首先大概介绍BT网络运行的整体流程:
开始阶段,一个节点加入到网络中,并向tracker节点发送信息,tracker返回若干个邻居的列表
得到列表后,向每个邻居发送bitfiled信息,来获取他们的文件状态。接着确定需要的piece,并向拥有该
piece的邻居发送关注的请求消息。本地节点根据过去20s内邻居节点的带宽传输表现,选出前3,并把它们置为疏通状态,向他们发送块的请求。
当收到请求信息时,返回一个piece信息,注意如果本地节点上传少于10个块,就把当前请求入队,按队列顺序一个个请求处理,直到上传了10个块。
每当一个节点完成了一个piece的下载,就会给所有邻居发送一个hava信息,表明自己有可以分享的piece
接下来贴上bittorent.java,附有自己添加的注释
1 /* 2 * Copyright (c) 2007-2008 Fabrizio Frioli, Michele Pedrolli 3 * 4 * This program is free software; you can redistribute it and/or modify 5 * it under the terms of the GNU Lesser General Public License version 2 as 6 * published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 11 * GNU Lesser General Public License for more details. 12 * 13 * You should have received a copy of the GNU Lesser General Public License 14 * along with this program; if not, write to the Free Software 15 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 16 * 17 * -- 18 * 19 * Please send your questions/suggestions to: 20 * {fabrizio.frioli, michele.pedrolli} at studenti dot unitn dot it 21 * 22 */ 23 24 package peersim.bittorrent; 25 26 import peersim.core.*; 27 import peersim.config.*; 28 import peersim.edsim.*; 29 import peersim.transport.*; 30 31 /** 32 * This is the class that implements the BitTorrent module for Peersim 33 */ 34 public class BitTorrent implements EDProtocol { 35 //用于配置文件接受数据的字符串 36 /** 37 * The size in Megabytes of the file being shared. 38 * @config 39 */ 40 private static final String PAR_SIZE="file_size"; 41 /** 42 * The Transport used by the the protocol. 43 * @config 44 */ 45 private static final String PAR_TRANSPORT="transport"; 46 /** 47 * The maximum number of neighbor that a node can have. 48 * @config 49 */ 50 private static final String PAR_SWARM="max_swarm_size"; 51 /** 52 * The maximum number of peers returned by the tracker when a new 53 * set of peers is requested through a <tt>TRACKER</tt> message. 54 * @config 55 */ 56 private static final String PAR_PEERSET_SIZE="peerset_size"; 57 /** 58 * Defines how much the network can grow with respect to the <tt>network.size</tt> 59 * when {@link NetworkDynamics} is used. 60 * @config 61 */ 62 private static final String PAR_MAX_GROWTH="max_growth"; 63 /** 64 * Is the number of requests of the same block sent to different peers. 65 * @config 66 */ 67 private static final String PAR_DUP_REQ = "duplicated_requests"; 68 69 //16种事件的代号定义 70 71 /** 72 * KEEP_ALIVE message. 73 * @see SimpleEvent#type "Event types" 74 */ 75 private static final int KEEP_ALIVE = 1; 76 77 /** 78 * CHOKE message. 79 * @see SimpleEvent#type "Event types" 80 */ 81 private static final int CHOKE = 2; 82 83 /** 84 * UNCHOKE message. 85 * @see SimpleEvent#type "Event types" 86 */ 87 private static final int UNCHOKE = 3; 88 89 /** 90 * INTERESTED message. 91 * @see SimpleEvent#type "Event types" 92 */ 93 private static final int INTERESTED = 4; 94 95 /** 96 * NOT_INTERESTED message. 97 * @see SimpleEvent#type "Event types" 98 */ 99 private static final int NOT_INTERESTED = 5; 100 101 /** 102 * HAVE message. 103 * @see SimpleEvent#type "Event types" 104 */ 105 private static final int HAVE = 6; 106 107 /** 108 * BITFIELD message. 109 * @see SimpleEvent#type "Event types" 110 */ 111 private static final int BITFIELD = 7; 112 113 /** 114 * REQUEST message. 115 * @see SimpleEvent#type "Event types" 116 */ 117 private static final int REQUEST = 8; 118 119 /** 120 * PIECE message. 121 * @see SimpleEvent#type "Event types" 122 */ 123 private static final int PIECE = 9; 124 125 /** 126 * CANCEL message. 127 * @see SimpleEvent#type "Event types" 128 */ 129 private static final int CANCEL = 10; 130 131 /** 132 * TRACKER message. 133 * @see SimpleEvent#type "Event types" 134 */ 135 private static final int TRACKER = 11; 136 137 /** 138 * PEERSET message. 139 * @see SimpleEvent#type "Event types" 140 */ 141 private static final int PEERSET = 12; 142 143 /** 144 * CHOKE_TIME event. 145 * @see SimpleEvent#type "Event types" 146 */ 147 private static final int CHOKE_TIME = 13; 148 149 /** 150 * OPTUNCHK_TIME event. 151 * @see SimpleEvent#type "Event types" 152 */ 153 private static final int OPTUNCHK_TIME = 14; 154 155 /** 156 * ANTISNUB_TIME event. 157 * @see SimpleEvent#type "Event types" 158 */ 159 private static final int ANTISNUB_TIME = 15; 160 161 /** 162 * CHECKALIVE_TIME event. 163 * @see SimpleEvent#type "Event types" 164 */ 165 private static final int CHECKALIVE_TIME = 16; 166 167 /** 168 * TRACKERALIVE_TIME event. 169 * @see SimpleEvent#type "Event types" 170 */ 171 private static final int TRACKERALIVE_TIME = 17; 172 173 /** 174 * DOWNLOAD_COMPLETED event. 175 * @see SimpleEvent#type "Event types" 176 */ 177 private static final int DOWNLOAD_COMPLETED = 18; 178 179 //一大堆的变量初始化定义,要仔细看,不记得回头查一下 180 181 /** 182 * The maxium connection speed of the local node. 183 */ 184 int maxBandwidth; //本地节点的最大带宽(连接速度) 185 186 /** 187 * Stores the neighbors ordered by ID. 188 * @see Element 189 */ 190 private peersim.bittorrent.Element byPeer[]; //按ID存储的邻居节点组 191 192 /** 193 * Contains the neighbors ordered by bandwidth as needed by the unchocking 194 * algorithm. 195 */ 196 private peersim.bittorrent.Element byBandwidth[]; //按带宽存储的邻居节点组 197 198 /** 199 * The Neighbors list. 200 */ 201 private Neighbor cache[]; //邻居节点列表,很常用 202 203 /** 204 * Reference to the neighbors that unchocked the local node. 205 */ 206 private boolean unchokedBy[]; //对本地疏通的节点组 207 /** 208 * Number of neighbors in the cache. When it decreases under 20, a new peerset 209 * is requested to the tracker. 210 */ 211 private int nNodes = 0; //邻居节点列表中的数量,降到20以下,需要重新向TRACKER发请求 212 213 /** 214 * Maximum number of nodes in the network. 215 */ 216 private int nMaxNodes; //网络中最大节点数 217 218 /** 219 * The status of the local peer. 0 means that the current peer is a leecher, 1 a seeder. 220 */ 221 private int peerStatus; //节点状态,0是非种子节点,1是种子 222 223 /** 224 * Defines how much the network can grow with respect to the <tt>network.size</tt> 225 * when {@link NetworkDynamics} is used. 226 */ 227 public int maxGrowth; 228 229 /** 230 * File status of the local node. Contains the blocks owned by the local node. 231 */ 232 private int status[]; //本地节点拥有的块组 233 234 /** 235 * Current number of Bitfield request sent. It must be taken into account 236 * before sending another one. 237 */ 238 private int nBitfieldSent = 0; 239 240 /** 241 * Current number of pieces in upload from the local peer. 242 */ 243 public int nPiecesUp = 0; 244 /** 245 * Current number of pieces in download to the local peer. 246 */ 247 public int nPiecesDown = 0; 248 249 /** 250 * Current number of piece completed. 251 */ 252 private int nPieceCompleted = 0; 253 254 /** 255 * Current downloading piece ID, the previous lastInterested piece. 256 */ 257 int currentPiece = -1; //正在下载的piece的ID 258 259 /** 260 * Used to compute the average download rates in choking algorithm. Stores the 261 * number of <tt>CHOKE</tt> events. 262 */ 263 int n_choke_time = 0; 264 265 /** 266 * Used to send the <tt>TRACKER</tt> message when the local node has 20 neighbors 267 * for the first time. 268 */ 269 boolean lock = false; 270 271 /** 272 * Number of peers interested to my pieces. 273 */ 274 int numInterestedPeers = 0; 275 276 /** 277 * Last piece for which the local node sent an <tt>INTERESTED</tt> message. 278 */ 279 int lastInterested = -1; 280 281 /** 282 * The status of the current piece in download. Length 16, every time the local node 283 * receives a PIECE message, it updates the corrisponding block's cell. The cell 284 * contains the ID for that block of that piece. If an already owned 285 * block is received this is discarded. 286 */ 287 private int pieceStatus[]; 288 289 /** 290 * Length of the file. Stored as number of pieces (256KB each one). 291 */ 292 int nPieces; //文件长度,有几个PIECE 293 294 /** 295 * Contains the neighbors's status of the file. Every row represents a 296 * node and every a cell has value O if the neighbor doesn't 297 * have the piece, 1 otherwise. It has {@link #swarmSize} rows and {@link #nPieces} 298 * columns. 299 */ 300 int [][]swarm; //节点的文件状态组,行代表每个节点,列代表每个piece 301 302 /** 303 * The summation of the swarm's rows. Calculated every time a {@link #BITFIELD} message 304 * is received and updated every time HAVE message is received. 