les: improved header fetcher and server statistics

9 files changed, 801 insertions, 481 deletions

diff --git a/les/fetcher.go b/les/fetcher.go
index 24cf51839..0c8be9e2a 100644
--- a/les/fetcher.go
+++ b/les/fetcher.go
@@ -23,173 +23,416 @@ import (
 	"time"
 
 	"github.com/ethereum/go-ethereum/common"
+	"github.com/ethereum/go-ethereum/common/mclock"
 	"github.com/ethereum/go-ethereum/core"
 	"github.com/ethereum/go-ethereum/core/types"
+	"github.com/ethereum/go-ethereum/light"
+	"github.com/ethereum/go-ethereum/logger"
+	"github.com/ethereum/go-ethereum/logger/glog"
 )
 
+const (
+	blockDelayTimeout = time.Second * 10 // timeout for a peer to announce a head that has already been confirmed by others
+	maxNodeCount      = 20               // maximum number of fetcherTreeNode entries remembered for each peer
+)
+
+// lightFetcher
 type lightFetcher struct {
 	pm    *ProtocolManager
 	odr   *LesOdr
-	chain BlockChain
-
-	headAnnouncedMu sync.Mutex
-	headAnnouncedBy map[common.Hash][]*peer
-	currentTd       *big.Int
-	deliverChn      chan fetchResponse
-	reqMu           sync.RWMutex
-	requested       map[uint64]fetchRequest
-	timeoutChn      chan uint64
-	notifyChn       chan bool // true if initiated from outside
-	syncing         bool
-	syncDone        chan struct{}
+	chain *light.LightChain
+
+	maxConfirmedTd  *big.Int
+	peers           map[*peer]*fetcherPeerInfo
+	lastUpdateStats *updateStatsEntry
+
+	lock       sync.Mutex // qwerqwerqwe
+	deliverChn chan fetchResponse
+	reqMu      sync.RWMutex
+	requested  map[uint64]fetchRequest
+	timeoutChn chan uint64
+	requestChn chan bool // true if initiated from outside
+	syncing    bool
+	syncDone   chan *peer
+}
+
+// fetcherPeerInfo holds fetcher-specific information about each active peer
+type fetcherPeerInfo struct {
+	root, lastAnnounced *fetcherTreeNode
+	nodeCnt             int
+	confirmedTd         *big.Int
+	bestConfirmed       *fetcherTreeNode
+	nodeByHash          map[common.Hash]*fetcherTreeNode
+	firstUpdateStats    *updateStatsEntry
 }
 
+// fetcherTreeNode is a node of a tree that holds information about blocks recently
+// announced and confirmed by a certain peer. Each new announce message from a peer
+// adds nodes to the tree, based on the previous announced head and the reorg depth.
+// There are three possible states for a tree node:
+// - announced: not downloaded (known) yet, but we know its head, number and td
+// - intermediate: not known, hash and td are empty, they are filled out when it becomes known
+// - known: both announced by this peer and downloaded (from any peer).
+// This structure makes it possible to always know which peer has a certain block,
+// which is necessary for selecting a suitable peer for ODR requests and also for
+// canonizing new heads. It also helps to always download the minimum necessary
+// amount of headers with a single request.
+type fetcherTreeNode struct {
+	hash             common.Hash
+	number           uint64
+	td               *big.Int
+	known, requested bool
+	parent           *fetcherTreeNode
+	children         []*fetcherTreeNode
+}
+
+// fetchRequest represents a header download request
 type fetchRequest struct {
-	hash   common.Hash
-	amount uint64
-	peer   *peer
+	hash    common.Hash
+	amount  uint64
+	peer    *peer
+	sent    mclock.AbsTime
+	timeout bool
 }
 
+// fetchResponse represents a header download response
 type fetchResponse struct {
 	reqID   uint64
 	headers []*types.Header
 	peer    *peer
 }
 
+// newLightFetcher creates a new light fetcher
 func newLightFetcher(pm *ProtocolManager) *lightFetcher {
 	f := &lightFetcher{
-		pm:              pm,
-		chain:           pm.blockchain,
-		odr:             pm.odr,
-		headAnnouncedBy: make(map[common.Hash][]*peer),
-		deliverChn:      make(chan fetchResponse, 100),
-		requested:       make(map[uint64]fetchRequest),
-		timeoutChn:      make(chan uint64),
-		notifyChn:       make(chan bool, 100),
-		syncDone:        make(chan struct{}),
-		currentTd:       big.NewInt(0),
+		pm:             pm,
+		chain:          pm.blockchain.(*light.LightChain),
+		odr:            pm.odr,
+		peers:          make(map[*peer]*fetcherPeerInfo),
+		deliverChn:     make(chan fetchResponse, 100),
+		requested:      make(map[uint64]fetchRequest),
+		timeoutChn:     make(chan uint64),
+		requestChn:     make(chan bool, 100),
+		syncDone:       make(chan *peer),
+		maxConfirmedTd: big.NewInt(0),
 	}
 	go f.syncLoop()
 	return f
 }
 
-func (f *lightFetcher) notify(p *peer, head *announceData) {
-	var headHash common.Hash
-	if head == nil {
-		// initial notify
-		headHash = p.Head()
-	} else {
-		if core.GetTd(f.pm.chainDb, head.Hash, head.Number) != nil {
-			head.haveHeaders = head.Number
-		}
-		//fmt.Println("notify", p.id, head.Number, head.ReorgDepth, head.haveHeaders)
-		if !p.addNotify(head) {
-			//fmt.Println("addNotify fail")
-			f.pm.removePeer(p.id)
+// syncLoop is the main event loop of the light fetcher
+func (f *lightFetcher) syncLoop() {
+	f.pm.wg.Add(1)
+	defer f.pm.wg.Done()
+
+	requestStarted := false
+	for {
+		select {
+		case <-f.pm.quitSync:
+			return
+		// when a new announce is received, request loop keeps running until
+		// no further requests are necessary or possible
+		case newAnnounce := <-f.requestChn:
+			f.lock.Lock()
+			s := requestStarted
+			requestStarted = false
+			if !f.syncing && !(newAnnounce && s) {
+				if peer, node, amount := f.nextRequest(); node != nil {
+					requestStarted = true
+					reqID, started := f.request(peer, node, amount)
+					if started {
+						go func() {
+							time.Sleep(softRequestTimeout)
+							f.reqMu.Lock()
+							req, ok := f.requested[reqID]
+							if ok {
+								req.timeout = true
+								f.requested[reqID] = req
+							}
+							f.reqMu.Unlock()
+							// keep starting new requests while possible
+							f.requestChn <- false
+						}()
+					}
+				}
+			}
+			f.lock.Unlock()
+		case reqID := <-f.timeoutChn:
+			f.reqMu.Lock()
+			req, ok := f.requested[reqID]
+			if ok {
+				delete(f.requested, reqID)
+			}
+			f.reqMu.Unlock()
+			if ok {
+				f.pm.serverPool.adjustResponseTime(req.peer.poolEntry, time.Duration(mclock.Now()-req.sent), true)
+				glog.V(logger.Debug).Infof("hard timeout by peer %v", req.peer.id)
+				go f.pm.removePeer(req.peer.id)
+			}
+		case resp := <-f.deliverChn:
+			f.reqMu.Lock()
+			req, ok := f.requested[resp.reqID]
+			if ok && req.peer != resp.peer {
+				ok = false
+			}
+			if ok {
+				delete(f.requested, resp.reqID)
+			}
+			f.reqMu.Unlock()
+			if ok {
+				f.pm.serverPool.adjustResponseTime(req.peer.poolEntry, time.Duration(mclock.Now()-req.sent), req.timeout)
+			}
+			f.lock.Lock()
+			if !ok || !(f.syncing || f.processResponse(req, resp)) {
+				glog.V(logger.Debug).Infof("failed processing response by peer %v", resp.peer.id)
+				go f.pm.removePeer(resp.peer.id)
+			}
+			f.lock.Unlock()
+		case p := <-f.syncDone:
+			f.lock.Lock()
+			glog.V(logger.Debug).Infof("done synchronising with peer %v", p.id)
+			f.checkSyncedHeaders(p)
+			f.syncing = false
+			f.lock.Unlock()
 		}
-		headHash = head.Hash
 	}
-	f.headAnnouncedMu.Lock()
-	f.headAnnouncedBy[headHash] = append(f.headAnnouncedBy[headHash], p)
-	f.headAnnouncedMu.Unlock()
-	f.notifyChn <- true
 }
 
-func (f *lightFetcher) gotHeader(header *types.Header) {
-	f.headAnnouncedMu.Lock()
-	defer f.headAnnouncedMu.Unlock()
+// addPeer adds a new peer to the fetcher's peer set
+func (f *lightFetcher) addPeer(p *peer) {
+	f.lock.Lock()
+	defer f.lock.Unlock()
+
+	f.peers[p] = &fetcherPeerInfo{nodeByHash: make(map[common.Hash]*fetcherTreeNode)}
+}
 
