CN114039891A

CN114039891A - Sketch-based software-defined network large-flow end-to-end time delay and packet loss rate estimation method

Info

Publication number: CN114039891A
Application number: CN202111309342.8A
Authority: CN
Inventors: 麻恩泽; 韩园峰
Original assignee: Individual
Current assignee: Individual
Priority date: 2021-11-06
Filing date: 2021-11-06
Publication date: 2022-02-11

Abstract

The method for estimating the end-to-end time delay and the packet loss rate of the software defined network big flow based on sketch comprises the following steps: s1, initializing an index array, a sketch array and a hash function; s2, stream information acquisition, wherein the stream information acquisition step is mainly to record all data stream information on a current data plane in each time period through a sketch data structure and send the recorded information to a controller of a control plane at the end of each time period; s3, estimating the packet loss rate, namely receiving data flow information periodically collected by the data plane in the step S2 by the control plane, and estimating the packet loss rate of the network by using a query algorithm; and S4, estimating the end-to-end time delay. The invention designs a large flow identification sketch algorithm on the basis of the traditional sketch, can periodically record large flow messages in a switch, periodically sends the recorded messages to a control plane, utilizes the control plane to inquire and search, realizes network measurement tasks such as end-to-end delay estimation, packet loss rate estimation and the like, and expands the sketch network measurement function.

Description

Sketch-based software-defined network large-flow end-to-end time delay and packet loss rate estimation method

Technical Field

The invention relates to the technical field of sketch network measurement methods, in particular to a sketch-based software-defined network large-flow end-to-end time delay and packet loss rate estimation method.

Background

Sketch is a hash-based network measurement technique with fixed spatial overhead, which usually consists of an array and a hash function. Each hash function generally corresponds to the content of a bin in the array, i.e., the index is mapped to the bin of the array one by one after being calculated by the hash function. Through the combination of the hash function and the array, the related information of the data stream can be recorded into the bucket (a certain element of the array) of the sketch. The state of the corresponding data stream can be inquired by using the related information recorded in the bucket, so that the measurement of the network parameters is realized.

The traditional sketch network measurement method has limited number of measurement tasks supported due to technical limitation, and although the traditional sketch network measurement method has good measurement effects on tasks such as super-point detection (Superspreader), change detection (Heavy-change) and Heavy-flow detection (Heavy-hitter), the traditional sketch network measurement method has no capability on complex network measurement tasks such as end-to-end delay statistics of a network, packet loss rate statistics of the network and the like, so that technical restriction is caused on effective application of the sketch network measurement method.

Disclosure of Invention

In order to overcome the technical limitation of the conventional sketch network measuring method, the method cannot be effectively applied to network measuring tasks such as end-to-end time delay statistics of a network, packet loss rate statistics of the network and the like, and causes technical restriction on the application of the network measuring tasks.

The technical scheme adopted by the invention for solving the technical problems is as follows:

the method for estimating the end-to-end time delay and the packet loss rate of the software defined network big flow based on sketch is characterized by comprising the following steps of: s1, initializing an index array, a sketch array and a hash function; s2, stream information acquisition, wherein the stream information acquisition step is mainly to record all data stream information on a current data plane in each time period through a sketch data structure and send the recorded information to a controller of a control plane at the end of each time period; s3, estimating the packet loss rate, namely receiving data flow information periodically collected by the data plane in the step S2 by the control plane, and estimating the packet loss rate of the network by using a query algorithm; and S4, estimating the end-to-end time delay.

Further, the step S1 includes the following substeps, S101: initializing a sketch two-dimensional array, specifically traversing all buckets of the sketch two-dimensional array, and setting the value of each domain in each bucket to be 0; s102: initializing an index array, and setting initial values of an index array stream identification domain, a timestamp domain and a counter domain; s103: initializing a hash function; d +1 hash functions are initialized, namely when the quintuple arrives, the hash functions automatically add a certain value behind the hash key, so that the hash result of each hash function is different.

Further, the step S2 includes the following substeps, S201: in the recording process, firstly, extracting a data stream identification quintuple of a data packet, inputting a hash function of an index array, and calculating to obtain a corresponding index value; s202: searching a bucket corresponding to the index value, and deciding how to update the counter information, the data stream information of the bucket and the timestamp information according to whether the data stream is the same as the current data stream in the bucket; s203: after the operation is finished, recording the index of the searched barrel; s204: for the two-dimensional array of sketch, traversing each row of the two-dimensional array, calculating an index by using a hash function corresponding to the row aiming at each row, searching a corresponding bucket according to the index, and deciding how to update the counter and the index of the current bucket according to whether the index in the bucket is the same as the index of the index array.

