eth, p2p/msgrate: move peer QoS tracking to its own package and use it for snap (#22876)

This change extracts the peer QoS tracking logic from eth/downloader, moving
it into the new package p2p/msgrate. The job of msgrate.Tracker is determining
suitable timeout values and request sizes per peer.

The snap sync scheduler now uses msgrate.Tracker instead of the hard-coded 15s
timeout. This should make the sync work better on network links with high latency.

eth/downloader/downloader.go

@@ -47,16 +47,6 @@ var (
 	MaxReceiptFetch = 256 // Amount of transaction receipts to allow fetching per request
 	MaxStateFetch   = 384 // Amount of node state values to allow fetching per request
 
-	rttMinEstimate   = 2 * time.Second  // Minimum round-trip time to target for download requests
-	rttMaxEstimate   = 20 * time.Second // Maximum round-trip time to target for download requests
-	rttMinConfidence = 0.1              // Worse confidence factor in our estimated RTT value
-	ttlScaling       = 3                // Constant scaling factor for RTT -> TTL conversion
-	ttlLimit         = time.Minute      // Maximum TTL allowance to prevent reaching crazy timeouts
-
-	qosTuningPeers   = 5    // Number of peers to tune based on (best peers)
-	qosConfidenceCap = 10   // Number of peers above which not to modify RTT confidence
-	qosTuningImpact  = 0.25 // Impact that a new tuning target has on the previous value
-
 	maxQueuedHeaders            = 32 * 1024                         // [eth/62] Maximum number of headers to queue for import (DOS protection)
 	maxHeadersProcess           = 2048                              // Number of header download results to import at once into the chain
 	maxResultsProcess           = 2048                              // Number of content download results to import at once into the chain
@@ -96,13 +86,6 @@ var (
 )
 
 type Downloader struct {
-	// WARNING: The `rttEstimate` and `rttConfidence` fields are accessed atomically.
-	// On 32 bit platforms, only 64-bit aligned fields can be atomic. The struct is
-	// guaranteed to be so aligned, so take advantage of that. For more information,
-	// see https://golang.org/pkg/sync/atomic/#pkg-note-BUG.
-	rttEstimate   uint64 // Round trip time to target for download requests
-	rttConfidence uint64 // Confidence in the estimated RTT (unit: millionths to allow atomic ops)
-
 	mode uint32         // Synchronisation mode defining the strategy used (per sync cycle), use d.getMode() to get the SyncMode
 	mux  *event.TypeMux // Event multiplexer to announce sync operation events
 
@@ -232,8 +215,6 @@ func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom,
 		checkpoint:     checkpoint,
 		queue:          newQueue(blockCacheMaxItems, blockCacheInitialItems),
 		peers:          newPeerSet(),
-		rttEstimate:    uint64(rttMaxEstimate),
-		rttConfidence:  uint64(1000000),
 		blockchain:     chain,
 		lightchain:     lightchain,
 		dropPeer:       dropPeer,
@@ -252,7 +233,6 @@ func New(checkpoint uint64, stateDb ethdb.Database, stateBloom *trie.SyncBloom,
 		},
 		trackStateReq: make(chan *stateReq),
 	}
-	go dl.qosTuner()
 	go dl.stateFetcher()
 	return dl
 }
@@ -310,8 +290,6 @@ func (d *Downloader) RegisterPeer(id string, version uint, peer Peer) error {
 		logger.Error("Failed to register sync peer", "err", err)
 		return err
 	}
-	d.qosReduceConfidence()
-
 	return nil
 }
 
@@ -670,7 +648,7 @@ func (d *Downloader) fetchHead(p *peerConnection) (head *types.Header, pivot *ty
 	}
 	go p.peer.RequestHeadersByHash(latest, fetch, fsMinFullBlocks-1, true)
 
-	ttl := d.requestTTL()
+	ttl := d.peers.rates.TargetTimeout()
 	timeout := time.After(ttl)
 	for {
 		select {
@@ -853,7 +831,7 @@ func (d *Downloader) findAncestorSpanSearch(p *peerConnection, mode SyncMode, re
 	// Wait for the remote response to the head fetch
 	number, hash := uint64(0), common.Hash{}
 
-	ttl := d.requestTTL()
+	ttl := d.peers.rates.TargetTimeout()
 	timeout := time.After(ttl)
 
 	for finished := false; !finished; {
@@ -942,7 +920,7 @@ func (d *Downloader) findAncestorBinarySearch(p *peerConnection, mode SyncMode,
 		// Split our chain interval in two, and request the hash to cross check
 		check := (start + end) / 2
 
-		ttl := d.requestTTL()
+		ttl := d.peers.rates.TargetTimeout()
 		timeout := time.After(ttl)
 
 		go p.peer.RequestHeadersByNumber(check, 1, 0, false)
@@ -1035,7 +1013,7 @@ func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error {
 	getHeaders := func(from uint64) {
 		request = time.Now()
 
-		ttl = d.requestTTL()
+		ttl = d.peers.rates.TargetTimeout()
 		timeout.Reset(ttl)
 
 		if skeleton {
@@ -1050,7 +1028,7 @@ func (d *Downloader) fetchHeaders(p *peerConnection, from uint64) error {
 		pivoting = true
 		request = time.Now()
 
-		ttl = d.requestTTL()
+		ttl = d.peers.rates.TargetTimeout()
 		timeout.Reset(ttl)
 
 		d.pivotLock.RLock()
@@ -1262,12 +1240,12 @@ func (d *Downloader) fillHeaderSkeleton(from uint64, skeleton []*types.Header) (
 			pack := packet.(*headerPack)
 			return d.queue.DeliverHeaders(pack.peerID, pack.headers, d.headerProcCh)
 		}
-		expire  = func() map[string]int { return d.queue.ExpireHeaders(d.requestTTL()) }
+		expire  = func() map[string]int { return d.queue.ExpireHeaders(d.peers.rates.TargetTimeout()) }
 		reserve = func(p *peerConnection, count int) (*fetchRequest, bool, bool) {
 			return d.queue.ReserveHeaders(p, count), false, false
 		}
 		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchHeaders(req.From, MaxHeaderFetch) }
-		capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.requestRTT()) }
+		capacity = func(p *peerConnection) int { return p.HeaderCapacity(d.peers.rates.TargetRoundTrip()) }
 		setIdle  = func(p *peerConnection, accepted int, deliveryTime time.Time) {
 			p.SetHeadersIdle(accepted, deliveryTime)
 		}
@@ -1293,9 +1271,9 @@ func (d *Downloader) fetchBodies(from uint64) error {
 			pack := packet.(*bodyPack)
 			return d.queue.DeliverBodies(pack.peerID, pack.transactions, pack.uncles)
 		}
-		expire   = func() map[string]int { return d.queue.ExpireBodies(d.requestTTL()) }
+		expire   = func() map[string]int { return d.queue.ExpireBodies(d.peers.rates.TargetTimeout()) }
 		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchBodies(req) }
-		capacity = func(p *peerConnection) int { return p.BlockCapacity(d.requestRTT()) }
+		capacity = func(p *peerConnection) int { return p.BlockCapacity(d.peers.rates.TargetRoundTrip()) }
 		setIdle  = func(p *peerConnection, accepted int, deliveryTime time.Time) { p.SetBodiesIdle(accepted, deliveryTime) }
 	)
 	err := d.fetchParts(d.bodyCh, deliver, d.bodyWakeCh, expire,
@@ -1317,9 +1295,9 @@ func (d *Downloader) fetchReceipts(from uint64) error {
 			pack := packet.(*receiptPack)
 			return d.queue.DeliverReceipts(pack.peerID, pack.receipts)
 		}
-		expire   = func() map[string]int { return d.queue.ExpireReceipts(d.requestTTL()) }
+		expire   = func() map[string]int { return d.queue.ExpireReceipts(d.peers.rates.TargetTimeout()) }
 		fetch    = func(p *peerConnection, req *fetchRequest) error { return p.FetchReceipts(req) }
-		capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.requestRTT()) }
+		capacity = func(p *peerConnection) int { return p.ReceiptCapacity(d.peers.rates.TargetRoundTrip()) }
 		setIdle  = func(p *peerConnection, accepted int, deliveryTime time.Time) {
 			p.SetReceiptsIdle(accepted, deliveryTime)
 		}
@@ -2031,78 +2009,3 @@ func (d *Downloader) deliver(destCh chan dataPack, packet dataPack, inMeter, dro
 		return errNoSyncActive
 	}
 }
-
-// qosTuner is the quality of service tuning loop that occasionally gathers the
-// peer latency statistics and updates the estimated request round trip time.
-func (d *Downloader) qosTuner() {
-	for {
-		// Retrieve the current median RTT and integrate into the previoust target RTT
-		rtt := time.Duration((1-qosTuningImpact)*float64(atomic.LoadUint64(&d.rttEstimate)) + qosTuningImpact*float64(d.peers.medianRTT()))
-		atomic.StoreUint64(&d.rttEstimate, uint64(rtt))
-
-		// A new RTT cycle passed, increase our confidence in the estimated RTT
-		conf := atomic.LoadUint64(&d.rttConfidence)
-		conf = conf + (1000000-conf)/2
-		atomic.StoreUint64(&d.rttConfidence, conf)
-
-		// Log the new QoS values and sleep until the next RTT
-		log.Debug("Recalculated downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
-		select {
-		case <-d.quitCh:
-			return
-		case <-time.After(rtt):
-		}
-	}
-}
-
-// qosReduceConfidence is meant to be called when a new peer joins the downloader's
-// peer set, needing to reduce the confidence we have in out QoS estimates.
-func (d *Downloader) qosReduceConfidence() {
-	// If we have a single peer, confidence is always 1
-	peers := uint64(d.peers.Len())
-	if peers == 0 {
-		// Ensure peer connectivity races don't catch us off guard
-		return
-	}
-	if peers == 1 {
-		atomic.StoreUint64(&d.rttConfidence, 1000000)
-		return
-	}
-	// If we have a ton of peers, don't drop confidence)
-	if peers >= uint64(qosConfidenceCap) {
-		return
-	}
-	// Otherwise drop the confidence factor
-	conf := atomic.LoadUint64(&d.rttConfidence) * (peers - 1) / peers
-	if float64(conf)/1000000 < rttMinConfidence {
-		conf = uint64(rttMinConfidence * 1000000)
-	}
-	atomic.StoreUint64(&d.rttConfidence, conf)
-
-	rtt := time.Duration(atomic.LoadUint64(&d.rttEstimate))
-	log.Debug("Relaxed downloader QoS values", "rtt", rtt, "confidence", float64(conf)/1000000.0, "ttl", d.requestTTL())
-}
-
-// requestRTT returns the current target round trip time for a download request
-// to complete in.
-//
-// Note, the returned RTT is .9 of the actually estimated RTT. The reason is that
-// the downloader tries to adapt queries to the RTT, so multiple RTT values can
-// be adapted to, but smaller ones are preferred (stabler download stream).
-func (d *Downloader) requestRTT() time.Duration {
-	return time.Duration(atomic.LoadUint64(&d.rttEstimate)) * 9 / 10
-}
-
-// requestTTL returns the current timeout allowance for a single download request
-// to finish under.
-func (d *Downloader) requestTTL() time.Duration {
-	var (
-		rtt  = time.Duration(atomic.LoadUint64(&d.rttEstimate))
-		conf = float64(atomic.LoadUint64(&d.rttConfidence)) / 1000000.0
-	)
-	ttl := time.Duration(ttlScaling) * time.Duration(float64(rtt)/conf)
-	if ttl > ttlLimit {
-		ttl = ttlLimit
-	}
-	return ttl
-}

eth/downloader/peer.go

@@ -32,11 +32,11 @@ import (
 	"github.com/ethereum/go-ethereum/eth/protocols/eth"
 	"github.com/ethereum/go-ethereum/event"
 	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/p2p/msgrate"
 )
 
 const (
 	maxLackingHashes = 4096 // Maximum number of entries allowed on the list or lacking items
-	measurementImpact = 0.1  // The impact a single measurement has on a peer's final throughput value.
 )
 
 var (
@@ -54,18 +54,12 @@ type peerConnection struct {
 	receiptIdle int32 // Current receipt activity state of the peer (idle = 0, active = 1)
 	stateIdle   int32 // Current node data activity state of the peer (idle = 0, active = 1)
 
-	headerThroughput  float64 // Number of headers measured to be retrievable per second
-	blockThroughput   float64 // Number of blocks (bodies) measured to be retrievable per second
-	receiptThroughput float64 // Number of receipts measured to be retrievable per second
-	stateThroughput   float64 // Number of node data pieces measured to be retrievable per second
-
-	rtt time.Duration // Request round trip time to track responsiveness (QoS)
-
 	headerStarted  time.Time // Time instance when the last header fetch was started
 	blockStarted   time.Time // Time instance when the last block (body) fetch was started
 	receiptStarted time.Time // Time instance when the last receipt fetch was started
 	stateStarted   time.Time // Time instance when the last node data fetch was started
 
+	rates   *msgrate.Tracker         // Tracker to hone in on the number of items retrievable per second
 	lacking map[common.Hash]struct{} // Set of hashes not to request (didn't have previously)
 
 	peer Peer
@@ -133,11 +127,6 @@ func (p *peerConnection) Reset() {
 	atomic.StoreInt32(&p.receiptIdle, 0)
 	atomic.StoreInt32(&p.stateIdle, 0)
 
-	p.headerThroughput = 0
-	p.blockThroughput = 0
-	p.receiptThroughput = 0
-	p.stateThroughput = 0
-
 	p.lacking = make(map[common.Hash]struct{})
 }
 
@@ -212,93 +201,72 @@ func (p *peerConnection) FetchNodeData(hashes []common.Hash) error {
 // requests. Its estimated header retrieval throughput is updated with that measured
 // just now.
 func (p *peerConnection) SetHeadersIdle(delivered int, deliveryTime time.Time) {
-	p.setIdle(deliveryTime.Sub(p.headerStarted), delivered, &p.headerThroughput, &p.headerIdle)
+	p.rates.Update(eth.BlockHeadersMsg, deliveryTime.Sub(p.headerStarted), delivered)
+	atomic.StoreInt32(&p.headerIdle, 0)
 }
 
 // SetBodiesIdle sets the peer to idle, allowing it to execute block body retrieval
 // requests. Its estimated body retrieval throughput is updated with that measured
 // just now.
 func (p *peerConnection) SetBodiesIdle(delivered int, deliveryTime time.Time) {
-	p.setIdle(deliveryTime.Sub(p.blockStarted), delivered, &p.blockThroughput, &p.blockIdle)
+	p.rates.Update(eth.BlockBodiesMsg, deliveryTime.Sub(p.blockStarted), delivered)
+	atomic.StoreInt32(&p.blockIdle, 0)
 }
 
 // SetReceiptsIdle sets the peer to idle, allowing it to execute new receipt
 // retrieval requests. Its estimated receipt retrieval throughput is updated
 // with that measured just now.
 func (p *peerConnection) SetReceiptsIdle(delivered int, deliveryTime time.Time) {
-	p.setIdle(deliveryTime.Sub(p.receiptStarted), delivered, &p.receiptThroughput, &p.receiptIdle)
+	p.rates.Update(eth.ReceiptsMsg, deliveryTime.Sub(p.receiptStarted), delivered)
+	atomic.StoreInt32(&p.receiptIdle, 0)
 }
 
 // SetNodeDataIdle sets the peer to idle, allowing it to execute new state trie
 // data retrieval requests. Its estimated state retrieval throughput is updated
 // with that measured just now.
 func (p *peerConnection) SetNodeDataIdle(delivered int, deliveryTime time.Time) {
-	p.setIdle(deliveryTime.Sub(p.stateStarted), delivered, &p.stateThroughput, &p.stateIdle)
+	p.rates.Update(eth.NodeDataMsg, deliveryTime.Sub(p.stateStarted), delivered)
-}
+	atomic.StoreInt32(&p.stateIdle, 0)
-
-// setIdle sets the peer to idle, allowing it to execute new retrieval requests.
-// Its estimated retrieval throughput is updated with that measured just now.
-func (p *peerConnection) setIdle(elapsed time.Duration, delivered int, throughput *float64, idle *int32) {
-	// Irrelevant of the scaling, make sure the peer ends up idle
-	defer atomic.StoreInt32(idle, 0)
-
-	p.lock.Lock()
-	defer p.lock.Unlock()
-
-	// If nothing was delivered (hard timeout / unavailable data), reduce throughput to minimum
-	if delivered == 0 {
-		*throughput = 0
-		return
-	}
-	// Otherwise update the throughput with a new measurement
-	if elapsed <= 0 {
-		elapsed = 1 // +1 (ns) to ensure non-zero divisor
-	}
-	measured := float64(delivered) / (float64(elapsed) / float64(time.Second))
-
-	*throughput = (1-measurementImpact)*(*throughput) + measurementImpact*measured
-	p.rtt = time.Duration((1-measurementImpact)*float64(p.rtt) + measurementImpact*float64(elapsed))
-
-	p.log.Trace("Peer throughput measurements updated",
-		"hps", p.headerThroughput, "bps", p.blockThroughput,
-		"rps", p.receiptThroughput, "sps", p.stateThroughput,
-		"miss", len(p.lacking), "rtt", p.rtt)
 }
 
 // HeaderCapacity retrieves the peers header download allowance based on its
 // previously discovered throughput.
 func (p *peerConnection) HeaderCapacity(targetRTT time.Duration) int {
-	p.lock.RLock()
+	cap := int(math.Ceil(p.rates.Capacity(eth.BlockHeadersMsg, targetRTT)))
-	defer p.lock.RUnlock()
+	if cap > MaxHeaderFetch {
-
+		cap = MaxHeaderFetch
-	return int(math.Min(1+math.Max(1, p.headerThroughput*float64(targetRTT)/float64(time.Second)), float64(MaxHeaderFetch)))
+	}
+	return cap
 }
 
 // BlockCapacity retrieves the peers block download allowance based on its
 // previously discovered throughput.
 func (p *peerConnection) BlockCapacity(targetRTT time.Duration) int {
-	p.lock.RLock()
+	cap := int(math.Ceil(p.rates.Capacity(eth.BlockBodiesMsg, targetRTT)))
-	defer p.lock.RUnlock()
+	if cap > MaxBlockFetch {
-
+		cap = MaxBlockFetch
-	return int(math.Min(1+math.Max(1, p.blockThroughput*float64(targetRTT)/float64(time.Second)), float64(MaxBlockFetch)))
+	}
+	return cap
 }
 
 // ReceiptCapacity retrieves the peers receipt download allowance based on its
 // previously discovered throughput.
 func (p *peerConnection) ReceiptCapacity(targetRTT time.Duration) int {
-	p.lock.RLock()
+	cap := int(math.Ceil(p.rates.Capacity(eth.ReceiptsMsg, targetRTT)))
-	defer p.lock.RUnlock()
+	if cap > MaxReceiptFetch {
-
+		cap = MaxReceiptFetch
-	return int(math.Min(1+math.Max(1, p.receiptThroughput*float64(targetRTT)/float64(time.Second)), float64(MaxReceiptFetch)))
+	}
+	return cap
 }
 
 // NodeDataCapacity retrieves the peers state download allowance based on its
 // previously discovered throughput.
 func (p *peerConnection) NodeDataCapacity(targetRTT time.Duration) int {
-	p.lock.RLock()
+	cap := int(math.Ceil(p.rates.Capacity(eth.NodeDataMsg, targetRTT)))
-	defer p.lock.RUnlock()
+	if cap > MaxStateFetch {
-
+		cap = MaxStateFetch
-	return int(math.Min(1+math.Max(1, p.stateThroughput*float64(targetRTT)/float64(time.Second)), float64(MaxStateFetch)))
+	}
+	return cap
 }
 
 // MarkLacking appends a new entity to the set of items (blocks, receipts, states)
@@ -331,8 +299,11 @@ func (p *peerConnection) Lacks(hash common.Hash) bool {
 // download procedure.
 type peerSet struct {
 	peers map[string]*peerConnection
+	rates *msgrate.Trackers // Set of rate trackers to give the sync a common beat
+
 	newPeerFeed  event.Feed
 	peerDropFeed event.Feed
+
 	lock sync.RWMutex
 }
 
@@ -340,6 +311,7 @@ type peerSet struct {
 func newPeerSet() *peerSet {
 	return &peerSet{
 		peers: make(map[string]*peerConnection),
+		rates: msgrate.NewTrackers(log.New("proto", "eth")),
 	}
 }
 
@@ -371,30 +343,15 @@ func (ps *peerSet) Reset() {
 // average of all existing peers, to give it a realistic chance of being used
 // for data retrievals.
 func (ps *peerSet) Register(p *peerConnection) error {
-	// Retrieve the current median RTT as a sane default
-	p.rtt = ps.medianRTT()
-
 	// Register the new peer with some meaningful defaults
 	ps.lock.Lock()
 	if _, ok := ps.peers[p.id]; ok {
 		ps.lock.Unlock()
 		return errAlreadyRegistered
 	}
-	if len(ps.peers) > 0 {
+	p.rates = msgrate.NewTracker(ps.rates.MeanCapacities(), ps.rates.MedianRoundTrip())
-		p.headerThroughput, p.blockThroughput, p.receiptThroughput, p.stateThroughput = 0, 0, 0, 0
+	if err := ps.rates.Track(p.id, p.rates); err != nil {
-
+		return err
-		for _, peer := range ps.peers {
-			peer.lock.RLock()
-			p.headerThroughput += peer.headerThroughput
-			p.blockThroughput += peer.blockThroughput
-			p.receiptThroughput += peer.receiptThroughput
-			p.stateThroughput += peer.stateThroughput
-			peer.lock.RUnlock()
-		}
-		p.headerThroughput /= float64(len(ps.peers))
-		p.blockThroughput /= float64(len(ps.peers))
-		p.receiptThroughput /= float64(len(ps.peers))
-		p.stateThroughput /= float64(len(ps.peers))
 	}
 	ps.peers[p.id] = p
 	ps.lock.Unlock()
@@ -413,6 +370,7 @@ func (ps *peerSet) Unregister(id string) error {
 		return errNotRegistered
 	}
 	delete(ps.peers, id)
+	ps.rates.Untrack(id)
 	ps.lock.Unlock()
 
 	ps.peerDropFeed.Send(p)
@@ -454,9 +412,7 @@ func (ps *peerSet) HeaderIdlePeers() ([]*peerConnection, int) {
 		return atomic.LoadInt32(&p.headerIdle) == 0
 	}
 	throughput := func(p *peerConnection) float64 {
-		p.lock.RLock()
+		return p.rates.Capacity(eth.BlockHeadersMsg, time.Second)
-		defer p.lock.RUnlock()
-		return p.headerThroughput
 	}
 	return ps.idlePeers(eth.ETH65, eth.ETH66, idle, throughput)
 }
@@ -468,9 +424,7 @@ func (ps *peerSet) BodyIdlePeers() ([]*peerConnection, int) {
 		return atomic.LoadInt32(&p.blockIdle) == 0
 	}
 	throughput := func(p *peerConnection) float64 {
-		p.lock.RLock()
+		return p.rates.Capacity(eth.BlockBodiesMsg, time.Second)
-		defer p.lock.RUnlock()
-		return p.blockThroughput
 	}
 	return ps.idlePeers(eth.ETH65, eth.ETH66, idle, throughput)
 }
@@ -482,9 +436,7 @@ func (ps *peerSet) ReceiptIdlePeers() ([]*peerConnection, int) {
 		return atomic.LoadInt32(&p.receiptIdle) == 0
 	}
 	throughput := func(p *peerConnection) float64 {
-		p.lock.RLock()
+		return p.rates.Capacity(eth.ReceiptsMsg, time.Second)
-		defer p.lock.RUnlock()
-		return p.receiptThroughput
 	}
 	return ps.idlePeers(eth.ETH65, eth.ETH66, idle, throughput)
 }
@@ -496,9 +448,7 @@ func (ps *peerSet) NodeDataIdlePeers() ([]*peerConnection, int) {
 		return atomic.LoadInt32(&p.stateIdle) == 0
 	}
 	throughput := func(p *peerConnection) float64 {
-		p.lock.RLock()
+		return p.rates.Capacity(eth.NodeDataMsg, time.Second)
-		defer p.lock.RUnlock()
-		return p.stateThroughput
 	}
 	return ps.idlePeers(eth.ETH65, eth.ETH66, idle, throughput)
 }
@@ -527,37 +477,6 @@ func (ps *peerSet) idlePeers(minProtocol, maxProtocol uint, idleCheck func(*peer
 	return sortPeers.p, total
 }
 
-// medianRTT returns the median RTT of the peerset, considering only the tuning
-// peers if there are more peers available.
-func (ps *peerSet) medianRTT() time.Duration {
-	// Gather all the currently measured round trip times
-	ps.lock.RLock()
-	defer ps.lock.RUnlock()
-
-	rtts := make([]float64, 0, len(ps.peers))
-	for _, p := range ps.peers {
-		p.lock.RLock()
-		rtts = append(rtts, float64(p.rtt))
-		p.lock.RUnlock()
-	}
-	sort.Float64s(rtts)
-
-	median := rttMaxEstimate
-	if qosTuningPeers <= len(rtts) {
-		median = time.Duration(rtts[qosTuningPeers/2]) // Median of our tuning peers
-	} else if len(rtts) > 0 {
-		median = time.Duration(rtts[len(rtts)/2]) // Median of our connected peers (maintain even like this some baseline qos)
-	}
-	// Restrict the RTT into some QoS defaults, irrelevant of true RTT
-	if median < rttMinEstimate {
-		median = rttMinEstimate
-	}
-	if median > rttMaxEstimate {
-		median = rttMaxEstimate
-	}
-	return median
-}
-
 // peerThroughputSort implements the Sort interface, and allows for
 // sorting a set of peers by their throughput
 // The sorted data is with the _highest_ throughput first

eth/downloader/peer_test.go

@@ -1,53 +0,0 @@
-// Copyright 2020 The go-ethereum Authors
-// This file is part of go-ethereum.
-//
-// go-ethereum is free software: you can redistribute it and/or modify
-// it under the terms of the GNU General Public License as published by
-// the Free Software Foundation, either version 3 of the License, or
-// (at your option) any later version.
-//
-// go-ethereum 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 General Public License for more details.
-//
-// You should have received a copy of the GNU General Public License
-// along with go-ethereum. If not, see <http://www.gnu.org/licenses/>.
-
-package downloader
-
-import (
-	"sort"
-	"testing"
-)
-
-func TestPeerThroughputSorting(t *testing.T) {
-	a := &peerConnection{
-		id:               "a",
-		headerThroughput: 1.25,
-	}
-	b := &peerConnection{
-		id:               "b",
-		headerThroughput: 1.21,
-	}
-	c := &peerConnection{
-		id:               "c",
-		headerThroughput: 1.23,
-	}
-
-	peers := []*peerConnection{a, b, c}
-	tps := []float64{a.headerThroughput,
-		b.headerThroughput, c.headerThroughput}
-	sortPeers := &peerThroughputSort{peers, tps}
-	sort.Sort(sortPeers)
-	if got, exp := sortPeers.p[0].id, "a"; got != exp {
-		t.Errorf("sort fail, got %v exp %v", got, exp)
-	}
-	if got, exp := sortPeers.p[1].id, "c"; got != exp {
-		t.Errorf("sort fail, got %v exp %v", got, exp)
-	}
-	if got, exp := sortPeers.p[2].id, "b"; got != exp {
-		t.Errorf("sort fail, got %v exp %v", got, exp)
-	}
-
-}

eth/downloader/statesync.go

@@ -433,8 +433,8 @@ func (s *stateSync) assignTasks() {
 	peers, _ := s.d.peers.NodeDataIdlePeers()
 	for _, p := range peers {
 		// Assign a batch of fetches proportional to the estimated latency/bandwidth
-		cap := p.NodeDataCapacity(s.d.requestRTT())
+		cap := p.NodeDataCapacity(s.d.peers.rates.TargetRoundTrip())
-		req := &stateReq{peer: p, timeout: s.d.requestTTL()}
+		req := &stateReq{peer: p, timeout: s.d.peers.rates.TargetTimeout()}
 
 		nodes, _, codes := s.fillTasks(cap, req)
 

eth/protocols/snap/sync.go

@@ -37,6 +37,7 @@ import (
 	"github.com/ethereum/go-ethereum/event"
 	"github.com/ethereum/go-ethereum/light"
 	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/p2p/msgrate"
 	"github.com/ethereum/go-ethereum/rlp"
 	"github.com/ethereum/go-ethereum/trie"
 	"golang.org/x/crypto/sha3"
@@ -51,14 +52,15 @@ var (
 )
 
 const (
-	// maxRequestSize is the maximum number of bytes to request from a remote peer.
+	// minRequestSize is the minimum number of bytes to request from a remote peer.
-	maxRequestSize = 128 * 1024
+	// This number is used as the low cap for account and storage range requests.
+	// Bytecode and trienode are limited inherently by item count (1).
+	minRequestSize = 64 * 1024
 
-	// maxStorageSetRequestCount is the maximum number of contracts to request the
+	// maxRequestSize is the maximum number of bytes to request from a remote peer.
-	// storage of in a single query. If this number is too low, we're not filling
+	// This number is used as the high cap for account and storage range requests.
-	// responses fully and waste round trip times. If it's too high, we're capping
+	// Bytecode and trienode are limited more explicitly by the caps below.
-	// responses and waste bandwidth.
+	maxRequestSize = 512 * 1024
-	maxStorageSetRequestCount = maxRequestSize / 1024
 
 	// maxCodeRequestCount is the maximum number of bytecode blobs to request in a
 	// single query. If this number is too low, we're not filling responses fully
@@ -74,7 +76,7 @@ const (
 	// a single query. If this number is too low, we're not filling responses fully
 	// and waste round trip times. If it's too high, we're capping responses and
 	// waste bandwidth.
-	maxTrieRequestCount = 256
+	maxTrieRequestCount = maxRequestSize / 512
 )
 
 var (
@@ -85,10 +87,6 @@ var (
 	// storageConcurrency is the number of chunks to split the a large contract
 	// storage trie into to allow concurrent retrievals.
 	storageConcurrency = 16
-
-	// requestTimeout is the maximum time a peer is allowed to spend on serving
-	// a single network request.
-	requestTimeout = 15 * time.Second // TODO(karalabe): Make it dynamic ala fast-sync?
 )
 
 // ErrCancelled is returned from snap syncing if the operation was prematurely
@@ -107,6 +105,7 @@ var ErrCancelled = errors.New("sync cancelled")
 type accountRequest struct {
 	peer string    // Peer to which this request is assigned
 	id   uint64    // Request ID of this request
+	time time.Time // Timestamp when the request was sent
 
 	deliver chan *accountResponse // Channel to deliver successful response on
 	revert  chan *accountRequest  // Channel to deliver request failure on
@@ -144,6 +143,7 @@ type accountResponse struct {
 type bytecodeRequest struct {
 	peer string    // Peer to which this request is assigned
 	id   uint64    // Request ID of this request
+	time time.Time // Timestamp when the request was sent
 
 	deliver chan *bytecodeResponse // Channel to deliver successful response on
 	revert  chan *bytecodeRequest  // Channel to deliver request failure on
@@ -175,6 +175,7 @@ type bytecodeResponse struct {
 type storageRequest struct {
 	peer string    // Peer to which this request is assigned
 	id   uint64    // Request ID of this request
+	time time.Time // Timestamp when the request was sent
 
 	deliver chan *storageResponse // Channel to deliver successful response on
 	revert  chan *storageRequest  // Channel to deliver request failure on
@@ -220,6 +221,7 @@ type storageResponse struct {
 type trienodeHealRequest struct {
 	peer string    // Peer to which this request is assigned
 	id   uint64    // Request ID of this request
+	time time.Time // Timestamp when the request was sent
 
 	deliver chan *trienodeHealResponse // Channel to deliver successful response on
 	revert  chan *trienodeHealRequest  // Channel to deliver request failure on
@@ -254,6 +256,7 @@ type trienodeHealResponse struct {
 type bytecodeHealRequest struct {
 	peer string    // Peer to which this request is assigned
 	id   uint64    // Request ID of this request
+	time time.Time // Timestamp when the request was sent
 
 	deliver chan *bytecodeHealResponse // Channel to deliver successful response on
 	revert  chan *bytecodeHealRequest  // Channel to deliver request failure on
@@ -396,6 +399,7 @@ type Syncer struct {
 	peers    map[string]SyncPeer // Currently active peers to download from
 	peerJoin *event.Feed         // Event feed to react to peers joining
 	peerDrop *event.Feed         // Event feed to react to peers dropping
+	rates    *msgrate.Trackers   // Message throughput rates for peers
 
 	// Request tracking during syncing phase
 	statelessPeers map[string]struct{} // Peers that failed to deliver state data
@@ -452,6 +456,7 @@ func NewSyncer(db ethdb.KeyValueStore) *Syncer {
 		peers:    make(map[string]SyncPeer),
 		peerJoin: new(event.Feed),
 		peerDrop: new(event.Feed),
+		rates:    msgrate.NewTrackers(log.New("proto", "snap")),
 		update:   make(chan struct{}, 1),
 
 		accountIdlers:  make(map[string]struct{}),
@@ -484,6 +489,7 @@ func (s *Syncer) Register(peer SyncPeer) error {
 		return errors.New("already registered")
 	}
 	s.peers[id] = peer
+	s.rates.Track(id, msgrate.NewTracker(s.rates.MeanCapacities(), s.rates.MedianRoundTrip()))
 
 	// Mark the peer as idle, even if no sync is running
 	s.accountIdlers[id] = struct{}{}
@@ -509,6 +515,7 @@ func (s *Syncer) Unregister(id string) error {
 		return errors.New("not registered")
 	}
 	delete(s.peers, id)
+	s.rates.Untrack(id)
 
 	// Remove status markers, even if no sync is running
 	delete(s.statelessPeers, id)
@@ -851,10 +858,24 @@ func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *ac
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// If there are no idle peers, short circuit assignment
+	// Sort the peers by download capacity to use faster ones if many available
-	if len(s.accountIdlers) == 0 {
+	idlers := &capacitySort{
+		ids:  make([]string, 0, len(s.accountIdlers)),
+		caps: make([]float64, 0, len(s.accountIdlers)),
+	}
+	targetTTL := s.rates.TargetTimeout()
+	for id := range s.accountIdlers {
+		if _, ok := s.statelessPeers[id]; ok {
+			continue
+		}
+		idlers.ids = append(idlers.ids, id)
+		idlers.caps = append(idlers.caps, s.rates.Capacity(id, AccountRangeMsg, targetTTL))
+	}
+	if len(idlers.ids) == 0 {
 		return
 	}
+	sort.Sort(sort.Reverse(idlers))
+
 	// Iterate over all the tasks and try to find a pending one
 	for _, task := range s.tasks {
 		// Skip any tasks already filling
@@ -864,20 +885,15 @@ func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *ac
 		// Task pending retrieval, try to find an idle peer. If no such peer
 		// exists, we probably assigned tasks for all (or they are stateless).
 		// Abort the entire assignment mechanism.
-		var idle string
+		if len(idlers.ids) == 0 {
-		for id := range s.accountIdlers {
-			// If the peer rejected a query in this sync cycle, don't bother asking
-			// again for anything, it's either out of sync or already pruned
-			if _, ok := s.statelessPeers[id]; ok {
-				continue
-			}
-			idle = id
-			break
-		}
-		if idle == "" {
 			return
 		}
-		peer := s.peers[idle]
+		var (
+			idle = idlers.ids[0]
+			peer = s.peers[idle]
+			cap  = idlers.caps[0]
+		)
+		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]
 
 		// Matched a pending task to an idle peer, allocate a unique request id
 		var reqid uint64
@@ -895,6 +911,7 @@ func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *ac
 		req := &accountRequest{
 			peer:    idle,
 			id:      reqid,
+			time:    time.Now(),
 			deliver: success,
 			revert:  fail,
 			cancel:  cancel,
@@ -903,8 +920,9 @@ func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *ac
 			limit:   task.Last,
 			task:    task,
 		}
-		req.timeout = time.AfterFunc(requestTimeout, func() {
+		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
 			peer.Log().Debug("Account range request timed out", "reqid", reqid)
+			s.rates.Update(idle, AccountRangeMsg, 0, 0)
 			s.scheduleRevertAccountRequest(req)
 		})
 		s.accountReqs[reqid] = req
@@ -915,7 +933,13 @@ func (s *Syncer) assignAccountTasks(success chan *accountResponse, fail chan *ac
 			defer s.pend.Done()
 
 			// Attempt to send the remote request and revert if it fails
-			if err := peer.RequestAccountRange(reqid, root, req.origin, req.limit, maxRequestSize); err != nil {
+			if cap > maxRequestSize {
+				cap = maxRequestSize
+			}
+			if cap < minRequestSize { // Don't bother with peers below a bare minimum performance
+				cap = minRequestSize
+			}
+			if err := peer.RequestAccountRange(reqid, root, req.origin, req.limit, uint64(cap)); err != nil {
 				peer.Log().Debug("Failed to request account range", "err", err)
 				s.scheduleRevertAccountRequest(req)
 			}
@@ -931,10 +955,24 @@ func (s *Syncer) assignBytecodeTasks(success chan *bytecodeResponse, fail chan *
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// If there are no idle peers, short circuit assignment
+	// Sort the peers by download capacity to use faster ones if many available
-	if len(s.bytecodeIdlers) == 0 {
+	idlers := &capacitySort{
+		ids:  make([]string, 0, len(s.bytecodeIdlers)),
+		caps: make([]float64, 0, len(s.bytecodeIdlers)),
+	}
+	targetTTL := s.rates.TargetTimeout()
+	for id := range s.bytecodeIdlers {
+		if _, ok := s.statelessPeers[id]; ok {
+			continue
+		}
+		idlers.ids = append(idlers.ids, id)
+		idlers.caps = append(idlers.caps, s.rates.Capacity(id, ByteCodesMsg, targetTTL))
+	}
+	if len(idlers.ids) == 0 {
 		return
 	}
+	sort.Sort(sort.Reverse(idlers))
+
 	// Iterate over all the tasks and try to find a pending one
 	for _, task := range s.tasks {
 		// Skip any tasks not in the bytecode retrieval phase
@@ -948,20 +986,15 @@ func (s *Syncer) assignBytecodeTasks(success chan *bytecodeResponse, fail chan *
 		// Task pending retrieval, try to find an idle peer. If no such peer
 		// exists, we probably assigned tasks for all (or they are stateless).
 		// Abort the entire assignment mechanism.
-		var idle string
+		if len(idlers.ids) == 0 {
-		for id := range s.bytecodeIdlers {
-			// If the peer rejected a query in this sync cycle, don't bother asking
-			// again for anything, it's either out of sync or already pruned
-			if _, ok := s.statelessPeers[id]; ok {
-				continue
-			}
-			idle = id
-			break
-		}
-		if idle == "" {
 			return
 		}
-		peer := s.peers[idle]
+		var (
+			idle = idlers.ids[0]
+			peer = s.peers[idle]
+			cap  = idlers.caps[0]
+		)
+		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]
 
 		// Matched a pending task to an idle peer, allocate a unique request id
 		var reqid uint64
@@ -976,17 +1009,21 @@ func (s *Syncer) assignBytecodeTasks(success chan *bytecodeResponse, fail chan *
 			break
 		}
 		// Generate the network query and send it to the peer
-		hashes := make([]common.Hash, 0, maxCodeRequestCount)
+		if cap > maxCodeRequestCount {
+			cap = maxCodeRequestCount
+		}
+		hashes := make([]common.Hash, 0, int(cap))
 		for hash := range task.codeTasks {
 			delete(task.codeTasks, hash)
 			hashes = append(hashes, hash)
-			if len(hashes) >= maxCodeRequestCount {
+			if len(hashes) >= int(cap) {
 				break
 			}
 		}
 		req := &bytecodeRequest{
 			peer:    idle,
 			id:      reqid,
+			time:    time.Now(),
 			deliver: success,
 			revert:  fail,
 			cancel:  cancel,
@@ -994,8 +1031,9 @@ func (s *Syncer) assignBytecodeTasks(success chan *bytecodeResponse, fail chan *
 			hashes:  hashes,
 			task:    task,
 		}
-		req.timeout = time.AfterFunc(requestTimeout, func() {
+		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
 			peer.Log().Debug("Bytecode request timed out", "reqid", reqid)
+			s.rates.Update(idle, ByteCodesMsg, 0, 0)
 			s.scheduleRevertBytecodeRequest(req)
 		})
 		s.bytecodeReqs[reqid] = req
@@ -1020,10 +1058,24 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// If there are no idle peers, short circuit assignment
+	// Sort the peers by download capacity to use faster ones if many available
-	if len(s.storageIdlers) == 0 {
+	idlers := &capacitySort{
+		ids:  make([]string, 0, len(s.storageIdlers)),
+		caps: make([]float64, 0, len(s.storageIdlers)),
+	}
+	targetTTL := s.rates.TargetTimeout()
+	for id := range s.storageIdlers {
+		if _, ok := s.statelessPeers[id]; ok {
+			continue
+		}
+		idlers.ids = append(idlers.ids, id)
+		idlers.caps = append(idlers.caps, s.rates.Capacity(id, StorageRangesMsg, targetTTL))
+	}
+	if len(idlers.ids) == 0 {
 		return
 	}
+	sort.Sort(sort.Reverse(idlers))
+
 	// Iterate over all the tasks and try to find a pending one
 	for _, task := range s.tasks {
 		// Skip any tasks not in the storage retrieval phase
@@ -1037,20 +1089,15 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 		// Task pending retrieval, try to find an idle peer. If no such peer
 		// exists, we probably assigned tasks for all (or they are stateless).
 		// Abort the entire assignment mechanism.
-		var idle string
+		if len(idlers.ids) == 0 {
-		for id := range s.storageIdlers {
-			// If the peer rejected a query in this sync cycle, don't bother asking
-			// again for anything, it's either out of sync or already pruned
-			if _, ok := s.statelessPeers[id]; ok {
-				continue
-			}
-			idle = id
-			break
-		}
-		if idle == "" {
 			return
 		}
-		peer := s.peers[idle]
+		var (
+			idle = idlers.ids[0]
+			peer = s.peers[idle]
+			cap  = idlers.caps[0]
+		)
+		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]
 
 		// Matched a pending task to an idle peer, allocate a unique request id
 		var reqid uint64
@@ -1067,9 +1114,17 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 		// Generate the network query and send it to the peer. If there are
 		// large contract tasks pending, complete those before diving into
 		// even more new contracts.
+		if cap > maxRequestSize {
+			cap = maxRequestSize
+		}
+		if cap < minRequestSize { // Don't bother with peers below a bare minimum performance
+			cap = minRequestSize
+		}
+		storageSets := int(cap / 1024)
+
 		var (
-			accounts = make([]common.Hash, 0, maxStorageSetRequestCount)
+			accounts = make([]common.Hash, 0, storageSets)
-			roots    = make([]common.Hash, 0, maxStorageSetRequestCount)
+			roots    = make([]common.Hash, 0, storageSets)
 			subtask  *storageTask
 		)
 		for account, subtasks := range task.SubTasks {
@@ -1096,7 +1151,7 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 				accounts = append(accounts, acccount)
 				roots = append(roots, root)
 
-				if len(accounts) >= maxStorageSetRequestCount {
+				if len(accounts) >= storageSets {
 					break
 				}
 			}
@@ -1109,6 +1164,7 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 		req := &storageRequest{
 			peer:     idle,
 			id:       reqid,
+			time:     time.Now(),
 			deliver:  success,
 			revert:   fail,
 			cancel:   cancel,
@@ -1122,8 +1178,9 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 			req.origin = subtask.Next
 			req.limit = subtask.Last
 		}
-		req.timeout = time.AfterFunc(requestTimeout, func() {
+		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
 			peer.Log().Debug("Storage request timed out", "reqid", reqid)
+			s.rates.Update(idle, StorageRangesMsg, 0, 0)
 			s.scheduleRevertStorageRequest(req)
 		})
 		s.storageReqs[reqid] = req
@@ -1138,7 +1195,7 @@ func (s *Syncer) assignStorageTasks(success chan *storageResponse, fail chan *st
 			if subtask != nil {
 				origin, limit = req.origin[:], req.limit[:]
 			}
-			if err := peer.RequestStorageRanges(reqid, root, accounts, origin, limit, maxRequestSize); err != nil {
+			if err := peer.RequestStorageRanges(reqid, root, accounts, origin, limit, uint64(cap)); err != nil {
 				log.Debug("Failed to request storage", "err", err)
 				s.scheduleRevertStorageRequest(req)
 			}
@@ -1157,10 +1214,24 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// If there are no idle peers, short circuit assignment
+	// Sort the peers by download capacity to use faster ones if many available
-	if len(s.trienodeHealIdlers) == 0 {
+	idlers := &capacitySort{
+		ids:  make([]string, 0, len(s.trienodeHealIdlers)),
+		caps: make([]float64, 0, len(s.trienodeHealIdlers)),
+	}
+	targetTTL := s.rates.TargetTimeout()
+	for id := range s.trienodeHealIdlers {
+		if _, ok := s.statelessPeers[id]; ok {
+			continue
+		}
+		idlers.ids = append(idlers.ids, id)
+		idlers.caps = append(idlers.caps, s.rates.Capacity(id, TrieNodesMsg, targetTTL))
+	}
+	if len(idlers.ids) == 0 {
 		return
 	}
+	sort.Sort(sort.Reverse(idlers))
+
 	// Iterate over pending tasks and try to find a peer to retrieve with
 	for len(s.healer.trieTasks) > 0 || s.healer.scheduler.Pending() > 0 {
 		// If there are not enough trie tasks queued to fully assign, fill the
@@ -1186,20 +1257,15 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 		// Task pending retrieval, try to find an idle peer. If no such peer
 		// exists, we probably assigned tasks for all (or they are stateless).
 		// Abort the entire assignment mechanism.
-		var idle string
+		if len(idlers.ids) == 0 {
-		for id := range s.trienodeHealIdlers {
-			// If the peer rejected a query in this sync cycle, don't bother asking
-			// again for anything, it's either out of sync or already pruned
-			if _, ok := s.statelessPeers[id]; ok {
-				continue
-			}
-			idle = id
-			break
-		}
-		if idle == "" {
 			return
 		}
-		peer := s.peers[idle]
+		var (
+			idle = idlers.ids[0]
+			peer = s.peers[idle]
+			cap  = idlers.caps[0]
+		)
+		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]
 
 		// Matched a pending task to an idle peer, allocate a unique request id
 		var reqid uint64
@@ -1214,10 +1280,13 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 			break
 		}
 		// Generate the network query and send it to the peer
+		if cap > maxTrieRequestCount {
+			cap = maxTrieRequestCount
+		}
 		var (
-			hashes   = make([]common.Hash, 0, maxTrieRequestCount)
+			hashes   = make([]common.Hash, 0, int(cap))
-			paths    = make([]trie.SyncPath, 0, maxTrieRequestCount)
+			paths    = make([]trie.SyncPath, 0, int(cap))
-			pathsets = make([]TrieNodePathSet, 0, maxTrieRequestCount)
+			pathsets = make([]TrieNodePathSet, 0, int(cap))
 		)
 		for hash, pathset := range s.healer.trieTasks {
 			delete(s.healer.trieTasks, hash)
@@ -1226,13 +1295,14 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 			paths = append(paths, pathset)
 			pathsets = append(pathsets, [][]byte(pathset)) // TODO(karalabe): group requests by account hash
 
-			if len(hashes) >= maxTrieRequestCount {
+			if len(hashes) >= int(cap) {
 				break
 			}
 		}
 		req := &trienodeHealRequest{
 			peer:    idle,
 			id:      reqid,
+			time:    time.Now(),
 			deliver: success,
 			revert:  fail,
 			cancel:  cancel,
@@ -1241,8 +1311,9 @@ func (s *Syncer) assignTrienodeHealTasks(success chan *trienodeHealResponse, fai
 			paths:   paths,
 			task:    s.healer,
 		}
-		req.timeout = time.AfterFunc(requestTimeout, func() {
+		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
 			peer.Log().Debug("Trienode heal request timed out", "reqid", reqid)
+			s.rates.Update(idle, TrieNodesMsg, 0, 0)
 			s.scheduleRevertTrienodeHealRequest(req)
 		})
 		s.trienodeHealReqs[reqid] = req
@@ -1267,10 +1338,24 @@ func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fai
 	s.lock.Lock()
 	defer s.lock.Unlock()
 
-	// If there are no idle peers, short circuit assignment
+	// Sort the peers by download capacity to use faster ones if many available
-	if len(s.bytecodeHealIdlers) == 0 {
+	idlers := &capacitySort{
+		ids:  make([]string, 0, len(s.bytecodeHealIdlers)),
+		caps: make([]float64, 0, len(s.bytecodeHealIdlers)),
+	}
+	targetTTL := s.rates.TargetTimeout()
+	for id := range s.bytecodeHealIdlers {
+		if _, ok := s.statelessPeers[id]; ok {
+			continue
+		}
+		idlers.ids = append(idlers.ids, id)
+		idlers.caps = append(idlers.caps, s.rates.Capacity(id, ByteCodesMsg, targetTTL))
+	}
+	if len(idlers.ids) == 0 {
 		return
 	}
+	sort.Sort(sort.Reverse(idlers))
+
 	// Iterate over pending tasks and try to find a peer to retrieve with
 	for len(s.healer.codeTasks) > 0 || s.healer.scheduler.Pending() > 0 {
 		// If there are not enough trie tasks queued to fully assign, fill the
@@ -1296,20 +1381,15 @@ func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fai
 		// Task pending retrieval, try to find an idle peer. If no such peer
 		// exists, we probably assigned tasks for all (or they are stateless).
 		// Abort the entire assignment mechanism.
-		var idle string
+		if len(idlers.ids) == 0 {
-		for id := range s.bytecodeHealIdlers {
-			// If the peer rejected a query in this sync cycle, don't bother asking
-			// again for anything, it's either out of sync or already pruned
-			if _, ok := s.statelessPeers[id]; ok {
-				continue
-			}
-			idle = id
-			break
-		}
-		if idle == "" {
 			return
 		}
-		peer := s.peers[idle]
+		var (
+			idle = idlers.ids[0]
+			peer = s.peers[idle]
+			cap  = idlers.caps[0]
+		)
+		idlers.ids, idlers.caps = idlers.ids[1:], idlers.caps[1:]
 
 		// Matched a pending task to an idle peer, allocate a unique request id
 		var reqid uint64
@@ -1324,18 +1404,22 @@ func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fai
 			break
 		}
 		// Generate the network query and send it to the peer
-		hashes := make([]common.Hash, 0, maxCodeRequestCount)
+		if cap > maxCodeRequestCount {
+			cap = maxCodeRequestCount
+		}
+		hashes := make([]common.Hash, 0, int(cap))
 		for hash := range s.healer.codeTasks {
 			delete(s.healer.codeTasks, hash)
 
 			hashes = append(hashes, hash)
-			if len(hashes) >= maxCodeRequestCount {
+			if len(hashes) >= int(cap) {
 				break
 			}
 		}
 		req := &bytecodeHealRequest{
 			peer:    idle,
 			id:      reqid,
+			time:    time.Now(),
 			deliver: success,
 			revert:  fail,
 			cancel:  cancel,
@@ -1343,8 +1427,9 @@ func (s *Syncer) assignBytecodeHealTasks(success chan *bytecodeHealResponse, fai
 			hashes:  hashes,
 			task:    s.healer,
 		}
-		req.timeout = time.AfterFunc(requestTimeout, func() {
+		req.timeout = time.AfterFunc(s.rates.TargetTimeout(), func() {
 			peer.Log().Debug("Bytecode heal request timed out", "reqid", reqid)
+			s.rates.Update(idle, ByteCodesMsg, 0, 0)
 			s.scheduleRevertBytecodeHealRequest(req)
 		})
 		s.bytecodeHealReqs[reqid] = req
@@ -2142,6 +2227,7 @@ func (s *Syncer) OnAccounts(peer SyncPeer, id uint64, hashes []common.Hash, acco
 		return nil
 	}
 	delete(s.accountReqs, id)
+	s.rates.Update(peer.ID(), AccountRangeMsg, time.Since(req.time), int(size))
 
 	// Clean up the request timeout timer, we'll see how to proceed further based
 	// on the actual delivered content
@@ -2253,6 +2339,7 @@ func (s *Syncer) onByteCodes(peer SyncPeer, id uint64, bytecodes [][]byte) error
 		return nil
 	}
 	delete(s.bytecodeReqs, id)
+	s.rates.Update(peer.ID(), ByteCodesMsg, time.Since(req.time), len(bytecodes))
 
 	// Clean up the request timeout timer, we'll see how to proceed further based
 	// on the actual delivered content
@@ -2361,6 +2448,7 @@ func (s *Syncer) OnStorage(peer SyncPeer, id uint64, hashes [][]common.Hash, slo
 		return nil
 	}
 	delete(s.storageReqs, id)
+	s.rates.Update(peer.ID(), StorageRangesMsg, time.Since(req.time), int(size))
 
 	// Clean up the request timeout timer, we'll see how to proceed further based
 	// on the actual delivered content
@@ -2487,6 +2575,7 @@ func (s *Syncer) OnTrieNodes(peer SyncPeer, id uint64, trienodes [][]byte) error
 		return nil
 	}
 	delete(s.trienodeHealReqs, id)
+	s.rates.Update(peer.ID(), TrieNodesMsg, time.Since(req.time), len(trienodes))
 
 	// Clean up the request timeout timer, we'll see how to proceed further based
 	// on the actual delivered content
@@ -2581,6 +2670,7 @@ func (s *Syncer) onHealByteCodes(peer SyncPeer, id uint64, bytecodes [][]byte) e
 		return nil
 	}
 	delete(s.bytecodeHealReqs, id)
+	s.rates.Update(peer.ID(), ByteCodesMsg, time.Since(req.time), len(bytecodes))
 
 	// Clean up the request timeout timer, we'll see how to proceed further based
 	// on the actual delivered content
@@ -2756,3 +2846,24 @@ func estimateRemainingSlots(hashes int, last common.Hash) (uint64, error) {
 	}
 	return space.Uint64() - uint64(hashes), nil
 }
+
+// capacitySort implements the Sort interface, allowing sorting by peer message
+// throughput. Note, callers should use sort.Reverse to get the desired effect
+// of highest capacity being at the front.
+type capacitySort struct {
+	ids  []string
+	caps []float64
+}
+
+func (s *capacitySort) Len() int {
+	return len(s.ids)
+}
+
+func (s *capacitySort) Less(i, j int) bool {
+	return s.caps[i] < s.caps[j]
+}
+
+func (s *capacitySort) Swap(i, j int) {
+	s.ids[i], s.ids[j] = s.ids[j], s.ids[i]
+	s.caps[i], s.caps[j] = s.caps[j], s.caps[i]
+}

eth/protocols/snap/sync_test.go

@@ -796,12 +796,6 @@ func TestMultiSyncManyUseless(t *testing.T) {
 
 // TestMultiSyncManyUseless contains one good peer, and many which doesn't return anything valuable at all
 func TestMultiSyncManyUselessWithLowTimeout(t *testing.T) {
-	// We're setting the timeout to very low, to increase the chance of the timeout
-	// being triggered. This was previously a cause of panic, when a response
-	// arrived simultaneously as a timeout was triggered.
-	defer func(old time.Duration) { requestTimeout = old }(requestTimeout)
-	requestTimeout = time.Millisecond
-
 	var (
 		once   sync.Once
 		cancel = make(chan struct{})
@@ -838,6 +832,11 @@ func TestMultiSyncManyUselessWithLowTimeout(t *testing.T) {
 		mkSource("noStorage", true, false, true),
 		mkSource("noTrie", true, true, false),
 	)
+	// We're setting the timeout to very low, to increase the chance of the timeout
+	// being triggered. This was previously a cause of panic, when a response
+	// arrived simultaneously as a timeout was triggered.
+	syncer.rates.OverrideTTLLimit = time.Millisecond
+
 	done := checkStall(t, term)
 	if err := syncer.Sync(sourceAccountTrie.Hash(), cancel); err != nil {
 		t.Fatalf("sync failed: %v", err)
@@ -848,10 +847,6 @@ func TestMultiSyncManyUselessWithLowTimeout(t *testing.T) {
 
 // TestMultiSyncManyUnresponsive contains one good peer, and many which doesn't respond at all
 func TestMultiSyncManyUnresponsive(t *testing.T) {
-	// We're setting the timeout to very low, to make the test run a bit faster
-	defer func(old time.Duration) { requestTimeout = old }(requestTimeout)
-	requestTimeout = time.Millisecond
-
 	var (
 		once   sync.Once
 		cancel = make(chan struct{})
@@ -888,6 +883,9 @@ func TestMultiSyncManyUnresponsive(t *testing.T) {
 		mkSource("noStorage", true, false, true),
 		mkSource("noTrie", true, true, false),
 	)
+	// We're setting the timeout to very low, to make the test run a bit faster
+	syncer.rates.OverrideTTLLimit = time.Millisecond
+
 	done := checkStall(t, term)
 	if err := syncer.Sync(sourceAccountTrie.Hash(), cancel); err != nil {
 		t.Fatalf("sync failed: %v", err)

p2p/msgrate/msgrate.go

@@ -0,0 +1,458 @@
+// Copyright 2021 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 msgrate allows estimating the throughput of peers for more balanced syncs.
+package msgrate
+
+import (
+	"errors"
+	"sort"
+	"sync"
+	"time"
+
+	"github.com/ethereum/go-ethereum/log"
+)
+
+// measurementImpact is the impact a single measurement has on a peer's final
+// capacity value. A value closer to 0 reacts slower to sudden network changes,
+// but it is also more stable against temporary hiccups. 0.1 worked well for
+// most of Ethereum's existence, so might as well go with it.
+const measurementImpact = 0.1
+
+// capacityOverestimation is the ratio of items to over-estimate when retrieving
+// a peer's capacity to avoid locking into a lower value due to never attempting
+// to fetch more than some local stable value.
+const capacityOverestimation = 1.01
+
+// qosTuningPeers is the number of best peers to tune round trip times based on.
+// An Ethereum node doesn't need hundreds of connections to operate correctly,
+// so instead of lowering our download speed to the median of potentially many
+// bad nodes, we can target a smaller set of vey good nodes. At worse this will
+// result in less nodes to sync from, but that's still better than some hogging
+// the pipeline.
+const qosTuningPeers = 5
+
+// rttMinEstimate is the minimal round trip time to target requests for. Since
+// every request entails a 2 way latency + bandwidth + serving database lookups,
+// it should be generous enough to permit meaningful work to be done on top of
+// the transmission costs.
+const rttMinEstimate = 2 * time.Second
+
+// rttMaxEstimate is the maximal round trip time to target requests for. Although
+// the expectation is that a well connected node will never reach this, certain
+// special connectivity ones might experience significant delays (e.g. satellite
+// uplink with 3s RTT). This value should be low enough to forbid stalling the
+// pipeline too long, but large enough to cover the worst of the worst links.
+const rttMaxEstimate = 20 * time.Second
+
+// rttPushdownFactor is a multiplier to attempt forcing quicker requests than
+// what the message rate tracker estimates. The reason is that message rate
+// tracking adapts queries to the RTT, but multiple RTT values can be perfectly
+// valid, they just result in higher packet sizes. Since smaller packets almost
+// always result in stabler download streams, this factor hones in on the lowest
+// RTT from all the functional ones.
+const rttPushdownFactor = 0.9
+
+// rttMinConfidence is the minimum value the roundtrip confidence factor may drop
+// to. Since the target timeouts are based on how confident the tracker is in the
+// true roundtrip, it's important to not allow too huge fluctuations.
+const rttMinConfidence = 0.1
+
+// ttlScaling is the multiplier that converts the estimated roundtrip time to a
+// timeout cap for network requests. The expectation is that peers' response time
+// will fluctuate around the estimated roundtrip, but depending in their load at
+// request time, it might be higher than anticipated. This scaling factor ensures
+// that we allow remote connections some slack but at the same time do enforce a
+// behavior similar to our median peers.
+const ttlScaling = 3
+
+// ttlLimit is the maximum timeout allowance to prevent reaching crazy numbers
+// if some unforeseen network events shappen. As much as we try to hone in on
+// the most optimal values, it doesn't make any sense to go above a threshold,
+// even if everything is slow and screwy.
+const ttlLimit = time.Minute
+
+// tuningConfidenceCap is the number of active peers above which to stop detuning
+// the confidence number. The idea here is that once we hone in on the capacity
+// of a meaningful number of peers, adding one more should ot have a significant
+// impact on things, so just ron with the originals.
+const tuningConfidenceCap = 10
+
+// tuningImpact is the influence that a new tuning target has on the previously
+// cached value. This number is mostly just an out-of-the-blue heuristic that
+// prevents the estimates from jumping around. There's no particular reason for
+// the current value.
+const tuningImpact = 0.25
+
+// Tracker estimates the throughput capacity of a peer with regard to each data
+// type it can deliver. The goal is to dynamically adjust request sizes to max
+// out network throughput without overloading either the peer or th elocal node.
+//
+// By tracking in real time the latencies and bandiwdths peers exhibit for each
+// packet type, it's possible to prevent overloading by detecting a slowdown on
+// one type when another type is pushed too hard.
+//
+// Similarly, real time measurements also help avoid overloading the local net
+// connection if our peers would otherwise be capable to deliver more, but the
+// local link is saturated. In that case, the live measurements will force us
+// to reduce request sizes until the throughput gets stable.
+//
+// Lastly, message rate measurements allows us to detect if a peer is unsuaully
+// slow compared to other peers, in which case we can decide to keep it around
+// or free up the slot so someone closer.
+//
+// Since throughput tracking and estimation adapts dynamically to live network
+// conditions, it's fine to have multiple trackers locally track the same peer
+// in different subsystem. The throughput will simply be distributed across the
+// two trackers if both are highly active.
+type Tracker struct {
+	// capacity is the number of items retrievable per second of a given type.
+	// It is analogous to bandwidth, but we deliberately avoided using bytes
+	// as the unit, since serving nodes also spend a lot of time loading data
+	// from disk, which is linear in the number of items, but mostly constant
+	// in their sizes.
+	//
+	// Callers of course are free to use the item counter as a byte counter if
+	// or when their protocol of choise if capped by bytes instead of items.
+	// (eg. eth.getHeaders vs snap.getAccountRange).
+	capacity map[uint64]float64
+
+	// roundtrip is the latency a peer in general responds to data requests.
+	// This number is not used inside the tracker, but is exposed to compare
+	// peers to each other and filter out slow ones. Note however, it only
+	// makes sense to compare RTTs if the caller caters request sizes for
+	// each peer to target the same RTT. There's no need to make this number
+	// the real networking RTT, we just need a number to compare peers with.
+	roundtrip time.Duration
+
+	lock sync.RWMutex
+}
+
+// NewTracker creates a new message rate tracker for a specific peer. An initial
+// RTT is needed to avoid a peer getting marked as an outlier compared to others
+// right after joining. It's suggested to use the median rtt across all peers to
+// init a new peer tracker.
+func NewTracker(caps map[uint64]float64, rtt time.Duration) *Tracker {
+	if caps == nil {
+		caps = make(map[uint64]float64)
+	}
+	return &Tracker{
+		capacity:  caps,
+		roundtrip: rtt,
+	}
+}
+
+// Capacity calculates the number of items the peer is estimated to be able to
+// retrieve within the alloted time slot. The method will round up any division
+// errors and will add an additional overestimation ratio on top. The reason for
+// overshooting the capacity is because certain message types might not increase
+// the load proportionally to the requested items, so fetching a bit more might
+// still take the same RTT. By forcefully overshooting by a small amount, we can
+// avoid locking into a lower-that-real capacity.
+func (t *Tracker) Capacity(kind uint64, targetRTT time.Duration) float64 {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	// Calculate the actual measured throughput
+	throughput := t.capacity[kind] * float64(targetRTT) / float64(time.Second)
+
+	// Return an overestimation to force the peer out of a stuck minima, adding
+	// +1 in case the item count is too low for the overestimator to dent
+	return 1 + capacityOverestimation*throughput
+}
+
+// Update modifies the peer's capacity values for a specific data type with a new
+// measurement. If the delivery is zero, the peer is assumed to have either timed
+// out or to not have the requested data, resulting in a slash to 0 capacity. This
+// avoids assigning the peer retrievals that it won't be able to honour.
+func (t *Tracker) Update(kind uint64, elapsed time.Duration, items int) {
+	t.lock.Lock()
+	defer t.lock.Unlock()
+
+	// If nothing was delivered (timeout / unavailable data), reduce throughput
+	// to minimum
+	if items == 0 {
+		t.capacity[kind] = 0
+		return
+	}
+	// Otherwise update the throughput with a new measurement
+	if elapsed <= 0 {
+		elapsed = 1 // +1 (ns) to ensure non-zero divisor
