diff --git a/experimental/tracer/vendor/github.com/msackman/skiplist/LICENSE b/experimental/tracer/vendor/github.com/msackman/skiplist/LICENSE
new file mode 100644
index 000000000..872daf759
--- /dev/null
+++ b/experimental/tracer/vendor/github.com/msackman/skiplist/LICENSE
@@ -0,0 +1,22 @@
+The MIT License (MIT)
+
+Copyright (c) 2015 Matthew Sackman
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
diff --git a/experimental/tracer/vendor/github.com/msackman/skiplist/README.md b/experimental/tracer/vendor/github.com/msackman/skiplist/README.md
new file mode 100644
index 000000000..4c5f8d87c
--- /dev/null
+++ b/experimental/tracer/vendor/github.com/msackman/skiplist/README.md
@@ -0,0 +1,2 @@
+Skiplist implementation for Go. Comes with tests. Currently no docs,
+but it's pretty obvious stuff.
\ No newline at end of file
diff --git a/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist.go b/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist.go
new file mode 100644
index 000000000..4f568f49e
--- /dev/null
+++ b/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist.go
@@ -0,0 +1,429 @@
+package skiplist
+
+import (
+	"fmt"
+	"math"
+	"math/rand"
+	"strings"
+)
+
+const (
+	p            = 0.3
+	defaultDepth = 2
+)
+
+type Comparable interface {
+	LessThan(Comparable) bool
+	Equal(Comparable) bool
+}
+
+type SkipList struct {
+	length             uint
+	terminus           *Node
+	levelProbabilities []float32
+	curCapacity        uint
+	curDepth           uint
+	nodes              []Node
+	localRand          *rand.Rand
+}
+
+type Node struct {
+	Key        Comparable
+	Value      interface{}
+	heightRand float32
+	prev       *Node
+	nexts      []*Node
+	skiplist   *SkipList
+}
+
+func New(rng *rand.Rand) *SkipList {
+	depth := defaultDepth
+
+	terminus := &Node{
+		heightRand: 0,
+		nexts:      make([]*Node, depth),
+	}
+	terminus.prev = terminus
+	for idx := 0; idx < len(terminus.nexts); idx++ {
+		terminus.nexts[idx] = terminus
+	}
+	s := &SkipList{
+		length:    0,
+		terminus:  terminus,
+		curDepth:  uint(depth),
+		localRand: rng,
+	}
+	s.levelProbabilities = []float32{p}
+	terminus.skiplist = s
+	s.determineCapacity()
+
+	// s.validate()
+
+	return s
+}
+
+func (s *SkipList) determineCapacity() {
+	base := float64(1.0) / p
+	capacity := math.Pow(base, float64(s.curDepth))
+	s.curCapacity = uint(math.Floor(capacity))
+}
+
+func (s *SkipList) chooseNumLevels() (float32, int) {
+	r := s.localRand.Float32()
+	max := len(s.levelProbabilities)
+	for idx := 0; idx < max; idx++ {
+		if r > s.levelProbabilities[idx] {
+			return r, idx + 1
+		}
+	}
+	return r, max + 1
+}
+
+func (s *SkipList) ensureCapacity() {
+	// defer s.validate()
+	if s.length < s.curCapacity {
+		return
+	}
+
+	threshold := p * s.levelProbabilities[s.curDepth-2]
+	s.curDepth++
+	s.levelProbabilities = append(s.levelProbabilities, threshold)
+
+	s.determineCapacity()
+
+	// cur and next are just used to walk through the list at lvl. prev
+	// records the last node that made it up to the new level.
