diff --git a/internal/coordinator/pool/ledger.go b/internal/coordinator/pool/ledger.go
new file mode 100644
index 0000000..105145c
--- /dev/null
+++ b/internal/coordinator/pool/ledger.go
@@ -0,0 +1,256 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.13
+// +build linux darwin
+
+package pool
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"sort"
+	"sync"
+	"time"
+
+	"golang.org/x/build/internal/cloud"
+)
+
+// entry contains the resource usage of an instance as well as
+// identifying information.
+type entry struct {
+	createdAt    time.Time
+	instanceID   string
+	instanceName string
+	vCPUCount    int64
+}
+
+// ledger contains a record of the instances and their resource
+// consumption. Before an instance is created, a call to the ledger
+// will ensure that there are available resources for the new instance.
+type ledger struct {
+	mu sync.RWMutex
+	// cpuLimit is the limit of how many on-demand vCPUs can be created on EC2.
+	cpuLimit int64
+	// cpuUsed is the current count of vCPUs reserved for on-demand instances.
+	cpuUsed int64
+	// entries contains a mapping of instance name to entries for each instance
+	// that has resources allocated to it.
+	entries map[string]*entry
+	// types contains a mapping of instance type names to instance types for each
+	// ARM64 EC2 instance.
+	types map[string]*cloud.InstanceType
+}
+
+// newLedger creates a new ledger.
+func newLedger() *ledger {
+	return &ledger{
+		entries: make(map[string]*entry),
+		types:   make(map[string]*cloud.InstanceType),
+	}
+}
+
+// ReserveResources attempts to reserve the resources required for an instance to be created.
+// It will attempt to reserve the resourses that an instance type would require. This will
+// attempt to reserve the resources until the context deadline is reached.
+func (l *ledger) ReserveResources(ctx context.Context, instName, vmType string) error {
+	instType, err := l.PrepareReservationRequest(instName, vmType)
+	if err != nil {
+		return err
+	}
+
+	t := time.NewTicker(2 * time.Second)
+	defer t.Stop()
+	for {
+		if l.allocateCPU(instType.CPU, instName) {
+			return nil
+		}
+		select {
+		case <-t.C:
+		case <-ctx.Done():
+			return ctx.Err()
+		}
+	}
+}
+
+// PrepareReservationRequest ensures all the preconditions necessary for a reservation request are
+// met. If the conditions are met then an instance type for the requested VM type is returned. If
+// not an error is returned.
+func (l *ledger) PrepareReservationRequest(instName, vmType string) (*cloud.InstanceType, error) {
+	l.mu.RLock()
+	defer l.mu.RUnlock()
+
+	instType, ok := l.types[vmType]
+	if !ok {
+		return nil, fmt.Errorf("unknown EC2 vm type: %s", vmType)
+	}
+	_, ok = l.entries[instName]
+	if ok {
+		return nil, fmt.Errorf("quota has already been allocated for %s of type %s", instName, vmType)
+	}
+	return instType, nil
+}
+
+// releaseResources deletes the entry associated with an instance. The resources associated to the
+// instance will also be released. An error is returned if the instance entry is not found.
+// Lock l.mu must be held by the caller.
+func (l *ledger) releaseResources(instName string) error {
+	e, ok := l.entries[instName]
+	if !ok {
+		return fmt.Errorf("instance not found for releasing quota: %s", instName)
+	}
+	l.deallocateCPU(e.vCPUCount)
+	return nil
+}
+
+// allocateCPU ensures that there is enough CPU to allocate below the CPU Quota
+// for the caller to create a resouce with the numCPU passed in. If there is enough
+// then the ammount of used CPU will increase by the requested ammount. If there is
+// not enough CPU available, then a false is returned. In the event that CPU is allocated
+// an entry will be added in the entries map for the instance.