Further, the step S3 includes the following substeps, S301: the control plane receives the sketch from different switches, the corresponding sketch is identified through the switch ID, and when the sketch with the same ID arrives, a new sketch is used for replacement; s302: when the control plane receives a packet loss statistical instruction of the application plane, the ske in the current controller is usedLoading the positioning index of tch into memory, randomly selecting one sketch positioning index to start traversing, and maintaining two counters in the traversing process, wherein the number of the counters is C_counterAnd packet loss number counter C_lostRespectively counting the number of the traversed data streams and the number of the packet-lost data streams; s303: the traversal process in the same controller is as follows: for the traversed current barrel, if the elements in the barrel are empty, skipping the traversal; if the elements in the bucket are not empty, extracting the data stream ID information f in the bucket₁And time stamp information t corresponding to the stream ID₁According to the data stream ID f₁Inquiring the flow table of the corresponding switch, acquiring the ID of the next-hop switch, and inquiring the sk corresponding to the next-hop switch by using the flow ID_et_ch positioning index structure; s304: if the current traversal result is packet loss, the counter C of the number of pairs is needed_counterAnd packet loss number counter C_lostRespectively adding one, if the current traversal result is no packet loss, only counting the number of the counter C_counterAdding one; step S30: for packet loss statistics among different controllers, the remote procedure call mode of restful is needed to be used for judging among the controllers, and the source controller sends a data stream ID and a time stamp t to a target controller₁Executing the same query in the same controller in the target controller, returning the query result in the target controller after the query is completed, and returning false if the packet loss is judged in the target controller, or returning true if the packet loss is not considered; s306: calculating the final packet loss rate, wherein the final packet loss rate calculation result formula should be:

where drop _ rate represents the packet loss rate.

Further, in the substep S303, in querying the location index structure of the sketch corresponding to the next-hop switch by using the stream ID, if the stream ID of the bucket corresponding to the query result is the same as the current stream ID, that is, f₁Equals f2, and the timestamp t of the query result₂-t₁Is greater than 0 and less than a thresholdIf so, indicating that the data packet corresponding to the current data stream ID is not lost in the current hop; if the flow ID of the bucket corresponding to the query result is different from the current flow ID, but t₂-t₁If the result is greater than 0 and less than the threshold, it cannot be determined whether packet loss occurs in the current hop, and there is a possibility that the flow ID is not a big flow and is covered, and the next hop route needs to be queried, and if the current hop ID exceeds the specified hop count threshold, it cannot be determined, and then packet loss is considered. Other cases than the above indicate that packet loss has occurred.

Further, the step S4 includes the following substeps, S401: for the time delay calculation of a single data stream, the controller loads the sketch positioning index of each subordinate switch at first, and then selects any non-empty data stream ID f from any index₁And its time stamp t₁Inquiring ID of next-hop switch according to the flow table, determining a positioning index of the corresponding multi-dimensional time sequence sketch by using the switch ID, inquiring, and extracting the timestamp t if the index has the corresponding flow ID₂(ii) a S402: if t is₂＜t₁Ending the current calculation process, traversing a stream ID again, and continuing to execute the process until t₂＜t₁(ii) a S403: if t is₂＜t₁But the flow ID of the inquired target bucket is different from the flow ID to be inquired, inquiring the flow table of the switch, and continuing the inquiry process on the next-hop switch until t₂＜t₁And the two stream IDs are the same or the number of inquired hops is greater than a threshold value; s404: if t is₂＜t₁And both stream IDs are the same, then t is calculated₂-t₁The value and the intermediate hop count seg, the end-to-end delay formula is