+	cur := s.terminus
+	lvl := len(cur.nexts) - 1
+	prev := cur
+	for {
+		next := cur.nexts[lvl]
+		if cur.heightRand <= threshold {
+			cur.nexts = append(cur.nexts, s.terminus)
+			prev.nexts[lvl+1] = cur
+			prev = cur
+		}
+		if next == s.terminus {
+			break
+		} else {
+			cur = next
+		}
+	}
+}
+
+func (s *SkipList) getNode() *Node {
+	l := len(s.nodes)
+	if l == 0 {
+		l = int(s.curCapacity)
+		s.nodes = make([]Node, l)
+	}
+	l--
+	n := &s.nodes[l]
+	s.nodes = s.nodes[:l]
+	return n
+}
+
+func (s *SkipList) getEqOrLessThan(cur *Node, k Comparable, captureDescent bool) (*Node, []*Node) {
+	// defer s.validate()
+
+	if s.length == 0 {
+		return s.terminus, nil
+	}
+	if cur != s.terminus {
+		if k.Equal(cur.Key) {
+			return cur, nil
+		}
+		if k.LessThan(cur.Key) {
+			return s.getEqOrLessThan(s.terminus, k, captureDescent)
+		}
+	}
+	// 1. Travel, not-descending, as far as possible
+	lvl := len(cur.nexts) - 1
+	for {
+		n := cur.nexts[lvl]
+		if n == s.terminus {
+			break
+		}
+		if n.Key.LessThan(k) {
+			cur = n
+			lvl = len(cur.nexts) - 1
+		} else if n.Key.Equal(k) {
+			return n, nil
+		} else {
+			break
+		}
+	}
+	// 2. Now descend as needed
+	var descent []*Node
+	if captureDescent {
+		descent = make([]*Node, lvl+1)
+		descent[lvl] = cur
+	}
+	for lvl--; lvl >= 0; lvl-- {
+		for {
+			n := cur.nexts[lvl]
+			if n == s.terminus {
+				break
+			}
+			if n.Key.LessThan(k) {
+				cur = n
+			} else if n.Key.Equal(k) {
+				return n, descent
+			} else {
+				break
+			}
+		}
+		if captureDescent {
+			descent[lvl] = cur
+		}
+	}
+	return cur, descent
+}
+
+func (s *SkipList) insert(cur *Node, k Comparable, v interface{}, n *Node) *Node {
+	// defer s.validate()
+
+	// do this first even though we may not need to - if we do it after
+	// the getEqOrLessThan call, we may break descent.
+	s.ensureCapacity()
+	cur, descent := s.getEqOrLessThan(cur, k, true)
+	if cur != s.terminus && cur.Key.Equal(k) {
+		cur.Value = v
+		return cur
+	}
+	// We didn't find k, so cur will be the node immediately prior to
+	// where k should go.
+	heightRand, height := s.chooseNumLevels()
+	if n == nil {
+		n = s.getNode()
+	}
+	n.Key = k
+	n.Value = v
+	n.heightRand = heightRand
+	n.nexts = make([]*Node, height)
+	n.prev = cur
+	n.skiplist = s
+
+	if len(cur.nexts) >= len(n.nexts) {
+		for idx := 0; idx < len(n.nexts); idx++ {
+			n.nexts[idx] = cur.nexts[idx]
+			cur.nexts[idx] = n
+		}
+	} else {
+		// Descent may capture only part of the path: it may be shorter
+		// than levels (in the case where the original cur is !=
+		// s.terminus) and we reached the correct location without
+		// travelling up very far. However, because we didn't find k, we
+		// know that all the "lower" levels of descent will be populated
+		// (where "lower" is "closer to [0]"), so we just need to fill in
+		// the "top".
+		if len(n.nexts) > len(descent) {
+			_, extra := s.getEqOrLessThan(s.terminus, descent[len(descent)-1].Key, true)
+			// Aside: because we know we'll find that Key, all the lower
+			// indices of extra will be nil.
+			descent = append(descent, extra[len(descent):]...)
+		}
+		for idx := 0; idx < len(n.nexts); idx++ {
+			n.nexts[idx] = descent[idx].nexts[idx]
+			descent[idx].nexts[idx] = n
+		}
+	}
+	n._next().prev = n
+	s.length++
+	return n
+}
+
+func (s *SkipList) remove(cur *Node, k Comparable) interface{} {
+	// defer s.validate()
+
+	n, _ := s.getEqOrLessThan(cur, k, false)
+	if n == s.terminus || !n.Key.Equal(k) {
+		return nil
+	}
+	s.removeNode(n)
+	n.nullify()
+	return n.Value
+}
+
+func (s *SkipList) removeNode(n *Node) {
+	// defer s.validate()
+
+	p := n.prev
+	n._next().prev = p
+	s.length--
+	for idx := 0; idx < len(p.nexts) && idx < len(n.nexts); idx++ {
+		p.nexts[idx] = n.nexts[idx]
+	}
+	if len(p.nexts) < len(n.nexts) {
+		_, descent := s.getEqOrLessThan(s.terminus, p.Key, true)
+		// because we know we're going to find Key, the lower indices
+		// of descent will be nil. But we know p == n.prev, so all of
+		// those pointers will be to n anyway, which we've already
+		// dealt with in the previous loop.
