From a1577dc1acfb1cdbc7924113929003c43b9c4187 Mon Sep 17 00:00:00 2001 From: Lv Date: Sat, 19 Mar 2022 09:52:30 +0800 Subject: [PATCH] Fixed some known issues, mainly in the forwarder --- scenario/extensions/mircc-strategy.cpp | 376 ------------------------- scenario/extensions/mircc-strategy.hpp | 124 -------- 2 files changed, 500 deletions(-) delete mode 100644 scenario/extensions/mircc-strategy.cpp delete mode 100644 scenario/extensions/mircc-strategy.hpp diff --git a/scenario/extensions/mircc-strategy.cpp b/scenario/extensions/mircc-strategy.cpp deleted file mode 100644 index 10a020d..0000000 --- a/scenario/extensions/mircc-strategy.cpp +++ /dev/null @@ -1,376 +0,0 @@ -#include "mircc-strategy.hpp" - -namespace nfd { - namespace fw { - - MIRCCStrategy::MIRCCStrategy(Forwarder &forwarder, const Name &name = getStrategyName()) - : Strategy(forwarder), ProcessNackTraits(this) - { - this->setInstanceName(makeInstanceName(name, getStrategyName())); - - FaceTable& faceTable = m_forwarder.m_faceTable; - for (auto i = faceTable.begin();i != faceTable.end();i++) - { - m_Rp_base[i] = getMaxR(); - m_Rp_ex[i] = 0; - m_Rs_base[i] = 0; - m_Rs_ex[i] = 0; - - m_Yp[i] = 0; - m_Yp_old[i] = 0; - m_Ys[i] = 0; - m_Ys_old[i] = 0; - - m_d[i] = 500; - m_ratio[i] = 3000/INTEREST_SIZE; - m_linkC[i] = 10 * 1000 * 1000 * 8; - sendInterest(i); - calcRt(i); - } - - //初始化hash - hashUnit[0] = 1; - for (int i = 1;i < 100;i++) - { - hashUnit[i] = hashUnit[i - 1] * hashNum % hashMo; - } - //初始化逆元 - invMod = fpow(hashNum,hashMo - 2,hashMo); - - - } - - void - MIRCCStrategy::sendQueueInterest(FaceEndpoint &ingress) - { - //设置下一次发送，根据兴趣包与数据包的大小之比，计算出兴趣包可用带宽，进而计算出每发送一个兴趣包需要的时间，最小为1ms - getScheduler().schedule(time::milliseconds(max(1,INTEREST_SIZE / (m_linkC[ingress] * (1.0 / (m_ratio[ingress] + 1))))), std::bind(&MIRCCStrategy::sendQueueInterest, this, ingress)); - - //清空队列中过期的兴趣包 - refreshQueueInterest(ingress); - - //分三种情况讨论，只有第一类兴趣包可以发，只有第二类兴趣包可以发，两类可以发（选择最早到达的） - if (m_Queue[ingress].qp.size() > 0 and m_Queue[ingress].qs.size() == 0) - { - //发送兴趣包 - sendInterest(m_forwarder.m_pit,ingress,m_Queue[ingress].qp.front()); - //这个位置是不是要设置pit记录的过期时间？似乎sendInterest里面实现了这个功能 - m_Queue[ingress].qp.pop_front(); - m_Queue[ingress].tp.pop_front(); - - }else if (m_Queue[ingress].qs.size() > 0 and m_Queue[ingress].qp.size() == 0) - { - //发送兴趣包 - sendInterest(m_forwarder.m_pit,ingress,m_Queue[ingress].qs.front()); - //这个位置是不是要设置pit记录的过期时间？似乎sendInterest里面实现了这个功能 - m_Queue[ingress].qs.pop_front(); - m_Queue[ingress].ts.pop_front(); - }else - { - //如果第一类兴趣包更早到达 - if (m_Queue[ingress].tp.front() < m_Queue[ingress].ts.front()) - { - //发送兴趣包 - sendInterest(m_forwarder.m_pit,ingress,m_Queue[ingress].qp.front()); - //这个位置是不是要设置pit记录的过期时间？