-	hash := header.Hash()
-	peerList := f.headAnnouncedBy[hash]
-	if peerList == nil {
+// removePeer removes a new peer from the fetcher's peer set
+func (f *lightFetcher) removePeer(p *peer) {
+	f.lock.Lock()
+	defer f.lock.Unlock()
+
+	// check for potential timed out block delay statistics
+	f.checkUpdateStats(p, nil)
+	delete(f.peers, p)
+}
+
+// announce processes a new announcement message received from a peer, adding new
+// nodes to the peer's block tree and removing old nodes if necessary
+func (f *lightFetcher) announce(p *peer, head *announceData) {
+	f.lock.Lock()
+	defer f.lock.Unlock()
+	glog.V(logger.Debug).Infof("received announce from peer %v  #%d  %016x  reorg: %d", p.id, head.Number, head.Hash[:8], head.ReorgDepth)
+
+	fp := f.peers[p]
+	if fp == nil {
+		glog.V(logger.Debug).Infof("announce: unknown peer")
+		return
+	}
+
+	if fp.lastAnnounced != nil && head.Td.Cmp(fp.lastAnnounced.td) <= 0 {
+		// announced tds should be strictly monotonic
+		glog.V(logger.Debug).Infof("non-monotonic Td from peer %v", p.id)
+		go f.pm.removePeer(p.id)
 		return
 	}
-	number := header.Number.Uint64()
-	td := core.GetTd(f.pm.chainDb, hash, number)
-	for _, peer := range peerList {
-		peer.lock.Lock()
-		ok := peer.gotHeader(hash, number, td)
-		peer.lock.Unlock()
-		if !ok {
-			//fmt.Println("gotHeader fail")
-			f.pm.removePeer(peer.id)
+
+	n := fp.lastAnnounced
+	for i := uint64(0); i < head.ReorgDepth; i++ {
+		if n == nil {
+			break
 		}
+		n = n.parent
 	}
-	delete(f.headAnnouncedBy, hash)
-}
+	if n != nil {
+		// n is now the reorg common ancestor, add a new branch of nodes
+		// check if the node count is too high to add new nodes
+		locked := false
+		for uint64(fp.nodeCnt)+head.Number-n.number > maxNodeCount && fp.root != nil {
+			if !locked {
+				f.chain.LockChain()
+				defer f.chain.UnlockChain()
+				locked = true
+			}
+			// if one of root's children is canonical, keep it, delete other branches and root itself
+			var newRoot *fetcherTreeNode
+			for i, nn := range fp.root.children {
+				if core.GetCanonicalHash(f.pm.chainDb, nn.number) == nn.hash {
+					fp.root.children = append(fp.root.children[:i], fp.root.children[i+1:]...)
+					nn.parent = nil
+					newRoot = nn
+					break
+				}
+			}
+			fp.deleteNode(fp.root)
+			if n == fp.root {
+				n = newRoot
+			}
+			fp.root = newRoot
+			if newRoot == nil || !f.checkKnownNode(p, newRoot) {
+				fp.bestConfirmed = nil
+				fp.confirmedTd = nil
+			}
 
-func (f *lightFetcher) nextRequest() (*peer, *announceData) {
-	var bestPeer *peer
-	bestTd := f.currentTd
-	for _, peer := range f.pm.peers.AllPeers() {
-		peer.lock.RLock()
-		if !peer.headInfo.requested && (peer.headInfo.Td.Cmp(bestTd) > 0 ||
-			(bestPeer != nil && peer.headInfo.Td.Cmp(bestTd) == 0 && peer.headInfo.haveHeaders > bestPeer.headInfo.haveHeaders)) {
-			bestPeer = peer
-			bestTd = peer.headInfo.Td
+			if n == nil {
+				break
+			}
 		}
-		peer.lock.RUnlock()
-	}
-	if bestPeer == nil {
-		return nil, nil
-	}
-	bestPeer.lock.Lock()
-	res := bestPeer.headInfo
-	res.requested = true
-	bestPeer.lock.Unlock()
-	for _, peer := range f.pm.peers.AllPeers() {
-		if peer != bestPeer {
-			peer.lock.Lock()
-			if peer.headInfo.Hash == bestPeer.headInfo.Hash && peer.headInfo.haveHeaders == bestPeer.headInfo.haveHeaders {
-				peer.headInfo.requested = true
+		if n != nil {
+			for n.number < head.Number {
+				nn := &fetcherTreeNode{number: n.number + 1, parent: n}
+				n.children = append(n.children, nn)
+				n = nn
+				fp.nodeCnt++
 			}
-			peer.lock.Unlock()
+			n.hash = head.Hash
+			n.td = head.Td
+			fp.nodeByHash[n.hash] = n
 		}
 	}
-	return bestPeer, res
-}
+	if n == nil {
+		// could not find reorg common ancestor or had to delete entire tree, a new root and a resync is needed
+		if fp.root != nil {
+			fp.deleteNode(fp.root)
+		}
+		n = &fetcherTreeNode{hash: head.Hash, number: head.Number, td: head.Td}
+		fp.root = n
+		fp.nodeCnt++
+		fp.nodeByHash[n.hash] = n
+		fp.bestConfirmed = nil
+		fp.confirmedTd = nil
+	}
 
-func (f *lightFetcher) deliverHeaders(peer *peer, reqID uint64, headers []*types.Header) {
-	f.deliverChn <- fetchResponse{reqID: reqID, headers: headers, peer: peer}
+	f.checkKnownNode(p, n)
+	p.lock.Lock()
+	p.headInfo = head
+	fp.lastAnnounced = n
+	p.lock.Unlock()
+	f.checkUpdateStats(p, nil)
+	f.requestChn <- true
 }
 
-func (f *lightFetcher) requestedID(reqID uint64) bool {
-	f.reqMu.RLock()
-	_, ok := f.requested[reqID]
-	f.reqMu.RUnlock()
-	return ok
+// peerHasBlock returns true if we can assume the peer knows the given block
+// based on its announcements
+func (f *lightFetcher) peerHasBlock(p *peer, hash common.Hash, number uint64) bool {
+	f.lock.Lock()
+	defer f.lock.Lock()
+
+	fp := f.peers[p]
+	if fp == nil || fp.root == nil {
+		return false
+	}
+
+	if number >= fp.root.number {
+		// it is recent enough that if it is known, is should be in the peer's block tree
+		return fp.nodeByHash[hash] != nil
+	}
+	f.chain.LockChain()
+	defer f.chain.UnlockChain()
+	// if it's older than the peer's block tree root but it's in the same canonical chain
+	// than the root, we can still be sure the peer knows it
+	return core.GetCanonicalHash(f.pm.chainDb, fp.root.number) == fp.root.hash && core.GetCanonicalHash(f.pm.chainDb, number) == hash
 }
 
-func (f *lightFetcher) request(p *peer, block *announceData) {
-	//fmt.Println("request", p.id, block.Number, block.haveHeaders)
-	amount := block.Number - block.haveHeaders
-	if amount == 0 {
-		return
+// request initiates a header download request from a certain peer
+func (f *lightFetcher) request(p *peer, n *fetcherTreeNode, amount uint64) (uint64, bool) {
+	fp := f.peers[p]
+	if fp == nil {
+		glog.V(logger.Debug).Infof("request: unknown peer")
+		return 0, false
 	}
-	if amount > 100 {
+	if fp.bestConfirmed == nil || fp.root == nil || !f.checkKnownNode(p, fp.root) {
 		f.syncing = true
 		go func() {
-			//fmt.Println("f.pm.synchronise(p)")
+			glog.V(logger.Debug).Infof("synchronising with peer %v", p.id)
 			f.pm.synchronise(p)
-			//fmt.Println("sync done")
-			f.syncDone <- struct{}{}
+			f.syncDone <- p
 		}()
-		return
+		return 0, false
 	}
 
