old-autovect-branch/libstdc++-v3/docs/html/ext/ballocator_doc.html

   1 <!DOCTYPE html PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
   2 <html>
   3 <head>
   4   <meta content="text/html; charset=ISO-8859-1"
   5  http-equiv="content-type">
   6   <title>Bitmap Allocator</title>
   7   <meta content="Dhruv Matani" name="author">
   8   <meta content="Bitmap Allocator" name="description">
   9 </head>
  10 <body>
  11 <h1 style="text-align: center;">Bitmap Allocator</h1>
  12 <em><br>
  13 <small><small>The latest version of this document is always available
  14 at <a
  15  href="http://gcc.gnu.org/onlinedocs/libstdc++/ext/ballocator_doc.html">
  16 http://gcc.gnu.org/onlinedocs/libstdc++/ext/ballocator_doc.html</a>.</small></small></em><br>
  17 <br>
  18 <em> To the <a href="http://gcc.gnu.org/libstdc++/">libstdc++-v3
  19 homepage</a>.</em><br>
  20 <br>
  21 <hr style="width: 100%; height: 2px;"><br>
  22 As this name suggests, this allocator uses a bit-map to keep track of
  23 the used and unused memory locations for it's book-keeping purposes.<br>
  24 <br>
  25 This allocator will make use of 1 single bit to keep track of whether
  26 it has been allocated or not. A bit 1 indicates free, while 0 indicates
  27 allocated. This has been done so that you can easily check a collection
  28 of bits for a free block. This kind of Bitmapped strategy works best
  29 for single object allocations, and with the STL type parameterized
  30 allocators, we do not need to choose any size for the block which will
  31 be represented by a single bit. This will be the size of the parameter
  32 around which the allocator has been parameterized. Thus, close to
  33 optimal performance will result. Hence, this should be used for node
  34 based containers which call the allocate function with an argument of 1.<br>
  35 <br>
  36 The bitmapped allocator's internal pool is exponentially growing.
  37 Meaning that internally, the blocks acquired from the Free List Store
  38 will double every time the bitmapped allocator runs out of memory.<br>
  39 <br>
  40 <hr style="width: 100%; height: 2px;"><br>
  41 The macro __GTHREADS decides whether to use Mutex Protection around
  42 every allocation/deallocation. The state of the macro is picked up
  43 automatically from the gthr abstration layer.<br>
  44 <br>
  45 <hr style="width: 100%; height: 2px;">
  46 <h3 style="text-align: center;">What is the Free List Store?</h3>
  47 <br>
  48 The Free List Store (referred to as FLS for the remaining part of this
  49 document) is the Global memory pool that is shared by all instances of
  50 the bitmapped allocator instantiated for any type. This maintains a
  51 sorted order of all free memory blocks given back to it by the
  52 bitmapped allocator, and is also responsible for giving memory to the
  53 bitmapped allocator when it asks for more.<br>
  54 <br>
  55 Internally, there is a Free List threshold which indicates the Maximum
  56 number of free lists that the FLS can hold internally (cache).
  57 Currently, this value is set at 64. So, if there are more than 64 free
  58 lists coming in, then some of them will be given back to the OS using
  59 operator delete so that at any given time the Free List's size does not
  60 exceed 64 entries. This is done because a Binary Search is used to
  61 locate an entry in a free list when a request for memory comes along.