+		for idx := len(p.nexts); idx < len(n.nexts); idx++ {
+			descent[idx].nexts[idx] = n.nexts[idx]
+		}
+	}
+}
+
+func (s *SkipList) reposition(cur *Node, k Comparable) {
+	// defer s.validate()
+
+	needsMove := false
+	if cur != s.terminus {
+		if cur.prev != s.terminus && !cur.prev.Key.LessThan(k) {
+			needsMove = true
+		} else if n := cur._next(); n != s.terminus && !k.LessThan(n.Key) {
+			needsMove = true
+		}
+	}
+	if needsMove {
+		s.removeNode(cur)
+		cur.Key = k
+		s.insert(cur.prev, cur.Key, cur.Value, cur)
+	}
+}
+
+func (s *SkipList) First() *Node {
+	return s.terminus.Next()
+}
+
+func (s *SkipList) Last() *Node {
+	return s.terminus.Prev()
+}
+
+func (s *SkipList) Insert(k Comparable, v interface{}) *Node {
+	return s.insert(s.terminus, k, v, nil)
+}
+
+func (s *SkipList) Get(k Comparable) *Node {
+	return s.terminus.Get(k)
+}
+
+func (s *SkipList) Remove(k Comparable) interface{} {
+	return s.remove(s.terminus, k)
+}
+
+func (s *SkipList) Len() uint {
+	return s.length
+}
+
+// NB: this destroys t. Do not use t after this.
+func (s *SkipList) Merge(t *SkipList) {
+	// defer s.validate()
+
+	cur := s.terminus
+	for n := t.First(); n != nil; {
+		m := n.Next() // need to save this out before we destroy it in the insert
+		cur = s.insert(cur, n.Key, n.Value, n)
+		n = m
+	}
+}
+
+func (s *SkipList) validate() {
+	visited := make(map[*Node]bool, int(s.length))
+	cur := s.terminus
+	visited[cur] = true
+	l := uint(0)
+	for {
+		if cur != s.terminus {
+			l++
+		}
+		if cur._next().prev != cur {
+			panic(fmt.Sprintf("Node (%v) has next pointer to %v, which has prev pointer to %v", cur, cur._next(), cur._next().prev))
+		}
+		if cur.prev._next() != cur {
+			panic(fmt.Sprintf("Node (%v) has prev pointer to %v, which has next pointer to %v", cur, cur.prev, cur.prev._next()))
+		}
+		for h, n := range cur.nexts {
+			if h >= len(n.nexts) {
+				panic(fmt.Sprintf("Node (%v) has next pointer at level %v pointing down to node (%v) which has %v height", cur, h, n, len(n.nexts)))
+			}
+		}
+		n := cur._next()
+		if n == s.terminus {
+			break
+		}
+		if visited[n] {
+			panic(fmt.Sprintf("Node (%v) has next as %v which is already visited!", cur, n))
+		}
+		if cur != s.terminus && !cur.Key.LessThan(n.Key) {
+			panic(fmt.Sprintf("Node keys in wrong order: expecting %v < %v", cur.Key, n.Key))
+		}
+		if n.prev != cur {
+			panic(fmt.Sprintf("Node (%v) has next (%v) which does not point back correctly", cur, n))
+		}
+		cur = n
+	}
+	if l != s.length {
+		panic(fmt.Sprintf("length is wrong: counted %v but length is %v", l, s.length))
+	}
+}
+
+func (n *Node) Get(k Comparable) *Node {
+	m, _ := n.skiplist.getEqOrLessThan(n, k, false)
+	if m != n.skiplist.terminus && m.Key.Equal(k) {
+		return m
+	} else {
+		return nil
+	}
+}
+
+func (n *Node) Insert(k Comparable, v interface{}) *Node {
+	return n.skiplist.insert(n, k, v, nil)
+}
+
+func (n *Node) Remove() interface{} {
+	return n.skiplist.remove(n, n.Key)
+}
+
+func (n *Node) _next() *Node {
+	return n.nexts[0]
+}
+
+func (n *Node) Next() *Node {
+	if m := n.nexts[0]; m != n.skiplist.terminus {
+		return m
+	} else {
+		return nil
+	}
+}
+
+func (n *Node) Prev() *Node {
+	if m := n.prev; m != n.skiplist.terminus {
+		return m
+	} else {
+		return nil
+	}
+}
+
+func (n *Node) Reposition(k Comparable) {
+	n.skiplist.reposition(n, k)
+}
+
+func (n *Node) nullify() {
+	// this is called when n is removed from the skiplist. It's really
+	// just to ensure that if someone has a reference to n lying
+	// around, they can't use it.