+func (l *ledger) allocateCPU(numCPU int64, instName string) bool {
+	// should never happen
+	if numCPU <= 0 {
+		log.Printf("invalid allocation requested: %d", numCPU)
+		return false
+	}
+
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	if numCPU+l.cpuUsed > l.cpuLimit {
+		return false
+	}
+	l.cpuUsed += numCPU
+	e, ok := l.entries[instName]
+	if ok {
+		e.vCPUCount = numCPU
+	} else {
+		l.entries[instName] = &entry{
+			instanceName: instName,
+			vCPUCount:    numCPU,
+		}
+	}
+	return true
+}
+
+// deallocateCPU releases the CPU allocated to an instance associated with an entry. When an instance
+// is deleted, the CPU allocated for the instance should not be counted against the CPU quota reserved
+// all on-demand instances. If an invalid CPU number is passed in the function will not lower the CPU count.
+// Lock l.mu must be held by the caller.
+func (l *ledger) deallocateCPU(numCPU int64) {
+	if numCPU <= 0 {
+		log.Printf("invalid deallocation requested: %d", numCPU)
+		return
+	}
+	if l.cpuUsed-numCPU < 0 {
+		log.Printf("attempting to deallocate more cpu than used: %d of %d", numCPU, l.cpuUsed)
+		return
+	}
+	l.cpuUsed -= numCPU
+}
+
+// UpdateReservation updates the entry for an instance with the id value for that instance. If
+// an entry for the instance does not exist then an error will be returned. Another mechanism should
+// be used to manage untracked instances. Updating the reservation acts as a signal that the instance
+// has actually been created since the instance ID is known.
+func (l *ledger) UpdateReservation(instName, instID string) error {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	e, ok := l.entries[instName]
+	if !ok {
+		return fmt.Errorf("unable to update reservation: instance not found %s", instName)
+	}
+	e.createdAt = time.Now()
+	e.instanceID = instID
+	return nil
+}
+
+// Remove releases any reserved resources for an instance and deletes the associated entry.
+// An error is returned if and entry does not exist for the instance.
+func (l *ledger) Remove(instName string) error {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	if err := l.releaseResources(instName); err != nil {
+		return fmt.Errorf("unable to remove instance: %w", err)
+	}
+	delete(l.entries, instName)
+	return nil
+}
+
+// InstanceID retrieves the instance ID for an instance by looking up the instance name.
+// If an instance is not found, an empty string is returned.
+func (l *ledger) InstanceID(instName string) string {
+	l.mu.RLock()
+	defer l.mu.RUnlock()
+
+	e, ok := l.entries[instName]
+	if !ok {
+		return ""
+	}
+	return e.instanceID
+}
+
+// SetCPULimit sets the vCPU limit used to determine if a CPU allocation would
+// cross the threshold for available CPU for on-demand instances.
+func (l *ledger) SetCPULimit(numCPU int64) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	l.cpuLimit = numCPU
+}
+
+// UpdateInstanceTypes updates the map of instance types used to map instance
+// type to the resources required for the instance.
+func (l *ledger) UpdateInstanceTypes(types []*cloud.InstanceType) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	for _, it := range types {
+		l.types[it.Type] = it
+	}
+}
+
+// resources contains the current limit and usage of instance related resources.
+type resources struct {
+	// InstCount is the count of how many on-demand instances are tracked in the ledger.
+	InstCount int64
+	// CPUUsed is a count of the vCPU's for on-demand instances are currently allocated in the ledger.
+	CPUUsed int64
+	// CPULimit is the limit of how many vCPU's for on-demand instances can be allocated.
+	CPULimit int64
+}
+
+// Resources retrives the resource usage and limits for instances in the
+// store.
+func (l *ledger) Resources() *resources {
+	l.mu.RLock()
+	defer l.mu.RUnlock()
+
+	return &resources{
+		InstCount: int64(len(l.entries)),
+		CPUUsed:   l.cpuUsed,
+		CPULimit:  l.cpuLimit,
+	}
+}
+
+// ResourceTime give a ResourceTime entry for each active instance.
+// The resource time slice is storted by creation time.