Otherwise, continuously searching the next flow ID until finding a flow ID capable of calculating the time delay; for the calculation of the time delay of a plurality of data streams, traversing from the positioning index of any sketch, and performing a single data stream calculation process for the stream ID of each bucket in the index; step S405: targeting controllers during cross-controller delay computationSending a request, transmitting current flow ID information and timestamp information, if the returned result of the target controller is 1, reselecting and sending the flow ID and the timestamp, if the target controller returns a time difference value and the hop count seg2 of the switch under the control of the target controller, wherein the end-to-end time delay is a formula:

the invention has the beneficial effects that: the invention is mainly applied to end-to-end time delay estimation and packet loss rate calculation in software defined network measurement, and the related main technologies comprise sketch, index compression and the like. The invention creates the traditional sketch algorithm, periodically records the large flow information of the switch by using fixed space overhead, transmits the data to the control plane, and queries a plurality of sketch channels in the control plane controller according to the acquired data, thereby estimating the packet loss rate and the end-to-end time delay in the current network. The invention can record the stream identification information of the big stream by using fixed space overhead without an auxiliary data structure; two different structures are designed for a sketch, the two different structures are divided into an index array and a sketch two-dimensional array, the index array stores stream identification information, and only the index inquired by the index array is stored in the sketch array, so that the space overhead of the sketch array is reduced, and the information recording efficiency is improved; the method can not only realize packet loss rate statistics and end-to-end time delay estimation in a network of a single controller, but also realize information calculation among a plurality of switches, has expansibility, and is compatible with application scenarios of distributed controllers. Based on the above, the invention has good application prospect.

Drawings

The invention is further illustrated below with reference to the figures and examples.

FIG. 1 is a diagram of an index array and a sketch array according to the present invention.

Fig. 2 is a flow chart of packet loss rate calculation according to the present invention.

Fig. 3 is a flow chart of the end-to-end delay computation of the present invention.

FIG. 4 is a flow chart of the deployment of the present invention.

FIG. 5 is a diagram of the deployment architecture of the present invention.

Fig. 6 is a schematic diagram of the end-to-end delay computation of the present invention.

Detailed Description

As shown in fig. 1, the method for estimating end-to-end delay and packet loss rate of a large flow in a sketch-based software-defined network includes the following steps, S1: initializing a data structure of a large stream identification sketch algorithm; specifically, a sketch two-dimensional array with the size of w multiplied by d, an index array with the size of w and d +1 hash functions are initialized, wherein w represents the number of columns of the two-dimensional array, and d represents the number of rows of the two-dimensional array. The steps are as follows, S101: initializing a sketch two-dimensional array; traversing all the buckets of the sketch two-dimensional array, and setting the value of each domain in each bucket to be 0. Step S102: initializing an index array; setting the values of a source IP address and a destination IP address in an index array to be 000000000000, setting the values of a source port domain and a destination port domain to be 0, and setting the value of a timestamp domain to be 1970-01T 00: 00: 00Z, setting the value of the protocol type field to 0, and setting the value of the current flow identification counter to 0. Step S103: initializing a hash function; initializing d +1 hash functions, setting all the hash functions as MD5 functions, calculating hash values, and then performing modulo calculation with ww to calculate corresponding index. In order to ensure that the hash functions of the index array and the sketch array are different for each row, 0 is added to each hash key for the hash function of the index array, and the value of 1 → d is added to each hash key in ascending order for the hash function of the two-dimensional array.

As shown in fig. 1, S2: and collecting flow information. The flow information acquisition step is mainly to record all data flow information on the current data plane in each time period through the sketch data structure, and send the recorded information to the controller of the control plane at the end of each time period. The steps are as follows, S201: in the recording process, firstly, extracting a data stream identification quintuple { source IP address, source port number, destination IP address, destination port number and protocol type } of a data packet, and converting information of the quintuple into a character string; the character string is firstly input into the hash function of the index array, and the corresponding index value is obtained through calculation. S202: searching a bucket corresponding to the index value, judging whether quintuple information of the current data stream is the same as quintuple information of data in the bucket or not, if so, updating timestamp information, and setting the value of a counter as the current value of the counter plus 1; if the current value of the counter is different from the current value of the counter, subtracting 1 from the current value of the counter, then judging whether the value of the counter is 0, if the value of the counter is 0, replacing each domain corresponding to the quintuple in the bucket by using the quintuple of the current data stream, updating the timestamp domain, setting the value of the counter to be 1, and if the value of the counter is greater than 0, only updating the timestamp information. S203: after the operation is completed, the index of the searched barrel is recorded. Step S204: for the two-dimensional array of the sketch, traversing each row of the two-dimensional array, calculating an index by using a hash function corresponding to the row aiming at each row, and searching a corresponding bucket according to the index; for each searched bucket, if the value of the index field in the bucket is the same as the index of the index bucket, setting the value of the counter field as the value of the current counter plus 1; if the value of the intra-bucket index and the value of the intra-bucket index are not the same as the index of the index bucket, setting the value of the counter field to be the current value minus 1, then judging whether the value of the current counter field is 0, if so, setting the value of the index field to be the index of the index bucket, setting the value of the counter to be 1, and if not, performing any operation.