+	n.prev = nil
+	n.nexts = nil
+	n.skiplist = nil
+}
+
+func (s *SkipList) String() string {
+	strs := make([]string, 1, s.length+1)
+	strs[0] = fmt.Sprint(s.terminus)
+	for cur := s.terminus._next(); cur != s.terminus; cur = cur._next() {
+		strs = append(strs, fmt.Sprint(cur))
+	}
+	return fmt.Sprintf("Skiplist of length %v (counted: %v), levelProbabilities %v, and nodes:\n\t[%v]",
+		s.length, len(strs)-1, s.levelProbabilities, strings.Join(strs, ",\n\t "))
+}
+
+func (n *Node) String() string {
+	strs := make([]string, len(n.nexts))
+	for idx := 0; idx < len(strs); idx++ {
+		strs[idx] = fmt.Sprint(n.nexts[idx].Key)
+	}
+	return fmt.Sprintf("%v -> %v (nexts: [%v])", n.Key, n.Value, strings.Join(strs, ", "))
+}
diff --git a/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist_test.go b/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist_test.go
new file mode 100644
index 000000000..84bf112ed
--- /dev/null
+++ b/experimental/tracer/vendor/github.com/msackman/skiplist/skiplist_test.go
@@ -0,0 +1,315 @@
+package skiplist
+
+import (
+	"math/rand"
+	"os"
+	"strconv"
+	"testing"
+)
+
+type intKey int
+
+func (sk intKey) LessThan(b Comparable) bool {
+	return sk < b.(intKey)
+}
+
+func (sk intKey) Equal(b Comparable) bool {
+	return sk == b.(intKey)
+}
+
+const (
+	keyValsLen = 1000000
+)
+
+var (
+	rng     = rand.New(rand.NewSource(0))
+	keyVals = make(map[intKey]string)
+	indices = make([]intKey, keyValsLen)
+)
+
+func TestAddFromEmptyRoot(t *testing.T) {
+	for idx := 0; idx < 10; idx++ {
+		s := New(rng)
+		for idy := 0; idy < idx; idy++ {
+			k := indices[idy]
+			v := keyVals[k]
+			n := s.Insert(k, v)
+			if n.Key != k || n.Value != v {
+				t.Fatal("Key or Value of inserted pair changed!", k, v, n)
+			}
+			if m := s.Get(k); n != m {
+				t.Fatal("Node changed for just inserted value:", n, m)
+			}
+			if s.Len() != uint(idy+1) {
+				t.Fatal("Incorrect length")
+			}
+		}
+	}
+}
+
+func TestAddFromEmptyRel(t *testing.T) {
+	for idx := 0; idx < 10; idx++ {
+		s := New(rng)
+		var n *Node
+		for idy := 0; idy < idx; idy++ {
+			k := indices[idy]
+			v := keyVals[k]
+			if n == nil {
+				n = s.Insert(k, v)
+			} else {
+				n = n.Insert(k, v)
+			}
+			if n.Key != k || n.Value != v {
+				t.Fatal("Key or Value of inserted pair changed!", k, v, n)
+			}
+			if m := n.Get(k); n != m {
+				t.Fatal("Node changed for just inserted value:", n, m)
+			}
+			if p := n.Prev(); p != nil {
+				if m := p.Get(k); n != m {
+					t.Fatal("Node changed for just inserted value:", n, m)
+				}
+			}
+			if s.Len() != uint(idy+1) {
+				t.Fatal("Incorrect length")
+			}
+		}
+	}
+}
+
+func TestDupInsert(t *testing.T) {
+	s := New(rng)
+	for idx := 0; idx < 10; idx++ {
+		for idy := 0; idy < idx; idy++ {
+			k := indices[idy]