+	}
+	measured := float64(items) / (float64(elapsed) / float64(time.Second))
+
+	t.capacity[kind] = (1-measurementImpact)*(t.capacity[kind]) + measurementImpact*measured
+	t.roundtrip = time.Duration((1-measurementImpact)*float64(t.roundtrip) + measurementImpact*float64(elapsed))
+}
+
+// Trackers is a set of message rate trackers across a number of peers with the
+// goal of aggregating certain measurements across the entire set for outlier
+// filtering and newly joining initialization.
+type Trackers struct {
+	trackers map[string]*Tracker
+
+	// roundtrip is the current best guess as to what is a stable round trip time
+	// across the entire collection of connected peers. This is derived from the
+	// various trackers added, but is used as a cache to avoid recomputing on each
+	// network request. The value is updated once every RTT to avoid fluctuations
+	// caused by hiccups or peer events.
+	roundtrip time.Duration
+
+	// confidence represents the probability that the estimated roundtrip value
+	// is the real one across all our peers. The confidence value is used as an
+	// impact factor of new measurements on old estimates. As our connectivity
+	// stabilizes, this value gravitates towards 1, new measurements havinng
+	// almost no impact. If there's a large peer churn and few peers, then new
+	// measurements will impact it more. The confidence is increased with every
+	// packet and dropped with every new connection.
+	confidence float64
+
+	// tuned is the time instance the tracker recalculated its cached roundtrip
+	// value and confidence values. A cleaner way would be to have a heartbeat
+	// goroutine do it regularly, but that requires a lot of maintenance to just
+	// run every now and again.
+	tuned time.Time
+
+	// The fields below can be used to override certain default values. Their
+	// purpose is to allow quicker tests. Don't use them in production.
+	OverrideTTLLimit time.Duration
+
+	log  log.Logger
+	lock sync.RWMutex
+}
+
+// NewTrackers creates an empty set of trackers to be filled with peers.
+func NewTrackers(log log.Logger) *Trackers {
+	return &Trackers{
+		trackers:         make(map[string]*Tracker),
+		roundtrip:        rttMaxEstimate,
+		confidence:       1,
+		tuned:            time.Now(),
+		OverrideTTLLimit: ttlLimit,
+		log:              log,
+	}
+}
+
+// Track inserts a new tracker into the set.
+func (t *Trackers) Track(id string, tracker *Tracker) error {
+	t.lock.Lock()
+	defer t.lock.Unlock()
+
+	if _, ok := t.trackers[id]; ok {
+		return errors.New("already tracking")
+	}
+	t.trackers[id] = tracker
+	t.detune()
+
+	return nil
+}
+
+// Untrack stops tracking a previously added peer.
+func (t *Trackers) Untrack(id string) error {
+	t.lock.Lock()
+	defer t.lock.Unlock()
+
+	if _, ok := t.trackers[id]; !ok {
+		return errors.New("not tracking")
+	}
+	delete(t.trackers, id)
+	return nil
+}
+
+// MedianRoundTrip returns the median RTT across all known trackers. The purpose
+// of the median RTT is to initialize a new peer with sane statistics that it will
+// hopefully outperform. If it seriously underperforms, there's a risk of dropping
+// the peer, but that is ok as we're aiming for a strong median.
+func (t *Trackers) MedianRoundTrip() time.Duration {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	return t.medianRoundTrip()
+}
+
+// medianRoundTrip is the internal lockless version of MedianRoundTrip to be used
+// by the QoS tuner.
+func (t *Trackers) medianRoundTrip() time.Duration {
+	// Gather all the currently measured round trip times
+	rtts := make([]float64, 0, len(t.trackers))
+	for _, tt := range t.trackers {
+		tt.lock.RLock()
+		rtts = append(rtts, float64(tt.roundtrip))
+		tt.lock.RUnlock()
+	}
+	sort.Float64s(rtts)
+
+	median := rttMaxEstimate
+	if qosTuningPeers <= len(rtts) {
+		median = time.Duration(rtts[qosTuningPeers/2]) // Median of our best few peers
+	} else if len(rtts) > 0 {
+		median = time.Duration(rtts[len(rtts)/2]) // Median of all out connected peers
+	}
+	// Restrict the RTT into some QoS defaults, irrelevant of true RTT
+	if median < rttMinEstimate {
+		median = rttMinEstimate
+	}
+	if median > rttMaxEstimate {
+		median = rttMaxEstimate
+	}
+	return median
+}
+
+// MeanCapacities returns the capacities averaged across all the added trackers.
+// The purpos of the mean capacities are to initialize a new peer with some sane
+// starting values that it will hopefully outperform. If the mean overshoots, the
+// peer will be cut back to minimal capacity and given another chance.
+func (t *Trackers) MeanCapacities() map[uint64]float64 {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	return t.meanCapacities()
+}
+
+// meanCapacities is the internal lockless version of MeanCapacities used for
+// debug logging.
+func (t *Trackers) meanCapacities() map[uint64]float64 {
+	capacities := make(map[uint64]float64)
+	for _, tt := range t.trackers {
+		tt.lock.RLock()
+		for key, val := range tt.capacity {
+			capacities[key] += val
+		}
+		tt.lock.RUnlock()
+	}
+	for key, val := range capacities {
+		capacities[key] = val / float64(len(t.trackers))
+	}
+	return capacities
+}
+
+// TargetRoundTrip returns the current target round trip time for a request to
+// complete in.The returned RTT is slightly under the estimated RTT. The reason
+// is that message rate estimation is a 2 dimensional problem which is solvable
+// for any RTT. The goal is to gravitate towards smaller RTTs instead of large
+// messages, to result in a stabler download stream.
+func (t *Trackers) TargetRoundTrip() time.Duration {
+	// Recalculate the internal caches if it's been a while
+	t.tune()
+
+	// Caches surely recent, return target roundtrip
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	return time.Duration(float64(t.roundtrip) * rttPushdownFactor)
+}
+
+// TargetTimeout returns the timeout allowance for a single request to finish
+// under. The timeout is proportional to the roundtrip, but also takes into
+// consideration the tracker's confidence in said roundtrip and scales it
+// accordingly. The final value is capped to avoid runaway requests.
+func (t *Trackers) TargetTimeout() time.Duration {
+	// Recalculate the internal caches if it's been a while
+	t.tune()
+
+	// Caches surely recent, return target timeout
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	return t.targetTimeout()
+}
+
+// targetTimeout is the internal lockless version of TargetTimeout to be used
+// during QoS tuning.
+func (t *Trackers) targetTimeout() time.Duration {
+	timeout := time.Duration(ttlScaling * float64(t.roundtrip) / t.confidence)
+	if timeout > t.OverrideTTLLimit {
+		timeout = t.OverrideTTLLimit
+	}
+	return timeout
+}
+
+// tune gathers the individual tracker statistics and updates the estimated
+// request round trip time.
+func (t *Trackers) tune() {
+	// Tune may be called concurrently all over the place, but we only want to
+	// periodically update and even then only once. First check if it was updated
+	// recently and abort if so.
+	t.lock.RLock()
+	dirty := time.Since(t.tuned) > t.roundtrip
+	t.lock.RUnlock()
+	if !dirty {
+		return
+	}
+	// If an update is needed, obtain a write lock but make sure we don't update
+	// it on all concurrent threads one by one.
+	t.lock.Lock()
+	defer t.lock.Unlock()
+
+	if dirty := time.Since(t.tuned) > t.roundtrip; !dirty {
+		return // A concurrent request beat us to the tuning
+	}
+	// First thread reaching the tuning point, update the estimates and return
+	t.roundtrip = time.Duration((1-tuningImpact)*float64(t.roundtrip) + tuningImpact*float64(t.medianRoundTrip()))
+	t.confidence = t.confidence + (1-t.confidence)/2
+
+	t.tuned = time.Now()
+	t.log.Debug("Recalculated msgrate QoS values", "rtt", t.roundtrip, "confidence", t.confidence, "ttl", t.targetTimeout(), "next", t.tuned.Add(t.roundtrip))
+	t.log.Trace("Debug dump of mean capacities", "caps", log.Lazy{Fn: t.meanCapacities})
+}
+
+// detune reduces the tracker's confidence in order to make fresh measurements
+// have a larger impact on the estimates. It is meant to be used during new peer
+// connections so they can have a proper impact on the estimates.
+func (t *Trackers) detune() {
+	// If we have a single peer, confidence is always 1
+	if len(t.trackers) == 1 {
+		t.confidence = 1
+		return
+	}
+	// If we have a ton of peers, don't drop the confidence since there's enough
+	// remaining to retain the same throughput
+	if len(t.trackers) >= tuningConfidenceCap {
+		return
+	}
+	// Otherwise drop the confidence factor
+	peers := float64(len(t.trackers))
+
+	t.confidence = t.confidence * (peers - 1) / peers
+	if t.confidence < rttMinConfidence {
+		t.confidence = rttMinConfidence
+	}
+	t.log.Debug("Relaxed msgrate QoS values", "rtt", t.roundtrip, "confidence", t.confidence, "ttl", t.targetTimeout())
+}
+
+// Capacity is a helper function to access a specific tracker without having to
+// track it explicitly outside.
+func (t *Trackers) Capacity(id string, kind uint64, targetRTT time.Duration) float64 {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	tracker := t.trackers[id]
+	if tracker == nil {
+		return 1 // Unregister race, don't return 0, it's a dangerous number
+	}
+	return tracker.Capacity(kind, targetRTT)
+}
+
+// Update is a helper function to access a specific tracker without having to
+// track it explicitly outside.
+func (t *Trackers) Update(id string, kind uint64, elapsed time.Duration, items int) {
+	t.lock.RLock()
+	defer t.lock.RUnlock()
+
+	if tracker := t.trackers[id]; tracker != nil {
+		tracker.Update(kind, elapsed, items)
+	}
+}