As shown in fig. 2 and 3, S3, estimating the packet loss rate, where the primary control plane receives data stream information periodically collected by the data plane in step S2, and then estimates the packet loss rate of the network by using an inquiry algorithm. The main substeps are as follows, S301: the control plane receives the sketch from different switches, and the corresponding sketch is identified through the switch ID; when the sketch with the same ID arrives, the new sketch is used for replacement. S302: when the control plane receives a packet loss counting instruction of the application plane, the positioning index of the sketch in the current controller is loaded into a memory, the positioning index of the sketch is randomly selected to start traversal, two counters are maintained in the traversal process, and a counter C is counted_counterAnd packet loss number counter C_lostAnd respectively counting the number of the traversed data streams and the number of the lost data streams. S303: the traversal process in the same controller is as follows: for the current bucket traversed, if a bucketIf the internal element is empty, skipping the traversal; if the elements in the bucket are not empty, extracting the data stream ID information f in the bucket₁And time stamp information t corresponding to the stream ID₁According to the data flow IDf₁Inquiring the flow table of the corresponding switch, acquiring the ID of the next-hop switch, inquiring the sketch positioning index structure corresponding to the next-hop switch by using the flow ID, and if the flow ID of the bucket corresponding to the inquiry result is the same as the current flow ID, namely f₁Is equal to f₂And a time stamp t of the query result₂-t₁If the difference is greater than 0 and smaller than the threshold, it indicates that the packet corresponding to the current data stream ID is not lost in the current hop. If the flow ID of the bucket corresponding to the query result is different from the current flow ID, but t₂-t₁If the result is greater than 0 and less than the threshold, it cannot be determined whether packet loss occurs in the current hop, and there is a possibility that the flow ID is not a big flow and is covered, and the next hop route needs to be queried, and if the current hop ID exceeds the specified hop count threshold, it cannot be determined, and then packet loss is considered. Other cases than the above indicate that packet loss has occurred. S304: if the current traversal result is packet loss, the counter C of the number of pairs is needed_counterAnd packet loss number counter C_lostRespectively adding one, if the current traversal result is no packet loss, only counting the number of the counter C_counterAnd adding one. S305: for packet loss statistics between different controllers, a remote procedure call mode of restful needs to be used for judgment between the controllers. The source controller sends a data stream ID and a time stamp t to the target controller₁And executing the same query in the target controller as that in the same controller, and returning a query result in the target controller after the query is finished. If the target controller judges that the packet loss returns false, if the target controller does not consider that the packet loss returns true. S306: and calculating the final packet loss rate. The final packet loss rate calculation result formula should be:

where drop _ rate represents the packet loss rate.

FIG. 6 shows, S4. terminalAnd estimating the time delay of the terminal. The positioning index in the sketch stores the timestamp information of the last data packet of the large stream, and the end-to-end time delay between two exchangers can be calculated through the skip path of the stream and the timestamp information on the path. The control plane receives the data collected by the data plane in step S2, and then performs an end-to-end query operation between the switch under the control of one controller and the switches under the control of different controllers by using a query algorithm, so as to estimate an end-to-end delay between two nodes on the link. Sketch-based delay estimation can be realized by delay calculation of a single data stream with high efficiency and relatively large error and average delay calculation of multiple data streams with low efficiency but relatively small error. The delay calculation for a single data stream and the delay calculation for multiple data streams operate differently. The estimation process is shown in fig. 2, and the steps are as follows. S401: for the time delay calculation of a single data stream, the controller loads the sketch positioning index of each subordinate switch at first, and then selects any non-empty data stream ID f from any index₁And its time stamp t₁Inquiring ID of next-hop switch according to the flow table, determining a positioning index of the corresponding multi-dimensional time sequence sketch by using the switch ID, inquiring, and extracting the timestamp t if the index has the corresponding flow ID₂. S402: if t is₂＜t₁Ending the current calculation process, traversing a stream ID again, and continuing to execute the process until t₂＜t₁. S403: if t is₂＜t₁But the flow ID of the inquired target bucket is different from the flow ID to be inquired, inquiring the flow table of the switch, and continuing the inquiry process on the next-hop switch until t₂＜t₁And the two stream IDs are the same or the number of queried hops is greater than a threshold. S404: if t is₂＜t₁And both stream IDs are the same, then t is calculated₂-t₁The value and the intermediate hop count seg, end-to-end delay of