+			v := keyVals[k]
+			if n := s.Insert(k, v); n.Key != k || n.Value != v {
+				t.Fatal("Key or Value of inserted pair changed!", k, v, n)
+			}
+		}
+	}
+}
+
+func TestGetMissingEmpty(t *testing.T) {
+	s := New(rng)
+	for idx := 0; idx < 10; idx++ {
+		k := indices[idx]
+		if n := s.Get(k); n != nil {
+			t.Fatal("Expected not to find elem")
+		}
+	}
+}
+
+func TestGetMissingNonEmpty(t *testing.T) {
+	s := New(rng)
+	for idx := 0; idx < 20; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		if idx%2 == 0 {
+			s.Insert(k, v)
+		} else {
+			if n := s.Get(k); n != nil {
+				t.Fatal("Expected not to find elem")
+			}
+		}
+	}
+}
+
+func TestRemoveRoot(t *testing.T) {
+	s := New(rng)
+	for idx := 0; idx < 20; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		s.Insert(k, v)
+	}
+	for idx := 0; idx < 20; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		if u := s.Remove(k); u != v {
+			t.Fatal("Wrong value returned:", u, v)
+		}
+		if int(s.Len()) != 19-idx {
+			t.Fatal("Wrong length")
+		}
+	}
+}
+
+func TestRemoveSelf(t *testing.T) {
+	s := New(rng)
+	ns := []*Node{}
+	for idx := 0; idx < 20; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		ns = append(ns, s.Insert(k, v))
+	}
+	for idx, n := range ns {
+		k := indices[idx]
+		v := keyVals[k]
+		if u := n.Remove(); u != v {
+			t.Fatal("Wrong value returned:", u, v)
+		}
+		if int(s.Len()) != len(ns)-idx-1 {
+			t.Fatal("Wrong length")
+		}
+	}
+}
+
+func TestRemoveMissing(t *testing.T) {
+	s := New(rng)
+	for idx := 0; idx < 20; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		s.Insert(k, v)
+	}
+	for idx := 0; idx < 20; idx++ {
+		for idy := 0; idy < idx; idy++ {
+			k := indices[idy]
+			v := keyVals[k]
+			u := s.Remove(k)
+			if idy+1 == idx {
+				if u != v {
+					t.Fatal("Wrong value returned:", u, v)
+				}
+			} else {
+				if u != nil {
+					t.Fatal("Wrong value returned - expected nil:", u)
+				}
+			}
+		}
+	}
+	if s.Len() != 1 {
+		t.Fatal("Wrong length")
+	}
+}
+
+func TestFirstLast(t *testing.T) {
+	s := New(rng)
+	if s.First() != nil || s.Last() != nil {
+		t.Fatal("Expected nil for First and Last on empty list")
+	}
+	var min, max intKey
+	for idx := 0; idx < 200; idx++ {
+		k := indices[idx]
+		v := keyVals[k]
+		s.Insert(k, v)
+		if idx == 0 {
+			min, max = k, k
+		} else {
+			if k < min {
+				min = k
+			}
+			if k > max {
+				max = k
+			}
+		}
+	}
+	if f := s.First(); f.Key != min {
+		t.Fatal("Did not get minimum key back for first", min, f)
+	}
+	if l := s.Last(); l.Key != max {
+		t.Fatal("Did not get maximum key back for last", max, l)
+	}
+}
+
+func TestMerge(t *testing.T) {
+	s0 := New(rng)
+	s1 := New(rng)
+	lim := 1000
+	for idx := 0; idx < lim; idx++ {
+		if idx%2 == 0 {
+			s0.Insert(intKey(idx), idx)
+		} else {
