Rough draft of fq-codel implementation

2018-07-10 16:50:12 -07:00 · 2018-07-10 16:50:12 -07:00 · 28cb40529d
commit 28cb40529d
parent bdcdccfcc3
7 changed files with 419 additions and 15 deletions
--- a/node/Switch.cpp
+++ b/node/Switch.cpp
@ -266,6 +266,8 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 		}
 	}

+	uint8_t qosBucket = ZT_QOS_DEFAULT_BUCKET;
+
 	if (to.isMulticast()) {
 		MulticastGroup multicastGroup(to,0);

@ -383,7 +385,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 			network->learnBridgedMulticastGroup(tPtr,multicastGroup,RR->node->now());

 		// First pass sets noTee to false, but noTee is set to true in OutboundMulticast to prevent duplicates.
-		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),Address(),from,to,(const uint8_t *)data,len,etherType,vlanId)) {
+		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),Address(),from,to,(const uint8_t *)data,len,etherType,vlanId,qosBucket)) {
 			RR->t->outgoingNetworkFrameDropped(tPtr,network,from,to,etherType,vlanId,len,"filter blocked");
 			return;
 		}
@ -407,7 +409,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 		Address toZT(to.toAddress(network->id())); // since in-network MACs are derived from addresses and network IDs, we can reverse this
 		SharedPtr<Peer> toPeer(RR->topology->getPeer(tPtr,toZT));

-		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),toZT,from,to,(const uint8_t *)data,len,etherType,vlanId)) {
+		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),toZT,from,to,(const uint8_t *)data,len,etherType,vlanId,qosBucket)) {
 			RR->t->outgoingNetworkFrameDropped(tPtr,network,from,to,etherType,vlanId,len,"filter blocked");
 			return;
 		}
@ -422,7 +424,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 			outp.append(data,len);
 			if (!network->config().disableCompression())
 				outp.compress();
-			send(tPtr,outp,true);
+			aqm_enqueue(tPtr,network,outp,true,qosBucket);
 		} else {
 			Packet outp(toZT,RR->identity.address(),Packet::VERB_FRAME);
 			outp.append(network->id());
@ -430,7 +432,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 			outp.append(data,len);
 			if (!network->config().disableCompression())
 				outp.compress();
-			send(tPtr,outp,true);
+			aqm_enqueue(tPtr,network,outp,true,qosBucket);
 		}

 	} else {
@ -439,7 +441,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 		// We filter with a NULL destination ZeroTier address first. Filtrations
 		// for each ZT destination are also done below. This is the same rationale
 		// and design as for multicast.
-		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),Address(),from,to,(const uint8_t *)data,len,etherType,vlanId)) {
+		if (!network->filterOutgoingPacket(tPtr,false,RR->identity.address(),Address(),from,to,(const uint8_t *)data,len,etherType,vlanId,qosBucket)) {
 			RR->t->outgoingNetworkFrameDropped(tPtr,network,from,to,etherType,vlanId,len,"filter blocked");
 			return;
 		}
@ -477,7 +479,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 		}

 		for(unsigned int b=0;b<numBridges;++b) {
-			if (network->filterOutgoingPacket(tPtr,true,RR->identity.address(),bridges[b],from,to,(const uint8_t *)data,len,etherType,vlanId)) {
+			if (network->filterOutgoingPacket(tPtr,true,RR->identity.address(),bridges[b],from,to,(const uint8_t *)data,len,etherType,vlanId,qosBucket)) {
 				Packet outp(bridges[b],RR->identity.address(),Packet::VERB_EXT_FRAME);
 				outp.append(network->id());
 				outp.append((uint8_t)0x00);
@ -487,7 +489,7 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 				outp.append(data,len);
 				if (!network->config().disableCompression())
 					outp.compress();
-				send(tPtr,outp,true);
+				aqm_enqueue(tPtr,network,outp,true,qosBucket);
 			} else {
 				RR->t->outgoingNetworkFrameDropped(tPtr,network,from,to,etherType,vlanId,len,"filter blocked (bridge replication)");
 			}
@ -495,6 +497,263 @@ void Switch::onLocalEthernet(void *tPtr,const SharedPtr<Network> &network,const
 	}
 }

+void Switch::aqm_enqueue(void *tPtr, const SharedPtr<Network> &network, Packet &packet,bool encrypt,int qosBucket)
+{
+	if(!network->QoSEnabled()) {
+		send(tPtr, packet, encrypt);
+		return;
+	}
+	NetworkQoSControlBlock *nqcb = _netQueueControlBlock[network->id()];
+	if (!nqcb) {
+		// DEBUG_INFO("creating network QoS control block (NQCB) for network %llx", network->id());
+		nqcb = new NetworkQoSControlBlock();
+		_netQueueControlBlock[network->id()] = nqcb;
+		// Initialize ZT_QOS_NUM_BUCKETS queues and place them in the INACTIVE list
+		// These queues will be shuffled between the new/old/inactive lists by the enqueue/dequeue algorithm
+		for (int i=0; i<ZT_QOS_NUM_BUCKETS; i++) {
+			nqcb->inactiveQueues.push_back(new ManagedQueue(i));
+		}
+	}
+
+	if (packet.verb() != Packet::VERB_FRAME && packet.verb() != Packet::VERB_EXT_FRAME) {
+		// DEBUG_INFO("skipping, no QoS for this packet, verb=%x", packet.verb());
+		// just send packet normally, no QoS for ZT protocol traffic
+		send(tPtr, packet, encrypt);
+	} 
+
+	_aqm_m.lock();
+
+	// Enqueue packet and move queue to appropriate list
+
+	const Address dest(packet.destination());
+	TXQueueEntry *txEntry = new TXQueueEntry(dest,RR->node->now(),packet,encrypt);
+	
+	ManagedQueue *selectedQueue = nullptr;
+	for (int i=0; i<ZT_QOS_NUM_BUCKETS; i++) {
+		if (i < nqcb->oldQueues.size()) { // search old queues first (I think this is best since old would imply most recent usage of the queue)
+			if (nqcb->oldQueues[i]->id == qosBucket) {
+				selectedQueue = nqcb->oldQueues[i];
+			}
+		} if (i < nqcb->newQueues.size()) { // search new queues (this would imply not often-used queues)
+			if (nqcb->newQueues[i]->id == qosBucket) {
+				selectedQueue = nqcb->newQueues[i];
+			}
+		} if (i < nqcb->inactiveQueues.size()) { // search inactive queues
+			if (nqcb->inactiveQueues[i]->id == qosBucket) {
+				selectedQueue = nqcb->inactiveQueues[i];
+				// move queue to end of NEW queue list
+				selectedQueue->byteCredit = ZT_QOS_QUANTUM;
+				// DEBUG_INFO("moving q=%p from INACTIVE to NEW list", selectedQueue);
+				nqcb->newQueues.push_back(selectedQueue);
+				nqcb->inactiveQueues.erase(nqcb->inactiveQueues.begin() + i);
+			}
+		}
+	}
+	if (!selectedQueue) {
+		return;
+	}
+
+	selectedQueue->q.push_back(txEntry);
+	selectedQueue->byteLength+=txEntry->packet.payloadLength();
+	nqcb->_currEnqueuedPackets++;
+
+	// DEBUG_INFO("nq=%2lu, oq=%2lu, iq=%2lu, nqcb.size()=%3d, bucket=%2d, q=%p", nqcb->newQueues.size(), nqcb->oldQueues.size(), nqcb->inactiveQueues.size(), nqcb->_currEnqueuedPackets, qosBucket, selectedQueue);
+
+	// Drop a packet if necessary
+	ManagedQueue *selectedQueueToDropFrom = nullptr;
+	if (nqcb->_currEnqueuedPackets > ZT_QOS_MAX_ENQUEUED_PACKETS)
+	{
+		// DEBUG_INFO("too many enqueued packets (%d), finding packet to drop", nqcb->_currEnqueuedPackets);
+		int maxQueueLength = 0;
+		for (int i=0; i<ZT_QOS_NUM_BUCKETS; i++) {
+			if (i < nqcb->oldQueues.size()) {
+				if (nqcb->oldQueues[i]->byteLength > maxQueueLength) {
+					maxQueueLength = nqcb->oldQueues[i]->byteLength;
+					selectedQueueToDropFrom = nqcb->oldQueues[i];
+				}
+			} if (i < nqcb->newQueues.size()) {
+				if (nqcb->newQueues[i]->byteLength > maxQueueLength) {
+					maxQueueLength = nqcb->newQueues[i]->byteLength;
+					selectedQueueToDropFrom = nqcb->newQueues[i];
+				}
+			} if (i < nqcb->inactiveQueues.size()) {
+				if (nqcb->inactiveQueues[i]->byteLength > maxQueueLength) {
+					maxQueueLength = nqcb->inactiveQueues[i]->byteLength;
+					selectedQueueToDropFrom = nqcb->inactiveQueues[i];
+				}
+			}
+		}
+		if (selectedQueueToDropFrom) {
+			// DEBUG_INFO("dropping packet from head of largest queue (%d payload bytes)", maxQueueLength);
+			int sizeOfDroppedPacket = selectedQueueToDropFrom->q.front()->packet.payloadLength();
+			delete selectedQueueToDropFrom->q.front();
+			selectedQueueToDropFrom->q.pop_front();
+			selectedQueueToDropFrom->byteLength-=sizeOfDroppedPacket;
+			nqcb->_currEnqueuedPackets--;
+		}
+	}
+	_aqm_m.unlock();
+	aqm_dequeue(tPtr);
+}
+
+uint64_t Switch::control_law(uint64_t t, int count)
+{
+	return t + ZT_QOS_INTERVAL / sqrt(count);
+}
+
+Switch::dqr Switch::dodequeue(ManagedQueue *q, uint64_t now) 
+{
+	dqr r;
+	r.ok_to_drop = false;
+	r.p = q->q.front();
+
+	if (r.p == NULL) {
+		q->first_above_time = 0;
+		return r;
+	}
+	uint64_t sojourn_time = now - r.p->creationTime;
+	if (sojourn_time < ZT_QOS_TARGET || q->byteLength <= ZT_DEFAULT_MTU) {
+		// went below - stay below for at least interval
+		q->first_above_time = 0;
+	} else {
+		if (q->first_above_time == 0) {
+			// just went above from below. if still above at
+			// first_above_time, will say it's ok to drop.
+			q->first_above_time = now + ZT_QOS_INTERVAL;
+		} else if (now >= q->first_above_time) {
+			r.ok_to_drop = true;
+		}
+	}
+	return r;
+}
+
+Switch::TXQueueEntry * Switch::CoDelDequeue(ManagedQueue *q, bool isNew, uint64_t now)
+{
+	dqr r = dodequeue(q, now);
+
+	if (q->dropping) {
+		if (!r.ok_to_drop) {
+			q->dropping = false;
+		}
+		while (now >= q->drop_next && q->dropping) {
+			q->q.pop_front(); // drop
+			r = dodequeue(q, now);
+			if (!r.ok_to_drop) {
+				// leave dropping state
+				q->dropping = false;
+			} else {
+				++(q->count);
+				// schedule the next drop.
+				q->drop_next = control_law(q->drop_next, q->count);
+			}
+		}
+	} else if (r.ok_to_drop) {
+		q->q.pop_front(); // drop
+		r = dodequeue(q, now);
+		q->dropping = true;
+		q->count = (q->count > 2 && now - q->drop_next < 8*ZT_QOS_INTERVAL)?
+		q->count - 2 : 1;
+		q->drop_next = control_law(now, q->count);
+	}
+	return r.p;
+}
+
+void Switch::aqm_dequeue(void *tPtr)
+{
+	// Cycle through network-specific QoS control blocks
+	for(std::map<uint64_t,NetworkQoSControlBlock*>::iterator nqcb(_netQueueControlBlock.begin());nqcb!=_netQueueControlBlock.end();) {
+		if (!(*nqcb).second->_currEnqueuedPackets) {
+			return;
+		}
+
+		uint64_t now = RR->node->now();
+		TXQueueEntry *entryToEmit = nullptr;
+		std::vector<ManagedQueue*> *currQueues = &((*nqcb).second->newQueues);
+		std::vector<ManagedQueue*> *oldQueues = &((*nqcb).second->oldQueues);
+		std::vector<ManagedQueue*> *inactiveQueues = &((*nqcb).second->inactiveQueues);
+
+		_aqm_m.lock();
+
+		// Attempt dequeue from queues in NEW list
+		bool examiningNewQueues = true;
+		while (currQueues->size()) {
+			ManagedQueue *queueAtFrontOfList = currQueues->front();
+			if (queueAtFrontOfList->byteCredit < 0) {
+				queueAtFrontOfList->byteCredit += ZT_QOS_QUANTUM;
+				// Move to list of OLD queues
+				// DEBUG_INFO("moving q=%p from NEW to OLD list", queueAtFrontOfList);
+				oldQueues->push_back(queueAtFrontOfList);
+				currQueues->erase(currQueues->begin());
+			} else {
+				entryToEmit = CoDelDequeue(queueAtFrontOfList, examiningNewQueues, now);
+				if (!entryToEmit) {
+					// Move to end of list of OLD queues
+					// DEBUG_INFO("moving q=%p from NEW to OLD list", queueAtFrontOfList);
+					oldQueues->push_back(queueAtFrontOfList);
+					currQueues->erase(currQueues->begin());
+				}
+				else {
+					int len = entryToEmit->packet.payloadLength();
+					queueAtFrontOfList->byteLength -= len;
+					queueAtFrontOfList->byteCredit -= len;
+					// Send the packet!
+					queueAtFrontOfList->q.pop_front();
+					send(tPtr, entryToEmit->packet, entryToEmit->encrypt);
+					(*nqcb).second->_currEnqueuedPackets--;
+				}
+				if (queueAtFrontOfList) {
+					//DEBUG_INFO("dequeuing from q=%p, len=%lu in NEW list (byteCredit=%d)", queueAtFrontOfList, queueAtFrontOfList->q.size(), queueAtFrontOfList->byteCredit);
+				}
+				break;
+			}
+		}
+
+		// Attempt dequeue from queues in OLD list
+		examiningNewQueues = false;
+		currQueues = &((*nqcb).second->oldQueues);
+		while (currQueues->size()) {
+			ManagedQueue *queueAtFrontOfList = currQueues->front();
+			if (queueAtFrontOfList->byteCredit < 0) {
+				queueAtFrontOfList->byteCredit += ZT_QOS_QUANTUM;
+				oldQueues->push_back(queueAtFrontOfList);
+				currQueues->erase(currQueues->begin());
+			} else {
+				entryToEmit = CoDelDequeue(queueAtFrontOfList, examiningNewQueues, now);
+				if (!entryToEmit) {
+					//DEBUG_INFO("moving q=%p from OLD to INACTIVE list", queueAtFrontOfList);
+					// Move to inactive list of queues
+					inactiveQueues->push_back(queueAtFrontOfList);
+					currQueues->erase(currQueues->begin());
+				}
+				else {
+					int len = entryToEmit->packet.payloadLength();
+					queueAtFrontOfList->byteLength -= len;
+					queueAtFrontOfList->byteCredit -= len;
+					queueAtFrontOfList->q.pop_front();
+					send(tPtr, entryToEmit->packet, entryToEmit->encrypt);
+					(*nqcb).second->_currEnqueuedPackets--;
+				}
+				if (queueAtFrontOfList) {
+					//DEBUG_INFO("dequeuing from q=%p, len=%lu in OLD list (byteCredit=%d)", queueAtFrontOfList, queueAtFrontOfList->q.size(), queueAtFrontOfList->byteCredit);
+				}
+				break;
+			}
+		}
+		nqcb++;
+		_aqm_m.unlock();
+	}
+}
+
+void Switch::removeNetworkQoSControlBlock(uint64_t nwid)
+{
+	NetworkQoSControlBlock *nq = _netQueueControlBlock[nwid];
+	if (nq) {
+		_netQueueControlBlock.erase(nwid);
+		delete nq;
+		nq = NULL;
+	}
+}
+
 void Switch::send(void *tPtr,Packet &packet,bool encrypt)
 {
 	const Address dest(packet.destination());
@ -550,6 +809,7 @@ void Switch::doAnythingWaitingForPeer(void *tPtr,const SharedPtr<Peer> &peer)

 	{
 		Mutex::Lock _l(_txQueue_m);
+
 		for(std::list< TXQueueEntry >::iterator txi(_txQueue.begin());txi!=_txQueue.end();) {
 			if (txi->dest == peer->address()) {
 				if (_trySend(tPtr,txi->packet,txi->encrypt)) {
@ -574,6 +834,7 @@ unsigned long Switch::doTimerTasks(void *tPtr,int64_t now)
 	std::vector<Address> needWhois;
 	{
 		Mutex::Lock _l(_txQueue_m);
+
 		for(std::list< TXQueueEntry >::iterator txi(_txQueue.begin());txi!=_txQueue.end();) {
 			if (_trySend(tPtr,txi->packet,txi->encrypt)) {
 				_txQueue.erase(txi++);