Documentation ¶
Index ¶
- func StatScaleToTime(r float64) time.Duration
- func TimeToStatScale(d time.Duration) float64
- type NodeValueTracker
- type PrivateClientAPI
- func (api *PrivateClientAPI) Distribution(nodeStr string, normalized bool) (RtDistribution, error)
- func (api *PrivateClientAPI) RequestStats() []RequestStatsItem
- func (api *PrivateClientAPI) Timeout(nodeStr string, failRate float64) (float64, error)
- func (api *PrivateClientAPI) Value(nodeStr string, timeout float64) (float64, error)
- type RequestInfo
- type RequestStatsItem
- type ResponseTimeStats
- func (rt *ResponseTimeStats) Add(respTime time.Duration, weight float64, expFactor utils.ExpirationFactor)
- func (rt *ResponseTimeStats) AddStats(s *ResponseTimeStats)
- func (rt *ResponseTimeStats) DecodeRLP(s *rlp.Stream) error
- func (rt ResponseTimeStats) Distribution(normalized bool, expFactor utils.ExpirationFactor) (res RtDistribution)
- func (rt *ResponseTimeStats) EncodeRLP(w io.Writer) error
- func (rt *ResponseTimeStats) SubStats(s *ResponseTimeStats)
- func (rt ResponseTimeStats) Timeout(failRatio float64) time.Duration
- func (rt ResponseTimeStats) Value(weights ResponseTimeWeights, expFactor utils.ExpirationFactor) float64
- type ResponseTimeWeights
- type RtDistribution
- type ServedRequest
- type ValueTracker
- func (vt *ValueTracker) GetNode(id enode.ID) *NodeValueTracker
- func (vt *ValueTracker) Register(id enode.ID) *NodeValueTracker
- func (vt *ValueTracker) RequestStats() []RequestStatsItem
- func (vt *ValueTracker) RtStats() ResponseTimeStats
- func (vt *ValueTracker) Served(nv *NodeValueTracker, reqs []ServedRequest, respTime time.Duration)
- func (vt *ValueTracker) StatsExpirer() *utils.Expirer
- func (vt *ValueTracker) Stop()
- func (vt *ValueTracker) TotalServiceValue(nv *NodeValueTracker, weights ResponseTimeWeights) float64
- func (vt *ValueTracker) Unregister(id enode.ID)
- func (vt *ValueTracker) UpdateCosts(nv *NodeValueTracker, reqCosts []uint64)
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func StatScaleToTime ¶
StatScaleToTime converts a distribution vector index to a response time. The index is represented by a float64 so that linear interpolation can be applied.
func TimeToStatScale ¶
TimeToStatScale converts a response time to a distribution vector index. The index is represented by a float64 so that linear interpolation can be applied.
Types ¶
type NodeValueTracker ¶
type NodeValueTracker struct {
// contains filtered or unexported fields
}
NodeValueTracker collects service value statistics for a specific server node
func (*NodeValueTracker) RtStats ¶
func (nv *NodeValueTracker) RtStats() ResponseTimeStats
RtStats returns the node's own response time distribution statistics
type PrivateClientAPI ¶
type PrivateClientAPI struct {
// contains filtered or unexported fields
}
PrivateClientAPI implements the lespay client side API
func NewPrivateClientAPI ¶
func NewPrivateClientAPI(vt *ValueTracker) *PrivateClientAPI
NewPrivateClientAPI creates a PrivateClientAPI
func (*PrivateClientAPI) Distribution ¶
func (api *PrivateClientAPI) Distribution(nodeStr string, normalized bool) (RtDistribution, error)
Distribution returns a distribution as a series of (X, Y) chart coordinates, where the X axis is the response time in seconds while the Y axis is the amount of service value received with a response time close to the X coordinate. The distribution is optionally normalized to a sum of 1. If nodeStr == "" then the global distribution is returned, otherwise the individual distribution of the specified server node.
func (*PrivateClientAPI) RequestStats ¶
func (api *PrivateClientAPI) RequestStats() []RequestStatsItem
RequestStats returns the current contents of the reference request basket, with request values meaning average per request rather than total.
func (*PrivateClientAPI) Timeout ¶
func (api *PrivateClientAPI) Timeout(nodeStr string, failRate float64) (float64, error)
Timeout suggests a timeout value based on either the global distribution or the distribution of the specified node. The parameter is the desired rate of timeouts assuming a similar distribution in the future. Note that the actual timeout should have a sensible minimum bound so that operating under ideal working conditions for a long time (for example, using a local server with very low response times) will not make it very hard for the system to accommodate longer response times in the future.
type RequestInfo ¶
type RequestInfo struct { // Name identifies the request type and is used for re-mapping the service vector if necessary Name string // InitAmount and InitValue are used to initialize the reference basket InitAmount, InitValue float64 }
RequestInfo is an initializer structure for the service vector.
type RequestStatsItem ¶
type ResponseTimeStats ¶
type ResponseTimeStats struct {
// contains filtered or unexported fields
}
ResponseTimeStats is the response time distribution of a set of answered requests, weighted with request value, either served by a single server or aggregated for multiple servers. It it a fixed length (timeStatLength) distribution vector with linear interpolation. The X axis (the time values) are not linear, they should be transformed with TimeToStatScale and StatScaleToTime.
func (*ResponseTimeStats) Add ¶
func (rt *ResponseTimeStats) Add(respTime time.Duration, weight float64, expFactor utils.ExpirationFactor)
Add adds a new response time with the given weight to the distribution.