 	reqID := f.odr.getNextReqID()
-	f.reqMu.Lock()
-	f.requested[reqID] = fetchRequest{hash: block.Hash, amount: amount, peer: p}
-	f.reqMu.Unlock()
+	n.requested = true
 	cost := p.GetRequestCost(GetBlockHeadersMsg, int(amount))
 	p.fcServer.SendRequest(reqID, cost)
-	go p.RequestHeadersByHash(reqID, cost, block.Hash, int(amount), 0, true)
+	f.reqMu.Lock()
+	f.requested[reqID] = fetchRequest{hash: n.hash, amount: amount, peer: p, sent: mclock.Now()}
+	f.reqMu.Unlock()
+	go p.RequestHeadersByHash(reqID, cost, n.hash, int(amount), 0, true)
 	go func() {
 		time.Sleep(hardRequestTimeout)
 		f.timeoutChn <- reqID
 	}()
+	return reqID, true
+}
+
+// requestAmount calculates the amount of headers to be downloaded starting
+// from a certain head backwards
+func (f *lightFetcher) requestAmount(p *peer, n *fetcherTreeNode) uint64 {
+	amount := uint64(0)
+	nn := n
+	for nn != nil && !f.checkKnownNode(p, nn) {
+		nn = nn.parent
+		amount++
+	}
+	if nn == nil {
+		amount = n.number
+	}
+	return amount
+}
+
+// requestedID tells if a certain reqID has been requested by the fetcher
+func (f *lightFetcher) requestedID(reqID uint64) bool {
+	f.reqMu.RLock()
+	_, ok := f.requested[reqID]
+	f.reqMu.RUnlock()
+	return ok
+}
+
+// nextRequest selects the peer and announced head to be requested next, amount
+// to be downloaded starting from the head backwards is also returned
+func (f *lightFetcher) nextRequest() (*peer, *fetcherTreeNode, uint64) {
+	var (
+		bestHash   common.Hash
+		bestAmount uint64
+	)
+	bestTd := f.maxConfirmedTd
+
+	for p, fp := range f.peers {
+		for hash, n := range fp.nodeByHash {
+			if !f.checkKnownNode(p, n) && !n.requested && (bestTd == nil || n.td.Cmp(bestTd) >= 0) {
+				amount := f.requestAmount(p, n)
+				if bestTd == nil || n.td.Cmp(bestTd) > 0 || amount < bestAmount {
+					bestHash = hash
+					bestAmount = amount
+					bestTd = n.td
+				}
+			}
+		}
+	}
+	if bestTd == f.maxConfirmedTd {
+		return nil, nil, 0
+	}
+
+	peer := f.pm.serverPool.selectPeer(func(p *peer) (bool, uint64) {
+		fp := f.peers[p]
+		if fp == nil || fp.nodeByHash[bestHash] == nil {
+			return false, 0
+		}
+		return true, p.fcServer.CanSend(p.GetRequestCost(GetBlockHeadersMsg, int(bestAmount)))
+	})
+	var node *fetcherTreeNode
+	if peer != nil {
+		node = f.peers[peer].nodeByHash[bestHash]
+	}
+	return peer, node, bestAmount
 }
 
+// deliverHeaders delivers header download request responses for processing
+func (f *lightFetcher) deliverHeaders(peer *peer, reqID uint64, headers []*types.Header) {
+	f.deliverChn <- fetchResponse{reqID: reqID, headers: headers, peer: peer}
+}
+
+// processResponse processes header download request responses
 func (f *lightFetcher) processResponse(req fetchRequest, resp fetchResponse) bool {
 	if uint64(len(resp.headers)) != req.amount || resp.headers[0].Hash() != req.hash {
 		return false
@@ -201,101 +444,248 @@ func (f *lightFetcher) processResponse(req fetchRequest, resp fetchResponse) boo
 	if _, err := f.chain.InsertHeaderChain(headers, 1); err != nil {
 		return false
 	}
-	for _, header := range headers {
-		td := core.GetTd(f.pm.chainDb, header.Hash(), header.Number.Uint64())
+	tds := make([]*big.Int, len(headers))
+	for i, header := range headers {
+		td := f.chain.GetTd(header.Hash(), header.Number.Uint64())
 		if td == nil {
 			return false
 		}
-		if td.Cmp(f.currentTd) > 0 {
-			f.currentTd = td
-		}
-		f.gotHeader(header)
+		tds[i] = td
 	}
+	f.newHeaders(headers, tds)
 	return true
 }
 
-func (f *lightFetcher) checkSyncedHeaders() {
-	//fmt.Println("checkSyncedHeaders()")
-	for _, peer := range f.pm.peers.AllPeers() {
-		peer.lock.Lock()
-		h := peer.firstHeadInfo
-		remove := false
-	loop:
-		for h != nil {
-			if td := core.GetTd(f.pm.chainDb, h.Hash, h.Number); td != nil {
-				//fmt.Println(" found", h.Number)
-				ok := peer.gotHeader(h.Hash, h.Number, td)
-				if !ok {
-					remove = true
-					break loop
-				}
-				if td.Cmp(f.currentTd) > 0 {
-					f.currentTd = td
-				}
-			}
-			h = h.next
+// newHeaders updates the block trees of all active peers according to a newly
+// downloaded and validated batch or headers
+func (f *lightFetcher) newHeaders(headers []*types.Header, tds []*big.Int) {
+	var maxTd *big.Int
+	for p, fp := range f.peers {
+		if !f.checkAnnouncedHeaders(fp, headers, tds) {
+			glog.V(logger.Debug).Infof("announce inconsistency by peer %v", p.id)
+			go f.pm.removePeer(p.id)
 		}
-		peer.lock.Unlock()
-		if remove {
-			//fmt.Println("checkSync fail")
-			f.pm.removePeer(peer.id)
+		if fp.confirmedTd != nil && (maxTd == nil || maxTd.Cmp(fp.confirmedTd) > 0) {
+			maxTd = fp.confirmedTd
 		}
 	}
+	if maxTd != nil {
+		f.updateMaxConfirmedTd(maxTd)
+	}
 }
 
-func (f *lightFetcher) syncLoop() {
-	f.pm.wg.Add(1)
-	defer f.pm.wg.Done()
+// checkAnnouncedHeaders updates peer's block tree if necessary after validating
+// a batch of headers. It searches for the latest header in the batch that has a
+// matching tree node (if any), and if it has not been marked as known already,
+// sets it and its parents to known (even those which are older than the currently
+// validated ones). Return value shows if all hashes, numbers and Tds matched
+// correctly to the announced values (otherwise the peer should be dropped).
+func (f *lightFetcher) checkAnnouncedHeaders(fp *fetcherPeerInfo, headers []*types.Header, tds []*big.Int) bool {
+	var (
+		n      *fetcherTreeNode
+		header *types.Header
+		td     *big.Int
+	)
 