+func (l *ledger) ResourceTime() []ResourceTime {
+	l.mu.RLock()
+	defer l.mu.RUnlock()
+
+	ret := make([]ResourceTime, 0, len(l.entries))
+	for name, data := range l.entries {
+		ret = append(ret, ResourceTime{
+			Name:     name,
+			Creation: data.createdAt,
+		})
+	}
+	sort.Sort(ByCreationTime(ret))
+	return ret
+}
diff --git a/internal/coordinator/pool/ledger_test.go b/internal/coordinator/pool/ledger_test.go
new file mode 100644
index 0000000..3349b71
--- /dev/null
+++ b/internal/coordinator/pool/ledger_test.go
@@ -0,0 +1,539 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// +build go1.13
+// +build linux darwin
+
+package pool
+
+import (
+	"context"
+	"sort"
+	"testing"
+	"time"
+
+	"golang.org/x/build/internal/cloud"
+)
+
+func canceledContext() context.Context {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	return ctx
+}
+
+func TestLedgerReserveResources(t *testing.T) {
+	testCases := []struct {
+		desc      string
+		ctx       context.Context
+		instName  string
+		vmType    string
+		instTypes []*cloud.InstanceType
+		cpuLimit  int64
+		cpuUsed   int64
+		wantErr   bool
+	}{
+		{
+			desc:     "success",
+			ctx:      context.Background(),
+			instName: "small-instance",
+			vmType:   "aa.small",
+			instTypes: []*cloud.InstanceType{
+				&cloud.InstanceType{
+					Type: "aa.small",
+					CPU:  5,
+				},
+			},
+			cpuLimit: 20,
+			cpuUsed:  5,
+			wantErr:  false,
+		},
+		{
+			desc:     "cancelled-context",
+			ctx:      canceledContext(),
+			instName: "small-instance",
+			vmType:   "aa.small",
+			instTypes: []*cloud.InstanceType{
+				&cloud.InstanceType{
+					Type: "aa.small",
+					CPU:  5,
+				},
+			},
+			cpuLimit: 20,
+			cpuUsed:  20,
+			wantErr:  true,
+		},
+		{
+			desc:      "unknown-instance-type",
+			ctx:       context.Background(),
+			instName:  "small-instance",
+			vmType:    "aa.small",
+			instTypes: []*cloud.InstanceType{},
+			cpuLimit:  20,
+			cpuUsed:   5,
+			wantErr:   true,
+		},
+		{
+			desc:     "instance-already-exists",
+			ctx:      context.Background(),
+			instName: "large-instance",
+			vmType:   "aa.small",
+			instTypes: []*cloud.InstanceType{
+				&cloud.InstanceType{
+					Type: "aa.small",
+					CPU:  5,
+				},
+			},
+			cpuLimit: 20,
+			cpuUsed:  5,
+			wantErr:  true,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				cpuLimit: tc.cpuLimit,
+				cpuUsed:  tc.cpuUsed,
+				entries: map[string]*entry{
+					"large-instance": &entry{},
+				},
+				types: make(map[string]*cloud.InstanceType),
+			}
+			l.UpdateInstanceTypes(tc.instTypes)
+			gotErr := l.ReserveResources(tc.ctx, tc.instName, tc.vmType)
+			if (gotErr != nil) != tc.wantErr {
+				t.Errorf("ledger.reserveResources(%+v, %s, %s) = %s; want error %t", tc.ctx, tc.instName, tc.vmType, gotErr, tc.wantErr)
+			}
+		})
+	}
+}
+
+func TestLedgerReleaseResources(t *testing.T) {
+	testCases := []struct {
+		desc        string
+		instName    string
+		entry       *entry
+		cpuUsed     int64
+		wantCPUUsed int64
+		wantErr     bool
+	}{
+		{
+			desc:     "success",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-x",
+				vCPUCount:    10,
+			},
+			cpuUsed:     20,
+			wantCPUUsed: 10,
+			wantErr:     false,