Otherwise, continue toThe next flow ID is looked up until a flow ID is found that can account for the delay. For the calculation of multiple data stream delays, traversal is required from the positioning index of any sketch, and the calculation process of a single data stream is performed once for the stream ID of each bucket in the index. S405: a request is sent to the target controller in a cross-controller latency calculation, passing current stream ID information and timestamp information. If the target controller returns a result of 1, the stream ID and timestamp are reselected and sent. If the target controller returns a time difference value and the hop count seg of the switch under the control of the target controller₂End-to-end delay of

The invention improves the traditional sketch algorithm, expands the application scene of the traditional sketch algorithm, is mainly applied to end-to-end time delay estimation and packet loss rate calculation in software defined network measurement, and relates to the following main technologies: sketch and index compression. The invention has a sketch two-dimensional array with the size of w multiplied by d, an index array with the size of w and d +1 hash functions. Each bucket of the two-dimensional array is a structure body, the structure body consists of two fields, one field is an index and is only corresponding to one index of the index array, and the other field is a counter and is used for filtering data streams stored in the current bucket. Each row of the two-dimensional array corresponds to one hash function, and the index array independently corresponds to one hash function. Each bucket of the index array is a structure body composed of 7 domains, and a source IP address, a destination IP address, a source port, a destination port, a timestamp, a protocol type and a current flow identification counter are respectively recorded. The extended sketch periodically records information of a data stream in a switch of the data plane by using the extended sketch and transmits the extended sketch recorded in the period to a controller of the control plane when the recording period is finished. The controller inquires the information recorded by the index array and the sketch array from each switch, and estimates the end-to-end time delay and the packet loss rate through the combination of the corresponding flow identification and the timestamp information and the inquiry information. And recording the stream identification through the index array, wherein the indexes of the index array correspond to the indexes of the sketch array one by one, and the stream identification information of the large stream is recorded through fixed space overhead. As shown in fig. 4 and 5, the extended sketch algorithm is deployed in a programmable switch, the index array, the sketch array, and the hash function are initialized periodically, then the sketch extended sketch algorithm is used to record data flow information of the data plane, after the sketch is transmitted to the control plane, the control plane may perform network end-to-end delay estimation and network packet loss rate estimation by combining sketch information of multiple switches, and at the same time, start data acquisition and estimation of the next period.

The sketch extension algorithm designed by the invention has the advantages that: 1. stream identification information for large streams may be recorded using a fixed spatial overhead without resorting to an auxiliary data structure. 2. Two different structures are designed for one sketch, the two different structures are divided into an index array and a sketch two-dimensional array, the index array stores the stream identification information, and only the index inquired by the index array is stored in the sketch array, so that the space overhead of the sketch array is reduced, and the information recording efficiency is improved. 3. The method provided by the invention can not only realize packet loss rate statistics and end-to-end time delay estimation in a network of a single controller, but also realize information calculation among a plurality of exchangers, has expansibility and can be compatible with application scenes of distributed controllers. The invention creates the traditional sketch algorithm, which comprises the following steps: the flow identification recording algorithm is improved on the basis of the traditional sketch, and the recording of the large flow identification is realized through fixed space overhead. And an index array is introduced, the storage content of the sketch array domain is modified, and the space overhead for storing the stream identification is compressed. Firstly, an estimation algorithm of end-to-end time delay and packet loss rate of a single controller is designed, and then cross-controller information query is realized by using a restful (presentation layer state transition) calling mode among a plurality of controllers, so that an end-to-end time delay estimation and packet loss rate estimation algorithm among a plurality of controllers is realized. In the invention, except for the traditional sketch, all sketch and related space compression methods, network measurement algorithms and technologies designed in the invention are within the protection scope of the invention.

While there have been shown and described what are at present considered the fundamental principles and essential features of the invention and its advantages, it will be apparent to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, but is capable of other specific forms without departing from the spirit or essential characteristics thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims

1. The method for estimating the end-to-end time delay and the packet loss rate of the software defined network big flow based on sketch is characterized by comprising the following steps of: s1, initializing an index array, a sketch array and a hash function; s2, stream information acquisition, wherein the stream information acquisition step is mainly to record all data stream information on a current data plane in each time period through a sketch data structure and send the recorded information to a controller of a control plane at the end of each time period; s3, estimating the packet loss rate, namely receiving data flow information periodically collected by the data plane in the step S2 by the control plane, and estimating the packet loss rate of the network by using a query algorithm; and S4, estimating the end-to-end time delay.