  62 Thus, the run-time complexity of the search would go up given an
  63 increasing size, for 64 entries however, lg(64) == 6 comparisons are
  64 enough to locate the correct free list if it exists.<br>
  65 <br>
  66 Suppose the free list size has reached it's threshold, then the largest
  67 block from among those in the list and the new block will be selected
  68 and given back to the OS. This is done because it reduces external
  69 fragmentation, and allows the OS to use the larger blocks later in an
  70 orderly fashion, possibly merging them later. Also, on some systems,
  71 large blocks are obtained via calls to mmap, so giving them back to
  72 free system resources becomes most important.<br>
  73 <br>
  74 The function _S_should_i_give decides the policy that determines
  75 whether the current block of memory should be given to the allocator
  76 for the request that it has made. That's because we may not always have
  77 exact fits for the memory size that the allocator requests. We do this
  78 mainly to prevent external fragmentation at the cost of a little
  79 internal fragmentation. Now, the value of this internal fragmentation
  80 has to be decided by this function. I can see 3 possibilities right
  81 now. Please add more as and when you find better strategies.<br>
  82 <br>
  83 <ol>
  84   <li>Equal size check. Return true only when the 2 blocks are of equal
  85 size.</li>
  86   <li>Difference Threshold: Return true only when the _block_size is
  87 greater than or equal to the _required_size, and if the _BS is &gt; _RS
  88 by a difference of less than some THRESHOLD value, then return true,
  89 else return false. </li>
  90   <li>Percentage Threshold. Return true only when the _block_size is
  91 greater than or equal to the _required_size, and if the _BS is &gt; _RS
  92 by a percentage of less than some THRESHOLD value, then return true,
  93 else return false.</li>
  94 </ol>
  95 <br>
  96 Currently, (3) is being used with a value of 36% Maximum wastage per
  97 Super Block.<br>
  98 <br>
  99 <hr style="width: 100%; height: 2px;"><span style="font-weight: bold;">1)
 100 What is a super block? Why is it needed?</span><br>
 101 <br>
 102 A super block is the block of memory acquired from the FLS from which
 103 the bitmap allocator carves out memory for single objects and satisfies
 104 the user's requests. These super blocks come in sizes that are powers
 105 of 2 and multiples of 32 (_Bits_Per_Block). Yes both at the same time!
 106 That's because the next super block acquired will be 2 times the
 107 previous one, and also all super blocks have to be multiples of the
 108 _Bits_Per_Block value. <br>
 109 <br>
 110 <span style="font-weight: bold;">2) How does it interact with the free
 111 list store?</span><br>
 112 <br>
 113 The super block is contained in the FLS, and the FLS is responsible for
 114 getting / returning Super Bocks to and from the OS using operator new
 115 as defined by the C++ standard.<br>
 116 <br>
 117 <hr style="width: 100%; height: 2px;">
 118 <h3 style="text-align: center;">How does the allocate function Work?</h3>
 119 <br>
 120 The allocate function is specialized for single object allocation ONLY.
 121 Thus, ONLY if n == 1, will the bitmap_allocator's specialized algorithm
 122 be used. Otherwise, the request is satisfied directly by calling
 123 operator new.<br>
 124 <br>
 125 Suppose n == 1, then the allocator does the following:<br>
 126 <br>
 127 <ol>
 128   <li>Checks to see whether the a free block exists somewhere in a
 129 region of memory close to the last satisfied request. If so, then that
 130 block is marked as allocated in the bit map and given to the user. If
 131 not, then (2) is executed.</li>
 132   <li>Is there a free block anywhere after the current block right upto
 133 the end of the memory that we have? If so, that block is found, and the
 134 same procedure is applied as above, and returned to the user. If not,
 135 then (3) is executed.</li>
 136   <li>Is there any block in whatever region of memory that we own free?