+			s1.Insert(intKey(idx), idx)
+		}
+	}
+	s0.Merge(s1)
+	if int(s0.Len()) != lim {
+		t.Fatal("Wrong len after merge", s0.Len())
+	}
+	cur := s0.First()
+	for idx := 0; idx < lim; idx++ {
+		if cur.Value.(int) != idx {
+			t.Fatal("Wrong value: ", cur.Value)
+		}
+		if cur != s0.Get(cur.Key) {
+			t.Fatal("Internal failure: ", cur)
+		}
+		cur = cur.Next()
+	}
+}
+
+func TestReposition(t *testing.T) {
+	s := New(rng)
+	lim := 200
+	for idx := 0; idx < lim; idx++ {
+		s.Insert(intKey(idx*5), idx)
+	}
+	for idx := lim - 1; idx >= 0; idx-- {
+		n := s.Get(intKey(idx * 5))
+		if n == nil {
+			t.Fatal("Unable to find node")
+		}
+		n.Reposition(intKey((idx * 5) - 11))
+	}
+	n := s.First()
+	for idx := 0; idx < lim; idx++ {
+		if n.Value != idx {
+			t.Fatal("Wrong value", idx, n)
+		}
+		n = n.Next()
+	}
+	if n != nil {
+		t.Fatal("Too many values!")
+	}
+}
+
+func BenchmarkAdd08192(b *testing.B) { benchmarkAdd(8192, b) }
+func BenchmarkAdd16384(b *testing.B) { benchmarkAdd(16384, b) }
+func BenchmarkAdd32768(b *testing.B) { benchmarkAdd(32768, b) }
+func BenchmarkAdd65536(b *testing.B) { benchmarkAdd(65536, b) }
+
+func benchmarkAdd(initial int, b *testing.B) {
+	s := populate(New(rng), 0, initial)
+	b.ResetTimer()
+	populateFast(s, initial, b.N)
+}
+
+func BenchmarkGet08192(b *testing.B) { benchmarkGet(8192, b) }
+func BenchmarkGet16384(b *testing.B) { benchmarkGet(16384, b) }
+func BenchmarkGet32768(b *testing.B) { benchmarkGet(32768, b) }
+func BenchmarkGet65536(b *testing.B) { benchmarkGet(65536, b) }
+
+func benchmarkGet(initial int, b *testing.B) {
+	s := populate(New(rng), 0, initial)
+	b.ResetTimer()
+	for idx := 0; idx < b.N; idx++ {
+		s.Get(indices[idx%initial])
+	}
+}
+
+func populate(s *SkipList, offset, lim int) *SkipList {
+	for idx := 0; idx < lim; idx++ {
+		k := indices[(offset+idx)%len(indices)]
+		v := keyVals[k]
+		s.Insert(k, v)
+	}
+	return s
+}
+
+func populateFast(s *SkipList, offset, lim int) *SkipList {
+	for idx := 0; idx < lim; idx++ {
+		n := idx + offset
+		s.Insert(intKey(n), n)
+	}
+	return s
+}
+
+func TestMain(m *testing.M) {
+	for idx := 0; idx < keyValsLen; idx++ {
+		keyVals[intKey(idx)] = strconv.FormatInt(int64(idx), 3)
+	}
+	for idx, k := range rand.Perm(keyValsLen) {
+		indices[idx] = intKey(k)
+	}
+	os.Exit(m.Run())
+}
diff --git a/experimental/tracer/vendor/manifest b/experimental/tracer/vendor/manifest
new file mode 100644
index 000000000..0005f2ddc
--- /dev/null
+++ b/experimental/tracer/vendor/manifest
@@ -0,0 +1,11 @@
+{
+	"version": 0,
+	"dependencies": [
+		{
+			"importpath": "github.com/msackman/skiplist",
+			"repository": "https://github.com/msackman/skiplist",
+			"revision": "57733164b18444c51f63e9a80f1693961dde8036",
+			"branch": "master"
+		}
+	]
+}
\ No newline at end of file