func (*ResponseTimeStats) AddStats ¶
func (rt *ResponseTimeStats) AddStats(s *ResponseTimeStats)
AddStats adds the given ResponseTimeStats to the current one.
func (*ResponseTimeStats) DecodeRLP ¶
func (rt *ResponseTimeStats) DecodeRLP(s *rlp.Stream) error
DecodeRLP implements rlp.Decoder
func (ResponseTimeStats) Distribution ¶
func (rt ResponseTimeStats) Distribution(normalized bool, expFactor utils.ExpirationFactor) (res RtDistribution)
Distribution returns a RtDistribution, optionally normalized to a sum of 1.
func (*ResponseTimeStats) EncodeRLP ¶
func (rt *ResponseTimeStats) EncodeRLP(w io.Writer) error
EncodeRLP implements rlp.Encoder
func (*ResponseTimeStats) SubStats ¶
func (rt *ResponseTimeStats) SubStats(s *ResponseTimeStats)
SubStats subtracts the given ResponseTimeStats from the current one.
func (ResponseTimeStats) Timeout ¶
func (rt ResponseTimeStats) Timeout(failRatio float64) time.Duration
Timeout suggests a timeout value based on the previous distribution. The parameter is the desired rate of timeouts assuming a similar distribution in the future. Note that the actual timeout should have a sensible minimum bound so that operating under ideal working conditions for a long time (for example, using a local server with very low response times) will not make it very hard for the system to accommodate longer response times in the future.
func (ResponseTimeStats) Value ¶
func (rt ResponseTimeStats) Value(weights ResponseTimeWeights, expFactor utils.ExpirationFactor) float64
Value calculates the total service value based on the given distribution, using the specified weight function.
type ResponseTimeWeights ¶
type ResponseTimeWeights [timeStatLength]float64
ResponseTimeStats is the response time distribution of a set of answered requests, weighted with request value, either served by a single server or aggregated for multiple servers. It it a fixed length (timeStatLength) distribution vector with linear interpolation. The X axis (the time values) are not linear, they should be transformed with TimeToStatScale and StatScaleToTime.
func TimeoutWeights ¶
func TimeoutWeights(timeout time.Duration) (res ResponseTimeWeights)
TimeoutWeights calculates the weight function used for calculating service value based on the response time distribution of the received service. It is based on the request timeout value of the system. It consists of a half cosine function starting with 1, crossing zero at timeout and reaching -1 at 2*timeout. After 2*timeout the weight is constant -1.
type RtDistribution ¶
type RtDistribution [timeStatLength][2]float64
RtDistribution represents a distribution as a series of (X, Y) chart coordinates, where the X axis is the response time in seconds while the Y axis is the amount of service value received with a response time close to the X coordinate.
type ServedRequest ¶
type ServedRequest struct {
ReqType, Amount uint32
}
type ValueTracker ¶
type ValueTracker struct {
// contains filtered or unexported fields
}
ValueTracker coordinates service value calculation for individual servers and updates global statistics
func NewValueTracker ¶
func NewValueTracker(db thordb.KeyValueStore, clock mclock.Clock, reqInfo []RequestInfo, updatePeriod time.Duration, transferRate, statsExpRate, offlineExpRate float64) *ValueTracker
NewValueTracker creates a new ValueTracker and loads its previously saved state from the database if possible.
func (*ValueTracker) GetNode ¶
func (vt *ValueTracker) GetNode(id enode.ID) *NodeValueTracker
GetNode returns an individual server node's value tracker. If it did not exist before then a new node is created.
func (*ValueTracker) Register ¶
func (vt *ValueTracker) Register(id enode.ID) *NodeValueTracker
Register adds a server node to the value tracker
func (*ValueTracker) RequestStats ¶
func (vt *ValueTracker) RequestStats() []RequestStatsItem
RequestStats returns the current contents of the reference request basket, with request values meaning average per request rather than total.
func (*ValueTracker) RtStats ¶
func (vt *ValueTracker) RtStats() ResponseTimeStats
RtStats returns the global response time distribution statistics
func (*ValueTracker) Served ¶
func (vt *ValueTracker) Served(nv *NodeValueTracker, reqs []ServedRequest, respTime time.Duration)
Served adds a served request to the node's statistics. An actual request may be composed of one or more request types (service vector indices).
func (*ValueTracker) StatsExpirer ¶
func (vt *ValueTracker) StatsExpirer() *utils.Expirer
StatsExpirer returns the statistics expirer so that other values can be expired with the same rate as the service value statistics.
func (*ValueTracker) Stop ¶
func (vt *ValueTracker) Stop()
Stop saves the value tracker's state and each loaded node's individual state and returns after shutting the internal goroutines down.
func (*ValueTracker) TotalServiceValue ¶
func (vt *ValueTracker) TotalServiceValue(nv *NodeValueTracker, weights ResponseTimeWeights) float64
TotalServiceValue returns the total service value provided by the given node (as a function of the weights which are calculated from the request timeout value).
func (*ValueTracker) Unregister ¶
func (vt *ValueTracker) Unregister(id enode.ID)
Unregister removes a server node from the value tracker
func (*ValueTracker) UpdateCosts ¶
func (vt *ValueTracker) UpdateCosts(nv *NodeValueTracker, reqCosts []uint64)
UpdateCosts updates the node value tracker's request cost table