-	srtoNotify := false
-	for {
-		select {
-		case <-f.pm.quitSync:
-			return
-		case ext := <-f.notifyChn:
-			//fmt.Println("<-f.notifyChn", f.syncing, ext, srtoNotify)
-			s := srtoNotify
-			srtoNotify = false
-			if !f.syncing && !(ext && s) {
-				if p, r := f.nextRequest(); r != nil {
-					srtoNotify = true
-					go func() {
-						time.Sleep(softRequestTimeout)
-						f.notifyChn <- false
-					}()
-					f.request(p, r)
+	for i := len(headers) - 1; ; i-- {
+		if i < 0 {
+			if n == nil {
+				// no more headers and nothing to match
+				return true
+			}
+			// we ran out of recently delivered headers but have not reached a node known by this peer yet, continue matching
+			td = f.chain.GetTd(header.ParentHash, header.Number.Uint64()-1)
+			header = f.chain.GetHeader(header.ParentHash, header.Number.Uint64()-1)
+		} else {
+			header = headers[i]
+			td = tds[i]
+		}
+		hash := header.Hash()
+		number := header.Number.Uint64()
+		if n == nil {
+			n = fp.nodeByHash[hash]
+		}
+		if n != nil {
+			if n.td == nil {
+				// node was unannounced
+				if nn := fp.nodeByHash[hash]; nn != nil {
+					// if there was already a node with the same hash, continue there and drop this one
+					nn.children = append(nn.children, n.children...)
+					n.children = nil
+					fp.deleteNode(n)
+					n = nn
+				} else {
+					n.hash = hash
+					n.td = td
+					fp.nodeByHash[hash] = n
 				}
 			}
-		case reqID := <-f.timeoutChn:
-			f.reqMu.Lock()
-			req, ok := f.requested[reqID]
-			if ok {
-				delete(f.requested, reqID)
+			// check if it matches the header
+			if n.hash != hash || n.number != number || n.td.Cmp(td) != 0 {
+				// peer has previously made an invalid announcement
+				return false
 			}
-			f.reqMu.Unlock()
-			if ok {
-				//fmt.Println("hard timeout")
-				f.pm.removePeer(req.peer.id)
+			if n.known {
+				// we reached a known node that matched our expectations, return with success
+				return true
 			}
-		case resp := <-f.deliverChn:
-			//fmt.Println("<-f.deliverChn", f.syncing)
-			f.reqMu.Lock()
-			req, ok := f.requested[resp.reqID]
-			if ok && req.peer != resp.peer {
-				ok = false
+			n.known = true
+			if fp.confirmedTd == nil || td.Cmp(fp.confirmedTd) > 0 {
+				fp.confirmedTd = td
+				fp.bestConfirmed = n
 			}
-			if ok {
-				delete(f.requested, resp.reqID)
+			n = n.parent
+			if n == nil {
+				return true
 			}
-			f.reqMu.Unlock()
-			if !ok || !(f.syncing || f.processResponse(req, resp)) {
-				//fmt.Println("processResponse fail")
-				f.pm.removePeer(resp.peer.id)
+		}
+	}
+}
+
+// checkSyncedHeaders updates peer's block tree after synchronisation by marking
+// downloaded headers as known. If none of the announced headers are found after
+// syncing, the peer is dropped.
+func (f *lightFetcher) checkSyncedHeaders(p *peer) {
+	fp := f.peers[p]
+	if fp == nil {
+		glog.V(logger.Debug).Infof("checkSyncedHeaders: unknown peer")
+		return
+	}
+	n := fp.lastAnnounced
+	var td *big.Int
+	for n != nil {
+		if td = f.chain.GetTd(n.hash, n.number); td != nil {
+			break
+		}
+		n = n.parent
+	}
+	// now n is the latest downloaded header after syncing
+	if n == nil {
+		glog.V(logger.Debug).Infof("synchronisation failed with peer %v", p.id)
+		go f.pm.removePeer(p.id)
+	} else {
+		header := f.chain.GetHeader(n.hash, n.number)
+		f.newHeaders([]*types.Header{header}, []*big.Int{td})
+	}
+}
+
+// checkKnownNode checks if a block tree node is known (downloaded and validated)
+// If it was not known previously but found in the database, sets its known flag
+func (f *lightFetcher) checkKnownNode(p *peer, n *fetcherTreeNode) bool {
+	if n.known {
+		return true
+	}
+	td := f.chain.GetTd(n.hash, n.number)
+	if td == nil {
+		return false
+	}
+
+	fp := f.peers[p]
+	if fp == nil {
+		glog.V(logger.Debug).Infof("checkKnownNode: unknown peer")
+		return false
+	}
+	header := f.chain.GetHeader(n.hash, n.number)
+	if !f.checkAnnouncedHeaders(fp, []*types.Header{header}, []*big.Int{td}) {
+		glog.V(logger.Debug).Infof("announce inconsistency by peer %v", p.id)
+		go f.pm.removePeer(p.id)
+	}
+	if fp.confirmedTd != nil {
+		f.updateMaxConfirmedTd(fp.confirmedTd)
+	}
+	return n.known
+}
+
+// deleteNode deletes a node and its child subtrees from a peer's block tree
+func (fp *fetcherPeerInfo) deleteNode(n *fetcherTreeNode) {
+	if n.parent != nil {
+		for i, nn := range n.parent.children {
+			if nn == n {
+				n.parent.children = append(n.parent.children[:i], n.parent.children[i+1:]...)
+				break
 			}
-		case <-f.syncDone:
-			//fmt.Println("<-f.syncDone", f.syncing)
-			f.checkSyncedHeaders()
-			f.syncing = false
+		}
+	}
+	for {
+		if n.td != nil {
+			delete(fp.nodeByHash, n.hash)
+		}
+		fp.nodeCnt--
+		if len(n.children) == 0 {
+			return
+		}
+		for i, nn := range n.children {
+			if i == 0 {
+				n = nn
+			} else {
+				fp.deleteNode(nn)
+			}
+		}
+	}
+}
+
+// updateStatsEntry items form a linked list that is expanded with a new item every time a new head with a higher Td
+// than the previous one has been downloaded and validated. The list contains a series of maximum confirmed Td values
+// and the time these values have been confirmed, both increasing monotonically. A maximum confirmed Td is calculated
+// both globally for all peers and also for each individual peer (meaning that the given peer has announced the head
+// and it has also been downloaded from any peer, either before or after the given announcement).
+// The linked list has a global tail where new confirmed Td entries are added and a separate head for each peer,
+// pointing to the next Td entry that is higher than the peer's max confirmed Td (nil if it has already confirmed
+// the current global head).
+type updateStatsEntry struct {
+	time mclock.AbsTime
+	td   *big.Int
+	next *updateStatsEntry
+}
+
+// updateMaxConfirmedTd updates the block delay statistics of active peers. Whenever a new highest Td is confirmed,
+// adds it to the end of a linked list together with the time it has been confirmed. Then checks which peers have
+// already confirmed a head with the same or higher Td (which counts as zero block delay) and updates their statistics.
+// Those who have not confirmed such a head by now will be updated by a subsequent checkUpdateStats call with a
+// positive block delay value.
+func (f *lightFetcher) updateMaxConfirmedTd(td *big.Int) {
+	if f.maxConfirmedTd == nil || td.Cmp(f.maxConfirmedTd) > 0 {
+		f.maxConfirmedTd = td
+		newEntry := &updateStatsEntry{
+			time: mclock.Now(),
+			td:   td,
+		}
+		if f.lastUpdateStats != nil {
+			f.lastUpdateStats.next = newEntry
+		}
+		f.lastUpdateStats = newEntry
+		for p, _ := range f.peers {
+			f.checkUpdateStats(p, newEntry)
+		}
+	}
+}
+
+// checkUpdateStats checks those peers who have not confirmed a certain highest Td (or a larger one) by the time it
+// has been confirmed by another peer. If they have confirmed such a head by now, their stats are updated with the
+// block delay which is (this peer's confirmation time)-(first confirmation time). After blockDelayTimeout has passed,
+// the stats are updated with blockDelayTimeout value. In either case, the confirmed or timed out updateStatsEntry
+// items are removed from the head of the linked list.
+// If a new entry has been added to the global tail, it is passed as a parameter here even though this function
+// assumes that it has already been added, so that if the peer's list is empty (all heads confirmed, head is nil),
+// it can set the new head to newEntry.
+func (f *lightFetcher) checkUpdateStats(p *peer, newEntry *updateStatsEntry) {
+	now := mclock.Now()
+	fp := f.peers[p]
+	if fp == nil {
+		glog.V(logger.Debug).Infof("checkUpdateStats: unknown peer")
+		return
+	}
+	if newEntry != nil && fp.firstUpdateStats == nil {
+		fp.firstUpdateStats = newEntry
+	}
+	for fp.firstUpdateStats != nil && fp.firstUpdateStats.time <= now-mclock.AbsTime(blockDelayTimeout) {
+		f.pm.serverPool.adjustBlockDelay(p.poolEntry, blockDelayTimeout)
+		fp.firstUpdateStats = fp.firstUpdateStats.next
+	}
+	if fp.confirmedTd != nil {
+		for fp.firstUpdateStats != nil && fp.firstUpdateStats.td.Cmp(fp.confirmedTd) <= 0 {
+			f.pm.serverPool.adjustBlockDelay(p.poolEntry, time.Duration(now-fp.firstUpdateStats.time))
+			fp.firstUpdateStats = fp.firstUpdateStats.next
 		}
 	}
 }
diff --git a/les/handler.go b/les/handler.go
index f1a8bc62c..4535aaeb9 100644
--- a/les/handler.go
+++ b/les/handler.go
@@ -24,10 +24,8 @@ import (
 	"math/big"
 	"net"
 	"sync"
-	"time"
 
 	"github.com/ethereum/go-ethereum/common"
-	"github.com/ethereum/go-ethereum/common/mclock"
 	"github.com/ethereum/go-ethereum/core"
 	"github.com/ethereum/go-ethereum/core/state"
 	"github.com/ethereum/go-ethereum/core/types"
@@ -60,7 +58,7 @@ const (
 	MaxHeaderProofsFetch = 64  // Amount of merkle proofs to be fetched per retrieval request
 	MaxTxSend            = 64  // Amount of transactions to be send per request
 
-	disableClientRemovePeer = true
+	disableClientRemovePeer = false
 )
 
 // errIncompatibleConfig is returned if the requested protocols and configs are
@@ -157,44 +155,27 @@ func NewProtocolManager(chainConfig *params.ChainConfig, lightSync bool, network
 			Length:  ProtocolLengths[i],
 			Run: func(p *p2p.Peer, rw p2p.MsgReadWriter) error {
 				var entry *poolEntry
+				peer := manager.newPeer(int(version), networkId, p, rw)
 				if manager.serverPool != nil {
 					addr := p.RemoteAddr().(*net.TCPAddr)
-					entry = manager.serverPool.connect(p.ID(), addr.IP, uint16(addr.Port))
+					entry = manager.serverPool.connect(peer, addr.IP, uint16(addr.Port))
 					if entry == nil {
 						return fmt.Errorf("unwanted connection")
 					}
 				}
-				peer := manager.newPeer(int(version), networkId, p, rw)
 				peer.poolEntry = entry
 				select {
 				case manager.newPeerCh <- peer:
 					manager.wg.Add(1)
 					defer manager.wg.Done()
-					start := mclock.Now()
 					err := manager.handle(peer)
 					if entry != nil {
-						connTime := time.Duration(mclock.Now() - start)
-						stopped := false
-						select {
-						case <-manager.quitSync:
-							stopped = true
-						default:
-						}
-						//fmt.Println("connTime", peer.id, connTime, stopped, err)
-						quality := float64(1)
-						setQuality := true
-						if connTime < time.Minute*10 {
-							quality = 0
-							if stopped {
-								setQuality = false
-							}
-						}
-						manager.serverPool.disconnect(entry, quality, setQuality)
+						manager.serverPool.disconnect(entry)
 					}
 					return err
 				case <-manager.quitSync:
 					if entry != nil {
-						manager.serverPool.disconnect(entry, 0, false)
+						manager.serverPool.disconnect(entry)
 					}
 					return p2p.DiscQuitting
 				}
@@ -224,7 +205,6 @@ func NewProtocolManager(chainConfig *params.ChainConfig, lightSync bool, network
 		manager.downloader = downloader.New(downloader.LightSync, chainDb, manager.eventMux, blockchain.HasHeader, nil, blockchain.GetHeaderByHash,
 			nil, blockchain.CurrentHeader, nil, nil, nil, blockchain.GetTdByHash,
 			blockchain.InsertHeaderChain, nil, nil, blockchain.Rollback, removePeer)
-		manager.fetcher = newLightFetcher(manager)
 	}
 