305 */ 306 int rarestPieceSet[]; //最少优先集合 307 308 /** 309 * The five pending block requests. 310 */ 311 int pendingRequest[]; //待处理组 312 313 /** 314 * The maximum swarm size (default is 80) 315 */ 316 int swarmSize; 317 318 /** 319 * The size of the peerset. This is the number of "friends" nodes 320 * sent from the tracker to each new node (default: 50) 321 */ 322 int peersetSize; 323 324 /** 325 * The ID of the current node 326 */ 327 private long thisNodeID; 328 329 /** 330 * Number of duplicated requests as specified in the configuration file. 331 * @see BitTorrent#PAR_DUP_REQ 332 */ 333 private int numberOfDuplicatedRequests; 334 335 /** 336 * The queue where the requests to serve are stored. 337 * The default dimension of the queue is 20. 338 */ 339 Queue requestToServe = null; 340 341 /** 342 * The queue where the out of sequence incoming pieces are stored 343 * waiting for the right moment to be processed. 344 * The default dimension of the queue is 100. 345 */ 346 Queue incomingPieces = null; 347 348 /** 349 * The Transport ID. 350 * @see BitTorrent#PAR_TRANSPORT 351 */ 352 int tid; 353 354 /** 355 * The reference to the tracker node. If equals to <tt>null</tt>, the local 356 * node is the tracker. 357 */ 358 private Node tracker = null; 359 360 /** 361 * The default constructor. Reads the configuration file and initializes the 362 * configuration parameters. 363 * @param prefix the component prefix declared in the configuration file 364 */ 365 public BitTorrent(String prefix){ // Used for the tracker's protocol 366 tid = Configuration.getPid(prefix+"."+PAR_TRANSPORT); 367 nPieces = (int)((Configuration.getInt(prefix+"."+PAR_SIZE))*1000000/256000); 368 swarmSize = (int)Configuration.getInt(prefix+"."+PAR_SWARM); 369 peersetSize = (int)Configuration.getInt(prefix+"."+PAR_PEERSET_SIZE); 370 numberOfDuplicatedRequests = (int)Configuration.getInt(prefix+"."+PAR_DUP_REQ); 371 maxGrowth = (int)Configuration.getInt(prefix+"."+PAR_MAX_GROWTH); 372 nMaxNodes = Network.getCapacity()-1; 373 } 374 375 /** 376 * Gets the reference to the tracker node. 377 * @return the reference to the tracker 378 */ 379 public Node getTracker(){ 380 return tracker; 381 } 382 383 /** 384 * Gets the number of neighbors currently stored in the cache of the local node. 385 * @return the number of neighbors in the cache 386 */ 387 public int getNNodes(){ 388 return this.nNodes; 389 } 390 391 /** 392 * Sets the reference to the tracker node. 393 * @param t the tracker node 394 */ 395 public void setTracker(Node t){ 396 tracker = t; 397 } 398 399 /** 400 * Sets the ID of the local node. 401 * @param id the ID of the node 402 */ 403 public void setThisNodeID(long id) { 404 this.thisNodeID = id; 405 } 406 407 /** 408 * Gets the ID of the local node. 409 * @return the ID of the local node 410 */ 411 public long getThisNodeID(){ 412 return this.thisNodeID; 413 } 414 415 /** 416 * Gets the file status of the local node. 417 * @return the file status of the local node 418 */ 419 public int[] getFileStatus(){ 420 return this.status; 421 } 422 423 /** 424 * Initializes the tracker node. This method 425 * only performs the initialization of the tracker's cache. 426 */ 427 public void initializeTracker() { 428 cache = new Neighbor[nMaxNodes+maxGrowth]; 429 for(int i=0; i<nMaxNodes+maxGrowth; i++){ 430 cache[i]= new Neighbor(); 431 } 432 } 433 434 /** 435 * <p>Checks the number of neighbors and if it is equal to 20 436 * sends a TRACKER messages to the tracker, asking for a new 437 * peer set.</p> 438 * 439 * <p>This method *must* be called after every call of {@link #removeNeighbor} 440 * in {@link #processEvent}. 441 * </p> 442 */ 443 private void processNeighborListSize(Node node, int pid) { 444 if (nNodes==20) { 445 Object ev; 446 long latency; 447 ev = new SimpleMsg(TRACKER, node); 448 Node tracker = ((BitTorrent)node.getProtocol(pid)).tracker; 449 if(tracker != null){ 450 latency = ((Transport)node.getProtocol(tid)).getLatency(node, tracker); 451 EDSimulator.add(latency,ev,tracker,pid); 452 } 453 } 454 } 455 456 /** 457 * The standard method that processes incoming events. 458 * @param node reference to the local node for which the event is going to be processed 459 * @param pid BitTorrent's protocol id 460 * @param event the event to process 461 */ 462 public void processEvent(Node node, int pid, Object event){ //核心函数,处理16种消息和事件,对照手册查看功能 463 464 Object ev; 465 long latency; 466 switch(((SimpleEvent)event).getType()){ 467 468 case KEEP_ALIVE: // 1 469 { 470 Node sender = ((IntMsg)event).getSender(); 471 int isResponse = ((IntMsg)event).getInt(); 472 //System.out.println("process, keep_alive: sender is "+sender.getID()+", local is "+node.getID()); 473 Element e = search(sender.getID()); 474 if(e!= null){ //if I know the sender 475 cache[e.peer].isAlive(); 476 if(isResponse==0 && alive(sender)){ 477 Object msg = new IntMsg(KEEP_ALIVE,node,1); 478 latency = ((Transport)node.getProtocol(tid)).getLatency(node, sender); 479 EDSimulator.add(latency,msg,sender,pid); 480 cache[e.peer].justSent(); 481 } 482 } 483 else{ 484 System.err.println("despite it should never happen, it happened"); 485 ev = new BitfieldMsg(BITFIELD, true, false, node, status, nPieces); 486 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 487 EDSimulator.add(latency,ev,sender,pid); 488 nBitfieldSent++; 489 } 490 491 };break; 492 493 case CHOKE: // 2, CHOKE message. 494 { 495 Node sender = ((SimpleMsg)event).getSender(); 496 //System.out.println("process, choke: sender is "+sender.getID()+", local is "+node.getID()); 497 Element e = search(sender.getID()); 498 if(e!= null){ //if I know the sender 499 cache[e.peer].isAlive(); 500 unchokedBy[e.peer]= false; // I'm choked by it 501 } 502 else{ 503 System.err.println("despite it should never happen, it happened"); 504 ev = new BitfieldMsg(BITFIELD, true, false, node, status, nPieces); 505 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 506 EDSimulator.add(latency,ev,sender,pid); 507 nBitfieldSent++; 508 } 509 };break; 510 511 case UNCHOKE: // 3, UNCHOKE message. 512 { 513 514 515 516 Node sender = ((SimpleMsg)event).getSender(); 517 //System.out.println("process, unchoke: sender is "+sender.getID()+", local is "+node.getID()); 518 Element e = search(sender.getID()); 519 if(e != null){ // If I know the sender 520 int senderIndex = e.peer; 521 cache[senderIndex].isAlive(); 522 /* I send to it some of the pending requests not yet satisfied. */ 523 int t = numberOfDuplicatedRequests; 524 for(int i=4;i>=0 && t>0;i--){ 525 if(pendingRequest[i]==-1) 526 break; 527 if(alive(cache[senderIndex].node) && swarm[senderIndex][decode(pendingRequest[i],0)]==1){ //If the sender has that piece 528 ev = new IntMsg(REQUEST, node,pendingRequest[i] ); 529 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 530 EDSimulator.add(latency,ev, sender,pid); 531 cache[senderIndex].justSent(); 532 } 533 if(!alive(cache[senderIndex].node)){ 534 System.out.println("unchoke1 rm neigh "+ cache[i].node.getID() ); 535 removeNeighbor(cache[senderIndex].node); 536 processNeighborListSize(node,pid); 537 return; 538 } 539 t--; 540 } 541 // I request missing blocks to fill the queue 542 int block = getBlock(); 543 int piece; 544 while(block != -2){ //while still available request to send 545 if(block < 0){ // No more block to request for the current piece 546 piece = getPiece(); 547 if(piece == -1){ // no more piece to request 548 break; 549 } 550 for(int j=0; j<swarmSize; j++){// send the interested message to those 551 // nodes which have that piece 552 lastInterested = piece; 553 if(alive(cache[j].node) && swarm[j][piece]==1){ 554 555 ev = new IntMsg(INTERESTED, node, lastInterested); 556 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[j].node); 557 EDSimulator.add(latency,ev,cache[j].node,pid); 558 cache[j].justSent(); 559 } 560 561 if(!alive(cache[j].node)){ 562 //System.out.println("unchoke2 rm neigh "+ cache[j].node.getID() ); 563 removeNeighbor(cache[j].node); 564 processNeighborListSize(node,pid); 565 } 566 } 567 block = getBlock(); 568 } 569 else{ // block value referred to a real block 570 if(alive(cache[senderIndex].node) && swarm[senderIndex][decode(block,0)]==1 && addRequest(block)){ // The sender has that block 571 ev = new IntMsg(REQUEST, node, block); 572 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 573 EDSimulator.add(latency,ev,sender,pid); 574 cache[senderIndex].justSent(); 575 } 576 else{ 577 if(!alive(cache[senderIndex].node)){ 578 System.out.println("unchoke3 rm neigh "+ cache[senderIndex].node.getID() ); 579 removeNeighbor(cache[senderIndex].node); 580 processNeighborListSize(node,pid); 581 } 582 return; 583 } 584 block = getBlock(); 585 } 586 } 587 unchokedBy[senderIndex] = true; // I add the sender to the list 588 } 589 else // It should never happen. 