2. The sketch-based software-defined network large-flow end-to-end delay and packet loss rate estimation method of claim 1, wherein the step S1 comprises the following substeps, S101: traversing all the buckets of the sketch two-dimensional array; s102: setting initial values of an index array stream identification domain, a timestamp domain and a counter domain; s103: d +1 hash functions are initialized, and when the quintuple arrives, the hash functions automatically add a certain value behind the hash key, so that the hash results of the hash functions are different.

3. The sketch-based software-defined network large-flow end-to-end delay and packet loss rate estimation method of claim 1, wherein the step S2 comprises the following substeps, S201: calculating to obtain a corresponding index value; s202: deciding how to update the counter information, the data stream information of the bucket, and the timestamp information; s203: recording the index of the searched barrel; s204: for the two-dimensional array of sketch, each row of the two-dimensional array is traversed, and a decision is made as to how to update the counter and index of the current bucket.

4. The sketch-based software-defined network large-flow end-to-end delay and packet loss rate estimation method of claim 1, wherein the step S3 comprises the following substeps, S301: when the sketch with the same ID arrives, replacing the sketch with a new sketch; s302: when the control plane receives a packet loss counting instruction of the application plane, respectively counting the number of the traversed data streams and the number of the packet-lost data streams; s303: if the elements in the barrel are empty, skipping the traversal; if the elements in the bucket are not empty, acquiring the ID of the next hop switch, and querying a sketch positioning index structure corresponding to the next hop switch by using the stream ID; s304: if the current traversal result is packet loss, the counter C of the number of pairs is needed_counterAnd packet loss number counter C_lostRespectively adding one, if the current traversal result is no packet loss, only counting the number of the counter C_counterAdding one; step S305: for packet loss statistics among different controllers, judging among the controllers by using a remote procedure calling mode of restful; s306: the final formula of the packet loss rate calculation result is as follows:

where drop _ rate represents the packet loss rate.

5. The base of claim 4The software defined network big stream end-to-end delay and packet loss rate estimation method in sketch is characterized in that in the substep S303, in a positioning index structure, if the stream ID of a bucket corresponding to an inquiry result is the same as the current stream ID, the packet loss of a data packet corresponding to the current data stream ID is not generated in the current hop; if the flow ID of the bucket corresponding to the query result is different from the current flow ID, but t₂-t₁If the result is greater than 0 and less than the threshold, the current hop cannot judge whether packet loss occurs, the possibility that the current flow ID is not a big flow and is covered is caused, the next hop route is inquired, if the current hop route exceeds the specified hop threshold, the next hop route cannot be determined, the packet loss is considered, and other situations occur to indicate the packet loss.

6. The sketch-based software-defined network large-flow end-to-end delay and packet loss rate estimation method of claim 1, wherein the step S4 comprises the following substeps, S401: the controller loads the sketch positioning index of each subordinate switch at first, and then selects any non-null data stream ID f from any index₁And its time stamp t₁Inquiring ID of next-hop switch according to the flow table, determining a positioning index of the corresponding multi-dimensional time sequence sketch by using the switch ID, inquiring, and extracting the timestamp t if the index has the corresponding flow ID₂(ii) a S402: if t is₂＜t₁Ending the current calculation process, traversing a stream ID again, and continuing to execute the process until t₂＜t₁(ii) a S403: if t is₂＜t₁But the flow ID of the inquired target bucket is different from the flow ID to be inquired, inquiring the flow table of the switch, and continuing the inquiry process on the next-hop switch until t₂＜t₁And the two stream IDs are the same or the number of inquired hops is greater than a threshold value; s404: if t is₂＜t₁And both stream IDs are the same, then t is calculated₂-t₁The value and the intermediate hop count seg, the end-to-end delay formula is

Otherwise, continue to search for the next flow ID untilFinding a stream ID for which the delay can be calculated; for the calculation of the time delay of a plurality of data streams, traversing from the positioning index of any sketch, and performing a single data stream calculation process for the stream ID of each bucket in the index; step S405: sending a request to a target controller during cross-controller time delay calculation, transmitting current flow ID information and timestamp information, if the returned result of the target controller is 1, reselecting and sending the flow ID and the timestamp, if the target controller returns a time difference value and the hop count seg of a switch under the control of the target controller₂The end-to-end delay is given by the formula: