1"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.1"></a>1</h5></div></div></div><p>
245 Q1) The "Data Layout" section is
246 cryptic. I have no idea of what you are trying to say. Layout of what?
247 The free-list? Each bitmap? The Super Block?
249 The layout of a Super Block of a given
250 size. In the example, a super block of size 32 x 1 is taken. The
251 general formula for calculating the size of a super block is
252 32 x sizeof(value_type) x 2^n, where n ranges from 0 to 32 for 32-bit
254 </p></div><div class="sect4" title="2"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.2"></a>2</h5></div></div></div><p>
255 And since I just mentioned the
256 term `each bitmap', what in the world is meant by it? What does each
257 bitmap manage? How does it relate to the super block? Is the Super
258 Block a bitmap as well?
260 Each bitmap is part of a Super Block which is made up of 3 parts
261 as I have mentioned earlier. Re-iterating, 1. The use count,
262 2. The bit-map for that Super Block. 3. The actual memory that
263 will be eventually given to the user. Each bitmap is a multiple
264 of 32 in size. If there are 32 x (2^3) blocks of single objects
265 to be given, there will be '32 x (2^3)' bits present. Each 32
266 bits managing the allocated / free status for 32 blocks. Since
267 each size_t contains 32-bits, one size_t can manage up to 32
268 blocks' status. Each bit-map is made up of a number of size_t,
269 whose exact number for a super-block of a given size I have just
271 </p></div><div class="sect4" title="3"><div class="titlepage"><div><div><h5 class="title"><a id="bitmap.impl.question.3"></a>3</h5></div></div></div><p>
272 How do the allocate and deallocate functions work in regard to
275 The allocate and deallocate functions manipulate the bitmaps and
276 have nothing to do with the memory that is given to the user. As
277 I have earlier mentioned, a 1 in the bitmap's bit field
278 indicates free, while a 0 indicates allocated. This lets us
279 check 32 bits at a time to check whether there is at lease one
280 free block in those 32 blocks by testing for equality with
281 (0). Now, the allocate function will given a memory block find
282 the corresponding bit in the bitmap, and will reset it (i.e.,
283 make it re-set (0)). And when the deallocate function is called,
284 it will again set that bit after locating it to indicate that
285 that particular block corresponding to this bit in the bit-map
286 is not being used by anyone, and may be used to satisfy future
289 e.g.: Consider a bit-map of 64-bits as represented below:
290 1111111111111111111111111111111111111111111111111111111111111111
292 Now, when the first request for allocation of a single object
293 comes along, the first block in address order is returned. And
294 since the bit-maps in the reverse order to that of the address
295 order, the last bit (LSB if the bit-map is considered as a
296 binary word of 64-bits) is re-set to 0.
298 The bit-map now looks like this:
299 1111111111111111111111111111111111111111111111111111111111111110
300 </p></div></div><div class="sect3" title="Locality"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.locality"></a>Locality</h4></div></div></div><p>
301 Another issue would be whether to keep the all bitmaps in a
302 separate area in memory, or to keep them near the actual blocks
303 that will be given out or allocated for the client. After some
304 testing, I've decided to keep these bitmaps close to the actual
305 blocks. This will help in 2 ways.
306 </p><div class="orderedlist"><ol class="orderedlist" type="1"><li class="listitem"><p>Constant time access for the bitmap themselves, since no kind of
307 look up will be needed to find the correct bitmap list or it's
308 equivalent.</p></li><li class="listitem"><p>And also this would preserve the cache as far as possible.</p></li></ol></div><p>
309 So in effect, this kind of an allocator might prove beneficial from a
310 purely cache point of view. But this allocator has been made to try and
311 roll out the defects of the node_allocator, wherein the nodes get
312 skewed about in memory, if they are not returned in the exact reverse
313 order or in the same order in which they were allocated. Also, the
314 new_allocator's book keeping overhead is too much for small objects and
315 single object allocations, though it preserves the locality of blocks
316 very well when they are returned back to the allocator.
317 </p></div><div class="sect3" title="Overhead and Grow Policy"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.grow_policy"></a>Overhead and Grow Policy</h4></div></div></div><p>
318 Expected overhead per block would be 1 bit in memory. Also, once
319 the address of the free list has been found, the cost for
320 allocation/deallocation would be negligible, and is supposed to be
321 constant time. For these very reasons, it is very important to
322 minimize the linear time costs, which include finding a free list
323 with a free block while allocating, and finding the corresponding
324 free list for a block while deallocating. Therefore, I have
325 decided that the growth of the internal pool for this allocator
326 will be exponential as compared to linear for
327 node_allocator. There, linear time works well, because we are
328 mainly concerned with speed of allocation/deallocation and memory
329 consumption, whereas here, the allocation/deallocation part does
330 have some linear/logarithmic complexity components in it. Thus, to
331 try and minimize them would be a good thing to do at the cost of a
332 little bit of memory.
334 Another thing to be noted is the pool size will double every time
335 the internal pool gets exhausted, and all the free blocks have
336 been given away. The initial size of the pool would be
337 sizeof(size_t) x 8 which is the number of bits in an integer,
338 which can fit exactly in a CPU register. Hence, the term given is
339 exponential growth of the internal pool.
340 </p></div></div></div><div class="navfooter"><hr /><table width="100%" summary="Navigation footer"><tr><td width="40%" align="left"><a accesskey="p" href="ext_allocators.html">Prev</a> </td><td width="20%" align="center"><a accesskey="u" href="ext_allocators.html">Up</a></td><td width="40%" align="right"> <a accesskey="n" href="ext_containers.html">Next</a></td></tr><tr><td width="40%" align="left" valign="top">Chapter 20. Allocators </td><td width="20%" align="center"><a accesskey="h" href="../spine.html">Home</a></td><td width="40%" align="right" valign="top"> Chapter 21. Containers</td></tr></table></div></body></html>
OSDN Git Service