590 { 591 System.err.println("despite it should never happen, it happened"); 592 for(int i=0; i<swarmSize; i++) 593 if(cache[i].node !=null) 594 System.err.println(cache[i].node.getID()); 595 ev = new BitfieldMsg(BITFIELD, true, false, node, status, nPieces); 596 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 597 EDSimulator.add(latency,ev,sender,pid); 598 nBitfieldSent++; 599 } 600 };break; 601 602 case INTERESTED: // 4, INTERESTED message. 603 { 604 numInterestedPeers++; 605 Node sender = ((IntMsg)event).getSender(); 606 //System.out.println("process, interested: sender is "+sender.getID()+", local is "+node.getID()); 607 int value = ((IntMsg)event).getInt(); 608 Element e = search(sender.getID()); 609 if(e!=null){ 610 cache[e.peer].isAlive(); 611 cache[e.peer].interested = value; 612 } 613 else{ 614 System.err.println("despite it should never happen, it happened"); 615 ev = new BitfieldMsg(BITFIELD, true, false, node, status, nPieces); 616 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 617 EDSimulator.add(latency,ev,sender,pid); 618 nBitfieldSent++; 619 } 620 621 }; break; 622 623 case NOT_INTERESTED: // 5, NOT_INTERESTED message. 624 { 625 numInterestedPeers--; 626 Node sender = ((IntMsg)event).getSender(); 627 //System.out.println("process, not_interested: sender is "+sender.getID()+", local is "+node.getID()); 628 int value = ((IntMsg)event).getInt(); 629 Element e = search(sender.getID()); 630 if(e!=null){ 631 cache[e.peer].isAlive(); 632 if(cache[e.peer].interested == value) 633 cache[e.peer].interested = -1; // not interested 634 } 635 }; break; 636 637 case HAVE: // 6, HAVE message. 638 { 639 640 641 Node sender = ((IntMsg)event).getSender(); 642 //System.out.println("process, have: sender is "+sender.getID()+", local is "+node.getID()); 643 int piece = ((IntMsg)event).getInt(); 644 Element e = search(sender.getID()); 645 if(e!=null){ 646 cache[e.peer].isAlive(); 647 swarm[e.peer][piece]=1; 648 rarestPieceSet[piece]++; 649 boolean isSeeder = true; 650 for(int i=0; i<nPieces; i++){ 651 isSeeder = isSeeder && (swarm[e.peer][i]==1); 652 } 653 e.isSeeder = isSeeder; 654 } 655 else{ 656 System.err.println("despite it should never happen, it happened"); 657 ev = new BitfieldMsg(BITFIELD, true, false, node, status, nPieces); 658 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 659 EDSimulator.add(latency,ev,sender,pid); 660 nBitfieldSent++; 661 } 662 }; break; 663 664 case BITFIELD: // 7, BITFIELD message 665 { 666 667 668 Node sender = ((BitfieldMsg)event).getSender(); 669 int []fileStatus = ((BitfieldMsg)event).getArray(); 670 /*Response with NACK*/ 671 if(!((BitfieldMsg)event).isRequest && !((BitfieldMsg)event).ack){ 672 Element e = search(sender.getID()); 673 if(e == null) // if is a response with nack that follows a request 674 nBitfieldSent--; 675 // otherwise is a response with ack that follows a duplicate 676 // insertion attempt 677 //System.out.println("process, bitfield_resp_nack: sender is "+sender.getID()+", local is "+node.getID()); 678 return; 679 } 680 /*Request with NACK*/ 681 if(((BitfieldMsg)event).isRequest && !((BitfieldMsg)event).ack){ 682 //System.out.println("process, bitfield_req_nack: sender is "+sender.getID()+", local is "+node.getID()); 683 if(alive(sender)){ 684 Element e = search(sender.getID()); 685 ev = new BitfieldMsg(BITFIELD, false, true, node, status, nPieces); //response with ack 686 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 687 EDSimulator.add(latency,ev,sender,pid); 688 cache[e.peer].justSent(); 689 } 690 } 691 /*Response with ACK*/ 692 if(!((BitfieldMsg)event).isRequest && ((BitfieldMsg)event).ack){ 693 nBitfieldSent--; 694 //System.out.println("process, bitfield_resp_ack: sender is "+sender.getID()+", local is "+node.getID()); 695 if(alive(sender)){ 696 if(addNeighbor(sender)){ 697 Element e = search(sender.getID()); 698 cache[e.peer].isAlive(); 699 swarm[e.peer] = fileStatus; 700 boolean isSeeder = true; 701 for(int i=0; i<nPieces; i++){ 702 rarestPieceSet[i]+= fileStatus[i]; 703 isSeeder = isSeeder && (fileStatus[i]==1); 704 } 705 e.isSeeder = isSeeder; 706 707 if(nNodes==10 && !lock){ // I begin to request pieces 708 lock = true; 709 int piece = getPiece(); 710 if(piece == -1) 711 return; 712 lastInterested = piece; 713 currentPiece = lastInterested; 714 ev = new IntMsg(INTERESTED, node, lastInterested); 715 for(int i=0; i<swarmSize; i++){// send the interested message to those 716 // nodes which have that piece 717 if(alive(cache[i].node) && swarm[i][piece]==1){ 718 719 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[i].node); 720 EDSimulator.add(latency,ev,cache[i].node,pid); 721 cache[i].justSent(); 722 } 723 } 724 725 } 726 727 } 728 } 729 else 730 System.out.println("Sender "+sender.getID()+" not alive"); 731 } 732 /*Request with ACK*/ 733 if(((BitfieldMsg)event).isRequest && ((BitfieldMsg)event).ack){ 734 //System.out.println("process, bitfield_req_ack: sender is "+sender.getID()+", local is "+node.getID()); 735 if(alive(sender)){ 736 if(addNeighbor(sender)){ 737 Element e = search(sender.getID()); 738 cache[e.peer].isAlive(); 739 swarm[e.peer] = fileStatus; 740 boolean isSeeder = true; 741 for(int i=0; i<nPieces; i++){ 742 rarestPieceSet[i]+= fileStatus[i]; // I update the rarestPieceSet with the pieces of the new node 743 isSeeder = isSeeder && (fileStatus[i]==1); // I check if the new node is a seeder 744 } 745 e.isSeeder = isSeeder; 746 ev = new BitfieldMsg(BITFIELD, false, true, node, status, nPieces); //response with ack 747 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 748 EDSimulator.add(latency,ev,sender,pid); 749 cache[e.peer].justSent(); 750 if(nNodes==10 && !lock){ // I begin to request pieces 751 int piece = getPiece(); 752 if(piece == -1) 753 return; 754 lastInterested = piece; 755 currentPiece = lastInterested; 756 ev = new IntMsg(INTERESTED, node, lastInterested); 757 for(int i=0; i<swarmSize; i++){// send the interested message to those 758 // nodes which have that piece 759 if(alive(cache[i].node) && swarm[i][piece]==1){ 760 761 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[i].node); 762 EDSimulator.add(latency,ev,cache[i].node,pid); 763 cache[i].justSent(); 764 } 765 } 766 767 } 768 } 769 else { 770 Element e; 771 if((e = search(sender.getID()))!=null){ // The sender was already in the cache 772 cache[e.peer].isAlive(); 773 ev = new BitfieldMsg(BITFIELD, false, true, node, status, nPieces); //response with ack 774 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 775 EDSimulator.add(latency,ev,sender,pid); 776 cache[e.peer].justSent(); 777 } 778 else{ // Was not be possible add the sender (nBitfield+nNodes > swarmSize) 779 ev = new BitfieldMsg(BITFIELD, false, false, node, status, nPieces); //response with nack 780 latency = ((Transport)node.getProtocol(tid)).getLatency(node,sender); 781 EDSimulator.add(latency,ev,sender,pid); 782 } 783 } 784 785 } 786 else 787 System.out.println("Sender "+sender.getID()+" not alive"); 788 } 789 };break; 790 791 case REQUEST: // 8, REQUEST message. 792 { 793 Object evnt; 794 Node sender = ((IntMsg)event).getSender(); 795 int value = ((IntMsg)event).getInt(); 796 Element e; 797 BitTorrent senderP; 798 int remoteRate; 799 int localRate; 800 int bandwidth; 801 int downloadTime; 802 803 e = search(sender.getID()); 804 if (e==null) 805 return; 806 cache[e.peer].isAlive(); 807 808 requestToServe.enqueue(value, sender); 809 810 /*I serve the enqueued requests until 10 uploding pieces or an empty queue*/ 811 while(!requestToServe.empty() && nPiecesUp <10){ 812 Request req = requestToServe.dequeue(); 813 e = search(req.sender.getID()); 814 if(e!