 	if odr != nil {
@@ -254,10 +234,12 @@ func (pm *ProtocolManager) removePeer(id string) {
 	glog.V(logger.Debug).Infof("LES: unregister peer %v", id)
 	if pm.lightSync {
 		pm.downloader.UnregisterPeer(id)
-		pm.odr.UnregisterPeer(peer)
 		if pm.txrelay != nil {
 			pm.txrelay.removePeer(id)
 		}
+		if pm.fetcher != nil {
+			pm.fetcher.removePeer(peer)
+		}
 	}
 	if err := pm.peers.Unregister(id); err != nil {
 		glog.V(logger.Error).Infoln("Removal failed:", err)
@@ -276,8 +258,10 @@ func (pm *ProtocolManager) Start(srvr *p2p.Server) {
 	lesTopic := discv5.Topic("LES@" + common.Bytes2Hex(pm.blockchain.Genesis().Hash().Bytes()[0:8]))
 	if pm.lightSync {
 		// start sync handler
-		if srvr != nil {
+		if srvr != nil { // srvr is nil during testing
 			pm.serverPool = newServerPool(pm.chainDb, []byte("serverPool/"), srvr, lesTopic, pm.quitSync, &pm.wg)
+			pm.odr.serverPool = pm.serverPool
+			pm.fetcher = newLightFetcher(pm)
 		}
 		go pm.syncer()
 	} else {
@@ -369,12 +353,17 @@ func (pm *ProtocolManager) handle(p *peer) error {
 			requestHeadersByHash, requestHeadersByNumber, nil, nil, nil); err != nil {
 			return err
 		}
-		pm.odr.RegisterPeer(p)
 		if pm.txrelay != nil {
 			pm.txrelay.addPeer(p)
 		}
 
-		pm.fetcher.notify(p, nil)
+		p.lock.Lock()
+		head := p.headInfo
+		p.lock.Unlock()
+		if pm.fetcher != nil {
+			pm.fetcher.addPeer(p)
+			pm.fetcher.announce(p, head)
+		}
 
 		if p.poolEntry != nil {
 			pm.serverPool.registered(p.poolEntry)
@@ -460,7 +449,9 @@ func (pm *ProtocolManager) handleMsg(p *peer) error {
 			return errResp(ErrDecode, "%v: %v", msg, err)
 		}
 		glog.V(logger.Detail).Infoln("AnnounceMsg:", req.Number, req.Hash, req.Td, req.ReorgDepth)
-		pm.fetcher.notify(p, &req)
+		if pm.fetcher != nil {
+			go pm.fetcher.announce(p, &req)
+		}
 
 	case GetBlockHeadersMsg:
 		glog.V(logger.Debug).Infof("<=== GetBlockHeadersMsg from peer %v", p.id)
@@ -558,7 +549,7 @@ func (pm *ProtocolManager) handleMsg(p *peer) error {
 			return errResp(ErrDecode, "msg %v: %v", msg, err)
 		}
 		p.fcServer.GotReply(resp.ReqID, resp.BV)
-		if pm.fetcher.requestedID(resp.ReqID) {
+		if pm.fetcher != nil && pm.fetcher.requestedID(resp.ReqID) {
 			pm.fetcher.deliverHeaders(p, resp.ReqID, resp.Headers)
 		} else {
 			err := pm.downloader.DeliverHeaders(p.id, resp.Headers)
diff --git a/les/helper_test.go b/les/helper_test.go
index a5428e954..2df3faab2 100644
--- a/les/helper_test.go
+++ b/les/helper_test.go
@@ -25,6 +25,7 @@ import (
 	"math/big"
 	"sync"
 	"testing"
+	"time"
 
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/core"
@@ -334,3 +335,13 @@ func (p *testPeer) handshake(t *testing.T, td *big.Int, head common.Hash, headNu
 func (p *testPeer) close() {
 	p.app.Close()
 }
+
+type testServerPool peer
+
+func (p *testServerPool) selectPeer(func(*peer) (bool, uint64)) *peer {
+	return (*peer)(p)
+}
+
+func (p *testServerPool) adjustResponseTime(*poolEntry, time.Duration, bool) {
+
+}
diff --git a/les/odr.go b/les/odr.go
index 444b1da2a..10ef928df 100644
--- a/les/odr.go
+++ b/les/odr.go
@@ -37,6 +37,11 @@ var (
 // peerDropFn is a callback type for dropping a peer detected as malicious.
 type peerDropFn func(id string)
 
+type odrPeerSelector interface {
+	selectPeer(func(*peer) (bool, uint64)) *peer
+	adjustResponseTime(*poolEntry, time.Duration, bool)
+}
+
 type LesOdr struct {
 	light.OdrBackend
 	db           ethdb.Database
@@ -44,7 +49,7 @@ type LesOdr struct {
 	removePeer   peerDropFn
 	mlock, clock sync.Mutex
 	sentReqs     map[uint64]*sentReq
-	peers        *odrPeerSet
+	serverPool   odrPeerSelector
 	lastReqID    uint64
 }
 
@@ -52,7 +57,6 @@ func NewLesOdr(db ethdb.Database) *LesOdr {
 	return &LesOdr{
 		db:       db,
 		stop:     make(chan struct{}),
-		peers:    newOdrPeerSet(),
 		sentReqs: make(map[uint64]*sentReq),
 	}
 }
@@ -77,16 +81,6 @@ type sentReq struct {
 	answered chan struct{} // closed and set to nil when any peer answers it
 }
 
-// RegisterPeer registers a new LES peer to the ODR capable peer set
-func (self *LesOdr) RegisterPeer(p *peer) error {
-	return self.peers.register(p)
-}
-
-// UnregisterPeer removes a peer from the ODR capable peer set
-func (self *LesOdr) UnregisterPeer(p *peer) {
-	self.peers.unregister(p)
-}
-
 const (
 	MsgBlockBodies = iota
 	MsgCode
@@ -142,29 +136,26 @@ func (self *LesOdr) requestPeer(req *sentReq, peer *peer, delivered, timeout cha
 
 	select {
 	case <-delivered:
-		servTime := uint64(mclock.Now() - stime)
-		self.peers.updateTimeout(peer, false)
-		self.peers.updateServTime(peer, servTime)
+		if self.serverPool != nil {
+			self.serverPool.adjustResponseTime(peer.poolEntry, time.Duration(mclock.Now()-stime), false)
+		}
 		return
 	case <-time.After(softRequestTimeout):
 		close(timeout)
-		if self.peers.updateTimeout(peer, true) {
-			self.removePeer(peer.id)
-		}
 	case <-self.stop:
 		return
 	}
 
 	select {
 	case <-delivered:
-		servTime := uint64(mclock.Now() - stime)
-		self.peers.updateServTime(peer, servTime)
-		return
 	case <-time.After(hardRequestTimeout):
-		self.removePeer(peer.id)
+		go self.removePeer(peer.id)
 	case <-self.stop:
 		return
 	}
+	if self.serverPool != nil {
+		self.serverPool.adjustResponseTime(peer.poolEntry, time.Duration(mclock.Now()-stime), true)
+	}
 }
 
 // networkRequest sends a request to known peers until an answer is received
@@ -193,7 +184,13 @@ func (self *LesOdr) networkRequest(ctx context.Context, lreq LesOdrRequest) erro
 
 	exclude := make(map[*peer]struct{})
 	for {
-		if peer := self.peers.bestPeer(lreq, exclude); peer == nil {
+		var p *peer
+		if self.serverPool != nil {
+			p = self.serverPool.selectPeer(func(p *peer) (bool, uint64) {
+				return true, p.fcServer.CanSend(lreq.GetCost(p))
+			})
+		}
+		if p == nil {
 			select {
 			case <-ctx.Done():
 				return ctx.Err()
@@ -202,17 +199,17 @@ func (self *LesOdr) networkRequest(ctx context.Context, lreq LesOdrRequest) erro
 			case <-time.After(retryPeers):
 			}
 		} else {
-			exclude[peer] = struct{}{}
+			exclude[p] = struct{}{}
 			delivered := make(chan struct{})
 			timeout := make(chan struct{})
 			req.lock.Lock()
-			req.sentTo[peer] = delivered
+			req.sentTo[p] = delivered
 			req.lock.Unlock()
 			reqWg.Add(1)
-			cost := lreq.GetCost(peer)
-			peer.fcServer.SendRequest(reqID, cost)
-			go self.requestPeer(req, peer, delivered, timeout, reqWg)
-			lreq.Request(reqID, peer)
+			cost := lreq.GetCost(p)
+			p.fcServer.SendRequest(reqID, cost)
+			go self.requestPeer(req, p, delivered, timeout, reqWg)
+			lreq.Request(reqID, p)
 
 			select {
 			case <-ctx.Done():
diff --git a/les/odr_peerset.go b/les/odr_peerset.go
deleted file mode 100644
index e9b7eec7f..000000000
--- a/les/odr_peerset.go
+++ /dev/null
@@ -1,120 +0,0 @@
-// Copyright 2016 The go-ethereum Authors
-// This file is part of the go-ethereum library.
-//
-// The go-ethereum library is free software: you can redistribute it and/or modify
-// it under the terms of the GNU Lesser General Public License as published by
-// the Free Software Foundation, either version 3 of the License, or
-// (at your option) any later version.
-//
-// The go-ethereum library is distributed in the hope that it will be useful,
-// but WITHOUT ANY WARRANTY; without even the implied warranty of
-// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
-// GNU Lesser General Public License for more details.
-//
-// You should have received a copy of the GNU Lesser General Public License
-// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
-
-package les
-
-import (
-	"sync"
-)
-
-const dropTimeoutRatio = 20
-
-type odrPeerInfo struct {
-	reqTimeSum, reqTimeCnt, reqCnt, timeoutCnt uint64
-}
-
-// odrPeerSet represents the collection of active peer participating in the block
-// download procedure.
-type odrPeerSet struct {
-	peers map[*peer]*odrPeerInfo
-	lock  sync.RWMutex
-}
-
-// newPeerSet creates a new peer set top track the active download sources.
-func newOdrPeerSet() *odrPeerSet {
-	return &odrPeerSet{
-		peers: make(map[*peer]*odrPeerInfo),
-	}
-}
-
-// Register injects a new peer into the working set, or returns an error if the
-// peer is already known.
-func (ps *odrPeerSet) register(p *peer) error {
-	ps.lock.Lock()
-	defer ps.lock.Unlock()
-
-	if _, ok := ps.peers[p]; ok {
-		return errAlreadyRegistered
-	}
-	ps.peers[p] = &odrPeerInfo{}
-	return nil
-}
-
-// Unregister removes a remote peer from the active set, disabling any further
-// actions to/from that particular entity.
-func (ps *odrPeerSet) unregister(p *peer) error {
-	ps.lock.Lock()
-	defer ps.lock.Unlock()
-
-	if _, ok := ps.peers[p]; !ok {
-		return errNotRegistered
-	}
-	delete(ps.peers, p)
-	return nil
-}
-
-func (ps *odrPeerSet) peerPriority(p *peer, info *odrPeerInfo, req LesOdrRequest) uint64 {
-	tm := p.fcServer.CanSend(req.GetCost(p))
-	if info.reqTimeCnt > 0 {
-		tm += info.reqTimeSum / info.reqTimeCnt
-	}
-	return tm
-}
-
-func (ps *odrPeerSet) bestPeer(req LesOdrRequest, exclude map[*peer]struct{}) *peer {
-	var best *peer
-	var bpv uint64
-	ps.lock.Lock()
-	defer ps.lock.Unlock()
-
-	for p, info := range ps.peers {
-		if _, ok := exclude[p]; !ok {
-			pv := ps.peerPriority(p, info, req)
-			if best == nil || pv < bpv {
-				best = p
-				bpv = pv
-			}
-		}
-	}
-	return best
-}
-
-func (ps *odrPeerSet) updateTimeout(p *peer, timeout bool) (drop bool) {
-	ps.lock.Lock()
-	defer ps.lock.Unlock()
-
-	if info, ok := ps.peers[p]; ok {
-		info.reqCnt++
-		if timeout {
-			// check ratio before increase to allow an extra timeout
-			if info.timeoutCnt*dropTimeoutRatio >= info.reqCnt {
-				return true
-			}
-			info.timeoutCnt++
-		}
-	}
-	return false
-}
-
-func (ps *odrPeerSet) updateServTime(p *peer, servTime uint64) {
-	ps.lock.Lock()
-	defer ps.lock.Unlock()
-
-	if info, ok := ps.peers[p]; ok {
-		info.reqTimeSum += servTime
-		info.reqTimeCnt++
-	}
-}
diff --git a/les/odr_test.go b/les/odr_test.go
index 27e3b283d..685435996 100644
--- a/les/odr_test.go
+++ b/les/odr_test.go
@@ -160,6 +160,8 @@ func testOdr(t *testing.T, protocol int, expFail uint64, fn odrTestFn) {
 	pm, db, odr := newTestProtocolManagerMust(t, false, 4, testChainGen)
 	lpm, ldb, odr := newTestProtocolManagerMust(t, true, 0, nil)
 	_, err1, lpeer, err2 := newTestPeerPair("peer", protocol, pm, lpm)
+	pool := (*testServerPool)(lpeer)
+	odr.serverPool = pool
 	select {
 	case <-time.After(time.Millisecond * 100):
 	case err := <-err1:
@@ -188,13 +190,13 @@ func testOdr(t *testing.T, protocol int, expFail uint64, fn odrTestFn) {
 	}
 
 	// temporarily remove peer to test odr fails
-	odr.UnregisterPeer(lpeer)
+	odr.serverPool = nil
 	// expect retrievals to fail (except genesis block) without a les peer
 	test(expFail)
-	odr.RegisterPeer(lpeer)
+	odr.serverPool = pool
 	// expect all retrievals to pass
 	test(5)
-	odr.UnregisterPeer(lpeer)
+	odr.serverPool = nil
 	// still expect all retrievals to pass, now data should be cached locally
 	test(5)
 }
diff --git a/les/peer.go b/les/peer.go
index 8eced439d..8ebbe3511 100644
--- a/les/peer.go
+++ b/les/peer.go
@@ -51,9 +51,8 @@ type peer struct {
 
 	id string
 
-	firstHeadInfo, headInfo *announceData
-	headInfoLen             int
-	lock                    sync.RWMutex
+	headInfo *announceData
+	lock     sync.RWMutex
 
 	announceChn chan announceData
 
@@ -111,67 +110,6 @@ func (p *peer) headBlockInfo() blockInfo {
 	return blockInfo{Hash: p.headInfo.Hash, Number: p.headInfo.Number, Td: p.headInfo.Td}
 }
 
-func (p *peer) addNotify(announce *announceData) bool {
-	p.lock.Lock()
-	defer p.lock.Unlock()
-
-	if announce.Td.Cmp(p.headInfo.Td) < 1 {
-		return false
-	}
-	if p.headInfoLen >= maxHeadInfoLen {
-		//return false
-		p.firstHeadInfo = p.firstHeadInfo.next
-		p.headInfoLen--
-	}
-	if announce.haveHeaders == 0 {
-		hh := p.headInfo.Number - announce.ReorgDepth
-		if p.headInfo.haveHeaders < hh {
-			hh = p.headInfo.haveHeaders
-		}
-		announce.haveHeaders = hh
-	}
-	p.headInfo.next = announce
-	p.headInfo = announce
-	p.headInfoLen++
-	return true
-}
-
-func (p *peer) gotHeader(hash common.Hash, number uint64, td *big.Int) bool {
-	h := p.firstHeadInfo
-	ptr := 0
-	for h != nil {
-		if h.Hash == hash {
-			if h.Number != number || h.Td.Cmp(td) != 0 {
-				return false
-			}
-			h.headKnown = true
-			h.haveHeaders = h.Number
-			p.firstHeadInfo = h
-			p.headInfoLen -= ptr
-			last := h
-			h = h.next
-			// propagate haveHeaders through the chain
-			for h != nil {
-				hh := last.Number - h.ReorgDepth
-				if last.haveHeaders < hh {
-					hh = last.haveHeaders
-				}
-				if hh > h.haveHeaders {
-					h.haveHeaders = hh
-				} else {
-					return true
-				}
-				last = h
-				h = h.next
-			}
-			return true
-		}
-		h = h.next
-		ptr++
-	}
-	return true
-}
-
 // Td retrieves the current total difficulty of a peer.
 func (p *peer) Td() *big.Int {
 	p.lock.RLock()
@@ -455,9 +393,7 @@ func (p *peer) Handshake(td *big.Int, head common.Hash, headNum uint64, genesis
 		p.fcCosts = MRC.decode()
 	}
 
-	p.firstHeadInfo = &announceData{Td: rTd, Hash: rHash, Number: rNum}
-	p.headInfo = p.firstHeadInfo
-	p.headInfoLen = 1
+	p.headInfo = &announceData{Td: rTd, Hash: rHash, Number: rNum}
 	return nil
 }
 
diff --git a/les/request_test.go b/les/request_test.go
index a6fbb06ce..03b946771 100644
--- a/les/request_test.go
+++ b/les/request_test.go
@@ -71,6 +71,8 @@ func testAccess(t *testing.T, protocol int, fn accessTestFn) {
 	pm, db, _ := newTestProtocolManagerMust(t, false, 4, testChainGen)
 	lpm, ldb, odr := newTestProtocolManagerMust(t, true, 0, nil)
 	_, err1, lpeer, err2 := newTestPeerPair("peer", protocol, pm, lpm)
+	pool := (*testServerPool)(lpeer)
+	odr.serverPool = pool
 	select {
 	case <-time.After(time.Millisecond * 100):
 	case err := <-err1:
@@ -100,11 +102,10 @@ func testAccess(t *testing.T, protocol int, fn accessTestFn) {
 	}
 
 	// temporarily remove peer to test odr fails
-	odr.UnregisterPeer(lpeer)
+	odr.serverPool = nil
 	// expect retrievals to fail (except genesis block) without a les peer
 	test(0)
-	odr.RegisterPeer(lpeer)
+	odr.serverPool = pool
 	// expect all retrievals to pass
 	test(5)
-	odr.UnregisterPeer(lpeer)
 }
diff --git a/les/serverpool.go b/les/serverpool.go
index 95968eda2..f5e880460 100644
--- a/les/serverpool.go
+++ b/les/serverpool.go
@@ -59,6 +59,9 @@ const (
 	targetKnownSelect = 3
 	// after dialTimeout, consider the server unavailable and adjust statistics
 	dialTimeout = time.Second * 30
+	// targetConnTime is the minimum expected connection duration before a server
+	// drops a client without any specific reason
+	targetConnTime = time.Minute * 10
 	// new entry selection weight calculation based on most recent discovery time:
 	// unity until discoverExpireStart, then exponential decay with discoverExpireConst
 	discoverExpireStart = time.Minute * 20
@@ -75,6 +78,17 @@ const (
 	// node address selection weight is dropped by a factor of exp(-addrFailDropLn) after
 	// each unsuccessful connection (restored after a successful one)
 	addrFailDropLn = math.Ln2
+	// responseScoreTC and delayScoreTC are exponential decay time constants for
+	// calculating selection chances from response times and block delay times
+	responseScoreTC = time.Millisecond * 100
+	delayScoreTC    = time.Second * 5
+	timeoutPow      = 10
+	// peerSelectMinWeight is added to calculated weights at request peer selection
+	// to give poorly performing peers a little chance of coming back
+	peerSelectMinWeight = 0.005
+	// initStatsWeight is used to initialize previously unknown peers with good
+	// statistics to give a chance to prove themselves
+	initStatsWeight = 1
 )
 
 // serverPool implements a pool for storing and selecting newly discovered and already
@@ -95,6 +109,7 @@ type serverPool struct {
 	entries              map[discover.NodeID]*poolEntry
 	lock                 sync.Mutex
 	timeout, enableRetry chan *poolEntry
+	adjustStats          chan poolStatAdjust
 
 	knownQueue, newQueue       poolEntryQueue
 	knownSelect, newSelect     *weightedRandomSelect
@@ -112,6 +127,7 @@ func newServerPool(db ethdb.Database, dbPrefix []byte, server *p2p.Server, topic
 		wg:           wg,
 		entries:      make(map[discover.NodeID]*poolEntry),
 		timeout:      make(chan *poolEntry, 1),
+		adjustStats:  make(chan poolStatAdjust, 100),
 		enableRetry:  make(chan *poolEntry, 1),
 		knownSelect:  newWeightedRandomSelect(),
 		newSelect:    newWeightedRandomSelect(),
@@ -139,18 +155,19 @@ func newServerPool(db ethdb.Database, dbPrefix []byte, server *p2p.Server, topic
 // Otherwise, the connection should be rejected.
 // Note that whenever a connection has been accepted and a pool entry has been returned,
 // disconnect should also always be called.
-func (pool *serverPool) connect(id discover.NodeID, ip net.IP, port uint16) *poolEntry {
+func (pool *serverPool) connect(p *peer, ip net.IP, port uint16) *poolEntry {
 	pool.lock.Lock()
 	defer pool.lock.Unlock()
-	entry := pool.entries[id]
+	entry := pool.entries[p.ID()]
 	if entry == nil {
 		return nil
 	}
-	glog.V(logger.Debug).Infof("connecting to %v, state: %v", id.String(), entry.state)
+	glog.V(logger.Debug).Infof("connecting to %v, state: %v", p.id, entry.state)
 	if entry.state != psDialed {
 		return nil
 	}
 	pool.connWg.Add(1)
+	entry.peer = p
 	entry.state = psConnected
 	addr := &poolEntryAddress{
 		ip:       ip,
@@ -172,42 +189,111 @@ func (pool *serverPool) registered(entry *poolEntry) {
 	defer pool.lock.Unlock()
 
 	entry.state = psRegistered
+	entry.regTime = mclock.Now()
 	if !entry.known {
 		pool.newQueue.remove(entry)
 		entry.known = true
 	}
 	pool.knownQueue.setLatest(entry)
 	entry.shortRetry = shortRetryCnt
-	entry.connectStats.add(1)
 }
 
 // disconnect should be called when ending a connection. Service quality statistics
 // can be updated optionally (not updated if no registration happened, in this case
 // only connection statistics are updated, just like in case of timeout)
-func (pool *serverPool) disconnect(entry *poolEntry, quality float64, setQuality bool) {
+func (pool *serverPool) disconnect(entry *poolEntry) {
 	glog.V(logger.Debug).Infof("disconnected %v", entry.id.String())
 	pool.lock.Lock()
 	defer pool.lock.Unlock()
 
-	if entry.state != psRegistered {
-		setQuality = false
+	if entry.state == psRegistered {
+		connTime := mclock.Now() - entry.regTime
+		connAdjust := float64(connTime) / float64(targetConnTime)
+		if connAdjust > 1 {
+			connAdjust = 1
+		}
+		stopped := false
+		select {
+		case <-pool.quit:
+			stopped = true
+		default:
+		}
+		if stopped {
+			entry.connectStats.add(1, connAdjust)
+		} else {
+			entry.connectStats.add(connAdjust, 1)
+		}
 	}
+
 	entry.state = psNotConnected
 	if entry.knownSelected {
 		pool.knownSelected--
 	} else {
 		pool.newSelected--
 	}
-	if setQuality {
-		glog.V(logger.Debug).Infof("update quality %v %v", quality, entry.id.String())
-		entry.qualityStats.add(quality)
-	} else {
-		glog.V(logger.Debug).Infof("do not update quality")
-	}
 	pool.setRetryDial(entry)
 	pool.connWg.Done()
 }
 
+const (
+	pseBlockDelay = iota
+	pseResponseTime
+	pseResponseTimeout
+)
+
+// poolStatAdjust records are sent to adjust peer block delay/response time statistics
+type poolStatAdjust struct {
+	adjustType int
+	entry      *poolEntry
+	time       time.Duration
+}
+
+// adjustBlockDelay adjusts the block announce delay statistics of a node
+func (pool *serverPool) adjustBlockDelay(entry *poolEntry, time time.Duration) {
+	pool.adjustStats <- poolStatAdjust{pseBlockDelay, entry, time}
+}
+
+// adjustResponseTime adjusts the request response time statistics of a node
+func (pool *serverPool) adjustResponseTime(entry *poolEntry, time time.Duration, timeout bool) {
+	if timeout {
+		pool.adjustStats <- poolStatAdjust{pseResponseTimeout, entry, time}
+	} else {
+		pool.adjustStats <- poolStatAdjust{pseResponseTime, entry, time}
+	}
+}
+
+type selectPeerItem struct {
+	peer   *peer
+	weight int64
+}
+
+func (sp selectPeerItem) Weight() int64 {
+	return sp.weight
+}
+
+// selectPeer selects a suitable peer for a request
+func (pool *serverPool) selectPeer(canSend func(*peer) (bool, uint64)) *peer {
+	pool.lock.Lock()
+	defer pool.lock.Unlock()
+
+	sel := newWeightedRandomSelect()
+	for _, entry := range pool.entries {
+		if entry.state == psRegistered {
+			p := entry.peer
+			ok, cost := canSend(p)
+			if ok {
+				w := int64(1000000000 * (peerSelectMinWeight + math.Exp(-(entry.responseStats.recentAvg()+float64(cost))/float64(responseScoreTC))*math.Pow((1-entry.timeoutStats.recentAvg()), timeoutPow)))
+				sel.update(selectPeerItem{peer: p, weight: w})
+			}
+		}
+	}
+	choice := sel.choose()
+	if choice == nil {
+		return nil
+	}
+	return choice.(selectPeerItem).peer
+}
+
 // eventLoop handles pool events and mutex locking for all internal functions
 func (pool *serverPool) eventLoop() {
 	lookupCnt := 0
@@ -230,6 +316,19 @@ func (pool *serverPool) eventLoop() {
 			}
 			pool.lock.Unlock()
 
+		case adj := <-pool.adjustStats:
+			pool.lock.Lock()
+			switch adj.adjustType {
+			case pseBlockDelay:
+				adj.entry.delayStats.add(float64(adj.time), 1)
+			case pseResponseTime:
+				adj.entry.responseStats.add(float64(adj.time), 1)
+				adj.entry.timeoutStats.add(0, 1)
+			case pseResponseTimeout:
+				adj.entry.timeoutStats.add(1, 1)
+			}
+			pool.lock.Unlock()
+
 		case node := <-pool.discNodes:
 			pool.lock.Lock()
 			now := mclock.Now()
@@ -244,6 +343,11 @@ func (pool *serverPool) eventLoop() {
 					shortRetry: shortRetryCnt,
 				}
 				pool.entries[id] = entry
+				// initialize previously unknown peers with good statistics to give a chance to prove themselves
+				entry.connectStats.add(1, initStatsWeight)
+				entry.delayStats.add(0, initStatsWeight)
+				entry.responseStats.add(0, initStatsWeight)
+				entry.timeoutStats.add(0, initStatsWeight)
 			}
 			entry.lastDiscovered = now
 			addr := &poolEntryAddress{
@@ -298,9 +402,8 @@ func (pool *serverPool) loadNodes() {
 		glog.V(logger.Debug).Infof("node list decode error: %v", err)
 		return
 	}
-	glog.V(logger.Debug).Infof("loaded node list")
 	for _, e := range list {
-		glog.V(logger.Debug).Infof(" adding node %v  fails: %v  connStats sum: %v cnt: %v  qualityStats sum: %v cnt: %v", e.id.String()+"@"+e.lastConnected.strKey(), e.lastConnected.fails, e.connectStats.sum, e.connectStats.cnt, e.qualityStats.sum, e.qualityStats.cnt)
+		glog.V(logger.Debug).Infof("loaded server stats %016x  fails: %v  connStats: %v / %v  delayStats: %v / %v  responseStats: %v / %v  timeoutStats: %v / %v", e.id[0:8], e.lastConnected.fails, e.connectStats.avg, e.connectStats.weight, time.Duration(e.delayStats.avg), e.delayStats.weight, time.Duration(e.responseStats.avg), e.responseStats.weight, e.timeoutStats.avg, e.timeoutStats.weight)
 		pool.entries[e.id] = e
 		pool.knownQueue.setLatest(e)
 		pool.knownSelect.update((*knownEntry)(e))
@@ -433,7 +536,7 @@ func (pool *serverPool) checkDialTimeout(entry *poolEntry) {
 	} else {
 		pool.newSelected--
 	}
-	entry.connectStats.add(0)
+	entry.connectStats.add(0, 1)
 	entry.dialed.fails++
 	pool.setRetryDial(entry)
 }
@@ -447,33 +550,36 @@ const (
 
 // poolEntry represents a server node and stores its current state and statistics.
 type poolEntry struct {
+	peer                  *peer
 	id                    discover.NodeID
 	addr                  map[string]*poolEntryAddress
 	lastConnected, dialed *poolEntryAddress
 	addrSelect            weightedRandomSelect
 
-	lastDiscovered             mclock.AbsTime
-	known, knownSelected       bool
-	connectStats, qualityStats poolStats
-	state                      int
-	queueIdx                   int
-	removed                    bool
+	lastDiscovered              mclock.AbsTime
+	known, knownSelected        bool
+	connectStats, delayStats    poolStats
+	responseStats, timeoutStats poolStats
+	state                       int
+	regTime                     mclock.AbsTime
+	queueIdx                    int
+	removed                     bool
 
 	delayedRetry bool
 	shortRetry   int
 }
 
 func (e *poolEntry) EncodeRLP(w io.Writer) error {
-	return rlp.Encode(w, []interface{}{e.id, e.lastConnected.ip, e.lastConnected.port, e.lastConnected.fails, &e.connectStats, &e.qualityStats})
+	return rlp.Encode(w, []interface{}{e.id, e.lastConnected.ip, e.lastConnected.port, e.lastConnected.fails, &e.connectStats, &e.delayStats, &e.responseStats, &e.timeoutStats})
 }
 
 func (e *poolEntry) DecodeRLP(s *rlp.Stream) error {
 	var entry struct {
-		ID           discover.NodeID
-		IP           net.IP
-		Port         uint16
-		Fails        uint
-		CStat, QStat poolStats
+		ID                         discover.NodeID
+		IP                         net.IP
+		Port                       uint16
+		Fails                      uint
+		CStat, DStat, RStat, TStat poolStats
 	}
 	if err := s.Decode(&entry); err != nil {
 		return err
@@ -486,7 +592,9 @@ func (e *poolEntry) DecodeRLP(s *rlp.Stream) error {
 	e.addrSelect.update(addr)
 	e.lastConnected = addr
 	e.connectStats = entry.CStat
-	e.qualityStats = entry.QStat
+	e.delayStats = entry.DStat
+	e.responseStats = entry.RStat
+	e.timeoutStats = entry.TStat
 	e.shortRetry = shortRetryCnt
 	e.known = true
 	return nil
@@ -516,7 +624,7 @@ func (e *knownEntry) Weight() int64 {
 	if e.state != psNotConnected || !e.known || e.delayedRetry {
 		return 0
 	}
-	return int64(1000000000 * e.connectStats.recentAvg() * (e.qualityStats.recentAvg() + 0.001) * math.Exp(-float64(e.lastConnected.fails)*failDropLn))
+	return int64(1000000000 * e.connectStats.recentAvg() * math.Exp(-float64(e.lastConnected.fails)*failDropLn-e.responseStats.recentAvg()/float64(responseScoreTC)-e.delayStats.recentAvg()/float64(delayScoreTC)) * math.Pow((1-e.timeoutStats.recentAvg()), timeoutPow))
 }
 
 // poolEntryAddress is a separate object because currently it is necessary to remember
@@ -544,18 +652,17 @@ func (a *poolEntryAddress) strKey() string {
 // pstatRecentAdjust with each update and also returned exponentially to the
 // average with the time constant pstatReturnToMeanTC
 type poolStats struct {
-	sum, avg, recent float64
-	cnt              uint
-	lastRecalc       mclock.AbsTime
+	sum, weight, avg, recent float64
+	lastRecalc               mclock.AbsTime
 }
 
 // init initializes stats with a long term sum/update count pair retrieved from the database
-func (s *poolStats) init(sum float64, cnt uint) {
+func (s *poolStats) init(sum, weight float64) {
 	s.sum = sum
-	s.cnt = cnt
+	s.weight = weight
 	var avg float64
-	if cnt > 0 {
-		avg = s.sum / float64(cnt)
+	if weight > 0 {
+		avg = s.sum / weight
 	}
 	s.avg = avg
 	s.recent = avg
@@ -566,16 +673,22 @@ func (s *poolStats) init(sum float64, cnt uint) {
 func (s *poolStats) recalc() {
 	now := mclock.Now()
 	s.recent = s.avg + (s.recent-s.avg)*math.Exp(-float64(now-s.lastRecalc)/float64(pstatReturnToMeanTC))
-	if s.cnt > 0 {
-		s.avg = s.sum / float64(s.cnt)
+	if s.sum == 0 {
+		s.avg = 0
+	} else {
+		if s.sum > s.weight*1e30 {
+			s.avg = 1e30
+		} else {
+			s.avg = s.sum / s.weight
+		}
 	}
 	s.lastRecalc = now
 }
 
 // add updates the stats with a new value
-func (s *poolStats) add(val float64) {
-	s.cnt++
-	s.sum += val
+func (s *poolStats) add(value, weight float64) {
+	s.weight += weight
+	s.sum += value * weight
 	s.recalc()
 }
 
@@ -586,18 +699,17 @@ func (s *poolStats) recentAvg() float64 {
 }
 
 func (s *poolStats) EncodeRLP(w io.Writer) error {
-	return rlp.Encode(w, []interface{}{math.Float64bits(s.sum), s.cnt})
+	return rlp.Encode(w, []interface{}{math.Float64bits(s.sum), math.Float64bits(s.weight)})
 }
 
 func (s *poolStats) DecodeRLP(st *rlp.Stream) error {
 	var stats struct {
-		SumUint uint64
-		Cnt     uint
+		SumUint, WeightUint uint64
 	}
 	if err := st.Decode(&stats); err != nil {
 		return err
 	}
-	s.init(math.Float64frombits(stats.SumUint), stats.Cnt)
+	s.init(math.Float64frombits(stats.SumUint), math.Float64frombits(stats.WeightUint))
 	return nil
 }