似乎sendInterest里面实现了这个功能 - m_Queue[ingress].qp.pop_front(); - m_Queue[ingress].tp.pop_front(); - }else - { - //发送兴趣包 - sendInterest(m_forwarder.m_pit,ingress,m_Queue[ingress].qs.front()); - //这个位置是不是要设置pit记录的过期时间？似乎sendInterest里面实现了这个功能 - m_Queue[ingress].qs.pop_front(); - m_Queue[ingress].ts.pop_front(); - } - } - } - - - void - MIRCCStrategy::calcRt(FaceEndpoint &ingress) - { - //每100ms自动重新计算Rt - getScheduler().schedule(time::milliseconds(100), std::bind(&MIRCCStrategy::calcRt, this, ingress)); - - uint64_t ia,ib,Rp,Rs; - Rp = m_Rp_base[ingress] + m_Rp_ex[ingress]; - Rs = m_Rs_base[ingress] + m_Rs_ex[ingress]; - - //Rp计算过程 - ia = std::max(m_linkC[ingress],m_Yp[ingress]); - ib = std::max(Rp,1); - uint64_t Np = ia / ib; - - ia = m_Yp[ingress] - m_Yp_old[ingress]; - ib = std::max(m_Yp[ingress],1); - double Bp = std::max(m_B,(double)ia / ib); - - ia = m_miu * m_linkC[ingress] - Bp * getQt() / std::max(m_d[ingress],0.001); - ib = std::max(Np,1); - m_Rp_base[ingress] = m_a * m_Rp_base[ingress] + (1 - m_a) * ia / ib; - - ia = m_Yp[ingress]; - ib = std::max(Np,1); - m_Rp_ex[ingress] = m_a * m_Rp_ex[ingress] + (1 - m_a) * (Rp - ia / ib); - - m_Yp_old[ingress] = m_Yp[ingress]; - - //Rs计算过程 - uint64_t Cs = std::max(m_linkC[ingress] - m_Yp[ingress],0); - uint64_t Ns = std::max(max(Cs,m_Ys[ingress]) / Rs,1); - - ia = m_Ys[ingress] - m_Ys_old[ingress]; - ib = std::max(m_Ys[ingress],1); - double Bs = std::max(m_B,(double)ia / ib); - - ia = m_miu * Cs - Bs * getQt() / std::max(m_d[ingress],0.001); - ib = std::max(Ns,1); - m_Rs_base[ingress] = m_a * m_Rs_base[ingress] + (1 - m_a) * ia / i b; - - ia = m_Ys[ingress]; - ib = std;:max(Ns,1); - m_Rs_ex[ingress] = m_a * m_Rs_ex[ingress] + (1 - m_a) * (Rs - ia / ib); - - m_Ys_old[ingress] = m_Ys[ingress]; - - m_Yp[ingress] = 0; - m_Ys[ingress] = 0; - - - } -//在队列大小限制范围内的兴趣包，转发过程不会超时，这个功能暂时没用。 -/* - void - MIRCCStrategy::refreshQueueInterest(FaceEndpoint &ingress) - { - //清空face队列中过期的兴趣包 - auto now = time::steady_clock::now(); - - while (m_Queue[ingress].qp.size() > 0 and m_Queue[ingress].tp.front() < now) - { - m_Queue[ingress].qp.pop_front(); - m_Queue[ingress].tp.pop_front(); - } - - while (m_Queue[ingress].qs.size() > 0 and m_Queue[ingress].ts.front() < now) - { - m_Queue[ingress].qs.pop_front(); - m_Queue[ingress].ts.pop_front(); - } - } -*/ - - uint64_t - MIRCCStrategy::getQt(FaceEndpoint &ingress) - { - //清除队列中过期兴趣包，避免造成干扰 - //refreshQueueInterest(ingress); - //返回兴趣包个数*兴趣包大小*数据包与兴趣包大小之比 - return (m_Queue[ingress].qp.size() + m_Queue[ingress].qs.size()) * INTEREST_SIZE * m_ratio[ingress]; - } - - void - MIRCCStrategy::pushQueueInterest(FaceEndpoint &ingress, const Interest &interest) - { - //更新Yt - updateYt(ingress,interest); - - uint64_t p = interest.getTag(); - //如果是第一类兴趣包 - if (p == 1) - { - //如果队列已经满了 - if (m_Queue[ingress].qp.size() + m_Queue[ingress].qs.size() >= QUEUE_MAX) - { - //如果存在第二类兴趣包可以删除 - if (m_Queue[ingress].qs.size() > 0) - { - //删掉第二类兴趣包最后一个，把当前第一类兴趣包加入 - m_Queue[ingress].qs.pop_back(); - m_Queue[ingress].ts.pop_back(); - m_Queue[ingress].qp.push_back(interest); - m_Queue[ingress].tp.push_back(time::steady_clock::now()); - }else - { - //否则当前兴趣包丢失 - - //这里应该回传Nack，还没有写完 - } - }else - { - //直接将当前兴趣包加入第一类兴趣包队列末尾 - m_Queue[ingress].qp.push_back(interest); - m_Queue[ingress].tp.push_back(time::steady_clock::now()); - } - }else - { - //如果队列已经满了 - if (m_Queue[ingress].qp.size() + m_Queue[ingress].qs.size() >= QUEUE_MAX) - { - //当前兴趣包丢失 - - //这里是否应该回传Nack不确定，需要回去看论文，第二类兴趣包处理方式似乎与第一类不同，还没有写完 - }else - { - //将但钱兴趣包加入第二类末尾 - m_Queue[ingress].qs.push_back(interest); - m_Queue[ingress].ts.push_back(time::steady_clock::now()); - } - } - } - - void - MIRCCStrategy::updateYt(FaceEndpoint &ingress, const Interest &interest) - { - uint64_t p = interest.getTag(); - if (p == 1) - { - m_Yp[ingress] += INTEREST_SIZE * m_ratio[ingress]; - }else - { - m_Ys[ingress] += INTEREST_SIZE * m_ratio[ingress]; - } - } - - void - MIRCCStrategy::afterReceiveInterest(const FaceEndpoint &ingress, const Interest &interest, const shared_ptr &pitEntry) - {//目前在按照标识路径转发，采用MIRCC论文中的方法2 - - - //MIRCC似乎不需要判断isNextHopEligible？？？ - //稍微有点不理解Out-record在干啥 - if (hasPendingOutRecords(*pitEntry)) - { - //已经有记录，等待即可。 - //这个位置要不要在其他路径继续转发（比如pathid不相同的情况），不太确定 - //我记得原文是说不考虑任何的中间的cs命中之类的情况 - return; - } - - //根据hash值获取一个参考端口号，还需要根据可用端口数量进一步取模来最终选取转发端口。 - uint64_t tmpFace = chooseFaceForInterestByHash(interest); - - const fib::Entry& fibEntry = this->lookupFib(*pitEntry); - - NextHopList outFaces = fibEntry.getNextHops(); - - if (outFace.size() != 0) - {//如果存在端口可以转发 - uint64_t choosedFace = tmpFace % outFace.size(); - Face& outFace = choosedFace[choosedFace].getFace(); - //放入对应端口队列进行排队 - this->pushQueueInterest(pitEntry, FaceEndpoint(outFace, 0), interest); - return; - } - - //如果发现没有端口可以转发，则设置立马删除该条pit - //后续Nack处理是否得当并不清楚，还需要研究 - lp::NackHeader nackHeader; - nackHeader.setReason(lp::NackReason::NO_ROUTE); - this->sendNack(pitEntry, ingress, nackHeader); - this->rejectPendingInterest(pitEntry); - return; - } - - void - MIRCCStrategy::afterReceiveNack(const FaceEndpoint &ingress, const lp::Nack &nack, const shared_ptr &pitEntry) - { - //nack具体怎么处理还需要研究 - //nack应该利用预留给data的带宽，返回给consumer - //Forwarder::onIncomingNack似乎已经把Nack返回给consumer了 - } - - void - MIRCCStrategy::beforeSatisfyInterest(const shared_ptr &pitEntry, const FaceEndpoint &ingress, const Data &data) - { - auto now = time::steady_clock::now(); - uint64_t rcd_cnt = 0; - //目前计算中，只考虑没有过期的并成功收到数据包的兴趣包对RTT和ratio的影响，超时的数据包是否对RTT有影响？。 - for (const pit::InRecord& inRecord : pitEntry->getInRecords()) { - if (inRecord.getExpiry() > now) - { - if (inRecord.getFace().getId() == ingress.face.getId() && inRecord.getFace().getLinkType() != ndn::nfd::LINK_TYPE_AD_HOC) - { - continue; - } - - //计算收到数据包对数据包/兴趣包大小比率的影响 - m_ratio[ingress] = m_ratio_a * m_ratio[ingress] + (1 - m_ratio_a) * (data.getContent().size() / INTEREST_SIZE); - - //计算收到数据包对RTT的影响。 - uint64_t tmpRTT = time::steady_clock::now() - inRecord.getLastRenewed();//不确定正确性，可能还要继续考虑 - m_d[ingress] = m_rtt_a * m_d[ingress] + (1 - m_rtt_a) * tmpRTT; - - } - } - } - - - void - MIRCCStrategy::fastpow(int64_t a,int64_t k,int64_t p) - { - int64_t = 1; - while (k != 0) - { - if (k % 2 != 0) - { - res = res * a % p; - } - k >>= 1; - a = a * a % p; - } - return res; - } - - void - MIRCCStrategy::setHashForData(const FaceEndpoint &ingress, const Data &data) - {//注意！！HopCountTag的工作情况目前不太了解，似乎ndnsim这个Tag没做处理，后续需要考虑 - //处理路径标识id - uint64_t hashOld = data.getTag(); - //uint64_t hashNew = (hashOld + hashUnit[data.getTag()] * ingress.getFace().getId() % hashMo) % hashMo; - //靠近producer的为hash值的高位 - uint64_t hashNew = (hashOld * hashNum % hashMo + ingress.getFace().getId() % hashMo) % hashMo; - //给data设置新的hash值 - data.setTag(make_shared(hashNew)); - - - } - - uint64_t - MIRCCStrategy::chooseFaceForInterestByHash(const Interest &interest) - { - uint64_t hashGuide = data.getTag(); - uint64_t Faceid = hashGuide % hashMo % hashNum; - return Faceid; - } - - void - MIRCCStrategy::afterReceiveData(const shared_ptr &pitEntry, const FaceEndpoint &ingress, const Data &data) - { - setHashForData(ingress, data) - - //修改Rt的值 - uint64_t dataRp = data.getTag(); - uint64_t dataRs = data.getTag(); - - if (dataRp > m_Rp_base[ingress] + m_Rp_ex[ingress]) - { - data.setTag(make_shared(m_Rp_base[ingress] + m_Rp_ex[ingress])); - } - - if (dataRs > m_Rs_base[ingress] + m_Rs_ex[ingress]) - { - data.setTag(make_shared(m_Rs_base[ingress] + m_Rs_ex[ingress])); - } - - - - beforeSatisfyInterest(pitEntry, ingress, data); - - sendDataToAll(pitEntry, ingress, data); - - } - - } -} \ No newline at end of file diff --git a/scenario/extensions/mircc-strategy.hpp b/scenario/extensions/mircc-strategy.hpp deleted file mode 100644 index 7e76b02..0000000 --- a/scenario/extensions/mircc-strategy.hpp +++ /dev/null @@ -1,124 +0,0 @@ - -#include "face/face.hpp" -#include "fw/strategy.hpp" -#include "fw/algorithm.hpp" -#include "fw/process-nack-traits.hpp" -#include -#include -#include -namespace nfd { - namespace fw { - - - class MIRCCQueue - { - public: - std:: deque qp,qs; - //记录的是过期时间，非达到时间 - std:: deque tp,ts; - - }; - - /* - MIRCC 策略实现 - */ - class MIRCCStrategy : public Strategy, public ProcessNackTraits - { - public: - explicit MIRCCStrategy(Forwarder &forwarder, const Name &name = getStrategyName()); - - ~MIRCCStrategy() override = default; - - void - afterReceiveInterest(const FaceEndpoint &ingress, const Interest &interest,const shared_ptr &pitEntry) override; - - void - afterReceiveNack(const FaceEndpoint &ingress, const lp::Nack &nack,const shared_ptr &pitEntry) override; - - void - beforeSatisfyInterest(const shared_ptr &pitEntry,const FaceEndpoint &ingress, const Data &data) override; - - void - afterReceiveData(const shared_ptr &pitEntry,const FaceEndpoint &ingress, const Data &data) override; - - /** - * 获取最大的目标速率 - * @return - */ - static uint64_t - getMaxR() - { - return std::numeric_limits::max(); - } - - static const Name & - getStrategyName(); - - protected: - friend ProcessNackTraits; - - private: - unordered_map m_Rp_base;//Rp的base速率 - unordered_map m_Rp_ex;//Rp的ex速率 - unordered_map m_Rs_base;//Rs的base速率 - unordered_map m_Rs_ex;//Rs的ex速率 - unordered_map m_Yp;//本周期的Yp(t) - unordered_map m_Yp_old;//上一个周期的Yp(t) - unordered_map m_Ys;//本周起的Ys(t) - unordered_map m_Ys_old;//上一个周期的Ys(t) - unordered_map m_d;//维护每一个face的RTT - unorderer_map m_Queue;//用于出口排队转发兴趣包 - unordered_map m_ratio;//维护每一个face的Data/Interest - //这个大小之比要按照第一类，第二类进行分类么？ - unordered_map m_linkC;//每一个链路的带宽，目前不知道怎么获取，都设置为10 * 1000 * 1000 * 8 - - //通过兴趣包标注的path来获得对应的转发端口 - //采用MIRCC论文中提出的第二个方案，由consumer随机生成hash值来引导转发，路由器无需在记录对应关系 - //unorderer_map pathid2face; - - //兴趣包大小常量 - const unint64_t INTEREST_SIZE = 300; - //队列大小常量 - const unint64_t QUEUE_MAX = 1000; - - const double m_B = 0.1,m_a = 0.5,m_miu = 0.95,m_ratio_a = 0.99 ,m_rtt_a = 0.99; - - const uint64_t = hashUnit[100],hashMo = 998244353,hashNum = 37,invMod = 0; - - - //发送对应端口1个正在排队的兴趣包，并设置下一次发送时间 - void - sendQueueInterest(FaceEndpoint &ingress); - - //清空对应端口队列中所有过期的兴趣包 - void - refreshQueueInterest(FaceEndpoint &ingress); - - //将兴趣包放入对应端口的队列 - void - pushQueueInterest(FaceEndpoint &ingress, const Interest &interest); - - //计算并返回q(t) - uint64_t - getQt(); - - //计算Rt - void - calcRt(FaceEndpoint &ingress); - - //更新Yt - void - updateYt(FaceEndpoint &ingress, const Interest &interest); - - void - setHashForData(const FaceEndpoint &ingress, const Data &data); - - uint64_t - fastpow(int64_t a,int64_t k,int64_t p); - }; - } -} - - - -