+		},
+		{
+			desc:     "entry-not-found",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-w",
+				vCPUCount:    10,
+			},
+			cpuUsed:     20,
+			wantCPUUsed: 20,
+			wantErr:     true,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				cpuUsed: tc.cpuUsed,
+				entries: map[string]*entry{
+					tc.entry.instanceName: tc.entry,
+				},
+			}
+			gotErr := l.releaseResources(tc.instName)
+			if (gotErr != nil) != tc.wantErr {
+				t.Errorf("ledger.releaseResources(%s) = %s; want error %t", tc.instName, gotErr, tc.wantErr)
+			}
+			if l.cpuUsed != tc.wantCPUUsed {
+				t.Errorf("ledger.cpuUsed = %d; wanted %d", l.cpuUsed, tc.wantCPUUsed)
+			}
+		})
+	}
+}
+
+func TestLedgerAllocateCPU(t *testing.T) {
+	testCases := []struct {
+		desc        string
+		numCPU      int64
+		cpuLimit    int64
+		cpuUsed     int64
+		instName    string
+		wantReserve bool
+		wantCPUUsed int64
+	}{
+		{
+			desc:        "reservation-success",
+			numCPU:      10,
+			cpuLimit:    10,
+			cpuUsed:     0,
+			instName:    "chacha",
+			wantReserve: true,
+			wantCPUUsed: 10,
+		},
+		{
+			desc:        "failed-to-reserve",
+			numCPU:      10,
+			cpuLimit:    5,
+			cpuUsed:     0,
+			instName:    "pasa",
+			wantReserve: false,
+			wantCPUUsed: 0,
+		},
+		{
+			desc:        "invalid-cpu-count",
+			numCPU:      0,
+			cpuLimit:    50,
+			cpuUsed:     20,
+			instName:    "double",
+			wantReserve: false,
+			wantCPUUsed: 20,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				entries:  make(map[string]*entry),
+				cpuLimit: tc.cpuLimit,
+				cpuUsed:  tc.cpuUsed,
+			}
+			gotReserve := l.allocateCPU(tc.numCPU, tc.instName)
+			if gotReserve != tc.wantReserve {
+				t.Errorf("ledger.allocateCPU(%d) = %v, want %v", tc.numCPU, gotReserve, tc.wantReserve)
+			}
+			if l.cpuUsed != tc.wantCPUUsed {
+				t.Errorf("ledger.cpuUsed = %d; want %d", l.cpuUsed, tc.wantCPUUsed)
+			}
+			if _, ok := l.entries[tc.instName]; tc.wantReserve && !ok {
+				t.Fatalf("ledger.entries[%s] = nil; want it to exist", tc.instName)
+			}
+			if e, _ := l.entries[tc.instName]; tc.wantReserve && e.vCPUCount != tc.numCPU {
+				t.Fatalf("ledger.entries[%s].vCPUCount = %d; want %d", tc.instName, e.vCPUCount, tc.numCPU)
+			}
+		})
+	}
+}
+
+func TestLedgerDeallocateCPU(t *testing.T) {
+	testCases := []struct {
+		desc        string
+		numCPU      int64
+		cpuUsed     int64
+		wantCPUUsed int64
+	}{
+		{
+			desc:        "release-success",
+			numCPU:      10,
+			cpuUsed:     10,
+			wantCPUUsed: 0,
+		},
+		{
+			desc:        "failed-to-release",
+			numCPU:      10,
+			cpuUsed:     5,
+			wantCPUUsed: 5,
+		},
+		{
+			desc:        "invalid-cpu-count",
+			numCPU:      0,
+			cpuUsed:     0,
+			wantCPUUsed: 0,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				cpuUsed: tc.cpuUsed,
+			}
+			l.deallocateCPU(tc.numCPU)
+			if l.cpuUsed != tc.wantCPUUsed {
+				t.Errorf("ledger.cpuUsed = %d; want %d", l.cpuUsed, tc.wantCPUUsed)
+			}
+		})
+	}
+}
+
+func TestLedgerUpdateReservation(t *testing.T) {
+	testCases := []struct {
+		desc     string
+		instName string
+		instID   string
+		entry    *entry
+		wantErr  bool
+	}{
+		{
+			desc:     "success",
+			instName: "inst-x",