 137 This is done by checking <br>
 138     <div style="margin-left: 40px;">
 139     <ul>
 140       <li>The use count for each super block, and if that fails then </li>
 141       <li>The individual bit-maps for each super block. </li>
 142     </ul>
 143     </div>
 144 Note: Here we are never touching any of the memory that the user will
 145 be given, and we are confining all memory accesses to a small region of
 146 memory! This helps reduce cache misses. If this succeeds then we apply
 147 the same procedure on that bit-map as (1), and return that block of
 148 memory to the user. However, if this process fails, then we resort to
 149 (4).</li>
 150   <li>This process involves Refilling the internal exponentially
 151 growing memory pool. The said effect is achieved by calling
 152 _S_refill_pool which does the following: <br>
 153     <div style="margin-left: 40px;">
 154     <ul>
 155       <li>Gets more memory from the Global Free List of the Required
 156 size. </li>
 157       <li>Adjusts the size for the next call to itself. </li>
 158       <li>Writes the appropriate headers in the bit-maps.</li>
 159       <li>Sets the use count for that super-block just allocated to 0
 160 (zero). </li>
 161       <li>All of the above accounts to maintaining the basic invariant
 162 for the allocator. If the invariant is maintained, we are sure that all
 163 is well. Now, the same process is applied on the newly acquired free
 164 blocks, which are dispatched accordingly.</li>
 165     </ul>
 166     </div>
 167   </li>
 168 </ol>
 169 <br>
 170 Thus, you can clearly see that the allocate function is nothing but a
 171 combination of the next-fit and first-fit algorithm optimized ONLY for
 172 single object allocations.<br>
 173 <br>
 174 <br>
 175 <hr style="width: 100%; height: 2px;">
 176 <h3 style="text-align: center;">How does the deallocate function work?</h3>
 177 <br>
 178 The deallocate function again is specialized for single objects ONLY.
 179 For all n belonging to &gt; 1, the operator delete is called without
 180 further ado, and the deallocate function returns.<br>
 181 <br>
 182 However for n == 1, a series of steps are performed:<br>
 183 <br>
 184 <ol>
 185   <li>We first need to locate that super-block which holds the memory
 186 location given to us by the user. For that purpose, we maintain a
 187 static variable _S_last_dealloc_index, which holds the index into the
 188 vector of block pairs which indicates the index of the last super-block
 189 from which memory was freed. We use this strategy in the hope that the
 190 user will deallocate memory in a region close to what he/she
 191 deallocated the last time around. If the check for belongs_to succeeds,
 192 then we determine the bit-map for the given pointer, and locate the
 193 index into that bit-map, and mark that bit as free by setting it.</li>
 194   <li>If the _S_last_dealloc_index does not point to the memory block
 195 that we're looking for, then we do a linear search on the block stored
 196 in the vector of Block Pairs. This vector in code is called
 197 _S_mem_blocks. When the corresponding super-block is found, we apply
 198 the same procedure as we did for (1) to mark the block as free in the
 199 bit-map.</li>
 200 </ol>
 201 <br>
 202 Now, whenever a block is freed, the use count of that particular super
 203 block goes down by 1. When this use count hits 0, we remove that super
 204 block from the list of all valid super blocks stored in the vector.
 205 While doing this, we also make sure that the basic invariant is
 206 maintained by making sure that _S_last_request and
 207 _S_last_dealloc_index point to valid locations within the vector.<br>
 208 <br>
 209 <hr style="width: 100%; height: 2px;"><br>
 210 <h3 style="text-align: center;">Data Layout for a Super Block:</h3>
 211 <br>
 212 Each Super Block will be of some size that is a multiple of the number
 213 of Bits Per Block. Typically, this value is chosen as Bits_Per_Byte x
 214 sizeof(size_t). On an x86 system, this gives the figure &nbsp;8 x
 215 4 = 32. Thus, each Super Block will be of size 32 x Some_Value. This
 216 Some_Value is sizeof(value_type). For now, let it be called 'K'. Thus,
 217 finally, Super Block size is 32 x K bytes.<br>
 218 <br>
 219 This value of 32 has been chosen because each size_t has 32-bits
 220 and Maximum use of these can be made with such a figure.<br>
 221 <br>
 222 Consider a block of size 64 ints. In memory, it would look like this:
 223 (assume a 32-bit system where, size_t is a 32-bit entity).<br>
 224 <br>
 225 <table cellpadding="0" cellspacing="0" border="1"
 226  style="text-align: left; width: 763px; height: 21px;">
 227   <tbody>
 228     <tr>
 229       <td style="vertical-align: top; text-align: center;">268<br>
 230       </td>
 231       <td style="vertical-align: top; text-align: center;">0<br>
 232       </td>
 233       <td style="vertical-align: top; text-align: center;">4294967295<br>
 234       </td>
 235       <td style="vertical-align: top; text-align: center;">4294967295<br>
 236       </td>
 237       <td style="vertical-align: top; text-align: center;">Data -&gt;
 238 Space for 64 ints<br>
 239       </td>
 240     </tr>
 241   </tbody>
 242 </table>
 243 <br>
 244 <br>
 245 The first Column(268) represents the size of the Block in bytes as seen
 246 by
 247 the Bitmap Allocator. Internally, a global free list is used to keep
 248 track of the free blocks used and given back by the bitmap allocator.