=null && alive(req.sender)){ 815 ev = new IntMsg(PIECE, node, req.id); 816 nPiecesUp++; 817 e.valueUP++; 818 senderP = ((BitTorrent)req.sender.getProtocol(pid)); 819 senderP.nPiecesDown++; 820 remoteRate = senderP.maxBandwidth/(senderP.nPiecesUp + senderP.nPiecesDown); 821 localRate = maxBandwidth/(nPiecesUp + nPiecesDown); 822 bandwidth = Math.min(remoteRate, localRate); 823 downloadTime = ((16*8)/(bandwidth))*1000; // in milliseconds 824 latency = ((Transport)node.getProtocol(tid)).getLatency(node,req.sender); 825 EDSimulator.add(latency+downloadTime,ev,req.sender,pid); 826 cache[e.peer].justSent(); 827 /*I send to me an event to indicate that the download is completed. 828 This prevent that, when the receiver death occurres, my value nPiecesUp 829 doesn't decrease.*/ 830 evnt = new SimpleMsg(DOWNLOAD_COMPLETED, req.sender); 831 EDSimulator.add(latency+downloadTime,evnt,node,pid); 832 } 833 } 834 }; break; 835 836 case PIECE: // 9, PIECE message. 837 { 838 Node sender = ((IntMsg)event).getSender(); 839 /* Set the correct value for the local uploading and remote 840 downloading number of pieces */ 841 nPiecesDown--; 842 843 if(peerStatus == 1)// To save CPU cycles 844 return; 845 //System.out.println("process, piece: sender is "+sender.getID()+", local is "+node.getID()); 846 Element e = search(sender.getID()); 847 848 if(e==null){ //I can't accept a piece not wait 849 return; 850 } 851 e.valueDOWN++; 852 853 cache[e.peer].isAlive(); 854 855 int value = ((IntMsg)event).getInt(); 856 int piece = decode(value,0); 857 int block = decode(value,1); 858 /* If the block has not been already downloaded and it belongs to 859 the current downloading piece.*/ 860 if(piece == currentPiece && decode(pieceStatus[block],0)!= piece){ 861 pieceStatus[block] = value; 862 status[piece]++; 863 removeRequest(value); 864 865 requestNextBlocks(node, pid, e.peer); 866 867 }else{ // Either a future piece or an owned piece 868 if(piece!=currentPiece && status[piece]!=16){ // Piece not owned, will be considered later 869 incomingPieces.enqueue(value, sender); 870 } 871 872 } 873 ev = new IntMsg(CANCEL, node, value); 874 /* I send a CANCEL to all nodes to which I previously requested the block*/ 875 for(int i=0; i<swarmSize; i++){ 876 if(alive(cache[i].node) && unchokedBy[i]==true && swarm[i][decode(block,0)]==1 && cache[i].node != sender){ 877 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[i].node); 878 EDSimulator.add(latency,ev,cache[i].node,pid); 879 cache[i].justSent(); 880 } 881 } 882 883 if(status[currentPiece]==16){ // if piece completed, I change the currentPiece to the next wanted 884 nPieceCompleted++; 885 ev = new IntMsg(HAVE, node, currentPiece); 886 for(int i=0; i<swarmSize; i++){ // I send the HAVE for the piece 887 if(alive(cache[i].node)){ 888 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[i].node); 889 EDSimulator.add(latency,ev,cache[i].node,pid); 890 cache[i].justSent(); 891 } 892 if(!alive(cache[i].node)){ 893 //System.out.println("piece3 rm neigh "+ cache[i].node.getID() ); 894 895 removeNeighbor(cache[i].node); 896 processNeighborListSize(node,pid); 897 } 898 } 899 ev = new IntMsg(NOT_INTERESTED, node, currentPiece); 900 for(int i=0; i<swarmSize; i++){ // I send the NOT_INTERESTED to which peer I sent an INTERESTED 901 if(swarm[i][piece]==1 && alive(cache[i].node)){ 902 latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[i].node); 903 EDSimulator.add(latency,ev,cache[i].node,pid); 904 cache[i].justSent(); 905 } 906 if(!alive(cache[i].node)){ 907 //System.out.println("piece4 rm neigh "+ cache[i].node.getID() ); 908 909 removeNeighbor(cache[i].node); 910 processNeighborListSize(node,pid); 911 } 912 } 913 if(nPieceCompleted == nPieces){ 914 System.out.println("FILE COMPLETED for peer "+node.getID()); 915 this.peerStatus = 1; 916 } 917 918 /* I set the currentPiece to the lastInterested. Then I extract 919 the queued received blocks 920 */ 921 922 currentPiece = lastInterested; 923 int m = incomingPieces.dim; 924 while(m > 0){ // I process the queue 925 m--; 926 Request temp = incomingPieces.dequeue(); 927 int p = decode(temp.id,0); // piece id 928 int b = decode(temp.id,1); // block id 929 Element s = search(temp.sender.getID()); 930 if(s==null) // if the node that sent the block in the queue is dead 931 continue; 932 if(p==currentPiece && decode(pieceStatus[b],0)!= p){ 933 pieceStatus[b] = temp.id; 934 status[p]++; 935 removeRequest(temp.id); 936 requestNextBlocks(node, pid, s.peer); 937 } 938 else{ // The piece not currently desired will be moved to the tail 939 if(p!= currentPiece) // If not a duplicate block but belongs to another piece 940 incomingPieces.enqueue(temp.id,temp.sender); 941 else // duplicate block 942 requestNextBlocks(node, pid, s.peer); 943 } 944 } 945 } 946 }; break; 947 948 case CANCEL: 949 { 950 Node sender = ((IntMsg)event).getSender(); 951 int value = ((IntMsg)event).getInt(); 952 requestToServe.remove(sender, value); 953 };break; 954 955 case PEERSET: // PEERSET message 956 { 957 Node sender = ((PeerSetMsg)event).getSender(); 958 //System.out.println("process, peerset: sender is "+sender.getID()+", local is "+node.getID()); 959 Neighbor n[] = ((PeerSetMsg)event).getPeerSet(); 960 961 for(int i=0; i<peersetSize; i++){ 962 if( n[i]!=null && alive(n[i].node) && search(n[i].node.getID())==null && nNodes+nBitfieldSent <swarmSize-2) { 963 ev = new BitfieldMsg(BITFIELD, true, true, node, status, nPieces); 964 latency = ((Transport)node.getProtocol(tid)).getLatency(node,n[i].node); 965 EDSimulator.add(latency,ev,n[i].node,pid); 966 nBitfieldSent++; 967 // Here I should call the Neighbor.justSent(), but here 968 // the node is not yet in the cache. 969 } 970 } 971 }; break; 972 973 case TRACKER: // TRACKER message 974 { 975 976 int j=0; 977 Node sender = ((SimpleMsg)event).getSender(); 978 //System.out.println("process, tracker: sender is "+sender.getID()+", local is "+node.getID()); 979 if(!alive(sender)) 980 return; 981 Neighbor tmp[] = new Neighbor[peersetSize]; 982 int k=0; 983 if(nNodes <= peersetSize){ 984 for(int i=0; i< nMaxNodes+maxGrowth; i++){ 985 if(cache[i].node != null && cache[i].node.getID()!= sender.getID()){ 986 tmp[k]=cache[i]; 987 k++; 988 } 989 } 990 ev = new PeerSetMsg(PEERSET, tmp, node); 991 latency = ((Transport)node.getProtocol(tid)).getLatency(node, sender); 992 EDSimulator.add(latency,ev,sender,pid); 993 return; 994 } 995 996 while(j < peersetSize){ 997 int i = CommonState.r.nextInt(nMaxNodes+maxGrowth); 998 for (int z=0; z<j; z++){ 999 if(cache[i].node==null || tmp[z].node.getID() == cache[i].node.getID() || cache[i].node.getID() == sender.getID()){ 1000 z=0; 1001 i= CommonState.r.nextInt(nMaxNodes+maxGrowth); 1002 } 1003 } 1004 if(cache[i].node != null){ 1005 tmp[j] = cache[i]; 1006 j++; 1007 } 1008 } 1009 ev = new PeerSetMsg(PEERSET, tmp, node); 1010 latency = ((Transport)node.getProtocol(tid)).getLatency(node, sender); 1011 EDSimulator.add(latency,ev,sender,pid); 1012 }; break; 1013 1014 case CHOKE_TIME: //Every 10 secs. 1015 { 1016 n_choke_time++; 1017 1018 ev = new SimpleEvent(CHOKE_TIME); 1019 EDSimulator.add(10000,ev,node,pid); 1020 int j=0; 1021 /*I copy the interested nodes in the byBandwidth array*/ 1022 for(int i=0;i< swarmSize && byPeer[i].peer != -1; i++){ 1023 if(cache[byPeer[i].peer].interested > 0){ 1024 byBandwidth[j]=byPeer[i]; //shallow copy 1025 j++; 1026 } 1027 } 1028 1029 /*It ensures that in the next 20sec, if there are less nodes interested 1030 than now, those in surplus will not be ordered. */ 1031 for(;j<swarmSize;j++){ 1032 byBandwidth[j]=null; 1033 } 1034 sortByBandwidth(); 1035 int optimistic = 3; 1036 int luckies[] = new int[3]; 1037 try{ // It takes the first three neighbors 1038 luckies[0] = byBandwidth[0].peer; 1039 optimistic--; 1040 luckies[1] = byBandwidth[1].peer; 1041 optimistic--; 1042 luckies[2] = byBandwidth[2].peer; 1043 } 1044 catch(NullPointerException e){ // If not enough peer in byBandwidth it chooses the other romdomly 1045 for(int z = optimistic; z>0;z--){ 1046 int lucky = CommonState.r.nextInt(nNodes); 1047 while(cache[byPeer[lucky].peer].status ==1 && alive(cache[byPeer[lucky].peer].node) && 1048 cache[byPeer[lucky].peer].interested == 0)// until the lucky peer is already unchoked or not interested 1049 lucky = CommonState.r.nextInt(nNodes); 1050 luckies[3-z]= byPeer[lucky].peer; 1051 } 1052 } 1053 for(int i=0; i<swarmSize; i++){ // I perform the chokes and the unchokes 1054 if((i==luckies[0] || i==luckies[1] || i==luckies[2]) && alive(cache[i].node) && cache[i].status != 2){ //the unchokes 1055 cache[i].status = 1; 1056 ev = new SimpleMsg(UNCHOKE, node); 1057 latency = ((Transport)node.getProtocol(tid)).getLatency(node, cache[i].node); 1058 EDSimulator.add(latency,ev,cache[i].node,pid); 1059 cache[i].justSent(); 1060 //System.out.println("average time, unchoked: "+cache[i].node.getID()); 1061 } 1062 else{ // the chokes 1063 if(alive(cache[i].node) && (cache[i].status == 1 || cache[i].status == 2)){ 1064 cache[i].status = 0; 1065 ev = new SimpleMsg(CHOKE, node); 1066 latency = ((Transport)node.getProtocol(tid)).getLatency(node, cache[i].node); 1067 EDSimulator.add(latency,ev,cache[i].node,pid); 1068 cache[i].justSent(); 1069 } 1070 } 1071 } 1072 1073 if(n_choke_time%2==0){ //every 20 secs. Used in computing the average download rates 1074 for(int i=0; i<nNodes; i++){ 1075 if(this.peerStatus == 0){ // I'm a leeacher 1076 byPeer[i].head20 = byPeer[i].valueDOWN; 1077 } 1078 else{ 1079 byPeer[i].head20 = byPeer[i].valueUP; 1080 } 1081 } 1082 } 1083 }; break; 1084 1085 case OPTUNCHK_TIME: 1086 { 1087 1088 //System.out.println("process, optunchk_time"); 1089 1090 ev = new SimpleEvent(OPTUNCHK_TIME); 1091 EDSimulator.add(30000,ev,node,pid); 1092 int lucky = CommonState.r.nextInt(nNodes); 1093 while(cache[byPeer[lucky].peer].status ==1)// until the lucky peer is already unchoked 1094 lucky = CommonState.r.nextInt(nNodes); 1095 if(!alive(cache[byPeer[lucky].peer].node)) 1096 return; 1097 cache[byPeer[lucky].peer].status = 1; 1098 Object msg = new SimpleMsg(UNCHOKE,node); 1099 latency = ((Transport)node.getProtocol(tid)).getLatency(node, cache[byPeer[lucky].peer].node); 1100 EDSimulator.add(latency,msg,cache[byPeer[lucky].peer].node,pid); 1101 cache[byPeer[lucky].peer].justSent(); 1102 }; break; 1103 1104 case ANTISNUB_TIME: 1105 { 1106 if(this.peerStatus == 1) // I'm a seeder, I don't update the event 1107 return; 1108 //System.out.println("process, antisnub_time"); 1109 for(int i=0; i<nNodes; i++){ 1110 if(byPeer[i].valueDOWN >0 && (byPeer[i].valueDOWN - byPeer[i].head60)==0){// No blocks downloaded in 1 min 1111 cache[byPeer[i].peer].status = 2; // I'm snubbed by it 1112 } 1113 byPeer[i].head60 = byPeer[i].valueDOWN; 1114 } 1115 ev = new SimpleEvent(ANTISNUB_TIME); 1116 EDSimulator.add(60000,ev,node,pid); 1117 long time = CommonState.getTime(); 1118 }; break; 1119 1120 case CHECKALIVE_TIME: 1121 { 1122 1123 //System.out.println("process, checkalive_time"); 1124 1125 long now = CommonState.getTime(); 1126 for(int i=0; i<swarmSize; i++){ 1127 /*If are at least 2 minutes (plus 1 sec of tolerance) that 1128 I don't send anything to it.*/ 1129 if(alive(cache[i].node) && (cache[i].lastSent < (now-121000))){ 1130 Object msg = new IntMsg(KEEP_ALIVE,node,0); 1131 latency = ((Transport)node.getProtocol(tid)).getLatency(node, cache[i].node); 1132 EDSimulator.add(latency,msg,cache[i].node,pid); 1133 cache[i].justSent(); 1134 } 1135 /*If are at least 2 minutes (plus 1 sec of tolerance) that I don't 1136 receive anything from it though I sent a keepalive 2 minutes ago*/ 1137 else{ 1138 if(cache[i].lastSeen <(now-121000) && cache[i].node != null && cache[i].lastSent < (now-121000)){ 1139 System.out.println("process, checkalive_time, rm neigh " + cache[i].node.getID()); 1140 if(cache[i].node.getIndex() != -1){ 1141 System.out.println("This should never happen: I remove a node that is not effectively died"); 1142 } 1143 removeNeighbor(cache[i].node); 1144 processNeighborListSize(node,pid); 1145 } 1146 } 1147 } 1148 ev = new SimpleEvent(CHECKALIVE_TIME); 1149 EDSimulator.add(120000,ev,node,pid); 1150 }; break; 1151 1152 case TRACKERALIVE_TIME: 1153 { 1154 //System.out.println("process, trackeralive_time"); 1155 if(alive(tracker)){ 1156 ev = new SimpleEvent(TRACKERALIVE_TIME); 1157 EDSimulator.add(1800000,ev,node,pid); 1158 } 1159 else 1160 tracker=null; 1161 1162 }; break; 1163 1164 case DOWNLOAD_COMPLETED: 1165 { 1166 nPiecesUp--; 1167 }; break; 1168 1169 } 1170 } 1171 1172 /** 1173 * Given a piece index and a block index it encodes them in an unique integer value. 1174 * @param piece the index of the piece to encode. 1175 * @param block the index of the block to encode. 1176 * @return the encoding of the piece and the block indexes. 1177 */ 1178 private int encode(int piece, int block){ //对piece和blockID编码,例如pieceID是1234,其中第2个块的ID就是123402 1179 return (piece*100)+block; 1180 1181 } 1182 /** 1183 * Returns either the piece or the block that contained in the <tt>value</tt> depending 1184 * on <tt>part</tt>: 0 means the piece value, 1 the block value. 1185 * @param value the ID of the block to decode. 1186 * @param part the information to extract from <tt>value</tt>. 0 means the piece index, 1 the block index. 1187 * @return the piece or the block index depending about the value of <tt>part</tt> 1188 */ 1189 private int decode(int value, int part){ //解码,0返回pieceID,1返回blockID 1190 if (value==-1) // Not a true value to decode 1191 return -1; 1192 if(part == 0) // I'm interested to the piece 1193 return value/100; 1194 else // I'm interested to the block 1195 return value%100; 1196 } 1197 1198 /** 1199 * Used by {@link NodeInitializer#choosePieces(int, BitTorrent) NodeInitializer} to set 1200 * the number of piece completed from the beginning in according with 1201 * the distribution in the configuration file. 1202 * @param number the number of piece completed 1203 */ 1204 public void setCompleted(int number){ 1205 this.nPieceCompleted = number; 1206 } 1207 1208 /** 1209 * Sets the status (the set of blocks) of the file for the current node. 1210 * Note that a piece is considered <i>completed</i> if the number 1211 * of downloaded blocks is 16. 1212 * @param index The index of the piece 1213 * @param value Number of blocks downloaded for the piece index. 1214 */ 1215 public void setStatus(int index, int value){ 1216 status[index]=value; 1217 } 1218 1219 /** 1220 * Sets the status of the local node. 1221 * @param status The status of the node: 1 means seeder, 0 leecher 1222 */ 1223 public void setPeerStatus(int status){ 1224 this.peerStatus = status; 1225 } 1226 1227 /** 1228 * Gets the status of the local node. 1229 * @return The status of the local node: 1 means seeder, 0 leecher 1230 */ 1231 public int getPeerStatus(){ 1232 return peerStatus; 1233 } 1234 1235 /** 1236 * Gets the number of blocks for a given piece owned by the local node. 1237 * @param index The index of the piece 1238 * @return Number of blocks downloaded for the piece index 1239 */ 1240 public int getStatus(int index){ 1241 return status[index]; 1242 } 1243 1244 /** 1245 * Sets the maximum bandwdith for the local node. 1246 * @param value The value of bandwidth in Kbps 1247 */ 1248 public void setBandwidth(int value){ 1249 maxBandwidth = value; 1250 } 1251 1252 /** 1253 * Checks if a node is still alive in the simulated network. 1254 * @param node The node to check 1255 * @return true if the node <tt>node</tt> is up, false otherwise 1256 * @see peersim.core.GeneralNode#isUp 1257 */ 1258 public boolean alive(Node node){ 1259 if(node == null) 1260 return false; 1261 else 1262 return node.isUp(); 1263 } 1264 1265 /** 1266 * Adds a neighbor to the cache of the local node. 1267 * The new neighbor is put in the first null position. 1268 * @param neighbor The neighbor node to add 1269 * @return <tt>false</tt> if the neighbor is already present in the cache (this can happen when the peer requests a 1270 * new peer set to the tracker an there is still this neighbor within) or no place is available. 1271 * Otherwise, returns true if the node is correctly added to the cache. 1272 */ 1273 public boolean addNeighbor(Node neighbor){ 1274 if(search(neighbor.getID()) !=null){// if already exists 1275 // System.err.println("Node "+neighbor.getID() + " not added, already exist."); 1276 return false; 1277 } 1278 if(this.tracker == null){ // I'm in the tracker's BitTorrent protocol 1279 for(int i=0; i< nMaxNodes+maxGrowth; i++){ 1280 if(cache[i].node == null){ 1281 cache[i].node = neighbor; 1282 cache[i].status = 0; //chocked 1283 cache[i].interested = -1; //not interested 1284 this.nNodes++; 1285 1286 //System.err.println("i: " + i +" nMaxNodes: " + nMaxNodes); 1287 return true; 1288 } 1289 } 1290 } 1291 else{ 1292 if((nNodes+nBitfieldSent) < swarmSize){ 1293 //System.out.println("I'm the node " + this.thisNodeID + ", trying to add node "+neighbor.getID()); 1294 for(int i=0; i<swarmSize; i++){ 1295 if(cache[i].node == null){ 1296 cache[i].node = neighbor; 1297 cache[i].status = 0; //choked 1298 cache[i].interested = -1; // not interested 1299 byPeer[nNodes].peer = i; 1300 byPeer[nNodes].ID = neighbor.getID(); 1301 sortByPeer(); 1302 this.nNodes++; 1303 //System.out.println(neighbor.getID()+" added!"); 1304 return true; 1305 } 1306 } 1307 System.out.println("Node not added, no places available"); 1308 } 1309 } 1310 return false; 1311 } 1312 1313 /** 1314 * Removes a neighbor from the cache of the local node. 1315 * @param neighbor The node to remove 1316 * @return true if the node is correctly removed, false otherwise. 1317 */ 1318 public boolean removeNeighbor(Node neighbor) { 1319 1320 if (neighbor == null) 1321 return true; 1322 1323 // this is the tracker's bittorrent protocol 1324 if (this.tracker == null) { 1325 for (int i=0; i< (nMaxNodes+maxGrowth); i++) { 1326 1327 // check the feasibility of the removal 1328 if ( (cache[i] != null) && (cache[i].node != null) && 1329 (cache[i].node.getID() == neighbor.getID()) ) { 1330 cache[i].node = null; 1331 this.nNodes--; 1332 return true; 1333 } 1334 } 1335 return false; 1336 } 1337 // this is the bittorrent protocol of a peer 1338 else { 1339 1340 Element e = search(neighbor.getID()); 1341 1342 if (e != null) { 1343 for (int i=0; i<nPieces; i++) { 1344 rarestPieceSet[i] -= swarm[e.peer][i]; 1345 swarm[e.peer][i] = 0; 1346 } 1347 1348 cache[e.peer].node = null; 1349 cache[e.peer].status = 0; 1350 cache[e.peer].interested = -1; 1351 unchokedBy[e.peer] = false; 1352 this.nNodes--; 1353 e.peer = -1; 1354 e.ID = Integer.MAX_VALUE; 1355 e.valueUP = 0; 1356 e.valueDOWN = 0; 1357 e.head20 = 0; 1358 e.head60 = 0; 1359 sortByPeer(); 1360 1361 return true; 1362 } 1363 } 1364 return false; 1365 } 1366 1367 /** 1368 * Adds a request to the pendingRequest queue. 1369 * @param block The requested block 1370 * @return true if the request has been successfully added to the queue, false otherwise 1371 */ 1372 private boolean addRequest(int block){ 1373 int i=4; 1374 while(i>=0 && pendingRequest[i]!=-1){ 1375 i--; 1376 } 1377 if(i>=0){ 1378 pendingRequest[i] = block; 1379 return true; 1380 } 1381 else { // It should never happen 1382 //System.err.println("pendingRequest queue full"); 1383 return false; 1384 } 1385 } 1386 1387 /** 1388 * Removes the block with the given <tt>id</tt> from the {@link #pendingRequest} queue 1389 * and sorts the queue leaving the empty cell at the left. 1390 * @param id the id of the requested block 1391 */ 1392 private void removeRequest(int id){ 1393 int i = 4; 1394 for(; i>=0; i--){ 1395 if(pendingRequest[i]==id) 1396 break; 1397 } 1398 for(; i>=0; i--){ 1399 if(i==0) 1400 pendingRequest[i] = -1; 1401 else 1402 pendingRequest[i] = pendingRequest[i-1]; 1403 } 1404 } 1405 1406 /** 1407 * Requests new block until the {@link #pendingRequest} is full to the sender of the just received piece. 1408 * It calls {@link #getNewBlock(Node, int)} to implement the <i>strict priority</i> strategy. 1409 * @param node the local node 1410 * @param pid the BitTorrent protocol id 1411 * @param sender the sender of the just received received piece. 1412 */ 1413 private void requestNextBlocks(Node node, int pid, int sender){ 1414 int block = getNewBlock(node, pid); 1415 while(block != -2){ 1416 if(unchokedBy[sender]==true && alive(cache[sender].node) && addRequest(block)){ 1417 Object ev = new IntMsg(REQUEST, node, block); 1418 long latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[sender].node); 1419 EDSimulator.add(latency,ev,cache[sender].node,pid); 1420 cache[sender].justSent(); 1421 } 1422 else{ // I cannot send request 1423 if(!alive(cache[sender].node) && cache[sender].node!=null){ 1424 System.out.println("piece2 rm neigh "+ cache[sender].node.getID() ); 1425 removeNeighbor(cache[sender].node); 1426 processNeighborListSize(node,pid); 1427 } 1428 return; 1429 } 1430 block = getNewBlock(node, pid); 1431 } 1432 } 1433 1434 /** 1435 * It returns the id of the next block to request. Sends <tt>INTERESTED</tt> if the new 1436 * block belongs to a new piece. 1437 * It uses {@link #getBlock()} to get the next block of a piece and calls {@link #getPiece()} 1438 * when all the blocks for the {@link #currentPiece} have been requested. 1439 * @param node the local node 1440 * @param pid the BitTorrent protocol id 1441 * @return -2 if no more places available in the <tt>pendingRequest</tt> queue;<br/> 1442 * the value of the next block to request otherwise</p> 1443 */ 1444 private int getNewBlock(Node node, int pid){ //返回下一个请求的块,包括了队列满等情况,申请新的piece,返回对应的块 1445 int block = getBlock(); 1446 if(block < 0){ // No more block to request for the current piece 1447 1448 if(block ==-2) // Pending request queue full 1449 return -2; 1450 1451 int newPiece = getPiece(); 1452 if(newPiece == -1){ // no more piece to request 1453 return -2; 1454 } 1455 1456 lastInterested = newPiece; 1457 Object ev = new IntMsg(INTERESTED, node, lastInterested); 1458 1459 for(int j=0; j<swarmSize; j++){// send the interested message to those 1460 // nodes which have that piece 1461 if(alive(cache[j].node) && swarm[j][newPiece]==1){ 1462 long latency = ((Transport)node.getProtocol(tid)).getLatency(node,cache[j].node); 1463 EDSimulator.add(latency,ev,cache[j].node,pid); 1464 cache[j].justSent(); 1465 } 1466 if(!alive(cache[j].node)){ 1467 //System.out.println("piece1 rm neigh "+ cache[j].node.getID() ); 1468 1469 removeNeighbor(cache[j].node); 1470 processNeighborListSize(node,pid); 1471 } 1472 } 1473 block = getBlock(); 1474 return block; 1475 } 1476 else{ 1477 // block value referred to a real block 1478 return block; 1479 } 1480 } 1481 1482 /** 1483 * Returns the next block to request for the {@link #currentPiece}. 1484 * @return an index of a block of the <tt>currentPiece</tt> if there are still 1485 * available places in the {@link #pendingRequest} queue;<br/> 1486 * -2 if the <tt>pendingRequest</tt> queue is full;<br/> 1487 * -1 if no more blocks to request for the current piece. 1488 */ 1489 private int getBlock(){ //返回请求的块,返回-2说明待处理队列满,-1说明当前piece没有需要下的块 1490 int i=4; 1491 while(i>=0 && pendingRequest[i]!=-1){ // i is the first empty position from the head 1492 i--; 1493 } 1494 if(i==-1){// No places in the pendingRequest available 1495 //System.out.println("Pendig request queue full!"); 1496 return -2; 1497 } 1498 int j; 1499 //The queue is not empty & last requested block belongs to lastInterested piece 1500 if(i!=4 && decode(pendingRequest[i+1],0)==lastInterested) 1501 j=decode(pendingRequest[i+1],1)+1; // the block following the last requested 1502 else // I don't know which is the next block, so I search it. 1503 j=0; 1504 /* I search another block until the current has been already received. 1505 * If in pieceStatus at position j there is a block that belongs to 1506 * lastInterested piece, means that the block j has been already 1507 * received, otherwise I can request it. 1508 */ 1509 while(j<16 && decode(pieceStatus[j],0)==lastInterested){ 1510 j++; 1511 } 1512 if(j==16) // No more block to request for lastInterested piece 1513 return -1; 1514 return encode(lastInterested,j); 1515 } 1516 1517 /** 1518 * Returns the next correct piece to download. It choose the piece by using the 1519 * <i>random first</i> and <i>rarest first</i> policy. For the beginning 4 pieces 1520 * of a file the first one is used then the pieces are chosen using <i>rarest first</i>. 1521 * @see "Documentation about the BitTorrent module" 1522 * @return the next piece to download. If the whole file has been requested 1523 * -1 is returned. 1524 */ 1525 private int getPiece(){ //返回下一个需要下载的合适的piece(随机优先和最少优先两种算法) 1526 int piece = -1; 1527 if(nPieceCompleted < 4){ //Uses random first piece //一开始,使用随机优先,随机选择piece,先下载一个完整的piece再说 1528 piece = CommonState.r.nextInt(nPieces); 1529 while(status[piece]==16 || piece == currentPiece) // until the piece is owned 1530 piece = CommonState.r.nextInt(nPieces); 1531 return piece; 1532 } 1533 else{ //Uses rarest piece first //否则就说明已经拥有完整的piece了,就可以使用最少优先算法 1534 int j=0; 1535 for(; j<nPieces; j++){ // I find the first not owned piece 1536 if(status[j]==0){ 1537 piece = j; 1538 if(piece != lastInterested) // teoretically it works because 1539 // there should be only one interested 1540 // piece not yet downloaded 1541 break; 1542 } 1543 } 1544 if(piece==-1){ // Never entered in the previous 'if' statement; for all 1545 // pieces an has been sent 1546 return -1; 1547 } 1548 1549 int rarestPieces[] = new int[nPieces-j]; // the pieces with the less number of occurences 1550 rarestPieces[0] = j; 1551 int nValues = 1; // number of pieces less distributed in the network 1552 for(int i=j+1; i<nPieces; i++){ // Finds the rarest piece not owned 1553 if(rarestPieceSet[i]< rarestPieceSet[rarestPieces[0]] && status[i]==0){ // if strictly less than the current one 1554 rarestPieces[0] = i; 1555 nValues = 1; 1556 } 1557 if(rarestPieceSet[i]==rarestPieceSet[rarestPieces[0]] && status[i]==0){ // if equal 1558 rarestPieces[nValues] = i; 1559 nValues++; 1560 } 1561 } 1562 1563 piece = CommonState.r.nextInt(nValues); // one of the less owned pieces 1564 return rarestPieces[piece]; 1565 } 1566 } 1567 1568 /** 1569 * Returns the file's size as number of pieces of 256KB. 1570 * @return number of pieces that compose the file. 1571 */ 1572 public int getNPieces(){ //返回文件大小(有多少piece) 1573 return nPieces; 1574 } 1575 /** 1576 * Clone method of the class. Returns a deep copy of the BitTorrent class. Used 1577 * by the simulation to initialize the {@link peersim.core.Network} 1578 * @return the deep copy of the BitTorrent class. 1579 */ 1580 public Object clone(){ 1581 Object prot = null; 1582 try{ 1583 prot = (BitTorrent)super.clone(); 1584 } 1585 catch(CloneNotSupportedException e){}; 1586 1587 ((BitTorrent)prot).cache = new Neighbor[swarmSize]; 1588 for(int i=0; i<swarmSize; i++){ 1589 ((BitTorrent)prot).cache[i] = new Neighbor(); 1590 } 1591 1592 ((BitTorrent)prot).byPeer = new Element[swarmSize]; 1593 for(int i=0; i<swarmSize; i++){ 1594 ((BitTorrent)prot).byPeer[i] = new Element(); 1595 } 1596 1597 ((BitTorrent)prot).unchokedBy = new boolean[swarmSize]; 1598 1599 ((BitTorrent)prot).byBandwidth = new Element[swarmSize]; 1600 ((BitTorrent)prot).status = new int[nPieces]; 1601 ((BitTorrent)prot).pieceStatus = new int[16]; 1602 for(int i=0; i<16;i++) 1603 ((BitTorrent)prot).pieceStatus[i] = -1; 1604 ((BitTorrent)prot).pendingRequest = new int[5]; 1605 for(int i=0; i<5;i++) 1606 ((BitTorrent)prot).pendingRequest[i] = -1; 1607 ((BitTorrent)prot).rarestPieceSet = new int[nPieces]; 1608 for(int i=0; i<nPieces;i++) 1609 ((BitTorrent)prot).rarestPieceSet[i] = 0; 1610 ((BitTorrent)prot).swarm = new int[swarmSize][nPieces]; 1611 ((BitTorrent)prot).requestToServe = new Queue(20); 1612 ((BitTorrent)prot).incomingPieces = new Queue(100); 1613 return prot; 1614 } 1615 1616 /** 1617 * Sorts {@link #byPeer} array by peer's ID. It implements the <i>InsertionSort</i> 1618 * algorithm. 1619 */ 1620 public void sortByPeer(){ //按节点顺序排序,插入排序算法 1621 int i; 1622 1623 for(int j=1; j<swarmSize; j++) // out is dividing line 1624 { 1625 Element key = new Element(); 1626 byPeer[j].copyTo(key) ; // remove marked item 1627 i = j-1; // start shifts at out 1628 while(i>=0 && (byPeer[i].ID > key.ID)) // until one is smaller, 1629 { 1630 byPeer[i].copyTo(byPeer[i+1]); // shift item right, 1631 i--; // go left one position 1632 } 1633 key.copyTo(byPeer[i+1]); // insert marked item 1634 } 1635 1636 } 1637 1638 /** 1639 * Sorts the array {@link #byBandwidth} using <i>QuickSort</i> algorithm. 1640 * <tt>null</tt> elements and seeders are moved to the end of the array. 1641 */ 1642 public void sortByBandwidth() { //按传输带宽给节点排序,使用快排算法 1643 quicksort(0, swarmSize-1); 1644 } 1645 1646 /** 1647 * Used by {@link #sortByBandwidth()}. It's the implementation of the 1648 * <i>QuickSort</i> algorithm. 1649 * @param left the leftmost index of the array to sort. 1650 * @param right the rightmost index of the array to sort. 1651 */ 1652 private void quicksort(int left, int right) { 1653 if (right <= left) return; 1654 int i = partition(left, right); 1655 quicksort(left, i-1); 1656 quicksort(i+1, right); 1657 } 1658 1659 /** 1660 * Used by {@link #quicksort(int, int)}, partitions the subarray to sort returning 1661 * the splitting point as stated by the <i>QuickSort</i> algorithm. 1662 * @see "The <i>QuickSort</i> algorithm". 1663 */ 1664 private int partition(int left, int right) { //快排算法的中间函数 1665 int i = left - 1; 1666 int j = right; 1667 while (true) { 1668 while (greater(byBandwidth[++i], byBandwidth[right])) // find item on left to swap 1669 ; // a[right] acts as sentinel 1670 while (greater(byBandwidth[right], byBandwidth[--j])) { // find item on right to swap 1671 if (j == left) break; // don't go out-of-bounds 1672 } 1673 if (i >= j) break; // check if pointers cross 1674 swap(i, j); // swap two elements into place 1675 } 1676 swap(i, right); // swap with partition element 1677 return i; 1678 } 1679 1680 /** 1681 * Aswers to the question "is x > y?". Compares the {@link Element}s given as 1682 * parameters. <tt>Element x</tt> is greater than <tt>y</tt> if isn't <tt>null</tt> 1683 * and in the last 20 seconds the local node has downloaded ("uploaded" if the local node is a 1684 * seeder) more blocks than from <tt>y</tt>. 1685 * @param x the first <tt>Element</tt> to compare. 1686 * @param y the second <tt>Element</tt> to compare 1687 * @return <tt>true</tt> if x > y;<br/> 1688 * <tt>false</tt> otherwise. 1689 */ 1690 private boolean greater(Element x, Element y) { //中间函数,用于比较两个节点的大小,用在排序算法里 1691 /* 1692 * Null elements and seeders are shifted at the end 1693 * of the array 1694 */ 1695 if (x==null) return false; 1696 if (y==null) return true; 1697 if (x.isSeeder) return false; 1698 if (y.isSeeder) return true; 1699 1700 // if the local node is a leecher 1701 if (peerStatus==0) { 1702 if ((x.valueDOWN - x.head20) > 1703 (y.valueDOWN -y.head20)) 1704 return true; 1705 else return false; 1706 } 1707 1708 // if peerStatus==1 (the local node is a seeder) 1709 else { 1710 if ((x.valueUP - x.head20) > 1711 (y.valueUP -y.head20)) 1712 return true; 1713 else return false; 1714 } 1715 } 1716 1717 /** 1718 * Swaps {@link Element} <tt>i</tt> with <tt>j</tt> in the {@link #byBandwidth}.<br/> 1719 * Used by {@link #partition(int, int)} 1720 * @param i index of the first element to swap 1721 * @param j index of the second element to swap 1722 */ 1723 private void swap(int i, int j) { //用下面的查询节点的中间函数 1724 Element swap = byBandwidth[i]; 1725 byBandwidth[i] = byBandwidth[j]; 1726 byBandwidth[j] = swap; 1727 } 1728 1729 /** Searches the node with the given ID. It does a dychotomic 1730 * search. 1731 * @param ID ID of the node to search. 1732 * @return the {@link Element} in {@link #byPeer} which represents the node with the 1733 * given ID. 1734 */ 1735 public Element search(long ID){ //二分法来查找节点 1736 int low = 0; 1737 int high = swarmSize-1; 1738 int p = low+((high-low)/2); //Initial probe position 1739 while ( low <= high) { 1740 if ( byPeer[p] == null || byPeer[p].ID > ID) 1741 high = p - 1; 1742 else { 1743 if( byPeer[p].ID < ID ) //Wasteful second comparison forced by syntax limitations. 1744 low = p + 1; 1745 else 1746 return byPeer[p]; 1747 } 1748 p = low+((high-low)/2); //Next probe position. 1749 } 1750 return null; 1751 } 1752 } 1753 1754 /** 1755 * This class is used to store the main informations about a neighbors regarding 1756 * the calculation of the Downloading/Uploading rates. Is the class of items in 1757 * {@link example.bittorrent.BitTorrent#byPeer} and {@link example.bittorrent.BitTorrent#byBandwidth}. 1758 */ 1759 class Element{ //用来存储邻居节点的一些关于上传下载速率信息的类 1760 /** 1761 * ID of the represented node. 1762 */ 1763 public long ID = Integer.MAX_VALUE; //索引,用来查询节点 1764 /** 1765 * Index position of the node in the {@link example.bittorrent.BitTorrent#cache} array. 1766 */ 1767 public int peer = -1; 1768 /** 1769 * Number of blocks uploaded to anyone since the beginning. 1770 */ 1771 public int valueUP = 0; //从开始上传过的块的数量 1772 /** 1773 * Number of blocks downloaded from anyone since the beginning. 1774 */ 1775 public int valueDOWN = 0; //从开始下载过的块的数量 1776 /** 1777 * Value of either {@link #valueUP} or {@link #valueDOWN} (depending by 1778 * {@link example.bittorrent.BitTorrent#peerStatus}) 20 seconds before. 1779 */ 1780 public int head20 = 0; //前20S内上传和下载总的流量,同下面60S前 1781 /** 1782 * Value of either {@link #valueUP} or {@link #valueDOWN} (depending by 1783 * {@link example.bittorrent.BitTorrent#peerStatus}) 60 seconds before. 1784 */ 1785 public int head60 = 0; 1786 /** 1787 * <tt>true</tt> if the node is a seeder, <tt>false</tt> otherwise. 1788 */ 1789 public boolean isSeeder = false; 1790 /** 1791 * Makes a deep copy of the Element to <tt>destination</tt> 1792 * @param destination Element instance where to make the copy 1793 */ 1794 public void copyTo(Element destination){ 1795 destination.ID = this.ID; 1796 destination.peer = this.peer; 1797 destination.valueUP = this.valueUP; 1798 destination.valueDOWN = this.valueDOWN; 1799 destination.head20 = this.head20; 1800 destination.head60 = this.head60; 1801 } 1802 } 1803 1804 /** 1805 * This class stores information about the neighbors regarding their status. It is 1806 * the type of the items in the {@link example.bittorrent.BitTorrent#cache}. 1807 */ 1808 class Neighbor{ //邻居节点类 1809 /** 1810 * Reference to the node in the {@link peersim.core.Network}. 1811 */ 1812 public Node node = null; 1813 /** 1814 * -1 means not interested<br/> 1815 * Other values means the last piece number for which the node is interested. 1816 */ 1817 public int interested; 1818 /** 1819 * 0 means CHOKED<br/> 1820 * 1 means UNCHOKED<br/> 1821 * 2 means SNUBBED_BY. If this value is set and the node is to be unchocked, 1822 * value 2 has the priority. 1823 */ 1824 public int status; //记录阻塞,疏通和拒绝状态 1825 /** 1826 * Last time a message from the node represented has been received. 1827 */ 1828 public long lastSeen = 0; 1829 /** 1830 * Last time a message to the node represented has been sent. 1831 */ 1832 public long lastSent = 0; 1833 1834 /** 1835 * Sets the last time the neighbor was seen. 1836 */ 1837 public void isAlive(){ //更新最后一次接受节点信息的时间 1838 long now = CommonState.getTime(); 1839 this.lastSeen = now; 1840 } 1841 1842 /* 1843 * Sets the last time the local peer sent something to the neighbor. 1844 */ 1845 public void justSent(){ //更新最后一次给节点发送信息的时间 1846 long now = CommonState.getTime(); 1847 this.lastSent = now; 1848 } 1849 1850 } 1851 1852 /** 1853 * Class type of the queues's items in {@link example.bittorrent.BitTorrent#incomingPieces} 1854 * and {@link example.bittorrent.BitTorrent#requestToServe}. 1855 */ 1856 class Queue{ //请求的队列类,包含基本的出入队,判空,查询移除功能 1857 int maxSize; 1858 int head = 0; 1859 int tail = 0; 1860 int dim = 0; 1861 Request queue[]; 1862 1863 /** 1864 * Public constructor. Creates a queue of size <tt>size</tt>. 1865 */ 1866 public Queue(int size){ 1867 maxSize = size; 1868 queue = new Request[size]; 1869 for(int i=0; i< size; i++) 1870 queue[i]= new Request(); 1871 } 1872 1873 /** 1874 * Enqueues the request of the block <tt>id</tt> and its <tt>sender</tt> 1875 * @param id the id of the block in the request 1876 * @param sender a reference to the sender of the request 1877 * @return <tt>true</tt> if the request has been correctly added, <tt>false</tt> 1878 * otherwise. 1879 */ 1880 public boolean enqueue(int id, Node sender){ 1881 if(dim < maxSize){ 1882 queue[tail%maxSize].id = id; 1883 queue[tail%maxSize].sender = sender; 1884 tail++; 1885 dim++; 1886 return true; 1887 } 1888 else return false; 1889 } 1890 1891 /** 1892 * Returns the {@link Request} in the head of the queue. 1893 * @return the element in the head.<br/> 1894 * <tt>null</tt> if the queue is empty. 1895 */ 1896 public Request dequeue(){ 1897 Request value; 1898 if(dim > 0){ 1899 value = queue[head%maxSize]; 1900 head++; 1901 dim--; 1902 return value; 1903 } 1904 else return null; //empty queue 1905 } 1906 1907 /** 1908 * Returns the status of the queue. 1909 * @return <tt>true</tt> if the queue is empty, <tt>false</tt> 1910 * otherwise. 1911 */ 1912 public boolean empty(){ 1913 return (dim == 0); 1914 } 1915 1916 /** 1917 * Returns <tt>true</tt> if block given as parameter is in. 1918 * @param value the id of the block to search. 1919 * @return <tt>true</tt> if the block <tt>value</tt> is in the queue, <tt>false</tt> 1920 * otherwise. 1921 */ 1922 public boolean contains(int value){ 1923 if(empty()) 1924 return false; 1925 for(int i=head; i<head+dim; i++){ 1926 if(queue[i%maxSize].id == value) 1927 return true; 1928 } 1929 return false; 1930 } 1931 1932 /** 1933 * Removes a request from the queue. 1934 * @param sender the sender of the request. 1935 * @param value the id of the block requested. 1936 * @return <tt>true</tt> if the request has been correctly removed, <tt>false</tt> 1937 * otherwise. 1938 */ 1939 public boolean remove(Node sender, int value){ 1940 if(empty()) 1941 return false; 1942 for(int i=head; i<head+dim; i++){ 1943 if(queue[i%maxSize].id == value && queue[i%maxSize].sender == sender){ 1944 for(int j=i; j>head; j--){ // Shifts the elements for the removal 1945 queue[j%maxSize]= queue[(j-1)%maxSize]; 1946 } 1947 head++; 1948 dim--; 1949 return true; 1950 } 1951 } 1952 return false; 1953 } 1954 } 1955 1956 /** 1957 * This class represent an enqueued request of a block. 1958 */ 1959 class Request{ //请求类,包括块的ID和请求的发送方 1960 /** 1961 * The id of the block. 1962 */ 1963 public int id; 1964 /** 1965 * The sender of the request. 1966 */ 1967 public Node sender; 1968 }
BT网络中涉及到了许多小的算法,例如块的选择,如何确定阻塞疏通状态等,这里附上
下载文件片断的选择(Piece Selection)
选择一个好的顺序来下载片断,对提高性能非常重要。一个差的片断选择算法可能导致所有的片断都处于下载中,或者另一种情况,没有任何片断被上载给其它 peers。
1)严格的优先级(Strict Priority)
片断选择的第一个策略是:一旦请求了某个片断的子片断,那么该片断剩下的子片断优先被请求。这样,可以尽可能快的获得一个完整的片断。
2)最少的优先(Rarest First)
对一个下载者来说,在选择下一个被下载的片断时,通常选择的是它的peers们所拥有的最少的那个片断,也就是所谓的“最少优先”。这种技术,确保了每个下载者都拥有它的peers们最希望得到的那些片断,从而一旦有需要,上载就可以开始。这也确保了那些越普通的片断越放在最后下载,从而减少了这样一种可能性,即某个peer当前正提供上载,而随后却没有任何的被别人感兴趣的片断了。也就说说,每个peer都优先选择整个系统中最少的那些片断去下载,而那些在系统中相对较多的片断,放在后面下载,这样,整个系统就趋向于一种更优的状态。如果不用这种算法,大家都去下载最多的那些片断,那么这些片断就会在系统中分布的越来越多,而那些在系统中相对较少的片断仍然很少,最后,某些 peer 就不再拥有其它 peer 感兴趣的片断了,那么系统的参与者越来越少,整个系统的性能就下降。
在BT系统中,充分考虑了经济学的概念,处处从整个系统的性能出发,参与者越多,系统越优化。
信息理论显示除非种子上传了文件的所有片断,否则没有任何下载者可以完成所有文件的下载。如果在一个部署中,只有一个种子,而且种子的上载能力比它的大多数下载者都要差,那么,不同的下载者从种子那里下载不同的片断,性能就会变得比较好,因为,重复的下载浪费了种子获取更多信息的机会。“最少优先”使得下载者只从种子处下载新的片断(也就是整个系统中其它peer都没有的片断),因为,下载者能够看到其它peers那里已经有了种子已经上传的片断。
在某些部署中,原始的种子由于某些原因最终关闭,只好由剩下的这些下载者们来负责上传。这样显然会带来一个风险:某些片断任何一个下载者都不拥有。“最少优先”也很好的处理了这种情况。通过尽快的复制最少的片断,减少了这种由于当前的peers停止上载后带来的风险。
3)随机的第一个片断(Random First Piece)
“ 最少优先”的一个例外是在下载刚开始的时候。此时,下载者没有任何片断可供上传,所以,需要尽快的获取一个完整的片断。而最少的片断,通常只有某一个peer拥有,所以,它可能比多个peers都拥有的那些片断下载的要慢。因此,第一个片断是随机选择的,直到第一个片断下载完成,才切换到“最少优先”的策略。
4)最后阶段模式(Endgame Mode)
有时候,从一个速率很慢的peer那里请求一个片断。在下载的中间阶段,这不是什么问题,但是却可能潜在的延迟下载的完成。为了防止这种情况,在最后阶段,peer向它的所有的peers们都发送某片断的子片断的请求,一旦某些子片断到了,那么就会向其它peer发送 cancel 消息,取消对这些子片断的请求,以避免带宽的浪费。实际上,用这种方法并没有浪费多少带宽,而文件的结束部分也一直下载的非常快。
2、阻塞算法(Choking Algorithms)
BT 并不集中分配资源。每个peer自己有责任来尽可能的提高它的下载速率。Peers从它可以连接的peers处下载文件,并根据对方提供的下载速率给予同等的上传回报(你敬我一尺,我敬你一丈)。对于合作者,提供上传服务,对于不合作的,就阻塞对方。所以说,阻塞是一种临时的拒绝上传策略,虽然上传停止了,但是下载仍然继续。在阻塞停止的时候,连接并不需要重新建立。
阻塞算法并不属于BT对等协议(指peers 之间交互的协议)的技术部分,但是对提高性能是必要的。一个好的阻塞算法应该利用所有可用的资源,为所有下载者提供一致可靠的下载速率,并适当惩罚那些只下载而不上传的peers。
1)帕累托效率(Pareto Efficiency)
在经济学里,帕累托效率可以这样来定义:一种状态(资源配置、社会制度等)被称为帕累托最优状态,如果不存在另外一种可选择的状态使得没有任何人的处境变差而至少有一个人的处境变得更好。这意味着,当满足给定的约束条件,一种资源配置的状态已经使没有人能够按照自己的偏好在不损害别人的条件下变得更好,那么就是达到了帕累托最优状态。可以通俗地理解为,如果处于这种状态:除非损人,就不能利己,这就达到了帕累托最优。在计算机领域,寻求帕累托有效是一种本地优化算法BitTorrent的阻塞算法,用一种针锋相对的方式来试图达到帕累托最优。(原文不太好翻译,我简化了)。Peers对那些向他提供上传服务的peers给予同样的回报,目的是希望在任何时候都有若干个连接正在进行着双向传输。
2)BitTorrent的阻塞算法
从技术层面上说,BT的每个peer一直与固定数量的其它 peers 保持疏通(通常是4个),所以问题就变成了哪些peers应该保持疏通?这种方法使得TCP的拥塞控制性能能够可靠的饱和上传容量。(也就是说,尽量让整个系统的上传能力达到最大)。
严格的根据当前的下载速率来决定哪些peers应该保持疏通。令人惊讶的是,计算当前下载速率是个大难题。当前的实现实质上是一个每隔20秒的轮询。而原来的算法是对一个长时间的网络传输进行总计,但这种方法很差劲,因为由于资源可用或者不可用,带宽会变化的很快。
为了避免因为频繁的阻塞和疏通 peers造成的资源浪费,BT每隔10秒计算一次哪个peer需要被阻塞,然后将这种状态保持到下一个10秒。10秒已经足够使得TCP来调整它的传输性能到最大。
3)开放检测(Optimistic Unchoking)
如果只是简单的为提供最好的下载速率的peers们提供上载,那么就没有办法来发现那些空闲的连接是否比当前正使用的连接更好。为了解决这个问题,在任何时候,每个peer都拥有一个称为“optimistic unchoking”的连接,这个连接总是保持疏通状态,而不管它的下载速率是怎样。每隔30秒,重新计算一次哪个连接应该是“optimistic unchoking”。30秒足以让上传能力达到最大,下载能力也相应的达到最大。这种和针锋相对类似的思想非常的伟大。“optimistic unchoking”非常和谐的与“囚徒困境”合作。
4)反对歧视(Anti-snubbing)
某些情况下,一个peer可能被它所有的peers都阻塞了,这种情况下,它将会保持较低的下载速率直到通过 “optimistic unchoking”找到更好peers。为了减轻这种问题,如果一段时间过后,从某个peer那里一个片断也没有得到,那么这个peer认为自己被对方 “怠慢”了,于是不再为对方提供上传,除非对方是“optimistic unchoking”。这种情况频繁发生,会导致多于一个的并发的“optimistic unchoking”。
5)仅仅上传(Upload Only)
一旦某个peer完成了下载,它不能再通过下载速率(因为下载速率已经为0了)来决定为哪些 peers 提供上载了。目前采用的解决办法是,优先选择那些从它这里得到更好的上载速率的peers。这样的理由是可以尽可能的利用上载带宽。