1 <?xml version="1.0" encoding="UTF-8" standalone="no"?>
2 <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
3 <html xmlns="http://www.w3.org/1999/xhtml"><head><title>Implementation</title><meta name="generator" content="DocBook XSL-NS Stylesheets V1.76.1"/><meta name="keywords" content=" ISO C++ , allocator "/><meta name="keywords" content=" ISO C++ , library "/><meta name="keywords" content=" ISO C++ , runtime , library "/><link rel="home" href="../index.html" title="The GNU C++ Library"/><link rel="up" href="bitmap_allocator.html" title="Chapter 21. The bitmap_allocator"/><link rel="prev" href="bitmap_allocator.html" title="Chapter 21. The bitmap_allocator"/><link rel="next" href="policy_data_structures.html" title="Chapter 22. Policy-Based Data Structures"/></head><body><div class="navheader"><table width="100%" summary="Navigation header"><tr><th colspan="3" align="center">Implementation</th></tr><tr><td align="left"><a accesskey="p" href="bitmap_allocator.html">Prev</a> </td><th width="60%" align="center">Chapter 21. The bitmap_allocator</th><td align="right"> <a accesskey="n" href="policy_data_structures.html">Next</a></td></tr></table><hr/></div><div class="section" title="Implementation"><div class="titlepage"><div><div><h2 class="title"><a id="allocator.bitmap.impl"/>Implementation</h2></div></div></div><div class="section" title="Free List Store"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.free_list_store"/>Free List Store</h3></div></div></div><p>
4 The Free List Store (referred to as FLS for the remaining part of this
5 document) is the Global memory pool that is shared by all instances of
6 the bitmapped allocator instantiated for any type. This maintains a
7 sorted order of all free memory blocks given back to it by the
8 bitmapped allocator, and is also responsible for giving memory to the
9 bitmapped allocator when it asks for more.
11 Internally, there is a Free List threshold which indicates the
12 Maximum number of free lists that the FLS can hold internally
13 (cache). Currently, this value is set at 64. So, if there are
14 more than 64 free lists coming in, then some of them will be given
15 back to the OS using operator delete so that at any given time the
16 Free List's size does not exceed 64 entries. This is done because
17 a Binary Search is used to locate an entry in a free list when a
18 request for memory comes along. Thus, the run-time complexity of
19 the search would go up given an increasing size, for 64 entries
20 however, lg(64) == 6 comparisons are enough to locate the correct
21 free list if it exists.
23 Suppose the free list size has reached its threshold, then the
24 largest block from among those in the list and the new block will
25 be selected and given back to the OS. This is done because it
26 reduces external fragmentation, and allows the OS to use the
27 larger blocks later in an orderly fashion, possibly merging them
28 later. Also, on some systems, large blocks are obtained via calls
29 to mmap, so giving them back to free system resources becomes most
32 The function _S_should_i_give decides the policy that determines
33 whether the current block of memory should be given to the
34 allocator for the request that it has made. That's because we may
35 not always have exact fits for the memory size that the allocator
36 requests. We do this mainly to prevent external fragmentation at
37 the cost of a little internal fragmentation. Now, the value of
38 this internal fragmentation has to be decided by this function. I
39 can see 3 possibilities right now. Please add more as and when you
40 find better strategies.
41 </p><div class="orderedlist"><ol class="orderedlist"><li class="listitem"><p>Equal size check. Return true only when the 2 blocks are of equal
42 size.</p></li><li class="listitem"><p>Difference Threshold: Return true only when the _block_size is
43 greater than or equal to the _required_size, and if the _BS is > _RS
44 by a difference of less than some THRESHOLD value, then return true,
45 else return false. </p></li><li class="listitem"><p>Percentage Threshold. Return true only when the _block_size is
46 greater than or equal to the _required_size, and if the _BS is > _RS
47 by a percentage of less than some THRESHOLD value, then return true,
48 else return false.</p></li></ol></div><p>
49 Currently, (3) is being used with a value of 36% Maximum wastage per
51 </p></div><div class="section" title="Super Block"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.super_block"/>Super Block</h3></div></div></div><p>
52 A super block is the block of memory acquired from the FLS from
53 which the bitmap allocator carves out memory for single objects
54 and satisfies the user's requests. These super blocks come in
55 sizes that are powers of 2 and multiples of 32
56 (_Bits_Per_Block). Yes both at the same time! That's because the
57 next super block acquired will be 2 times the previous one, and
58 also all super blocks have to be multiples of the _Bits_Per_Block
61 How does it interact with the free list store?
63 The super block is contained in the FLS, and the FLS is responsible for
64 getting / returning Super Bocks to and from the OS using operator new
65 as defined by the C++ standard.
66 </p></div><div class="section" title="Super Block Data Layout"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.super_block_data"/>Super Block Data Layout</h3></div></div></div><p>
67 Each Super Block will be of some size that is a multiple of the
68 number of Bits Per Block. Typically, this value is chosen as
69 Bits_Per_Byte x sizeof(size_t). On an x86 system, this gives the
70 figure 8 x 4 = 32. Thus, each Super Block will be of size 32
71 x Some_Value. This Some_Value is sizeof(value_type). For now, let
72 it be called 'K'. Thus, finally, Super Block size is 32 x K bytes.
74 This value of 32 has been chosen because each size_t has 32-bits
75 and Maximum use of these can be made with such a figure.
77 Consider a block of size 64 ints. In memory, it would look like this:
78 (assume a 32-bit system where, size_t is a 32-bit entity).
79 </p><div class="table"><a id="id515279"/><p class="title"><strong>Table 21.1. Bitmap Allocator Memory Map</strong></p><div class="table-contents"><table summary="Bitmap Allocator Memory Map" border="1"><colgroup><col style="text-align: left" class="c1"/><col style="text-align: left" class="c2"/><col style="text-align: left" class="c3"/><col style="text-align: left" class="c4"/><col style="text-align: left" class="c5"/></colgroup><tbody><tr><td style="text-align: left">268</td><td style="text-align: left">0</td><td style="text-align: left">4294967295</td><td style="text-align: left">4294967295</td><td style="text-align: left">Data -> Space for 64 ints</td></tr></tbody></table></div></div><br class="table-break"/><p>
80 The first Column(268) represents the size of the Block in bytes as
81 seen by the Bitmap Allocator. Internally, a global free list is
82 used to keep track of the free blocks used and given back by the
83 bitmap allocator. It is this Free List Store that is responsible
84 for writing and managing this information. Actually the number of
85 bytes allocated in this case would be: 4 + 4 + (4x2) + (64x4) =
86 272 bytes, but the first 4 bytes are an addition by the Free List
87 Store, so the Bitmap Allocator sees only 268 bytes. These first 4
88 bytes about which the bitmapped allocator is not aware hold the
91 What do the remaining values represent?</p><p>
92 The 2nd 4 in the expression is the sizeof(size_t) because the
93 Bitmapped Allocator maintains a used count for each Super Block,
94 which is initially set to 0 (as indicated in the diagram). This is
95 incremented every time a block is removed from this super block
96 (allocated), and decremented whenever it is given back. So, when
97 the used count falls to 0, the whole super block will be given
98 back to the Free List Store.
100 The value 4294967295 represents the integer corresponding to the bit
101 representation of all bits set: 11111111111111111111111111111111.
103 The 3rd 4x2 is size of the bitmap itself, which is the size of 32-bits
105 which is 8-bytes, or 2 x sizeof(size_t).
106 </p></div><div class="section" title="Maximum Wasted Percentage"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.max_wasted"/>Maximum Wasted Percentage</h3></div></div></div><p>
107 This has nothing to do with the algorithm per-se,
108 only with some vales that must be chosen correctly to ensure that the
109 allocator performs well in a real word scenario, and maintains a good
110 balance between the memory consumption and the allocation/deallocation
113 The formula for calculating the maximum wastage as a percentage:
115 (32 x k + 1) / (2 x (32 x k + 1 + 32 x c)) x 100.
117 where k is the constant overhead per node (e.g., for list, it is
118 8 bytes, and for map it is 12 bytes) and c is the size of the
119 base type on which the map/list is instantiated. Thus, suppose the
120 type1 is int and type2 is double, they are related by the relation
121 sizeof(double) == 2*sizeof(int). Thus, all types must have this
122 double size relation for this formula to work properly.
124 Plugging-in: For List: k = 8 and c = 4 (int and double), we get:
127 For map/multimap: k = 12, and c = 4 (int and double), we get: 37.524%
129 Thus, knowing these values, and based on the sizeof(value_type), we may
130 create a function that returns the Max_Wastage_Percentage for us to use.
131 </p></div><div class="section" title="allocate"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.allocate"/><code class="function">allocate</code></h3></div></div></div><p>
132 The allocate function is specialized for single object allocation
133 ONLY. Thus, ONLY if n == 1, will the bitmap_allocator's
134 specialized algorithm be used. Otherwise, the request is satisfied
135 directly by calling operator new.
137 Suppose n == 1, then the allocator does the following:
138 </p><div class="orderedlist"><ol class="orderedlist"><li class="listitem"><p>
139 Checks to see whether a free block exists somewhere in a region
140 of memory close to the last satisfied request. If so, then that
141 block is marked as allocated in the bit map and given to the
142 user. If not, then (2) is executed.
143 </p></li><li class="listitem"><p>
144 Is there a free block anywhere after the current block right
145 up to the end of the memory that we have? If so, that block is
146 found, and the same procedure is applied as above, and
147 returned to the user. If not, then (3) is executed.
148 </p></li><li class="listitem"><p>
149 Is there any block in whatever region of memory that we own
150 free? This is done by checking
151 </p><div class="itemizedlist"><ul class="itemizedlist"><li class="listitem"><p>
152 The use count for each super block, and if that fails then
153 </p></li><li class="listitem"><p>
154 The individual bit-maps for each super block.
155 </p></li></ul></div><p>
156 Note: Here we are never touching any of the memory that the
157 user will be given, and we are confining all memory accesses
158 to a small region of memory! This helps reduce cache
159 misses. If this succeeds then we apply the same procedure on
160 that bit-map as (1), and return that block of memory to the
161 user. However, if this process fails, then we resort to (4).
162 </p></li><li class="listitem"><p>
163 This process involves Refilling the internal exponentially
164 growing memory pool. The said effect is achieved by calling
165 _S_refill_pool which does the following:
166 </p><div class="itemizedlist"><ul class="itemizedlist"><li class="listitem"><p>
167 Gets more memory from the Global Free List of the Required
169 </p></li><li class="listitem"><p>
170 Adjusts the size for the next call to itself.
171 </p></li><li class="listitem"><p>
172 Writes the appropriate headers in the bit-maps.
173 </p></li><li class="listitem"><p>
174 Sets the use count for that super-block just allocated to 0
176 </p></li><li class="listitem"><p>
177 All of the above accounts to maintaining the basic invariant
178 for the allocator. If the invariant is maintained, we are
179 sure that all is well. Now, the same process is applied on
180 the newly acquired free blocks, which are dispatched
182 </p></li></ul></div></li></ol></div><p>
183 Thus, you can clearly see that the allocate function is nothing but a
184 combination of the next-fit and first-fit algorithm optimized ONLY for
185 single object allocations.
186 </p></div><div class="section" title="deallocate"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.deallocate"/><code class="function">deallocate</code></h3></div></div></div><p>
187 The deallocate function again is specialized for single objects ONLY.
188 For all n belonging to > 1, the operator delete is called without
189 further ado, and the deallocate function returns.
191 However for n == 1, a series of steps are performed:
192 </p><div class="orderedlist"><ol class="orderedlist"><li class="listitem"><p>
193 We first need to locate that super-block which holds the memory
194 location given to us by the user. For that purpose, we maintain
195 a static variable _S_last_dealloc_index, which holds the index
196 into the vector of block pairs which indicates the index of the
197 last super-block from which memory was freed. We use this
198 strategy in the hope that the user will deallocate memory in a
199 region close to what he/she deallocated the last time around. If
200 the check for belongs_to succeeds, then we determine the bit-map
201 for the given pointer, and locate the index into that bit-map,
202 and mark that bit as free by setting it.
203 </p></li><li class="listitem"><p>
204 If the _S_last_dealloc_index does not point to the memory block
205 that we're looking for, then we do a linear search on the block
206 stored in the vector of Block Pairs. This vector in code is
207 called _S_mem_blocks. When the corresponding super-block is
208 found, we apply the same procedure as we did for (1) to mark the
209 block as free in the bit-map.
210 </p></li></ol></div><p>
211 Now, whenever a block is freed, the use count of that particular
212 super block goes down by 1. When this use count hits 0, we remove
213 that super block from the list of all valid super blocks stored in
214 the vector. While doing this, we also make sure that the basic
215 invariant is maintained by making sure that _S_last_request and
216 _S_last_dealloc_index point to valid locations within the vector.
217 </p></div><div class="section" title="Questions"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.questions"/>Questions</h3></div></div></div><div class="section" title="1"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.question.1"/>1</h4></div></div></div><p>
218 Q1) The "Data Layout" section is
219 cryptic. I have no idea of what you are trying to say. Layout of what?
220 The free-list? Each bitmap? The Super Block?
222 The layout of a Super Block of a given
223 size. In the example, a super block of size 32 x 1 is taken. The
224 general formula for calculating the size of a super block is
225 32 x sizeof(value_type) x 2^n, where n ranges from 0 to 32 for 32-bit
227 </p></div><div class="section" title="2"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.question.2"/>2</h4></div></div></div><p>
228 And since I just mentioned the
229 term `each bitmap', what in the world is meant by it? What does each
230 bitmap manage? How does it relate to the super block? Is the Super
231 Block a bitmap as well?
233 Each bitmap is part of a Super Block which is made up of 3 parts
234 as I have mentioned earlier. Re-iterating, 1. The use count,
235 2. The bit-map for that Super Block. 3. The actual memory that
236 will be eventually given to the user. Each bitmap is a multiple
237 of 32 in size. If there are 32 x (2^3) blocks of single objects
238 to be given, there will be '32 x (2^3)' bits present. Each 32
239 bits managing the allocated / free status for 32 blocks. Since
240 each size_t contains 32-bits, one size_t can manage up to 32
241 blocks' status. Each bit-map is made up of a number of size_t,
242 whose exact number for a super-block of a given size I have just
244 </p></div><div class="section" title="3"><div class="titlepage"><div><div><h4 class="title"><a id="bitmap.impl.question.3"/>3</h4></div></div></div><p>
245 How do the allocate and deallocate functions work in regard to
248 The allocate and deallocate functions manipulate the bitmaps and
249 have nothing to do with the memory that is given to the user. As
250 I have earlier mentioned, a 1 in the bitmap's bit field
251 indicates free, while a 0 indicates allocated. This lets us
252 check 32 bits at a time to check whether there is at lease one
253 free block in those 32 blocks by testing for equality with
254 (0). Now, the allocate function will given a memory block find
255 the corresponding bit in the bitmap, and will reset it (i.e.,
256 make it re-set (0)). And when the deallocate function is called,
257 it will again set that bit after locating it to indicate that
258 that particular block corresponding to this bit in the bit-map
259 is not being used by anyone, and may be used to satisfy future
262 e.g.: Consider a bit-map of 64-bits as represented below:
263 1111111111111111111111111111111111111111111111111111111111111111
265 Now, when the first request for allocation of a single object
266 comes along, the first block in address order is returned. And
267 since the bit-maps in the reverse order to that of the address
268 order, the last bit (LSB if the bit-map is considered as a
269 binary word of 64-bits) is re-set to 0.
271 The bit-map now looks like this:
272 1111111111111111111111111111111111111111111111111111111111111110
273 </p></div></div><div class="section" title="Locality"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.locality"/>Locality</h3></div></div></div><p>
274 Another issue would be whether to keep the all bitmaps in a
275 separate area in memory, or to keep them near the actual blocks
276 that will be given out or allocated for the client. After some
277 testing, I've decided to keep these bitmaps close to the actual
278 blocks. This will help in 2 ways.
279 </p><div class="orderedlist"><ol class="orderedlist"><li class="listitem"><p>Constant time access for the bitmap themselves, since no kind of
280 look up will be needed to find the correct bitmap list or its
281 equivalent.</p></li><li class="listitem"><p>And also this would preserve the cache as far as possible.</p></li></ol></div><p>
282 So in effect, this kind of an allocator might prove beneficial from a
283 purely cache point of view. But this allocator has been made to try and
284 roll out the defects of the node_allocator, wherein the nodes get
285 skewed about in memory, if they are not returned in the exact reverse
286 order or in the same order in which they were allocated. Also, the
287 new_allocator's book keeping overhead is too much for small objects and
288 single object allocations, though it preserves the locality of blocks
289 very well when they are returned back to the allocator.
290 </p></div><div class="section" title="Overhead and Grow Policy"><div class="titlepage"><div><div><h3 class="title"><a id="bitmap.impl.grow_policy"/>Overhead and Grow Policy</h3></div></div></div><p>
291 Expected overhead per block would be 1 bit in memory. Also, once
292 the address of the free list has been found, the cost for
293 allocation/deallocation would be negligible, and is supposed to be
294 constant time. For these very reasons, it is very important to
295 minimize the linear time costs, which include finding a free list
296 with a free block while allocating, and finding the corresponding
297 free list for a block while deallocating. Therefore, I have
298 decided that the growth of the internal pool for this allocator
299 will be exponential as compared to linear for
300 node_allocator. There, linear time works well, because we are
301 mainly concerned with speed of allocation/deallocation and memory
302 consumption, whereas here, the allocation/deallocation part does
303 have some linear/logarithmic complexity components in it. Thus, to
304 try and minimize them would be a good thing to do at the cost of a
305 little bit of memory.
307 Another thing to be noted is the pool size will double every time
308 the internal pool gets exhausted, and all the free blocks have
309 been given away. The initial size of the pool would be
310 sizeof(size_t) x 8 which is the number of bits in an integer,
311 which can fit exactly in a CPU register. Hence, the term given is
312 exponential growth of the internal pool.
313 </p></div></div><div class="navfooter"><hr/><table width="100%" summary="Navigation footer"><tr><td align="left"><a accesskey="p" href="bitmap_allocator.html">Prev</a> </td><td align="center"><a accesskey="u" href="bitmap_allocator.html">Up</a></td><td align="right"> <a accesskey="n" href="policy_data_structures.html">Next</a></td></tr><tr><td align="left" valign="top">Chapter 21. The bitmap_allocator </td><td align="center"><a accesskey="h" href="../index.html">Home</a></td><td align="right" valign="top"> Chapter 22. Policy-Based Data Structures</td></tr></table></div></body></html>
OSDN Git Service