 249 It is this Free List Store that is responsible for writing and managing
 250 this information. Actually the number of bytes allocated in this case
 251 would be: 4 + 4 + (4x2) + (64x4) = 272 bytes, but the first 4 bytes are
 252 an
 253 addition by the Free List Store, so the Bitmap Allocator sees only 268
 254 bytes. These first 4 bytes about which the bitmapped allocator is not
 255 aware hold the value 268.<br>
 256 <br>
 257 <span style="font-weight: bold;">What do the remaining values represent?</span><br>
 258 <br>
 259 The 2nd 4 in the expression is the sizeof(size_t) because the
 260 Bitmapped Allocator maintains a used count for each Super Block, which
 261 is initially set to 0 (as indicated in the diagram). This is
 262 incremented every time a block is removed from this super block
 263 (allocated), and decremented whenever it is given back. So, when the
 264 used count falls to 0, the whole super block will be given back to the
 265 Free List Store.<br>
 266 <br>
 267 The value 4294967295 represents the integer corresponding to the bit
 268 representation of all bits set: 11111111111111111111111111111111.<br>
 269 <br>
 270 The 3rd 4x2 is size of the bitmap itself, which is the size of 32-bits
 271 x 2,
 272 which is 8-bytes, or 2 x sizeof(size_t).<br>
 273 <br>
 274 <hr style="width: 100%; height: 2px;"><br>
 275 Another issue would be whether to keep the all bitmaps in a separate
 276 area in memory, or to keep them near the actual blocks that will be
 277 given out or allocated for the client. After some testing, I've decided
 278 to keep these bitmaps close to the actual blocks. this will help in 2
 279 ways. <br>
 280 <br>
 281 <ol>
 282   <li>Constant time access for the bitmap themselves, since no kind of
 283 look up will be needed to find the correct bitmap list or it's
 284 equivalent.</li>
 285   <li>And also this would preserve the cache as far as possible.</li>
 286 </ol>
 287 <br>
 288 So in effect, this kind of an allocator might prove beneficial from a
 289 purely cache point of view. But this allocator has been made to try and
 290 roll out the defects of the node_allocator, wherein the nodes get
 291 skewed about in memory, if they are not returned in the exact reverse
 292 order or in the same order in which they were allocated. Also, the
 293 new_allocator's book keeping overhead is too much for small objects and
 294 single object allocations, though it preserves the locality of blocks
 295 very well when they are returned back to the allocator.<br>
 296 <br>
 297 <hr style="width: 100%; height: 2px;"><br>
 298 Expected overhead per block would be 1 bit in memory. Also, once the
 299 address of the free list has been found, the cost for
 300 allocation/deallocation would be negligible, and is supposed to be
 301 constant time. For these very reasons, it is very important to minimize
 302 the linear time costs, which include finding a free list with a free
 303 block while allocating, and finding the corresponding free list for a
 304 block while deallocating. Therefore, I have decided that the growth of
 305 the internal pool for this allocator will be exponential as compared to
 306 linear for node_allocator. There, linear time works well, because we
 307 are mainly concerned with speed of allocation/deallocation and memory
 308 consumption, whereas here, the allocation/deallocation part does have
 309 some linear/logarithmic complexity components in it. Thus, to try and
 310 minimize them would be a good thing to do at the cost of a little bit
 311 of memory.<br>
 312 <br>
 313 Another thing to be noted is the the pool size will double every time
 314 the internal pool gets exhausted, and all the free blocks have been
 315 given away. The initial size of the pool would be sizeof(size_t) x 8
 316 which is the number of bits in an integer, which can fit exactly
 317 in a CPU register. Hence, the term given is exponential growth of the
 318 internal pool.<br>
 319 <br>
 320 <hr style="width: 100%; height: 2px;">After reading all this, you may
 321 still have a few questions about the internal working of this
 322 allocator, like my friend had!<br>
 323 <br>
 324 Well here are the exact questions that he posed:<br>
 325 <br>
 326 <span style="font-weight: bold;">Q1) The "Data Layout" section is
 327 cryptic. I have no idea of what you are trying to say. Layout of what?
 328 The free-list? Each bitmap? The Super Block?</span><br>
 329 <br>
 330 <div style="margin-left: 40px;"> The layout of a Super Block of a given
 331 size. In the example, a super block of size 32 x 1 is taken. The
 332 general formula for calculating the size of a super block is
 333 32 x sizeof(value_type) x 2^n, where n ranges from 0 to 32 for 32-bit
 334 systems.<br>
 335 </div>
 336 <br>
 337 <span style="font-weight: bold;">Q2) And since I just mentioned the
 338 term `each bitmap', what in the world is meant by it? What does each
 339 bitmap manage? How does it relate to the super block? Is the Super
 340 Block a bitmap as well?</span><br style="font-weight: bold;">
 341 <br>
 342 <div style="margin-left: 40px;"> Good question! Each bitmap is part of
 343 a
 344 Super Block which is made up of 3 parts as I have mentioned earlier.
 345 Re-iterating, 1. The use count, 2. The bit-map for that Super Block. 3.
 346 The actual memory that will be eventually given to the user. Each
 347 bitmap is a multiple of 32 in size. If there are 32 x (2^3) blocks of
 348 single objects to be given, there will be '32 x (2^3)' bits present.