+			instID:   "id-foo-x",
+			entry: &entry{
+				instanceName: "inst-x",
+			},
+			wantErr: false,
+		},
+		{
+			desc:     "success",
+			instName: "inst-x",
+			instID:   "id-foo-x",
+			entry: &entry{
+				instanceName: "inst-w",
+			},
+			wantErr: true,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				entries: map[string]*entry{
+					tc.entry.instanceName: tc.entry,
+				},
+			}
+			if gotErr := l.UpdateReservation(tc.instName, tc.instID); (gotErr != nil) != tc.wantErr {
+				t.Errorf("ledger.updateReservation(%s, %s) = %s; want error %t", tc.instName, tc.instID, gotErr, tc.wantErr)
+			}
+			e, ok := l.entries[tc.instName]
+			if !tc.wantErr && !ok {
+				t.Fatalf("ledger.entries[%s] does not exist", tc.instName)
+			}
+			if !tc.wantErr && e.createdAt.IsZero() {
+				t.Errorf("ledger.entries[%s].createdAt = %s; time not set", tc.instName, e.createdAt)
+			}
+		})
+	}
+}
+
+func TestLedgerRemove(t *testing.T) {
+	testCases := []struct {
+		desc        string
+		instName    string
+		entry       *entry
+		cpuUsed     int64
+		wantCPUUsed int64
+		wantErr     bool
+	}{
+		{
+			desc:     "success",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-x",
+				vCPUCount:    10,
+			},
+			cpuUsed:     100,
+			wantCPUUsed: 90,
+			wantErr:     false,
+		},
+		{
+			desc:     "entry-does-not-exist",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-w",
+				vCPUCount:    10,
+			},
+			cpuUsed:     100,
+			wantCPUUsed: 100,
+			wantErr:     true,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				cpuUsed: tc.cpuUsed,
+				entries: map[string]*entry{
+					tc.entry.instanceName: tc.entry,
+				},
+			}
+			if gotErr := l.Remove(tc.instName); (gotErr != nil) != tc.wantErr {
+				t.Errorf("ledger.remove(%s) = %s; want error %t", tc.instName, gotErr, tc.wantErr)
+			}
+			if gotE, ok := l.entries[tc.instName]; ok {
+				t.Errorf("ledger.entries[%s] = %+v; want it not to exist", tc.instName, gotE)
+			}
+			if l.cpuUsed != tc.wantCPUUsed {
+				t.Errorf("ledger.cpuUsed = %d; want %d", l.cpuUsed, tc.wantCPUUsed)
+			}
+		})
+	}
+}
+
+func TestInstanceID(t *testing.T) {
+	testCases := []struct {
+		desc        string
+		instName    string
+		entry       *entry
+		cpuUsed     int64
+		wantCPUUsed int64
+		wantErr     bool
+	}{
+		{
+			desc:     "success",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-x",
+				vCPUCount:    10,
+			},
+			cpuUsed:     100,
+			wantCPUUsed: 90,
+			wantErr:     false,
+		},
+		{
+			desc:     "entry-does-not-exist",
+			instName: "inst-x",
+			entry: &entry{
+				instanceName: "inst-w",
+				vCPUCount:    10,
+			},
+			cpuUsed:     100,
+			wantCPUUsed: 100,
+			wantErr:     true,
+		},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				cpuUsed: tc.cpuUsed,
+				entries: map[string]*entry{
+					tc.entry.instanceName: tc.entry,
+				},
+			}
+			if gotErr := l.Remove(tc.instName); (gotErr != nil) != tc.wantErr {
+				t.Errorf("ledger.remove(%s) = %s; want error %t", tc.instName, gotErr, tc.wantErr)
+			}
+			if gotE, ok := l.entries[tc.instName]; ok {
+				t.Errorf("ledger.entries[%s] = %+v; want it not to exist", tc.instName, gotE)
+			}
+			if l.cpuUsed != tc.wantCPUUsed {
+				t.Errorf("ledger.cpuUsed = %d; want %d", l.cpuUsed, tc.wantCPUUsed)
+			}
+		})
+	}
+}
+
+func TestLedgerSetCPULimit(t *testing.T) {
+	l := &ledger{}
+	var want int64 = 300
+	l.SetCPULimit(300)
+	if l.cpuLimit != want {
+		t.Errorf("ledger.cpuLimit = %d; want %d", l.cpuLimit, want)
+	}
+}
+
+func TestLedgerUpdateInstanceTypes(t *testing.T) {
+	testCases := []struct {
+		desc  string
+		types []*cloud.InstanceType
+	}{
+		{"no-type", []*cloud.InstanceType{}},
+		{"single-type", []*cloud.InstanceType{&cloud.InstanceType{"x", 15}}},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := newLedger()
+			l.UpdateInstanceTypes(tc.types)
+			for _, it := range tc.types {
+				if gotV, ok := l.types[it.Type]; !ok || gotV != it {
+					t.Errorf("ledger.types[%s] = %v; want %v", it.Type, gotV, it)
+				}
+			}
+			if len(l.types) != len(tc.types) {
+				t.Errorf("len(ledger.types) = %d; want %d", len(l.types), len(tc.types))
+			}
+		})
+	}
+}
+
+func TestLedgerResources(t *testing.T) {
+	testCases := []struct {
+		desc          string
+		entries       map[string]*entry
+		cpuCount      int64
+		cpuLimit      int64
+		wantInstCount int64
+	}{
+		{"no-instances", map[string]*entry{}, 2, 3, 0},
+		{"single-instance", map[string]*entry{"x": &entry{}}, 2, 3, 1},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				entries:  tc.entries,
+				cpuLimit: tc.cpuLimit,
+				cpuUsed:  tc.cpuCount,
+			}
+			gotR := l.Resources()
+			if gotR.InstCount != tc.wantInstCount {
+				t.Errorf("ledger.instCount = %d; want %d", gotR.InstCount, tc.wantInstCount)
+			}
+			if gotR.CPUUsed != tc.cpuCount {
+				t.Errorf("ledger.cpuCount = %d; want %d", gotR.CPUUsed, tc.cpuCount)
+			}
+			if gotR.CPULimit != tc.cpuLimit {
+				t.Errorf("ledger.cpuLimit = %d; want %d", gotR.CPULimit, tc.cpuLimit)
+			}
+		})
+	}
+}
+
+func TestLedgerResourceTime(t *testing.T) {
+	ct := time.Now()
+
+	testCases := []struct {
+		desc    string
+		entries map[string]*entry
+	}{
+		{"no-instances", map[string]*entry{}},
+		{"single-instance", map[string]*entry{
+			"inst-x": &entry{
+				createdAt:    ct,
+				instanceID:   "id-x",
+				instanceName: "inst-x",
+				vCPUCount:    1,
+			},
+		}},
+		{"multiple-instances", map[string]*entry{
+			"inst-z": &entry{
+				createdAt:    ct.Add(2 * time.Second),
+				instanceID:   "id-z",
+				instanceName: "inst-z",
+				vCPUCount:    1,
+			},
+			"inst-y": &entry{
+				createdAt:    ct.Add(time.Second),
+				instanceID:   "id-y",
+				instanceName: "inst-y",
+				vCPUCount:    1,
+			},
+			"inst-x": &entry{
+				createdAt:    ct,
+				instanceID:   "id-x",
+				instanceName: "inst-x",
+				vCPUCount:    1,
+			},
+		}},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.desc, func(t *testing.T) {
+			l := &ledger{
+				entries: tc.entries,
+			}
+			gotRT := l.ResourceTime()
+			if !sort.SliceIsSorted(gotRT, func(i, j int) bool { return gotRT[i].Creation.Before(gotRT[j].Creation) }) {
+				t.Errorf("resource time is not sorted")
+			}
+			if len(l.entries) != len(gotRT) {
+				t.Errorf("mismatch in items returned")
+			}
+			for _, rt := range gotRT {
+				delete(l.entries, rt.Name)
+			}
+			if len(l.entries) != 0 {
+				t.Errorf("mismatch")
+			}
+		})
+	}
+}