 349 Each
 350 32 bits managing the allocated / free status for 32 blocks. Since each
 351 size_t contains 32-bits, one size_t can manage upto 32
 352 blocks' status. Each bit-map is made up of a number of size_t,
 353 whose exact number for a super-block of a given size I have just
 354 mentioned.<br>
 355 </div>
 356 <br>
 357 <span style="font-weight: bold;">Q3) How do the allocate and deallocate
 358 functions work in regard to bitmaps?</span><br>
 359 <br>
 360 <div style="margin-left: 40px;"> The allocate and deallocate functions
 361 manipulate the bitmaps and have nothing to do with the memory that is
 362 given to the user. As I have earlier mentioned, a 1 in the bitmap's bit
 363 field indicates free, while a 0 indicates allocated. This lets us check
 364 32 bits at a time to check whether there is at lease one free block in
 365 those 32 blocks by testing for equality with (0). Now, the allocate
 366 function will given a memory block find the corresponding bit in the
 367 bitmap, and will reset it (ie. make it re-set (0)). And when the
 368 deallocate function is called, it will again set that bit after
 369 locating it to indicate that that particular block corresponding to
 370 this bit in the bit-map is not being used by anyone, and may be used to
 371 satisfy future requests.<br>
 372 <br>
 373 eg: Consider a bit-map of 64-bits as represented below:<br>
 374 1111111111111111111111111111111111111111111111111111111111111111<br>
 375 <br>
 376 Now, when the first request for allocation of a single object comes
 377 along, the first block in address order is returned. And since the
 378 bit-maps in the reverse order to that of the address order, the last
 379 bit(LSB if the bit-map is considered as a binary word of 64-bits) is
 380 re-set to 0.<br>
 381 <br>
 382 The bit-map now looks like this:<br>
 383 1111111111111111111111111111111111111111111111111111111111111110<br>
 384 </div>
 385 <br>
 386 <br>
 387 <hr style="width: 100%; height: 2px;"><br>
 388 (Tech-Stuff, Please stay out if you are not interested in the selection
 389 of certain constants. This has nothing to do with the algorithm per-se,
 390 only with some vales that must be chosen correctly to ensure that the
 391 allocator performs well in a real word scenario, and maintains a good
 392 balance between the memory consumption and the allocation/deallocation
 393 speed).<br>
 394 <br>
 395 The formula for calculating the maximum wastage as a percentage:<br>
 396 <br>
 397 (32 x k + 1) / (2 x (32 x k + 1 + 32 x c)) x 100.<br>
 398 <br>
 399 Where,<br>
 400 k =&gt; The constant overhead per node. eg. for list, it is 8 bytes,
 401 and for map it is 12 bytes.<br>
 402 c =&gt; The size of the base type on which the map/list is
 403 instantiated. Thus, suppose the the type1 is int and type2 is double,
 404 they are related by the relation sizeof(double) == 2*sizeof(int). Thus,
 405 all types must have this double size relation for this formula to work
 406 properly.<br>
 407 <br>
 408 Plugging-in: For List: k = 8 and c = 4 (int and double), we get:<br>
 409 33.376%<br>
 410 <br>
 411 For map/multimap: k = 12, and c = 4 (int and double), we get:<br>
 412 37.524%<br>
 413 <br>
 414 Thus, knowing these values, and based on the sizeof(value_type), we may
 415 create a function that returns the Max_Wastage_Percentage for us to use.<br>
 416 <br>
 417 <hr style="width: 100%; height: 2px;"><small><small><em> See <a
 418  href="file:///home/dhruv/projects/libstdc++-v3/gcc/libstdc++-v3/docs/html/17_intro/license.html">license.html</a>
 419 for copying conditions. Comments and suggestions are welcome, and may
 420 be
 421 sent to <a href="mailto:libstdc++@gcc.gnu.org">the libstdc++ mailing
 422 list</a>.</em><br>
 423 </small></small><br>
 424 <br>
 425 </body>
 426 </html>